PERSUASIVE MESSAGE TYPICALITY AND SOURCE CREDIBILITY: A SCHEMA-COPY-PLUS-TAG MODEL WITH SLEEPER EFFECTS By JOON SOO LIM A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2006
Copyright 2006 by Joon Soo Lim
This dissertation is dedicated to my late parents Mr. Sang Man Lim and Mrs. Suk Hee Lee, my wife Hyo Suk Kang, and to our two sons, Hosan and Hohyun.
iv ACKNOWLEDGMENTS This dissertation leaves me indebted to many people. First and foremost, I would like to thank my wonderful advisor and mentor, Dr. Ma ry Ann Ferguson, for her inspiring way in deftly guiding me to im prove this dissertation. Her mentorship was paramount in providing the invaluable advice necessary to pursue my long-term career goals. I wish to thank heartily Dr. Lynda Lee Kaid and Dr. Marilyn Roberts for their invaluable help and support during my enti re degree process, especially for their compassionate support when I took a leave of ab sence for the painful loss of my parents. Dr. Spiro Kiousis deserves considerable th anks for suggesting insightful modifications and for directing me in the ways I could improve my final draft. I have benefited from taking graduate seminars in both cognitive and social psychology. The idea of this study originated when I took Dr. Fischl erÂ’s intellectually challenging course on memory and sensory processing. I cannot thank Dr. Fischler enough for his brilliant input. This dissertation wo uld not be half of what it is without his guidance. I also would like to thank Dr. AlbarracÃn whose expert ise in theories of attitude and depth of knowledge about the sleeper effect were most beneficial in completing this dissertation. I would also like to extend my sincere gratitude to Dr. Peg Hall and Dr. Linda Perry. Though my research veered away from fund-raising research, I never forget what I learned from Dr. Hall. Her expertise in th e nonprofit sector helped me broaden my
v knowledge and experiences in th e field of public relations. I thank Dr. Perry for being an outstanding teacher who instructed me in how to teach visual communication. I have had many good friends in my life each of whom has enriched me in many ways. Many thanks go to Sung-Un Yang, Moon Lee, Matthew Tedder, Jae Woong Shim, Yonghwan Kim, Seong-Mo Yim, Hyun Jung Yun, Eyun-Jung Ki, Andrew Bagley, and Guy Golan for their friendship and for all that I learned from each of them. I would like to give special thanks to Matty Wegeha upt, a great friend, whose intellectual and progressive views inspired and stimulated my wa ys of thinking for the past nine years. I also thank him for his super-meticulous atte ntion to detail in ir oning out all kinds of mistakes in my writing duri ng my entire degree work. Most importantly, words fall short in desc ribing my gratitude and thanks toward my wife, Hyo Suk Kang and our sons, Hosa n and Hohyun. They have suffered through this dissertation almost as much as I ha ve; maybe even more. Through the many nights I spent at the lab, they have shown endless patie nce, love, and understanding. I truly cannot imagine having gone through this work w ithout such amazing love and support. Across oceans or states, the prayers and en couragement of my family and relatives were a driving force and endless motivation to finish this journey. Among them, I especially thank my two brothers, Sung Soo and Jun Young, and my cousin, Moon Soo for their support, encouragement, and prayers. I also thank my pare nts-in-law for their love and care during my degree work. In addition, Mrs.Carlene Harmon and Mr. James Harmon are very special to me and their presen ce is the pillar I have always leaned on here in Gainesville, Florida.
vi I reserve the final words for closing this acknowledgment to my late parents, both of whom passed away together in the wi nter of 2002. I owe so much to them, but regrettably I can never find a way to pay back their immeasurable love and sacrifice. The very least I can do is to dedica te this dissertation to my la te parents, Sang Man Lim and Sook Hee Lee.
vii TABLE OF CONTENTS page ACKNOWLEDGMENTS.................................................................................................iv LIST OF TABLES...............................................................................................................x LIST OF FIGURES..........................................................................................................xii ABSTRACT.....................................................................................................................xi ii CHAPTER 1 INTRODUCTION........................................................................................................1 The Issue Context of the Current Study.......................................................................4 The Purpose of the Study..............................................................................................8 2 REVIEW OF SCHEMA THEORY............................................................................10 Overview of Theoretical Backgrounds.......................................................................10 BartlettÂ’s Legacy and Schema Theory........................................................................15 Message Typicality Effect: The Sche ma-Copy-Plus-Tag (SC + T) Model................21 Theoretical Underpinning of The SC + T Model................................................36 Differential processing views.......................................................................39 Representational views.................................................................................43 Memory for Persuasive Messages: A Test of the SC + T Model........................50 3 REVIEW OF THEORIES IN ATTITUDES..............................................................61 The Effect of Message Typicality on Persuasion.......................................................62 The Effect of Source Credibility on Attitude Change................................................73 Ethos and Source Credibility...............................................................................73 Factors Influencing Message Recipi entsÂ’ Credibility Judgment About a Source..............................................................................................................76 The Schema-Congruency Effect..........................................................................80 The Sleeper Effect......................................................................................................92 Theoretical Definition of Sleeper Effect.............................................................92 Original findings..........................................................................................92 Forgetting hypothesis...................................................................................93 Dissociation hypothesis................................................................................93
viii Conditions for the sleeper effect..................................................................94 Deferential decay hypothesis.......................................................................95 Availability-valence hypothesis...................................................................97 The Absolute Sleeper Effect................................................................................99 The Relative Sleeper Effect.................................................................................99 A Necessary Condition to Observe a Sleeper Effect.........................................100 Some Experimental (But Not Necessary) Conditions to Observe the Sleeper Effect..............................................................................................................101 4 METHODS...............................................................................................................105 Overview of the Experiment.....................................................................................105 Design.......................................................................................................................10 6 Stimuli: Typical versus Atypical Message........................................................107 Stimuli: Endorse r Credibility............................................................................110 Subjects..............................................................................................................114 Procedure..................................................................................................................114 Independent Variables..............................................................................................116 Typicality...........................................................................................................116 Endorser credibility...........................................................................................117 Dependent Variables.................................................................................................117 Post-attitude Measures.......................................................................................117 Recognition Performance..................................................................................117 5 RESULTS.................................................................................................................121 Demographics...........................................................................................................121 Group Size.........................................................................................................121 RespondentsÂ’ Demographics.............................................................................122 Manipulation Checks................................................................................................123 Endorser credibility...........................................................................................123 Argument typicality...........................................................................................124 Reliability..........................................................................................................126 Source credibility.......................................................................................126 Post-attitudes..............................................................................................126 Overall Review of Findings......................................................................................127 Tests of Hypotheses..................................................................................................129 Test of Hypothesis 1..........................................................................................129 Test of Hypothesis 2..........................................................................................131 Test of Hypothesis 3..........................................................................................131 Test of Hypothesis 4..........................................................................................133 Tests of H5a and H5b........................................................................................135 Test of Hypothesis 6..........................................................................................136 Test of Hypothesis 7a........................................................................................136 Tests of H7b and H7c........................................................................................139 Summary of Tests of Hypotheses......................................................................141
ix 6 DISCUSSIONS AND LIMITATIONS....................................................................142 Post-Hoc Analyses....................................................................................................142 Post-Hoc Analyses for H1a, H1b & H3............................................................142 Supplementary Analysis for H4........................................................................145 Post-Hoc Analyses for H5.................................................................................145 Post-Hoc Analyses for H6.................................................................................146 Post-Hoc Analyses for H7.................................................................................148 Discussions...............................................................................................................151 Discussions about the SC + T Model................................................................151 Discussions about the Sleeper Effect................................................................157 Limitations................................................................................................................161 Recall/Recognition............................................................................................161 External Validity...............................................................................................162 Untested Underlying Assumptions....................................................................162 Typicality of Foils.............................................................................................162 Individual Differences (Motivations and Prior Attitudes).................................163 The Order of Source-Message Presentation......................................................164 7 IMPLICATIONS AND SUGGESTIONS FOR FUTURE RESEARCH.................165 The SC + T Model....................................................................................................165 The Illusion of The Truth Effect...............................................................................166 Sleeper Effect............................................................................................................167 Practical Implications...............................................................................................169 APPENDIX A STIMULUS: LOW CREDIBILITY.........................................................................172 B STIMULUS: HIGH CREDIBILITY........................................................................174 C STIMULUS: TYPICAL MESSAGE........................................................................177 D STIMULUS: ATYPICAL MESSAGE.....................................................................181 E QUESTIONNAIRE..................................................................................................185 F CROSSTABULATION FOR CONFIDENCE RATINGS......................................193 LIST OF REFERENCES.................................................................................................206 BIOGRAPHICAL SKETCH...........................................................................................231
x LIST OF TABLES Table page 1. Manipulation check of endor ser credibility in a pretest.............................................111 2. Number of participants by experimental groups........................................................121 3. ParticipantsÂ’ majors...................................................................................................122 4. ParticipantsÂ’ household income.................................................................................123 5. Most frequently used Web browser among participants............................................123 6. Manipulation check of endorser cr edibility in the main experiment.........................124 7. Manipulation check of t ypicality in a pretest (N=26)................................................125 8.. Factor loadings for items in the source credibility measures....................................126 9.. Factor loadings for items in the attitude measures....................................................126 10.The effects of typicality, time, and credibility on recognition performance.............127 11. Means (SDs in parenthesis)for d by time, typicality and credibility.......................128 12.The effects of typicality, time, and cr edibility on attitudes toward Firefox..............128 13. Means ( SD s in parenthesis) for attitude toward Firefox by group............................129 14. Recognition performance by ty picality at an immediate test....................................130 15.Recognition performance by ty picality at a delayed test...........................................130 16. Recognition performance by retention intervals.......................................................131 17. The effect of retention interval and typicality on d (sensitivity).............................132 18. Interaction of typicality and time interval on recognition performance...................132 19. Interaction of typicality and reten tion interval on confidence of recognition judgments for hits...................................................................................................134
xi 20. Mean Â“SureÂ” responses for recogn ition judgments about typical/atypical arguments with retention intervals.........................................................................134 21. Attitudes by credibility in an im mediate and a delayed test group...........................136 22. Comparison of means between low cred ibility groups and a control group at the immediate test........................................................................................................136 23. The effects of time and credibility on attitudes under typical message condition....137 24. Comparison of mean post-attitudes by credibility and time under typical message condition.................................................................................................................138 25. The effects of time and credibility on attitudes under atypical message condition..139 26. Comparison of mean post-attitudes by credibility and time under atypical message condition.................................................................................................................140 27. The effect of retention in terval and typicality on HR...............................................142 28. The effect of retention in terval and typi cality on FAR.............................................143 29. Interaction of typicality and time interval on recognition performance...................143 30. Comparison of means between different levels of perceived credibility..................146 31. Comparison of means for low credibility groups and for a control group at the immediate test........................................................................................................147 32.The effects of typicality, time, and credibility on attitude change.............................148 33.Analysis of covariance on attitude change with the involvement and perceived credibility index as covariates................................................................................149 34.Hierarchical regression predic ting attitudes toward Firefox......................................150
xii LIST OF FIGURES Figure page 1. Representation of a schema-based pa ssage described in Schank and AbelsonÂ’s (1977) SP + T model................................................................................................23 2. Recall performance for typical vs. atypical actions over time.....................................30 3. Recognition performance for typi cal vs. atypical actions over time...........................31 4. Recognition memory performance at different retention intervals..............................32 5. A comparison of memory scores (MS) of recall and recognition at different retention intervals.....................................................................................................33 6. Pictorial summary of Hypotheses 1-3..........................................................................59 7. Evaluative responses (on a multiple 8-poin t semantic differential scale) toward the speaker over time intervals.......................................................................................75 8. Persistence of persuasi on. A: non-persisting boomerang effect. B: absolute sleeper effect. C: relative sleeper effect..............................................................................100 9. A pictorial summary of H5-H7..................................................................................104 10. The pretest-posttes t control group design extended in time.....................................106 11. Interaction of Typicality X Time on confidence for recognition judgments. Confidence was measured by the number of Â“SureÂ” responses for either typical or atypical hits........................................................................................................135 12. Effects of Time and Credibility on att itudes toward Firefox under typical message condition.................................................................................................................138 13. Interaction of Time X Credibility on attitudes toward Firefox under atypical message condition..................................................................................................140 14. Recognition performance as a functi on of typicality and time interval....................145
xiii Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy PERSUASIVE MESSAGE TYPICALITY AND SOURCE CREDIBILITY: A SCHEMA-COPY-PLUS-TAG MODEL WITH SLEEPER EFFECTS By Joon Soo Lim August 2006 Chair: Mary Ann Ferguson Major Department: Journalism and Communications This study investigated how a personÂ’s memories for persuasive messages were influenced by the messageÂ’s typicality over time. Then, it examined how a sleeper effect Â– a delayed persuasive effect from the low cr edibility source Â– could be influenced by the typicality of the persuasive message. In a 2 (typicality) X 2 (cre dibility) X 2 (time) between-subjects factorial design with tw o control groups, par ticipants read an endorsement message for Firefox on the Web. Each version of the endorsement message contained an equal number of typical and atypical arguments that were presented by either a highly-credible or a less-credi ble endorser. After reading the message, participantsÂ’ attitudes toward Firefox were measured either instantaneously or after a two-week delay. Following the attitudes ev aluations, participants were given an unexpected memory recognition test with target sentences that had been presented in the endorsement message along with an equal num ber of foil sentences that had not been
xiv presented. A measure of sensitivity ( d) was compared across the typicality and credibility combinations. Tests of hypotheses on the schema-copy-plus -tag (SC + T) model revealed that recognition performance dropped over time more rapidly for atypical arguments than for typical arguments. In add ition, there was also a typical ity-by-time interaction on the confidence-of-recognition judgments for hit it ems. These findings resonate with the primary assumption of the SC + T model that predicts a reversed memory performance over time between typical and atypical messages. Tests of hypotheses on the sleeper effect s howed that the sleeper effect would be observed only when the message was atypical. When the message was atypical, attitude change induced by a high-credibility endorser at a delayed test was significantly lower than the change made by a low-credibility e ndorser. These findings taken together with no effect of credibility on at titude change suggest that a motivation factor may have caused participants to scrutinize the message and change their attitudes based on the message itself instead of the source of the information.
1 CHAPTER 1 INTRODUCTION Despite the criticism on the credibility of information sources on the Web, people these days increasingly rely on diverse Intern et sources to gather information and to monitor the opinions of others (Flanagin & Metzger, 2000; Garrison, 2000). With more and more people getting their primary info rmation from the Web, both for-profit and nonprofit organizations are adap ting their business to the Web. Increasingly, traditional forms of marketing and public relations are also shifting to the Web. One that reflects such transformation is business organizati onsÂ’ use of ordinary voices in Web-based marketing, and public relations campaigns. It wa s rare in traditional marketing and public relations to use the voice of an average Joe because one of the golden rules of persuasion theory was to ensure that an endorser was cr edible. Under this regime, organizations with scarce resources would rely on endorsement s from credible sour ces or celebrities. However, in the Internet era, organizations are less likely to earn endorsement from such celebrities or influential sources such as Oprah Winfrey or Walter Mossberg . Instead, business people are increasingly paying atte ntion to individual We b users or bloggers (Story, 2005). Consider some salient cases: Â• Apple computerÂ’s famous Â“switchÂ” ad s presented testimonials from ordinary people who switched from PCs to a Mac. Â• MozillaÂ’s Firefox Â“switchÂ” campaigns al so presented five ordinary peopleÂ’s voices. Â• Google Inc.was recently sued by some publishers and copyright holders concerning its Google Book Search. To deal with this issue, Google launched a
2 grassroots public re lations campaign to attain more endorsements among Web users, presenting brief excerpts from numerous ordinary Web users and bloggers. One critical issue related to this tren d of using ordinary peopleÂ’s voices in persuasive campaigns is whether or not thos e messages will ultimately lead to persuasion. A subsequent question emerges: If persua sion is induced by messages from ordinary people, can they bring about as much attit ude or belief change as do highly credible sources? Before exploring this research quest ion, another issue needs to be addressed. Arguments in ordinary peopl eÂ’s endorsements and testim onials often violate another golden rule in persuasion Â– th at is, to induce persuasion, argu ments must be strong rather than weak to in terms of argument quality (Areni & Lutz, 1988; Axsom, Yates, & Chaiken, 1987; Coulter & Punj, 2004; Petty & Cacioppo, 1984; Petty, Cacioppo, & Goldman, 1981; Petty, Cacioppo, & Heesack er, 1981; Petty & Wegener, 1991). Argument quality, which will be discussed in Chapter 3 of this dissertation, is briefly defined as Â“a recipientÂ’s per ception that a messageÂ’s arguments are strong and cogent as opposed to weal and speciousÂ” (Eagly and Chaiken, 1993, p.311). Consider the following examples: Â• Â“I made my mom cry with the first iMovi e I made. I knew it was gorgeous, but to have my mom be so emotional with a simple five minute slideshow presentation of our ChristmasÂ” [a Mac userÂ’s testimonial which is excerpted from AppleÂ’s Â“SwitchÂ” campaign] Â• "My gob is well and truly smacked! G oogle [Book Search] is up and running, looks very impressive, and has already introduce d me to two facts th at blew me away. I did a search for books referring to my home to wn of Strathaven in Lanarkshire and came
3 up with two real surprises" [an endorseme nt message from a blogger in support of Google Print]. In the operational definition of a strong or a weak argument that will be discussed in a later chapter of this dissertation, the above messages apparently consist of a set of weak arguments. Not only are arguments in the two messages weak, but some of the sentences are also conceptually irrelevant to the message advocated and/or they are less logically sequenced in the flow of message. When an argument is irrelevant, incongruent and unexpected from the main topic or categor y of the issue, the ar gument is considered Â“schema-inconsistent.Â” Schemas here re fer to Â“generic knowledge structure or expectancies that are developed through experiencesÂ” (Koriat , Goldsmith, and Pansky, 2000). When an argument about an issue/object is schema-inconsistent, it is considered as an Â“atypicalÂ” argument in the present study. Wh en a persuasive message is composed of a set of atypical arguments rather than t ypical arguments, it is called an Â“atypicalÂ” message in the present study. The use of average peopleÂ’s voices in Web-based persuasive campaigns has brought forth a va riety of testimonials and/or endorsements that reflect each individual Â’s unique experiences or b ackground knowledge. However, the more diverse the individualsÂ’ voices are, the more an audience member reads atypical messages. This phenomenon raises some important research questions to communication researchers. It is possible th at Web users who rely more on getting information from the Web may develop more atypical thoughts regard ing an issue that is discussed. They may remember more atypical arguments, actions and examples than typical ones, and ultimately tend to regard those atypical thoughts as typical ones. Especially, when they
4 remember just the message and forget who th e source of the message was, then messagedriven persuasion could occur regardless of th e credibility of information obtained on the Web. The trend may also hold true when the me ssage from the low-credibility source was typical. When the source-memory decay phenomenon is taken for granted, a typical message from a low credibility source could exert a more persuasive power over time, because a typical message has more persuasi ve advantage over an atypical message. Scholars who examined human memory and text comprehensions (Goldman, Graesser, & van Den Broek, 1999; Graesser, 1981; Graesser, Mills, & Zwaan, 1997; Kintsch & van Dijk, 1978; Ki ntsch, 1988; McKoon & Ratcliff, 1992; van Dijk & Kintsch, 1983) have argued that a personÂ’ s schematic processing of text in prose leads to schemadriven encoding, inferences, comprehensi on and memory construction, which could either facilitate or hinder their comprehe nsion and/or memory representation of the contents. An unanswered question is how th is schema-driven information processing affects peopleÂ’s affective evalua tions about the attit udinal objects that are described in the message, especially when their memory and at titudes are measured at different retention intervals. This dissertation is an atte mpt to answer this research inquiry. The Issue Context of the Current Study The advent of Web technology and its rapid growth have been changing the way of how business organizations deliver a persuasive message. The most radical shift in this Web-based communication strategy is obser ved in the so-called grassroots public relations campaigns. For example, the Mo zilla Foundation, a nonprofit organization that develops and coordinates the release of the Firefox Web browser, benefited from worldwide grassroots public relations campaigns in creating a buzz about Firefox and
5 placing Firefox stories in numerous national media (Stross, 2004). Firefox also launched a Web-based Â“switchÂ” campaign in which it presented ordinary usersÂ’ supporting messages. There are also some endorseme nts from traditionally highly-regarded technology voices (e.g., Walt Mossberg, an infl uential technology columnist of the Wall Street Journal ). Certainly, there is a difference in the quality and typicality of endorsement between Walt Mossberg and ordi nary Web users. As discussed, messages from ordinary people tend to be atypical in that they sometimes emphasize a userÂ’s personal unique experiences, make some id iosyncratic examples or assertions. This study was based on this observation, and this observation was applied to previous theories in psychology and co mmunication research. There are several theoretical assumptions that penetrate this present study. The first assumption is that human memo ry is not literal reproduction of an incoming message but a reconstruction of the original message based on an individualÂ’s schema. This schema is assumed to guide our memory representation of what we saw, read, and heard. The second assumption that underlies the present study is that there are two different memory systems Â– that is, semantic and episodic (Tulving, 1972, 1983, 1985, 1993, 1995, 2002a, 2002b). In a nutshell, source memory constitutes the episodic memory since recollection of specific sour ce information necessarily involves some retrieval of the temporal and spatial context associated with the sourceÂ’s presentation of the information (Craik, L. W. Morris, R. G. Morris, & Loewen, 1990). Assume that a person watching Larry King Live on CNN in 2004 listened to the following message: Â“Today I call upon the Congress to promptly pass and to send to the states for ratification
6 an amendment to our Constitution defining and protecting marriage as a union of a man and woman as husband and wifeÂ” (Â“Debate ove r anti-gay marriage, Â” February 24, 2004). When asked to remember the source of information, one must recall the specific context in which the message was heard. In other word s, in order to retrie ve the source of the speech, the person has to remember where he/she heard the message (e.g., ABC News, CNN or Fox News) and what he/she did or how he/she responded to the message when it was heard. To answer the source of informati on, one should recollect any context that has spatiotemporal cues (e.g., Â“Yes, I remember I was traveling to San Fr ancisco at that time, and gosh, I never forget the City Hall...it was really crowded with gay peopleÂ…and yesÂ….I remember that President Bush made a speech a few days laterÂ”) with regard to the speech. Thus, memory researchers in s ource-monitoring framework (M. K. Johnson, Hashtroudi, & Lindsey, 1993) often considered source memory as being different from memory for contents (e.g., Batchelder & Ri efer, 1999; Bayen, Murnane, & Erdfelder, 2000; Yonlinas, 1999). Schacter, Kaszniak, Kilstrom and Vald iiserri (1991) explicitly referred to the source memory as "recollection of the episodic source from which a specific item or fact was acquired (e.g., from a person, a book, or television)Â” (p. 559). Some compelling evidence for the content-sour ce distinction can be found in the aging research. Researchers of aging research often found that meas ures of source memory had typically larger negative age relations than measures of memory for content (Siedlecki, Salthouse, & Berish, 2005). Sch acter et al. (1991) revealed that the elderly were less successful in identifying the presenter of the information than the younger. Some neuropsychological evidence (Craik et al., 1990; Gilsky, Polster, & Routhieaux, 1995; Gilsky, Rubin & Davidson, 2001; Shimamura & Squire, 1987) also implied that memory
7 for source information could be dissociate d from memory for facts (i.e., semantic memory). This theoretical assumption has rare ly been mentioned in previous research on the sleeper effect. The third assumption is related to th e second assumption. In this study, it is assumed that Web usersÂ’ failure in remember ing the source over time results from the source-memory decay phenomenon rather than from neglecting the source verification. Receiving information and opinions from a vari ety of Web-linked sources often leads to the Â“I-read-somewhere -thatÂ” phenomenon whereby Web users may accept the messages they read without adequately questioning the va lidity or credibility of the source (Sundar, 1998). Previous studies on the credibility of materials on the Web often referred to Web usersÂ’ tendency to neglect s ource verification as the main reason for their failure in remembering the source of information or opinions (e.g., Eysenbach & KÃ¶hler, 2002). However, these poor memories should not ne cessarily be attribut ed to Web usersÂ’ negligence to check their sources; as such an assumption may distort the nature of Webbased opinion monitoring. This assumption is undermined by the fact that as people are increasingly accustomed to using the Intern et as an information source; they are becoming increasingly savvy about which sour ces to believe and which to discredit (Eveland & Dunwoody, 2002; Flanagin & Metz ger, 2001; T. J. Johnson & Kaye, 2002). Even with this being true, Web users conti nue to experience the Â“I-read-somewhere-thatÂ” phenomenon. In the case of these Web users, source memory failure occurs not because they have skipped the sourceverification process but rather because their source memory has decayed over time as is found in the source-memory-failure phenomenon of traditional-media audiences.
8 Whether researchers attribute the Â“I-r ead-somewhere-thatÂ” phenomenon to the source verification failure or if they attr ibute it to the source-memory-decay phenomenon leads to different research que stions. From the perspective of source verification, Flangin and Metzger asked, Â“To what exte nt do users of the Internet verify the information they receive through this medium?Â” (Flanagin & Metzger, 2000, p.515). However, researchers who follow the source-memory decay phenomen on ask, Â“In what situation do people tend to forget or misattribute the source of information?Â” (Jacoby, Kelley, & Dywan, 1989; Zaragoza & Lane, 1994). The Purpose of the Study On the basis of the aforementioned three theoretical assumpti ons, the purposes of the current study are twofold. First, in an e xperimentally controlled setting, retention intervals were varied in an attempt to measure individualsÂ’ memory performance for either a typical or an atyp ical endorsement message about the Firefox Web browser. Second, in the assumption that a source memory decays over time faster than a persuasive message, this study attempted to test the sleeper effect. Unlike previous studies that examined the sleeper effect, th e persuasive message was varied with its typicality. It is a simulation of the case in which a Web us er could encounter either a typical or an atypical endorsement message from an ordinary person. To compare the effect of the low credibility, those two versi ons of a message were also projected from a highly credible source, and there were cont rol groups whose attitudes were measured without the presentation of the endorsement message. Investigating how Web users remember and how they believe the source of information they garnered from the Web is of great interest to researchers and professionals in communication fields (D utta-Bergman, 2004; Eysenbach & KÃ¶hler,
9 2002; Huh, DeLorme, & Reid, 2005; T. J. Johnson & Kaye, 2002; Warnick, 2004). In particular, these issues of source schemas and source credibility are also closely related to public relations matters, because trust or believability of information could be guided not only by a trustworthy source (i.e., source cred ibility), but also by an audience's prior knowledge (i.e., source schemas) of individua ls or institutions behind the information (Austin & Dong, 1994).
10 CHAPTER 2 REVIEW OF SCHEMA THEORY Overview of Theoretical Backgrounds Â“Who said what?Â” has been of multidiscipl inary interest among researchers in such diverse fields as communication (e.g., Sl ater & Rouner, 1997; Sundar, 1998), social psychology (e.g., Hovland, Janis, & Kelley, 1953; Hovland & Weiss, 1951), cognitive psychology (e.g., Graesser, Bowers, Olde, & Pomeroy, 1999; M. Johnson, 1993; Bayen, Murnane & Erdfelder, 1996), and consumer /marketing research (e.g., Pham & Johar, 1997). While the pursuit of the academic inquiry was somewhat different across disciplines, the same principle was shared in those diverse fields; oneÂ’s source memory or source attribution to the original message tends to deteriorat e over time. Notably, Hovland and his colleagues (Hovland & Wei ss, 1951; Hovland et al., 1953) found that individuals Â“remember and accept what wa s communicated but not remember who communicated itÂ” as time goes by (Hovland & Weiss, 1951, p.636). As a consequence, a message from a low-credibility source ultimately leads to attitude ch ange after a lapse of time, known as the sleeper effect. The finding of the sleeper effect was so counterintuitive1 that it prodded many researchers to replicate the effect. Indeed, throughout the 1950s and early 1960s, the sleeper effect took a respectful position in attitude research (McGuire, 1969; Priester, Wegener, Petty, & Fabrigar, 1999). However, subsequent studies during th e 1970s reported failure in repl icating the original sleeper 1 It is counterintuitive because the impact of persuasive communication in general is considered to be greater when it is measured at a moment closer to the initial exposure (Cook & Flay, 1978; Eagly & Chaiken, 1993; Kumkale & AlbarracÃn, 2004).
11 effect that Hovland & Weiss (1951) demonstr ated. A harsh criticism came from Gilling and Greenwald (1974) who conducted seven ex periments based on a thorough review of the extant literature. Because none of these ex periments resulted in the sleeper effect, the researchers argued that the sleeper effect s hould be put to rest. The same criticism was earlier raised by Capon and Hulbert (1973) who thoroughly reviewed and compared all the data in published research articles availa ble at that time, and concluded Â“there is no strong evidence for a generalized sleeper eff ect and we doubt that one exists Â” (p.358). Some researchers (e.g., Cook & Flay, 1978; Cook, Gruder, Hennigan, & Flay, 1979; Gruder, Cook, Hennigan, Flay, & Halamaj, 1978; Pratkanis, Greenwald, Leippe, & Baumgardnedr, 1988) have cautioned against an impetuous conclusion and have further provided theoretical and empirical work to demonstrate the presence of the sleeper effect. Recapitulated, those researchers who defended the sleeper effect came up with the socalled relative sleeper effect which included the non-significant incr ease or decrease in the amount of attitude change over time indu ced by a less credible source. In addition, by providing experimental conditions for replicating the sleeper effect, they suggested that the sleeper effect would occu r when the information about the discountin g cue and the message were poorly integrated and en coded into different memory systems2 (see Tulving, 1983 for different memory systems; see Pratkanis et al., 1988 for the discussion about the differential decay hypothesis). But some researchers (Allen & Stiff, 1998; Petty & Cacioppo, 1996) maintained that the assumption of the source-message separation hypothesis should not always be 2 In terms of two memory systems of Endel Tulving (1983), memory for the discounting cue may be represented in episodic memory, whereas the memory for the message (meaning) may be reserved in the semantic memory. Priester et al. (1 999) also discussed that the cue is forgotten sooner than the message because the memory decay is different for the ep isodic (cue) and the meaning (message) systems.
12 true. Since a message receiver possesses pers onal schemas that integrate past messages and source evaluation experience (Hamilton, 1998), there is a possibility that a certain type of argument is better integrated w ith the source information. Sometimes people do not pay much attention to what they consid er already familiar arguments because they consider them redundant. But sometimes th ey may be surprised to see unexpected arguments that violate their schematic expe ctation about the sour ce. Mazursky and Schul (1988) raised the possibility that, in the face of a discounting cue, a recipient may make a different judgment drawing an inferen ce about the source based on the message. In the present study, it is argued that th e prior knowledge and experiences a person has about the messageÂ’s source influence th e way the receiver processes the incoming messages such that they may influence th e delayed effect of messages from a lowcredibility source on attitude ch ange. In other words, it was assumed that the effect size of persuasion induced by a low-credibility source would be dependent upon the extent that the message meets what a message recipi ent expected to hear from the source of a persuasive message. Here, a persuasive me ssage is defined as Â“a set of arguments concerning beliefs that link the object w ith positive and negative consequences and evidence in support of those argumentsÂ” (Areni & Lutz, 1988, p.198). With regard to the source, a message does not comply with a reci pientÂ’s expectation in two cases: (1) when the source speaks against his or her self in terest; or (2) when message quality and/or message typicality violates the expectations about the source credibility. While some studies (Walster, Aronson, & Abrahams, 1966; Eagly, Wood, & Chaiken, 1978; Ziegler, Diehl, Zigon, & Fett, 2004) have investigated the first case, little is known about the joint effect of perceived source credibility and me ssage typicality on me mory and persistence
13 of attitude change. The typicality of a messa ge with respect to a central schema (e.g., smoking) refers to the extent to which the me ssage is structured, organized, and described in a semantically coherent way or in well-organized logical sequences3. In this conceptualization, an atypical argument4 for a certain issue/topic (e.g., an anti-smoking campaign) refers to an argument that is irrele vant from the central issue/topic and/or that is inconsistent with oneÂ’s es tablished schemas about the i ssue/topic. For instance, Â“I couldnÂ’t sleep last night Â– and you know whyÂ” is a typical sentence in laypeopleÂ’s Â“love letterÂ” schema. But would one expect Â“I no longer kill mos quitoes. I let dogs sniff my ankles. I line up the bars of soap on the shelves at K-Ma rtÂ” in a love letter? Certainly, those are least expected sentences in a love letter5. As seen in this everyday example, a perceiverÂ’s schemata and typicality percepti ons are the essential constituents of text, prose, narrative, and discourse compre hension (Goldman, Graesser, & Broek, 1999; Graesser, 1981; Graesser, Mills, & Zwaan, 1997; Kintsch & van Dijk, 1978; Kintsch, 1988; McKoon & Ratcliff, 1992; van Dijk & Kintsch, 1983). The main goal of this resear ch is to investigate whether the delayed effect of source credibility on attitude change occurs when messa ge typicality is varie d. It is assumed that the sleeper effect will occur when the message is typical rather than when it is atypical. When a low-credibility source makes an atypi cal message, not only will that message be discounted at once but it will not induce an in crease in attitude change even after a passage of time. On the other hand, if the lo w-credibility source makes a typical message, 3 Much of the information that people encounter in the course of their daily lives consists of sequences of temporally related events or episodes (Trafimow & Wyer, 1993) 4 The conceptualization will be further elaborated in the following chapters. 5 However, it could become a love letter with the subsequent sentences: Â“The world is mine, and IÂ’m its, and love it. Maybe this is a way of saying that I love you a whole worldÂ’s worth!Â” (Maggio, 2001, p.279)
14 the message will be elaborated, because of the inherent merit of the message itself. In this case, although the immediate opinion change may be s uppressed in the audienceÂ’ awareness of the discounting cue, the in itial source information may decay or be dissociated from the typical message, which ultimately results in an increased opinion change at delayed time intervals. With these assumptions, the sleeper effect may be enlightened with the schema theory because the generic schemata a person has about a source and/or an issue tend to guide memory representation of persuasive messages. Particularly, the schema-copyplus-tag (SC + T) model of a schema theo ry will help broaden our understanding of the sleeper effect in that it considers the possibi lity of differential me mory decay for typical versus atypical arguments. The SC + T m odel proposed by Graesser and his colleagues (1981; Graesser, Gordon, & Sawyer, 1979; Gr aesser, Kassler, Kreuz, & McLain-Alle, 1998; Graesser & Nakamura, 1982; Graesser, Woll, Kowalski, & Smith, 1980; Nakamura et al., 1985; Smith & Graesse r, 1981) has explicated th e role of schemas in comprehension and memory for prose, na rrative action, and stories. Schmidt and Sherman (1984) tested the SC + T model usin g persuasive arguments on the premise that arguments that are frequently encountered in persuasive messages on a particular issue may also be represented as a generic positi on schema. Results of their study provided strong support for the generality of the SC + T model. While both psychologists and consumer researchers have long argued that comprehension and memory for the contents of persuasive messages play an important role in the persistence of attitude change (Cook & Flay, 1978; Mick, 1992; Hunt et al., 1992), it is still unknow n whether attitudinal evaluations about an object coul d be influenced by a differe ntial retrieval context. In
15 addition, it has also not been investigated as to whet her the differential memory representation over time by message typicality could be also m oderated in the presence of different levels of source credibility, and whether these variations in memorability influence a recipientÂ’s attitudinal evaluations about the object. Relevant theories in memory and attitude research are reviewed in the following chapters. BartlettÂ’s Legacy and Schema Theory Schema theory represents perhaps th e most general framework for accuracyoriented memory research and it holds th at what people remember stems from an interaction between the input information and pre-exis ting Â“schemasÂ” -Â“generic knowledge structure or expectan cies that are developed thr ough experiencesÂ” (Koriat et al., 2000). Alba and Hasher (1983), in their wide ly cited review of the schema theory of memory, state, Â… what is encoded, or stored in me mory, is heavily determined by a guiding schema or knowledge framework that select s and actively modifies experience in order to arrive at a coherent, unifi ed, expectation-confirming and knowledgeconsistent representation of an experience (Alba & Hasher, 1983, p.203). Mandler (1992) detailed th e concept and structure of schemas as follows: A schema is a coherent and structures representation that organizes experience. Schemas are not carbon copies of experien ces, but generalized representations of experience regularities. Schemas range from the very concrete, involving the categorization of perceptual experience, to the very abstract, representing general levels of meaning such as Â‘loveÂ’ or Â‘j ustice.Â’ Abstract schemas subsume more concrete schemas; the resulting structure is hierarchical. Schemas are built up in the course of interaction with the social a nd physical environment. They organize and interpret in terms of the schemas laid own by past cognate e xperiences. Currently active schemas define what we are likely to see, hear, and remember, and also determine what we are unlikely to hear or see (Mandler, 1992, p.93). This knowledge structure in memory, or sc hema, thus, may facilitate remembering by providing mental cues with which new in formation can be asso ciated (Bellezza,
16 1996). However, the downside of schema-based information processing is found when people are asked to remember details which th ey process as a part of general episodes. When no particularly unusual information is found in the episodes th ey received, people process the input using general knowledge of th at type of situation, which may yield no specific remembering (Schank, 1999), and they may attempt to reconstruct the missing details by replacing the original input inform ation with generic information from the schemas, yielding schema-cons istent distortions (Koriat et al., 2000). For example, Brown and Park (2002) presented adults with medical information about either familiar (i.e., breast cancer) or unfamiliar (i.e., acromegaly, a rare pituitary disorder) illness. Then they asked their subjects questions asse ssing their recollection of the information, correcting for prior knowledge. They found that subjects learned more about an unfamiliar disease than a familiar disease on both a recall and a multiple-choice test that measured knowledge for new information. They also found that much of the information that participants recollected for a familiar disease came from what subjects already knew. The roots of schema theory can be tr aced back to Sir Frederic BartlettÂ’s (1932/1995) famous experiment, The War of the Ghosts . To simulate peopleÂ’s cognitive processes involving real-life language comp rehension and memory, Bartlett (1932/1995) conducted several field experiments with comple x literary material like brief stories or prose passages, which was reported in his remarkable book, Remembering: A study in experiment and social psychology . Subjects in his experiments read a short story and freely recalled what they could remember after a lapse of time. In other cases, a subject told another subject what they heard from the original source, and it was relayed to others by mouth. From this experiment, Bartlett found th at the original story became shorter and
17 more informal in style. In addition, he found that there were several types of transformations6 in message relay or in remembering after a delay. Finding these transformations and constructive processes in memory, he summarized the repeated reproduction results with this point: Â“Accuracy of reproduction, in a literal sense, is the rare exception and not the ruleÂ” (Bartlett, 1932/1995, p.93). Analyzing what he observed from his field experiment, Bartlett introduced the concept of schema ( schemata in plural). Bartlett attributed his use of the term sche ma to Cambridge neurologist Henry Head who used this word to explain Â“the models of body position that serves to re late past to present experience in the recognition of postura l changeÂ” (Johnston, 2001, p.352). Like Head, Bartlett introduced this concept exemp lifying a motor function of a stroke: Suppose I am making a stroke in a quick ga me, such as tennis or cricket. [Â…] When I make the stroke I do not, as a matte r of fact, produce something absolutely new, and I never merely repeat something old. The stroke is li terally manufactured out of the living visual and postural Â“schemataÂ” of the moment and their interrelations. I may say, I may think that I reproduce exactly a series of text-book movements, but demonstrably I do not; just as, under some other circumstances, I may say and think that I reproduce exactly some isolated event which I want to remember, and again demonstrably I do not. (Bartlett, 1932/1995, pp.201-202, italics are added for emphasis) It was the schemata that made BartlettÂ’s participants alter The War of the Ghosts by adding statements that were not presented in the original story, reflecting their culture, experiences and attitudes. Indeed, the gi st of BartlettÂ’s findings reported in Remembering was that a reader often did not make distinctions between wh at had been actually read in a passage and what was a plausible infe rence or guess at the retrieval time. 6 They include omissions, condensations, rationalization (to help explain certain incongruous passages), dominant theme (some themes became prominent, and other f eatures are related to the dominant theme), transformation of unfamiliar words to more familiar ones (e.g., Â‘hunting sealsÂ’ Â‘fishingÂ’), and transpositions (transformation of sequences) (Johnston, 2001; Solso, 1991).
18 Following BartlettÂ’s experiment, many rese archers have investigated psychological processes in remembering meaningful prose that closely resemble s real-life language experiences. It was also from BartlettÂ’s le gacy, inherited in ps ychology and in memory research, that psychologists began to recognize that Â“memor y is not a literal reproduction of the past but instead depends on construc tive processes that are sometimes prone to errors, distortions, and illusi onsÂ” (Schacter et al., 1998, 290). Under the rubric of the schema theory (B artlett, 1932), relate d issues have long been investigated by cognitive psychologists working on memory for narrative texts such as prose, sentences and disc ourse (Bower, Black, & Turner, 1979; Bransford, Barclay, & Franks, 1972; Graesser, Singer, & Trabass o, 1994; Kintsch, 1988; Kintsch & van Dijk, 1978; Owens, Bower, & Black, 1979; Scha nk & Abelson, 1977; Thorndyke, 1977; van Dijk & Kintsch, 1983). These studies addresse d, Â“how are memory schemata activated and used to guide to organization of in coming information?Â” and Â“how is that information represented in memory?Â” (Thor ndyke & Yekovich, 1980). In the core of their reasoning is an assumption that the gene ric knowledge structure which a person has acquired from prior experiences influences what they remember from the current narrative events. Thorndyke (1977) assumed that comprehensio n of certain prose passages is closely related to the use of abstr act structural schemata wh ich guide comprehension and encoding of the narrative structures in part icular passages of pros e. In other words, memory for events of the story is postulated to be dependent upon peopleÂ’s ability to map those events into familiar patterns or story forms that incorporate structural information about narratives. Then a break in the main thr ead of events can make a difference in story
19 comprehension. To test this assumption, Thorndyke first identified four hierarchies from a story entitled Circle Island according to how propositions in the story are situated. In this story structure, propositions at a higher le vel were more important to the reader than propositions at a lower level. Subjects in his ex periment then read either the original story or constructed versions of the basic Circle Island story. One version called Â“narrativeafter-themeÂ” contained the main theme at th e end of the story. Another version called Â“narrative-no themeÂ” did not carry the theme, and the Â“descriptiveÂ” story was presented without local causal im plication or any temporal conti nuity. What he hypothesized from this experiment was that subjectsÂ’ ability to recall would decrease with the alteration of structural information in the narratives. Re sults showed that recall percents for those altered versions of the stor y were poorer than those for the original story, and the difference was most noticeable at the hi ghest level of stru ctural hierarchy. Bower, Black and Turner (1979) showed how script-based text led to memory errors by interacting with a generic script in peopleÂ’s memory (experiment 3 & 4). In a recall test (experiment 3), they found that the percentage of unstated script intrusions increased when other story instances of the same script were present. In a recognition test (experiment 4), they found salient differences between different instance conditions only occurred with unstated script fillers: false-positive recognition ratings for an unstated script filler increased as more instances were put into memory. Owens, Bower, and Black (1979) showed that readersÂ’ beliefs about a characterÂ’s motivation guide their comprehension and r econstruction of text recall. In their experiments, those who were given b ackground information (e.g., an unwanted
20 pregnancy) about the characte rs in neutral scripts produ ced four times as many new propositions, most of which were guided by belief schemas. However, BartlettÂ’s theory of schema ha s often been criticized on the ground that it is too vague and complex to be empirica lly testable (Solso, 1991). Thorndyke and Yekovich (1980) voiced a concer n that Â“while schema theo ry provides a plausible and descriptive framework for understanding hum an knowledge processing, it provides few detailed processing assumptionsÂ” (p.23). Some even have argued that the concept of schema itself is abstract and has no fixed definition, simply referring to the Â“general knowledge a person possesses about a partic ular domainÂ” (Alba & Hasher, 1983, p.203). Despite the criticism of the general theory of schema, the concept of schema provided an important frame of reference for scholars across diverse academic disciplines. Numerous studies have tested the schema theory using a variety of objects such as prose, narratives, events, persons, actions, me ssages and so forth. Most of all, schema theory has developed to the extent that it explicates some patterns observed in oneÂ’s memory reconstruction processes over time . The general pattern in schema-driven memory error over time was also derived from BartlettÂ’s original experiment. Subjects in BartlettÂ’s study tended to reconstruct the stor y in a way that thei r prior knowledge or experiences thought plausible or fill the mi ssing details using schema-based inference processes at the time of remembering (Koriat et al., 2000). Thus, along with the schematic inference in the comprehension of prose, another legacy of BartlettÂ’s study was the delayed effect of the retrie val of information from memory over time; that is, as time passed, stories that conflicted with the main points of the theme were forgotten more rapidly.
21 Graesser and his colleagues (1981; Grae sser, Gordon, & Sawyer, 1979; Graesser, Kassler, Kreuz, & McLain-Alle, 1998; Gr aesser & Nakamura, 1982; Graesser, Woll, Kowalski, & Smith, 1980; Nakamura et al., 198 5; Smith & Graesser, 1981) elaborated BartlettÂ’s schema theory of prose comprehens ion in a testable theoretical model. In the following chapter, Graesser and his colleague sÂ’ schema-copy-plus-tag (SC + T) model, which attempted to reveal the patterns of schema-based memory retrieval in the passage of time, is reviewed. Message Typicality Effect: The Sche ma-Copy-Plus-Tag (SC + T) Model Memory researchers have long assumed that a schema guides a perceiverÂ’s interpretations, inference, expectations, attention, and later retrieval of an event (Nakamura et al., 1985). Particularly, psychologi sts who study the role of schemata in the acquisition, representati on, and retrieval of prose in me mory have agreed that those schemata interrelate statements in passages (Smith & Graesser, 1981). With regard to prose comprehension, psychologists have found an important phenomenon about laypeopleÂ’s schema-driven information processi ng and their later retrieval of acquired information. In other words, as an individua lÂ’s experiences or knowle dge about an object, concept, or situation broaden and become extensive, their schemata allow them to form relatively automatic perceptions and encoding of similar objec ts, concepts or situations. In this process, effortful processing of each element is not necessary, and only a small number of typical elements are needed to ac tivate the appropriate schema (Vakil, Sharot, Markowitz, Aberbusch & Groswasser, 2003). T hus, the level of typicality in upcoming information has a differential influence on memory performance for the information. When such information is specifically rela ted to a sequence of activities or actions for an event or a situation, it comprises a sc ript (Schank & Abelson, 1977). A script is a
22 class of a schema, and it repres ents the interrelationships amo ng its described activities or actions (Nakamura et al., 1985). For example, an ordinary personÂ’s Â“morningÂ” script may consist of a number of typi cal actions such as brushing teeth, getting dressed, eating breakfast, etc. In this definition of a script , Erdfelder and Bredenkamp (1998) referred to Â“(script)-atypical informationÂ” as Â“unexpected information Â– that is, information that is relevant to the script but ha s a low a priori probability of occurring when the script is activatedÂ” (p.922). Analogously, they defined Â“( script)-typical informationÂ” as Â“scriptrelevant information that is frequently associated with the scriptÂ” (p.922). Much research on scripts has examined how individuals used such script-based knowledge structures in making inferences a nd retrieving the action sequences that had occurred in a script-based text (e.g., A bbott, Black, & Smith, 1985; Bower et al., 1979; Graesser et al., 1979; Owens et al., 1979; Smith & Graesser, 1981). A theoretical model that buttressed the studies on the script-b ased memory representation was Schank and AbelsonÂ’s (1977) so called Â‘script pointe r + tagÂ’ (SP + T) hypothesis. Schank and Abelson (1977) proposed the SP + T hypothesi s to describe the nature of memory representation of typical/atypical actions in a certain script . The following two sequences of actions exemplified in Abbott et al.'s ( 1985, p.180) illustrate the examples of either a typical or atypical script fo r different everyday situati ons: (1) George entered the department store. He picked out some shoes. He paid the cashier; (2) George entered the doctor's office. He ordered a salad. He took careful notes. Schank and Abelson (1977) assumed that each typical statem ent (e.g., entered the department store; picked out some shoes, pa id the cahier) is represented in memory by a single pointer called a Â‘scrip t pointerÂ’ that relates it to the generic schema (e.g.,
23 shopping), whereas every atypi cal action is represented in a separate functionalorganizational unit called a Â‘tag.Â’ Figure 1. Representation of a schema-based passage described in Schank and AbelsonÂ’s (1977) SP + T model. When the generic schema is activated, it is likely that highly pr obable actions come to mind spontaneously. As a consequence, what the perceiver remembers from the scriptbased actions includes both previously pr esented and non-presented (e.g., George asked the salesperson if he could tr y it on) typical actions. For ex ample, while reading a passage about eating at a restaurant, a reader may infer by default that George ate even when this is never explicitly mentione d in the text. As a consequence, when requested to tell whether GeorgeÂ’s eating had or had not been explicitly me ntioned in the text, the reader could have difficulty in determining it (G raesser, Kassler, Kreuz, & McLain-Allen, 1998). In this way, the intrusion rate (false alar m rate in recognition) for typical actions is greater than that for atypical actions. In fact, it reflects a general assumption of the
24 schema theory since BartlettÂ’s (1932) fiel d experiment: peopleÂ’s preexisting schemata tend to lead their memories in schema-consiste nt directions such th at schema-consistent stimuli generate more false memories than do schema-inconsistent ones (Brewer, 2000; David, 1996; Davidson & Hoe, 1993; Johnston, 2001; Macrae et al., 2002; Neuschatz et al., 2002; Shapiro & Fox, 2002; Wicks & Drew, 1991). In other words, recall of typical items was accompanied by greater proportions of intrusion by unseen items than was the recall of atypical items (Smith and Gr aesser, 1981). This phenomenon is called a Â“typicality effect.Â” (Grae sser & Nakamura, 1982; Graesser et al., 1979, 1998; MÃ¤ntylÃ¤, 1997; Mantwill et al., 1995; Nakamura et al., 1985; Smith & Graesser, 1981; Vakil et al., 2003). In Matwill et al. (1995)Â’s experiment, pa rticipants with and without the experience of blood donation watched a videotape that c ontained video clips of both typical and atypical scenes regarding blood donation events . Results of Matwill et al.Â’s (1995) study showed that confabulation7 rates were significantly hi gher among blood donors than nondonors, yielding significantly more intrus ion errors among experienced donors than among non-donors. In other words, interviewees who were highly familiar with the blood donation process more likely filled in relevant details that were not seen in the videotape than did non-donors. In a nutshell, the typicality effect is a cousin of the phenomenon of schema-driven memory construction. Roediger and his coll eagues (2001) expanded how this typicality effect could be implanted by social influe nce using subject-confederate experimental design. An important finding from this st udy was that intrusion effects (i.e., false 7 After watching short films about the blood donation, participants were requested to remember what they had seen in the films. If a detail was mentioned that was not presented in the videotaped films, it was coded as a confabulation.
25 memories induced by a confederate) occurred in high-expectancy items (e.g., toothbrushsoap items [a studied vs. a contagion item] in a bathroom scene) than low-expectancy items (e.g., hair brush-lens solution in a ba throom scene). In addition, falsely recalled items in a high-expectancy condition were mo re likely to be given a Â“knowÂ” response Â– which reflects feeling of familiarity Â– th an a Â“rememberÂ” response Â– which reflects conscious recollection. An underlying assu mption of this gap-filling memory phenomenon is that oneÂ’s inferences of epis ode are governed by generic scripts that do more than just Â“bridging inferenceÂ” in te rms of psycholinguists (Abelson, 1981). Thus, when a person reads a story such as, Â“Feeling very hungry, John went to the restaurant. . . . . Suddenly, he realized that heÂ’d forgotten his glasses,Â” the reader must search typical restaurant events to understand why John suddenl y realized that he had left his glasses. Then, a reader, using a script that activates restaurant scen es, could think that John must have tried to see the menu. A relevant research paradigm has examined so-called Â“memory for place ( aka ., room schemas)Â” in which items in a real space are used to test schema-consistent memory performance (Brewer & Treyens, 1981; Lamp inen et al., 2001; MÃ¤ntylÃ¤ & BÃ¤ckman, 1992; Pedzek et al., 1989; Salmaso et al., 1983). A core issue in this paradigm is that schema expectancy or schema consistency causes perceivers to better remember something unique, unusual, novel, or atypical as opposed to those things common, usual, normal or typical. For example, subjects in Brewer and TreyensÂ’ s (1981) experiment were taken into an experimental room wh ich resembled a typical graduate studentÂ’s office. Some items that were not consistent with the office schema were included for experimental purposes. After introducing th e room as a graduate assistantÂ’s room,
26 subjects were taken out of the experime ntal room, and administered a recall and recognition test. The experimentÂ’s most in sightful finding was that schema-expectancy has a positive correlation with recognition for fo il items. In other word s, subjects in their experiment tended to recognize a non-present it em as a present member of the office. A slightly different version of the Â“room schemaÂ” has been tested in the Â“action schemaÂ” paradigm (Graesser et al., 1979; Gr aesser et al., 1980; Na kamura et al., 1985; Smith & Graesser, 1981). In Nakamura et al .Â’s experiment (1985), participants were given a recognition memory test following tw o lectures performed for experimental purposes. The recognition test queried ten typi cal actions (five from one lecture session and five from the other) and te n atypical actions (five from one lecture session and five from the other lecture session). Results showed that memory discrimination8 was better for atypical actions and that the false-alarm ra te was three times higher for typical actions than for atypical actions. Pedzek and her colleagues (1989) added one more place (i.e., a preschool classroom) to the original graduate studentÂ’s office sett ing. Again, this study reported a schema-consistency effect: items inconsistent with expectations were significantly recalled and recogni zed better than those consis tent with expectations. Based on Schank and AbelsonÂ’s (1977) Â‘Scr ipt Pointer + TagÂ’ hypothesis, Graesser and his colleagues (Graesser et al., 1 979, 1980, 1998; Graesser & Nakamura, 1982; Smith & Graesser, 1981) have devel oped the schema-copy-plus-tag (SC+T) model to explain memory for actions occurring in script-based pr ose. Nakamura et al. (1985) describes the SC + T model as follows: 8 Recognition performance is generally measured by hit rates, false alarm rates, and d scores. When it comes to d scores for recognition performance, it is often called memory discrimination (Smith and Graesser, 1981). This ability to memory discrimina tion is also called memory accuracy. Thus, memory accuracy and memory discrimina tion is the same concept.
27 According to the SC + T model, a speci fic memory trace is constructed by the comprehender for every scripted activity that is read or heard. This memory trace contains a pointer to an instantiated scri pt that has been copied from a permanent generic script. The instantiated script contai ns a subset of the sc ript-relevant actions whose variable slots were activated by e xplicit information that was mentioned in the passage. In addition, there is a subset of script-relevant ac tions whose variable slots were activated as the result of script-based inferences occurring during comprehension of the text. Thus, the instan tiated script contai ns activated explicit and inferred script-relevant actions. The memo ry trace also contains a set of tags, each of which corresponds to an irrelevant ac tion. Each of these tags is a distinctive unit that is automatically incorporated into the memory trace. (Nakamura et al., 1985, p.141) In other words, a reader of prose co mprehends or stores messages by first identifying a generic schema that best fits the main theme of the prose. For instance, when a person reads medical information about breast cancer, he or she will activate their generic schema for breast cance r (Brown and Park, 2002). In th is process, it is assumed that sentences that are typical of a relevant schema are to be copied from the generic schema and to be saved in memory represen tation, whereas atypical sentences are to be Â“tagged,Â” and thus they are sa ved in a separate unit in me mory representation. Thus, the SC + T model predicted that atypical senten ces were better remembered than typical sentences, at least initiall y. Graesser, Kassler, Kreuz and McLain-Allen (1998) and Woll and Graesser (1982) maintained that, in the SC+T hypothesis, the prediction of memory performance in an immediate test is directly contrary to those of a competing Â“perceptual filteringÂ” model, one which is not explic itly formulated in the areas of cognitive psychology, but one that clearly in fluenced several studiesÂ’ a ccounts. In this view of schematic processing of information, information that mismatches a personÂ’s schemas will be less likely to be encoded, because he or she may Â‘filterÂ’ the schema-incongruent information (Neisser, 1976; Stangor & McMi llan, 1992). Tesser and Leone (1977) also assumed that schema-inconsistent items fr om newly-updated information are hard to
28 retrieve since peopleÂ’s cognition is led to a schema-consistent direction. This assumption is directly opposite to th e Graesser et al.Â’s represen tational account for memory superiority of at ypical items. Stangor and McMillanÂ’s (1992) meta-ana lytic review of these two competing models provides strong support for the supe rior memorability of schema-incongruent information; the weighted effect of the recall and the recognition-sensitivity data correlated strongly with th e proportion of incongruent items. Rojahn and PettigrewÂ’s (1992) meta-analysis also resulted in a similar result; schema-inconsistent information had better recognition (less false alarm ra te) than schema-consistent information. Numerous studies replicated the aforementioned predictions of the SC + T model, and relatively robust results have been obtaine d through an array of related applications (Erdfelder & Bredenkamp, 1998; Ford & Smith, 1991; Graesser et al., 1979, 1980, 1998; Hunt et al., 1992; Lampinen et al., 2001; Lampinen et al., 2000; MÃ¤ntylÃ¤ & BÃ¤ckman, 1992; Nakamura et al., 1985; Neuschatz et al., 2002; O'Sullivan & Durso, 1984; D. F. Schmidt & Sherman, 1984; Shapiro & F ox, 2002; Smith & Graesser, 1981; Woll & Graesser, 1982). In addition, Graesser and his colleag ues (1981; Graesser & Nakamura, 1982; Nakamura et al., 1985; Smith & Graesser, 1981) assumed that memory for atypical information declines at a faster rate over ti me than memory for typical actions. However, in predicting the delayed memory performa nce, Graesser and his colleagues postulated two distinct processes Â– c onceptually-driven retrieval9 (recall) and data-driven retrieval10 9 In conceptually driven retrieval, organized retrie val strategies access a specific item that is part of a contextually specific memory trace. In this conceptua lly driven retrieval, an item would not be retrieved without its context being reinstated. Recall is guided exclusively by conceptually driven retrieval (Smith & Graesser, 1981, p.551).
29 (recognition) Â– to explain an individualÂ’s co rrect retrieval of info rmation acquired during the encoding stage. In their first study (Graesser et al., 1979) on the schema-copy-plustag model11, they did not propose any hypothesis w ith regard to the delayed memory performance. In the subsequent study, Grae sser et al. (1980) collected both recall and recognition data in different retention intervals (30 minutes v. one-week) and two trends emerged: First, memory performance was better for atypical information initially than for typical information in both the recall and r ecognition test. Second, atypical information decayed faster than typical information fo r both recall and recogn ition. In this memory decay process, recall was better for typical than for atypical information after a one-week retention interval. In Figure 2 belo w, MI indicates Â“memory improvement12Â” scores that corrected for guessing by intrusi on. They reported a main effect13 of typicality, a main effect of retention interval14, and an interaction15 between typicality and retention interval, concluding Â“this antagonistic intera ction clearly supports the hypothesis that atypical actions are forgotten at a faster rate than the typical actionsÂ” (Graesser et al., 1980, pp.509-510). 10 The test item on the recognition test contains a number of informational cues that provide relatively direct access to the item in memory. Data-driven retrie val is analogous to what has been called Â“detection of familiarity,Â” and is not always guided by organized retrieval strategies and does not always critically depend on reinstating the organized context in which a particular item is embedded. Recognition involves both data-driven and conceptually driven retrieval (Smith & Graesser, 1981, p.552). 11 In this study, they called their model a Â“script point + tagÂ” hypothesis. 12 MI (recall) = p(recall) Â– p (intrusion) / 1 Â– p (intrusion) Graesser et al., 1980, p.509 13 F (1, 54) = 8.03, p < .01 14 F (1, 54) = 89.97, p < .01 15 F (1, 54) = 22.60, p < .01
30 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 30 Min-Typical1 Week-Typical30 Min-Atypical1 Week-Atypical p(recall) p(intrusion) MI score Figure 2. Recall performance for typical vs. atypical actions over time. However, memory discrimination for scri pted actions on the recognition test was uniformly higher for atypical than for typi cal actions after both 30 minutes and one-week retention intervals. The d scores16 for atypical actions were significantly higher than those for typical ones. Their analysis of variance on the d scores revealed a significant main effect of re tention intervals, F (1, 54) = 57.75, p < .01 and a significant interaction effect between typicality and retention intervals, F (1, 54) = 53.65, p < .01, supporting their hypothesis that atypical acti ons were forgotten at a faster rate than typical actions. In summary, Graesser et al. (1980) concluded that the generic schema tends to increasingly guide conceptually driven memory test recall. 16 The d index is most widely used in memory research to assess subjectsÂ’ ability to discriminate between presented and nonpresented items in recognition memory tasks (see Murdock, 1974; 1982). The higher the scores, the more accurate oneÂ’s recognition memory discrimination.
31 0 0.5 1 1.5 2 2.5 30 Min-Typical1 Week-Typical30 Min-Atypical1 Week-Atypical p(hit) p(false alarm) d' score Figure 3. Recognition performance for typi cal vs. atypical actions over time In a follow-up study, Smith and Graesser (1981) examined this schema-guided memory performance, varying the retenti on intervals (two day, one week, and three weeks). The following Figure 4 was drawn for this present study based on the findings of Smith and GraesserÂ’s (1981) study.
32 Figure 4. Recognition memory performance at different rete ntion intervals As Figure 4 indicates, the analysis of d scores of Smith a nd GraesserÂ’s (1981) study showed that memory discrimination wa s uniformly higher for atypical than for typical information at each retention interval . Only significant main effects of typicality and retention intervals on d were found. Since d score was not appl icable to recall scores, Smith and Graesser (1981) conducte d a comparison of recall and recognition memory by computing Â“memory scoresÂ” (MS)17. The memory scores in recall and recognition reflected a measure of memory that corrected for guessing. In other words, false alarm rates served as an estimate of guessing rates on the recognition test, whereas 17 Memory scores for recall and recognition were computed using the following formulas (Smith & Graesser, 1981, p.554). MS (recall) = [p (recall) Â– p (intrusion)] / [1 Â– p (intrusion)] MS (recognition) = [p(hit) Â– p (false alarm)] / [1 Â– p (false alarm)]
33 the intrusion proportion served as an estim ate of the guessing rates for recall. They reported the following trends by comparing th e MS for recall and recognition (see Figure 5). Figure 5. A comparison of memory scores (M S) of recall and rec ognition at different retention intervals Initially memory scores were higher fo r atypical information than for typical information in both recall and recognition. Wh ereas the recognition curves for atypical and typical information maintained roughl y equal slopes across the four retention intervals, a crossover occurred between 30 mi nutes and one week for the recall curves. Thus, Smith and Graesser (1981) concluded the reversed memory scores sometime around two-day retention interval and therea fter corroborated th eir assumption that Â“schemata become progressively more important in guiding conceptually driven retrieval
34 as the retention interval incr easesÂ” (p.555). They also asserted , Â“the recall of information is directed by conceptually driven retrieva l, whereas recognition is the product of both conceptually driven and data -driven retrieval mechanismsÂ” (p.557). They discussed that the finding of exponential decay for conceptua lly driven retrieval is not surprising and indeed is consistent with most theories of forgetting that have been postulated ever since EbbinghausÂ’ (1885) seminal study. They further argued: The linear function, which best describes da ta-driven retrieval, is not common in memory theories. Most studies that plot the forgetting ra te for recognition memory have focused on short-term memory, in whic h decay is very rapid. Our results show that the rate of forgetting is much more gradual and constant at longer retention intervals. This gradual loss of item retrievability is perhaps due to the multiple dimensions that are encoded and retrie ved in data-driven retrieval (Smith & Graesser, 1981, p.557). The results18 of Smith and GrasserÂ’s (1981) resear ch indicate that any attempt to make general propositions about memory for typical vs. atypical information without the consideration of the retention intervals and the type of retrieval task (i.e., recall and recognition) is inappropriate. N onetheless, some studies that tested the SC + T model did not consider retention intervals for measur ing memory accuracy (e.g., Nakamura et al., 1985; Woll & Graesser, 1982; Vakil et al., 2003 ). Even when the time factor was considered, some studies (Davidson & Hoe, 1993; Ford & Smith, 1991; Lampinen et al., 2000, 2001; OÂ’Sullivan & Durso, 1984; Schmidt & Sherman, 1984; Shapiro & Fox, 2002; Pezdek et al., 1989) measured the rec ognition memory only in a short-term delay (less than or equal to 48 hours19). A few studies (e.g., Davidson, 1994; Davidson, Shari, 18 The same results were reported in Graesser and Nakamura (1982). However, Graesser and NakamuraÂ’s (1982) results were based on the same data that were used in Smith and Graesser (1981) and Graesser (1981). 19 Given that the advantage of atypical information in recall tended to be reversed by typical information around a week or thereafter, a delayed recognition test should be conducted at l east in a week interval.
35 Zupei, & Burden, 2000; Neuschat z, 2002) varied the retention intervals more than a week from the initial presentation of experimental stimuli. After digging into both academic databases20 and the Google Scholar21 search engine, all publishe d articles in refereed journals in the U.S. that empirically tested the SC + T model or the typicality effect were retrieved22. Among those studies retrieved, onl y one study by Rice and Okun (1994) allowed for the retention interval for more than a week (i.e., a two-week interval). However, a closer look at Rice and OkunÂ’s study revealed that their application of the SC + T model is somewhat different from other studies that examined the SC + T model. Strictly speaking, what they tested was not Â“typi cality effect,Â” even though they explicitly stated that their study was grounded on the SC + T model. What they investigated was memory for medical information that either affirmed a readerÂ’s belief or disconfirmed a readerÂ’s initial belief. Considering that the revers ed recall performance (see Fi gure 5) reported in Smith and GraesserÂ’s (1981) research o ccurred after a one-week reten tion interval, the lack of a long-delayed research design leaves room for further investigation of the SC + T model in a longitudinal research design. With rega rd to memory performance over time, two issues emerge. First, considering that little research ha s measured recall perf ormance after a week from the experimental treatment, the reve rsed recall performan ce over time is not yet 20 Databases that were searched included PsycINFO, EBSCOHost, JSTOR, and th e Social Science Citation Index. Searched keywords were Graesser , Script-Pointer , Schema-Copy-Plus-Tag , SP + T, SC + T, narrative comprehension, prose comprehension, typicality effect, schema-(in)consistent, schema(ir)relevant, schema-(in)congruent, and schema-(un)expected . 21 http://scholar.google.com 22 Full text of all empirical studies were electronically retrieved or hard-copied.
36 conclusive. Second, if atypical items have memory advantage over typical items, what underlies this phenomenon? In f act, numerous studies corrob orated a hypothesis for the SC + T model Â– that is, atypical items are be tter recalled and recogni zed at least in an immediate test. In addition, most research that examined the delayed effect in recognition memory found better memory discrimination fo r atypical items than for typical items. As discussed earlier, Graesser and associates have long argue d that this advantage for atypical over typical information in memory performance should be attributed to the distinctive representational tag that is intrin sic to the atypical information itself, which constitutes a Â“representationalÂ” hypothesis. However, many researchers have suggested that the superior recognition memory for atypical information may be due to a perceiverÂ’s elaborative cognitive processing. Graesser and his colleagues re petitiously rejected the latter view, arguing that it could not explain the reversed recall performance at the delayed test. However, as shown in the Figure 5, memory scores for recall between typical and at ypical items at long retention intervals (i.e., from one-week to three-week delay) are not distinguishable, whereas the memory scores for recognition show an apparent a dvantage of atypical items. In the next section, competing acc ounts for explaining the SC + T model were reviewed. Theoretical Underpinning of The SC + T Model Despite its intuitive predictions, the SC + T model has not delv ed into underlying mechanisms of memory for atypical informa tion and the different patterns of memory performance over time. Most of all, the SC + T model does not c onsider alternative possibilities of the typicality effect other than the assumption of separate memory organization of the atypical information. In ot her words, it attributes the better memory
37 discriminability of the atypical information to the schema-incongruent nature of the distinctive information per se. Thus, attenti onal and elaborative proc esses are completely irrelevant in this theoreti cal account (Erdfelder & Bredenkamp, 1998). However, some researchers (Erdfelder & Bredenkamp , 1998; O'Sullivan & Durso, 1984; Wyer & Gordon, 1982) have raised an alternative hypothesis either exp licitly or implicitly referring to the levels of processing concept (Craik & Lockhart, 1972). To answer whether better recognition and recall for atypical information at least at a short-term memory measure is a result of memory repres entation or deeper processing, one needs to ruminate on the nature of atypi cal information. Indeed, a unique explanation for the SC + T model comes from the Â“distin ctiveness effectÂ” (Eys enck, 1979; Jacoby & Craik, 1979; McDaniel & Einste in, 1986; Roediger & Guynn, 1996; Pezdek et al., 1989). According to the distinctiveness effect, Â“if th e distinctive features of an item are stored, the item may Â‘stand outÂ’ from other info rmation in memoryÂ” (S. R. Schmidt, 1991, p.535). Briefly, distinctive and unusual events are generally well remembered Â“because they represent powerful retrieval cuesÂ” (Roediger & Guynn, 1996, p.205). As a result, atypical information is remembered in a vi vid and detailed manner (Lampinen, Copeland, & Neuschatz, 2001). The assumption that superi or memory discriminability of atypical components leads to a qualitatively different status of memory representation (i.e., Remember vs. Know) was corroborated by Lamp inen et al. (2000). In a replication of Graesser et al.Â’s (1979) researc h, Lampinen et al. (2000) had s ubjects listen to narrative stories of Â“JackÂ” who engaged in six scripted activities. Each Jack story contained both typical (e.g., washing his car with soapy wa ter and turning off the hose) and bizarre actions (e.g., washing his nei ghborÂ’s car and spraying a neigh borÂ’s child with the hose).
38 By employing the Â“remember/know (R/K)Â” j udgments, Lampinen and his associatesÂ’ study evidenced that memory for atypical information was not only more correctly recognized than typical information, but it al so was more experientially vivid than was memory for typical actions. The distinctiveness effect as the theoretical underpinning for the SC + T model is Â“a direct descendent of the von Restorff effectÂ” (S. R. Schmidt, 1991, p.523). The von Restorff effect, known as the generic label for distinctiveness effect, stems from von RestorffÂ’s experiments in which her subjects showed better memory for the syllable in the isolation list (i.e., a syllable embedded in the numbers) than for the same item in the homogeneous list (Hunt, 1995). In a theoretical aspect, Grae sser et al. (1979) affirmed th at findings in the SC + T model are compatible with the von Restorff effe ct. Graesser et al. ( 1998), in explaining a representational view of the SC + T model, also argued that th e episodic representation of an event consists of Â“a full or partial copy of the schema as a whole, plus a set of distinctive tags for each piece of explicit information that deviates from the schema (much like a Â‘weird listÂ’)Â” (p.255, italics added). For example, the typical items in the script-based story may form the homoge neous background whereby atypical items may stand out from the typical items (Dav idson, Larson, Luo, & Burden, 2000). The distinctiveness effect is assumed to be th e theoretical underpinning of the SC + T model in some aspects. First, the theoretical accounts for the dis tinctiveness effect have explained the underlying mechanisms of the relatively robus t findings of the better memory for the atypical events over typical ones. Two paradigm s have contributed to the explanations for
39 the distinctiveness effect: differential pro cessing views and repres entational views. Graesser and his colleagues (1979; Woll & Gr aesser, 1982) explain that the different memory discriminability by typicality is a re sult of the natu re of memory representation, that is; atypical events are tagged as a func tionally separate organizational unit, being discriminated well in memory. Second, as Davidson et al. (2000) aptly explained, the ways in which atypical actions in a narrative story ar e represented in a personÂ’s memory are analogous to the ways distinctive items are represented in memo ry; that is, atypical information stands out from the background typical information. Third, the theoretical boundaries of distinctivene ss are larger than those of the SC + T model. Being atypical is a sufficient condi tion to being distinctive, but the former cannot satisfy a necessary condition for the latt er. Thus, the properties and characteristics of the distinctiveness effect include not only the representati onal assumption of the SC + T model but also other assumptions (e.g., elabor ation in distinctiveness) (Wallace, 1965). As discussed, much of explanation on dis tinctiveness has been generally classified into two general classes of theoretical view s: 1) Differential pr ocessing views and 2) representational views (S. R. Schmidt, 1985, 1991, 1994; Waddill & McDaniel, 1998). Graesser et al. (1998) also recently a ddressed those competing accounts (i.e., a representation view and a processing view) for the better memory discriminability for the atypical information. Differential processing views The theoretical underpinning of these differential processi ng views originated from the levels-of-processing framework that assu mes deeper and more meaningful processing (e.g., semantic processing) should increase re tention compared to more shallow or
40 superficial processing (J. R. Anderson & Re der, 1979; Craik & Lo ckhart, 1972; Craik & Tulving, 1975; Glover, Plake, R oberts, Zimmer, & Palmere, 1981). Waddill and McDaniel (1998) recapitulate d the differential processing accounts with three mechanisms in which the atypica l material may elicit more processing that aids memory performance. The first approach comes from spontaneous elaboration hypothesis which assumes, Â“the unusual material requires the activati on of a wider variety of background knowledge than does common material in order to create an adequate interpretation or understanding,Â” and it Â“prompts more s pontaneous elaboration than does common materialÂ” (Waddill & McDaniel, 1998, p.109). According to S. R. Schmidt (2002), this type of elaboration can be derived from eith er a specific encoding context or an attempt to interpret atypicality. The former is labele d as Â“context dependent elaboration,Â” and the latter as Â“interpretive ela borationÂ” (S. R. Schmidt, 2002, p.128). Recently, researchers in SC + T model also addressed the possibil ity of cognitive elaboration of atypical information (Erdfelder & Bredenkamp, 1998; Graesser et al., 1998). Based on a spontaneous elaboration hypothesis, Erdf elder and Bredenkamp (1998) tested a hypothesis that proposed th at the generation effect23 would not be effective in atypical information. The assumption is that if atypi cal information receives more elaborative processing, then the extra processing implied by missing letters are less effective. Results showed that while the typicality effect wa s stronger for complete items than for fragmented items, presenting actions in fragmented sentences improved recognition memory mostly for highly t ypical items. They argue that these results are readily 23 A task generating missing word leads to an increase in elaborative semantic encoding. E.g., Peter re_ov_s his glasses from the soup; generate removes.
41 explained by an attentional-elaboration framew ork; thus, explaining t ypicality effect only with a representational view poses some pr oblems for the predictions of the SC + T model. A second explanation was derived from the Â“selective displaced rehearsalÂ” hypothesis (Slamecka & Katsaiti, 1987). The ma in argument of this account is that time to process atypical items displaces the tim e to process typical items. Thus, more elaboration is given to the atypical items at the expense of typical items (Waddill & McDaniel, 1998). However, Bruce and Gaines (1976) disagree with this hypothesis in that it cannot be a necessar y condition even though it may be a sufficient condition for the von Restorff effect. S.R. Schmidt (1991) also argued that little ev idence supports that distinctive items somehow Â“stealÂ” atten tion surrounding common ite ms or Â“suppressÂ” recall of common items. Another account for differential processi ng is based on the expectation-violation hypothesis in which items with weak semantic associations are expect ed to have better recall than those with strong associations (Hirshman, 1988). This hypothesis assumes that a novel and unexpected stimulus may encourag e a more elaborate encoding because it leads more readily to a surprise response. So me researchers who tested the SC + T model agreed with this view that at ypical items may cause different attention to be paid to the stimuli to make sense out of the discrepancy which may be arousing a feeling of surprise or a mild Â“perturbationÂ” (Erdfelder & Bredenkamp, 1998; Graesser & Kassler, 1998; Warlaumont, 1997, Pezdek et al., 1989). Thus , atypical information may receive more elaborative processing or extra processing during the encoding process (Erdfelder & Bredenkamp, 1998; O'Sullivan & Durso, 1984; Wyer & Gordon, 1982).
42 A great deal of research has supported the differential-processing view in the assumption that schema-inconsistent stimuli elicited more elaborative processing (Hess, 1985; Johnston & Hawley, 1994; Sherman, Lee, Bessenoff, & Frost, 1998; Neuschatz et al., 2002; Pezdek, 1989). Neuschatz et al. (2002) maintained that atypical events are more memorable because they require more explan ations that lead to additional processing about the causes of such unusualness. Sh erman and his colleagues (1998) tested the differential-processing view of schema-relate d information processing by predicting that perceivers would attend more carefully to stereotype-inconsistent than stereotypeconsistent information when attentional reso urces were low. With a dependent measure of reading times for either stereotype-consis tent or stereotype-inc onsistent information about a target person, they found that subjects in a low capacity c ondition spent a longer amount of time reading inconsistent rather than consistent information. Pezdek and her colleagues (1989) also discussed this differe ntial-processing view in explaining their finding that items inconsistent with expectations were better recalled and recognized than items that were consistent with participan tsÂ’ expectations: Â“The schema inconsistent items were thus processed for a longer pe riod of time, and more important, schema inconsistent items were encoded qualitatively differently than schema consistent itemsÂ” (Pezdek et al., 1989, p.593). However, this differential-processing view was rejected in several studies (Erdfelder & Bredenkamp, 1998; Graesser et al., 1979; Graesser & Nakamura, 1982) that measured the typicality effect with varyi ng attentional resources. Graesser et al. (1979) found that variations in av ailable processing resources had no effect on recognition memory for both typical and atypical inform ation. Moreover, Graesser and associates
43 repeatedly argued that the assumption of deep er processing of atypi cal information could not account for the crossover in the recall curves for typical and atypical intervals presented in Figure 5 earlier in this dissertation. In other words, a common counterargument for these differential processi ng views was that if the deeper-processing view is correct, atypical information must be recalled better regardle ss of the retention intervals (Smith & Graesser, 1981). Representational views Representational views assume that ever y event leaves the properties in the memory record. In general, the distinctiv e stimuli produce more unusual memory records compared to other memory records, most of which reflect common experiences (Waddill & McDaniel, 1998). In other words, those memory records sta nding out from common memory records are intrinsically rooted in the stimuliÂ’s distinctive properties per se. According to Waddill and McDaniel (1998), ther e are two versions of representational views. The first version of a representati onal view, an Â“organizational hypothesis,Â” attributes better recall of distinctive item s to Â“an accessible isolation mnemonic unitÂ” (Bruce & Gaines, 1976). In other words, distin ctive items may be placed in a special unit containing only the distinctive features, wh ereas background items may be organized into larger units, decreasing the probability of retrieval of those items compared to the distinctive items (S. R. Schmidt, 1985). Thus , the distinctive items may enjoy a special privileged status at recall (S. R. Schm idt, 2002; Waddill & McDaniel, 1998). This hypothesis is closely related to the interference effect that re fers to Â“the impaired ability to remember an item when it is similar to ot her items stored in memoryÂ” (M. C. Anderson & Neely, 1996, p.237). In the interference accoun ts of retrieval processes, successful memory representation hinges not only on the strength of the associations between the
44 cue and the target event but also on other even ts related to the retrieval cue. When many competing events are linked to the same cue, this competition results in decreased memory performance, which is known as th e cue-overload principle (M. C. Anderson & Neely, 1996; Watkins & Watkins, 1975). This explanation is more appropriate for explaining the recognition memory than fo r explaining recall (Schmidt, 1991). As mentioned earlier, the SC + T model is base d on the assumption of the representational views of distinctiveness. Gr aesser and his colleagues (1979; D. A. Smith & Graesser, 1981) state that the model is explained well with this representational model in two reasons. First, D. A. Smith & Graesser (1981) argued that the in terruption condition in their experiment that was expected to steal cognitive resources did not affect discriminative accuracy for either typical or atypical actions in prose. Second, the reverse effect of typicality in a long-te rm retention interval that the SC + T model predicts is also related to the representational explanation of distinctiveness. In other words, Â“the retrievability will decline faster for atypi cal actions because they are not strongly associated with the organizational sche maÂ” (D. F. Schmidt & Sherman, 1984, p.19). Another version of representational vi ew can be called the Â“retrieval cue hypothesis.Â” The key idea of that hypothesis is that a person forms a set of retrieval cues when events are distinctive, which helps f acilitate the re call (Waddill & McDaniel, 1998; Wyer & Gordon, 1982). This retrieval cue hypot hesis emphasizes the functional value of the distinctive item itself of which a dvantage can be observed only when the distinctiveness is used to guide the recall process (S. R. Schmidt, 2002). However, the differences in the two versions of representa tional views are not so large in that both
45 perspectives can be summarized in terms of Â“trace distinctivenessÂ” (Eysenck, 1979; Fabiani & Donchin, 1995). An emerging question is what kinds of experimental manipulations can be employed to test the representa tional view that Â“the advantag e of atypical items is due to a memory representation that is organized into distinctive and nondi stinctive categoriesÂ” (Waddill & McDaniel, 1998, p.113). Unlike the diverse methods to manipulate the depth or levels of processing, the methods to redu ce the trace distinctiven ess are relatively few and somewhat complex. One of the popular me thods to demonstrate the organizational nature of the schematic memory representa tion is to examine clustering during recall. This method is developed to show the relationship between items in a common conceptual category (Bousfield & Bousfield, 1966; Jenkins, Mink, & Russell, 1958). This clustering method has been employed to char acterize the schematic organization of an event (Rabinowitz & Mandler, 1983) or to dem onstrate the organizational nature (i.e., organized in a separately isolated mnemonic unit) of distinctive items (Bruce & Gaines, 1976; Fabiani & Donchin, 1995). For inst ance, Rabinowitz and Mandler (1983) instructed participants to recall the arguments of stimulus materials either in any order or exactly in the same way as they were heard from a tape. They found that subjects in the schematically blocked condition clustered thei r recall almost exclusively into schematic categorization in which items have verti cal connections to higher levels of nodes24 which consisted of Â‘themesÂ’ or Â‘scenes.Â’ For exampl e, the Â“going to a partyÂ” scene consisted of several subordinate schematic actions (e.g., buy present, go to part y, put on paper, eat birthday cake, etc.). 24 Higher levels of nodes are close to Â“Higher levels in structural hierarchyÂ” in ThorndykeÂ’s (1977) study (see p.19 in this dissertation)
46 Fabiani and Donchin (1995) tried to dete rmine whether the memory for distinctive items was stochastically independent from that of the other items. As a way to demonstrate it, they examined the positions in which the distinctive items were written during the free recall test. The underlying assump tion was that distinct ive items would be more likely to be written either at the beginning or at the e nd of the recall sheet if they were organized separately from the others. In addition, they also expected that this pattern would be more evident in the case of semantically distinctive words because they belonged to a different cate gory than did non-distinctive words. Results supported the prediction that distinctive items would be separately organi zed from the other words. A drawback of this clustering method is that it cannot be applied to recognition memory test. An alternative method to demonstrate the separate memory organization for the atypical information is to reduce the trace dist inctiveness of the atypical information. To determine whether a main effect of better recognition memory for rare words (atypical information) over common words (typical information) could be modified under preexposure to the rare words prior to th e study session, Kinsbourne and George (1974) compared the recognition scores between expe rimental treatment (i.e., preexposure) and the control condition (i.e., no preexposure). Results showed that the preexposure had a significant effect on recogniti on score for rare words, but not on common words. While there is a possibility that this measure also can influe nce the depth of processing, Eysenck (1979) reasons two possibilities for the eff ect of preexposure on memory for atypical information: 1) the atypical information has a much more unusual frequency of occurrence in everyday life compared to the typical information; and 2) atypical information has little proactive interference25 from pre-experimental encodings compared 25 Proactive interference refers to Â“previously encoded materials hurting our memory for more recently
47 to typical information that has frequent en codings even in the absence of a preexposure encoding condition. In his experiments, some subjects were given preexposure to to-belearned words prior to the learning task for ostensibly unrelated purposes. However, the manipulation of the trace dist inctiveness did not influen ce the effects of depth of processing and word frequency on recognition. Although Waddill and McDa niel (1998), who thoroughly compared each competing hypothesis, are seemingly in support of the privileged stat us of retrieval of distinctive items over differen tial processing, the pure repres entational views have also been challenged by counter-evidence. First, some studies reported findings in which certain types of physiological responses (e.g., the event relate d brain potentials, or ERPs, in Fabiani & Donchin, 1995; heart-rate data in Schmidt, 2002) influenced the differences in encoding of distinctive vs . non-distinctive events, which may imply surprise responses or selective displaced rehearsal. Second, Fabiani and D onchin (1995) had subject biased their attention toward processi ng specific distinctive featur es by manipulations of the orienting tasks (semantic/lexical vs. physic al decision tasks). T hus, the interaction between the orienting task and the distinctiv eness (i.e., the type of isolation) helped clarify the extent to which encoding processe s accompanying distinctive events can be influenced by attentional resources. Results showed that orienting tasks (attentional bias) played a critical role in determining subjectsÂ’ responses to the semantic isolates. Third, a recent study revealed that patients with cl osed-head injury (CHI) showed impaired recognition performance for atypical actions compared with people in a control group learned items. For example, we suffer from proactive interference when we fail to recall our new phone number momentarily because our old number intrudes during recall processÂ” (M. C. Anderson & Neely, 1996, p.246)
48 (Vakil, Sharot, Markowitz, Aberbusch & Gr oswasser, 2003). Vakil et al. (2003) argued that if atypical actions, like typical actions, were proce ssed automatically, recognition performance of CHI patients should not di ffer from the control group. Thus, they concluded: Therefore, the present findings that CHI patientsÂ’ memory of atypical actions was primarily impaired is more readily expl ained by theories assuming that atypical actions are mediated by effortful processe s than automatic processes. Similarly, previous studies have demonstr ated that the typicality e ffect is affected by divided attention26 and by cognitive elaboration27 (p. 831). But a closer look at their data shows that the atypical actions were better discriminated than were typical actions am ong those CHI patients, while the CHI groupÂ’s memory discrimination for atypical actions wa s poorer than their counterparts in the control group. Therefore, it is most li kely that each competing hypothesis of representational views and diffe rential processing views is supplementary rather than complete in the explanation for the distinctiv eness effect. From the schema theory point of view, those two accounts for the distinctiv eness are not conflicting, because a schema guides both encoding and retrieval . If a distinctive item is observed from incoming information, a perceiver automatically detects th e distinctiveness that is intrinsic to the stimulus. Then, he or she may try to find th e cause of what makes the item distinctive, resulting in deeper cognitive processing. This present study could cont ribute to answering some unr esolved issues that were discussed so far. In genera l, the schema-copy-plus-tag model of schema theory is appropriate to explain the memo ry discriminability of statements derived from a source 26 MÃ¤ntylÃ¤ & BÃ¤ckman 27 Erdfelder & Bredenkamp, 1998
49 (e.g., a place or an actor), two of which are conceptually inte rrelated by virtue of generic schemata (Graesser et al., 1979). Thus, in traditional experiments testing the SC + T model, participants were first provided with a source (e.g., graduate assistantÂ’s office or an instructor); then, they were asked to disc riminate appropriate scenes for the schema of an office or a person. The SC + T model is contrary to the common-sense expectation about the human memory function, because it is difficult for indi viduals to presume that initially betterremembered items will be less available over time than initially less-remembered items due to faster decay for the former than for the latter. One may argue, Â“If this is true, why do so many advertisers attempt to position thei r product or service in an advertisement message that is atypical28 of conventional though ts related to the pa rticular product or service?Â” Indeed, if it is true that memory decays faster for atypical information than for typical information, then, there is no reason for persuasive communicators to develop an atypical message as an attempt to make a memorable message. This study can answer such question partly because the general pred iction of SC + T model is only effective when other factors are assumed to be controlled. Those fact ors include message repetition29, an individualÂ’s attention level an d/or oneÂ’s motivation to process the message. While the predictions made in the SC + T model are useful in explaining an individualÂ’s text processing in prose, few studies have atte mpted to apply this model to other real life materials (e.g., novels, movie tr ailers, and advertisem ents) that go beyond 28 A message is atypical when it sounds uncommon, rare, and/or unfamiliar. 29 Repetitious exposure to an atypical message before oneÂ’s memory decays may make the prediction of SC + T model void.
50 laboratory artifacts. In this regard, Graesser and his colleag ues (1998) tested the SC + T model with a real novel, and Smith and Sherman (1984) applied the model to memory for persuasive messages. A few researchers applie d this model to memory representation for medical information that was either familiar or unfamiliar (Brown & Park, 2002), or that was belief-affirming or be lief-disconfirming (Rice & Okun, 1994). Some marketing communication researchers have examined the ef fects of the level of congruency between a brand and its attributes as a product cat egory (schema) on consumersÂ’ memory (e.g., Sujan & Bettman, 1989; Mason & Lee, 1999; Stayman, Alden & Smith, 1992; MeyersLevy & Tybout, 1989; Rifon, Choi, Trimble, & Li ). To explicate the differential memory decay for persuasive messages by its typicalit y, relevant literature was reviewed in the subsequent section. Memory for Persuasive Messages: A Test of the SC + T Model When reading an opinion column from an op-ed page of the finest US newspaper, one perhaps feels that the column is very fa miliar as if he or she had read it before, although the title and examples are somewhat new: a feeling of dÃ©jÃ vu. Our subjective experiences or general knowledge gained fr om earlier exposure to messages give us a Â“feeling of familiarityÂ” to current messages though they are phenomenologically different from the earlier encounters. By repeated e xposure to persuasive speeches, messages and advertisements, a person also learns prototyp ical arguments with respect to objects or issues. Schmidt and Sherman (1984) stated, Â“arguments that are frequently encountered in persuasive messages on a particular issue may be represented by a kind of generic position schemaÂ” (p.18). For instance, regard less of their position on the gay marriage issue, an ordinary American may know wh at constitutes the typical arguments from
51 either side on the issue. Individuals living in America also may develop some archetypical examples of anti-abortion messa ges from repeated exposure to various speeches, sermons, essays, columns and media advertisements (Schmidt & Sherman, 1984). This feeling of familiarity makes our information processing faster and easier by playing a role of heuristics. On the other hand, it also makes recipients process and comprehend incoming messages within the boundaries of their prior knowledge and experiences, often resulting in a schema-driven memory error. It was Schmidt and Sherman (1984) who fi rst investigated individualsÂ’ schematic processing of persuasive arguments with a theoretical model. Criticizing the lack of consistency and theoretical power in explai ning schema-driven memory construction of persuasive messages, Schmidt and Sherman (1984) pointed out three problems in this area of research. First, previous studies have not fully sp ecified how different types of persuasive messages make a difference in remembering the messages. Second, little research has considered the time factor in the research desi gn. Third, it has not been examined in what process people retrieve unseen arguments that have not actually be en contained in the persuasive message. More than two decades after their seminal study, however, these three criticisms on the memory research for persuasive messages are still valid. Not only did few studies (e.g., Ford & Smith, 1991; H unt et al., 1992) repl icate Schmidt and ShermanÂ’s (1984) study, but also there ha s not been much further theoretical advancement since their study. Hunt et al. (1992) argued that unde rstanding how people represent persuasive information in memory is essential in persuasion research because this influences not only their immediate res ponse to persuasive communications but also
52 the residual cognitive frame they are lik ely to access over time. Considering the importance of message typicality in persuasive communication, this lack of interest in this line of research is lamentable. A reason for this lack of research in the typicality effect regarding memory may be attributed to a difficulty of conceptualizing typicality. In other words, the concept of message typicality is not only abstract and broa d, but also hard to de fine operationally. For instance, Â“atypicalÂ” in advertising is analogous to Â“unique,Â” Â“humorous,Â” or Â“distinctive,Â” and they are often used inte rchangeably. Typicality has also been used interchangeably with schema-congruency in many studies, even though the former is a term emphasizing a messageÂ’s generic char acteristics while the latter is a term emphasizing a recipientÂ’s perceptual judgmen t on the probability and likelihood of the messageÂ’s occurrence in the current contex t. However, both typi cality and schemaconsistency are apparently derived from th e same theoretical root, but are labeled differently. Indeed, Grasser and his colleague s used the term of Â“typicalÂ” and Â“atypicalÂ” interchangeably with Â“schema-consistent (o r congruent)Â” or Â“schema-inconsistent (or incongruent).Â” In addition, those researchers w ho applied schema theories to typicality (e.g., Loken & Wards, 1990) and schema c ongruity (e.g., Meyers-Levy & Tybout, 1989; Mandler, 1982) shared the same literature (i.e., Rosch, 1978; Rosch et al., 1976) in defining each term. Typicality judgment is gene rally regarded as subjectiv e because it is dependent upon an individualÂ’s knowledge, experiences, intelligence, and so on. Consider the following examples. The tabbed browsing is a great feature that is supported in most contemporary Web browsers but Microsoft.
53 The tabbed browsing is a great feature that is used by Homer Simpson in the latest episode of the Simpsons. These two arguments were produced duri ng the generation phase of the current studyÂ’s experimental stimuli for typical or atypical message that endorsed the Firefox Web Browser. The first argument was suggest ed as a typical one during the generation phase, while the second one was proposed as an atypical one. However, in a pretest with a separate pool of subjects (college junior s), the second sentence wa s rated more typical than atypical. To those who were grown up w ith television episodes of the Simpsons, the argument featuring a very familiar characte r may have sounded typical even though it is not likely that such an argu ment is common in describi ng the superiority of a Web browser. OneÂ’s typicality perceptions on an issue also could change as the issue evolves and debates begin to intensify. For instance, as the same-sex marriage debate rages on, a slippery-slope argument Â“Do you support polygamy? How about group marriage?Â” (Â“Debate over anti-gay marriage,Â” February 24, 2004) became the most frequently and widely used rhetorical qu estions among opponents of the same-sex marriage. Assume that a person who had no prior knowledge of th e history related to the same-sex marriage debate. The person may think, Â“Why did he (t he debater) talk about polygamy or group marriage when debating gay marriage?Â” But the slippery-slope argument may sound more Â“familiar,Â” Â“more banal,Â” and Â“more freq uent,Â” once a person is involved in the issue. Despite the aforementioned disadvantage in defining typicality, Loken and Ward (1990) greatly contributed to the conceptualiza tion of typicality. On th e basis of literature in cognitive psychology and cons umer research, they proposed four dimensions that
54 could help make a typicality measuremen t. According to Loken and Ward (1990), typicality refers to Â“the degree to which an item is perceived to represent a categoryÂ” (p.112). Four dimensions they identified as determinants of typical ity judgment were as follows: (1) whether an item shares or does not share attributes with other items, (2) familiarity with an itemÂ’s meaning, (3) freque ncy of exposure to the item, and (4) attitude toward the item, that is, the positive or nega tive value associated with the item. (p.112). Although Loken and WardÂ’s (1990) dimensi ons are conceptualized to explain product typicality, their dimensions can be a pplied to identifying what is typical or atypical in persuasive communication. An ar gumentÂ’s typicality can be measured by, (i) whether it shares or does not shar e thoughts, ideas, or themes with other arguments in a message, (ii) whether it sounds familiar or unfamiliar with regard to the issue in debate, (iii) whether it may be heard or read frequently from othe r sources with regard to the issue. The first dimension is buttressed by the von Restorff effect. The generic label for a distinctiveness effect stems from von Rest orffÂ’s experiments in which her subjects showed better memory for the syllable in th e isolation list (i.e., a syllable embedded in the numbers) than for the same item in the homogeneous list (Wallace, 1965). This von Restorff effect predicts that dissimilar or in congruent items standing out from a given set are better remembered than are the background items (Jacoby & Craik, 1979). Consider the following examples taken from an issu e ad projected by a co alition of anti-drug activist groups (the New York Times , Oct 10, 2005):
55 Â“Marijuana could threaten your teenÂ’s success.Â” (1) Marijuana can limit your t eenÂ’s academic achievement. (2) Marijuana can hinder a t eenÂ’s ability to learn. (3) Marijuana use is linked to poor grades. (4) Some frequent, long-term marijuan a users complain of taste change and lack of appetite. (5) Marijuana and underage drinking are linked to higher dropout rates One argument stands out from the message. In the arguments above, the fourth argument was changed from the original ar gument Â“Some frequent, long-term marijuana users show signs of lack of motivation.Â” Aside from its distinctiveness in syntax structure and in argument length, the argument Â“(4)Â” does not share the main idea with other arguments in the same message. As seen in this example, when a reader encounters a segment of text that does not share argument overlap with any of the current contents, effortful processes must be initiated (Kin tsch and van Dijk, 1978; OÂ’Brien & Meyers, 1985). Researchers found that the in crease in the use of effortfu l processes leaves a richer and more distinctive memory record for the initially unmatched item (Jacoby, Craik, and Begg, 1979; Jacoby & Craik, 1979; Anderson & Reder, 1979; OÂ’Brien & Meyers, 1985; Hunt et al., 1992). In other words, the rece iver confronted with an unexpected and distinctive persuasive argument may attempt to establish coherence that requires a more elaborative form of processing, which eventua lly leads to an enhanced memory for the distinctive argument (Hunt et al., 1992). In general, the distinctiveness effect unde rlies the whole mechanism of the SC + T model; therefore, it can be considered as th e theoretical underpinning of the SC+T model. According to S. R. Schmidt (1991), an event is distinctive if it is Â“incongruous with
56 active conceptual frameworks, or that contai ns salient features not present in active memoryÂ” (p.537). Familiarity, the second dimension of typicality judgment, has also been postulated as a key element of typi cality in numerous st udies (e.g., Mantwill, KÃ¶hnken, & Aschermann, 1995; Malt & Sm ith, 1982). Some researchers also corroborated a positive relationship between fr equency perceptions a nd typicality ratings (e.g., Ashcraft, 1978; Loken & Ward, 1990; Malt & Smith, 1982; Bourne, 1982). Little research has examined memory for t ypical or atypical pe rsuasive messages in terms of the SC + T model. In the most vali d application of the SC + T model to date, Schmidt and Sherman (1984) presented particip ants with a persuasive message in the form of an interview transcript of a hypothe tical candidate. The tran script contained both typical and atypical arguments regarding four controversial social issues: abortion, gun control, capital punishment and legalization of marijuana. After reading the interview transcript, participants were given a memory te st of either a recall or a recognition either immediately after their exposure to the stim uli or after a two-day delay. Results of Schmidt and ShermanÂ’s (1984) study provided support for the SC + T model. Recall was guided by schema, yielding less forgetting ov er time for typical than for atypical arguments. In addition, intrusions of unseen arguments and recognition of non-presented arguments were more frequent for typical than for atypical arguments. However, recognition memory were better for atypica l arguments both in an immediate and a delayed test, while the recognition accuracy dropped more rapidly over time for atypical arguments than for typical arguments. Ford and Smith (1991) applied the SC+T m odel to examine the effects of either typical or atypical messages for organ donati on on recognition accuracy. Contrary to their
57 prediction, results showed that the typical (i.e., one-sided) ar guments were more correctly recognized than the atypical (i .e., refutational two-sided) ar guments. However, a closer look at these studies with deviating results shows possible conceptual and methodological shortcomings. From the conceptual point of view, there is no theoretical ground that a refutational message is equivalent to an at ypical message. As noted, an argument sounds atypical when it is distinctive, not familiar, and rarely heard from the background theme that is discussed. Moreover, Ford and SmithÂ’s (1991) studies did not even consider the typicality ratings, which is a necessary proce dure to claim the validity of the typicality manipulations. Hunt et al., (1992) also test ed the SC + T model in a 2 (typical vs. atypical) X 2 (immediate vs. delayed) between-subject desi gn. Participants in Hunt et al.Â’s (1992) study were given a print advertisement embedded in a bogus magazine supposedly developed for college students. In the ad, a new ballpoint pen brand was promoted with a picture and an advertising message that was either ty pical or atypical of a normal ball-point pen ad. They failed in finding a significant main effect of typicality on recall. On the other hand, the recognition memory was affected by typicality; participants who received an atypical version of the ad scored significantly higher recognition than those who received a typical version both in an immediate and in a delayed test. The prediction that the atypical argument would decay faster than the typical one was rejecte d, with participants in the atypical version exhibiting increased r ecognition in a delayed test. However, there are also problems in Hunt et al.Â’s (1992) study. Even though their conceptualization and way of generating typical/atypi cal arguments are more reliab le than those of Ford and SmithÂ’s (1991), it is still argua ble as to whether the arguments used in their experiments
58 are valid. That is, no manipulation check was re ported regarding their examples for either the typical or atypical ad me ssage. In the typical ad versi on, the pen was represented as one that would Â“fit your handwriting,Â” while in the atypical ad condi tion it was depicted as one that would Â“actually improve your handwriting.Â” (Hunt et al., 1992, p.349). As discussed earlier, the two day delay is also insufficient for them to claim generality of their findings in the delayed test. The lack of cumulative research findings in the delayed effect of a persuasive messageÂ’s typicality on the recipientsÂ’ memo ry motivated the author of the current research to test the Schmidt and Sherma nÂ’s (1984) findings that correspond to the predictions of the SC + T model. Thus , the following hypotheses were proposed: H1a: participantsÂ’ recognition performance will be greater for atypical than for typical arguments in an immediate test. H1b: ParticipantsÂ’ recognition performance will be greater for atypical than for typical arguments in a delayed test. H2. There will be a main effect of times of measurement on recognition performance. In other words, participantsÂ’ recognition performance will decrease over time. H3. The decrease of recognition performance over time will be greater for atypical arguments than for typical arguments. Shapior and Fox (2002) found that particip ants tested immediately were more confident in their judgment about atypical items than typical items. However, as memory becomes less certain over time, people tend to be more conservative in their recognition judgment about atypical items than typical items, but more liberal about typical than atypical items (Shapiro & Fox, 2002). Thus , Hypotheses 4 was proposed to find an interaction of typicality and retention inte rval on participantsÂ’ confidence in their recognition judgments.
59 H4. There will be an interaction effect of typicality and retention interval on participantsÂ’ confidence about their recognition judgments. H4a. Participants tested immediately w ill be more confident in their recognition judgments about atypical hits than typical hits. Therefore, there will be more Â“SureÂ” responses for atypical hits than for typical hits at the immediate test. H4b. Participants tested at a delay wi ll be less confident in their recognition judgments about atypical hits than typical hits. Therefor e, there will be less Â“SureÂ” responses for atypical hits than for typical hits at the delay. The following Figure 6 is a pictorial summa ry of proposed hypotheses (H1 to H3). This pictorial summary was drawn based on a simulation, slightly changing data from Smith and Graesser (1981). Figure 6. Pictorial summary of Hypotheses 1-3
60 If memory for a persuasive message show s different patterns of decay over time by message typicality, it renders an important resear ch implication with regard to the sleeper effect. Previous studies that examined th e sleeper effect did not examine whether presenting either typical or atypical arguments following a discounting cue may enhance or reduce persuasion over time. However, memory superiority does not necessa rily lead to persuasion. It is possible that peculiar arguments may succeed in leavi ng a distinctive memory trace, but they may be less effective in persuasion than typical arguments. Attitude researchers have seldom examined the message typicality of attitude change. To address this issue, relevant literature is reviewed in the subsequent chapter.
61 CHAPTER 3 REVIEW OF THEORIES IN ATTITUDES The previous chapter discussed how comprehension and memory of persuasive campaign messages tends to be transformed and changed from the original ones. Attitude researchers have realized that schematic processing in the message comprehension of persuasive messages often result s in an individualÂ’s remember ing a gist, not the verbatim text of the persuasive messages (Tourangeau & Rasinski, 1988). Wh atever the precise form of an individualÂ’s genera l knowledge about an issue or an object, it influences oneÂ’s attitudes toward the issue or the object by assisting the individual in comprehending a persuasive message and by influencing th e representation of message in memory (Schmidt & Sherman, 1984). Although the importance of comprehens ion and memory representation of persuasive message has been emphasized ever since Hovland and his colleagues (Hovland & Weiss, 1951; Hovland et al., 1953), little scholarship ha s delved into how oneÂ’s schema-driven memory performance coul d make a difference in attitude change with the passage of time. Considering that the schema guides a recipientÂ’s encoding and remembering of persuasive arguments, a persua sive messageÂ’s typicality should play an important role in an audien ceÂ’s long-term memory repres entation of the persuasive message. Whether or not the memory advant age of an atypical message leads to the persuasive advantage is not know n. To answer this question, re lated theories in attitude literature were reviewed.
62 The Effect of Message Typicality on Persuasion Typicality, along with argument quality, is on e of the key factors that determine the effectiveness of persuasive communication messages since it influences oneÂ’s message comprehension and the memory representati on of the message. An emerging question is which one of the two will lead to more persua sion. If either leads to more persuasion than the other, will the effect persist ove r time, or will it be changed? Answering these questions is not easy b ecause it is hard to conclude whether a typical message is more positively evaluated than an atypical one, or vice versa, let alone the effect of message typicality on att itudes toward the object endorsed. Several researchers have argued that c ognitive responses to different levels of typicality could influence valence and extrem ity of affective responses (e.g., Mandler, 1982; MeyersLevey & Tybout, 1989). Mandler (1982) assumed th at attitudes toward attributes that are either typical or atypical of the original product category could be influenced by how smoothly the receiver processed those items. In general, items that are congruent with a category lead to favorable responses because pe ople tend to have more favorable attitudes toward objects that conform to their expecta tions and allow predictability (Meyers-Levey & Tybout, 1989). However, other researcher s propose an opposite scenario. MeyersLevey and Tybout (1989) explicated the opposit e hypothesis: Â“The novelty of the object increases arousal, and greater cognitive elaborat ion may occur in an effort to resolve the incongruity. [...] Moderate incongruities ar e regarded as Â“interesting and positively valued, thereby leading to more positive responses than ones elicited by schema congruity (p.40).Â” Meyers-Levy and Tybout (1989) also contra sted the effect of moderate item incongruity (or typicality) w ith extreme incongruity that they assumed generated more
63 negative evaluations because people were frustr ated when they failed in resolving such incongruities. There are at least three obs tacles to generalizin g those assumptions, especially when the typicality is related to persuasive messages rather than product attributes. First, it is more difficult to dis tinguish a moderately atypi cal argument from an extremely atypical argument than to separa te an atypical one from a typical one. Second, even though one could succeed in di stinguishing typicality of persuasive messages in three levels, he or she still need s to control favorability to the persuasive messages. If an extremely atypical argument is genetically evaluated as more negative than others (i.e., moderately atypical and typical argument), one cannot argue that the typicality effect comes from elaborative cognitiv e processing, but rather it comes from its negativity. Finally, researchers could agree with the assertion that moderate/extreme incongruity leads to more el aboration in cognitive processi ng, however, whether or not resolving discrepancy in atypical items leads to more favorable attitudes is inconclusive. Presumably, it depends on other factors such as prior attitudes, source credibility, involvement and so forth. Nonetheless, so me researchers assumed that an atypical message could induce more positive attitude ch ange than a typical message simply on the grounds that the atypical one enhances attenti on, which in turn leads to more elaborative processing. But it is simplistic to say that an atypical message has a persuasive advantage due to elaborative processing, because attitu de or decision making is not entirely a function of the amount of processing (e.g., Chattopadhyay & Alba, 1988; Chattopadhyay & Basu, 1990). Although there is re liable evidence that an at ypical argument leaves more of a distinctive memory representation than a typical argument, wh ether that advantage
64 leads to persuasion is quite doubtful. In summary, such a non-monotonic relationship between typicality and persuasion is not only di fficult to test but it is also contingent upon other factors. Affective evaluations of argument typicality Some scholars proposed the po ssibility that a typical argum ent is more persuasive because the typical message is perceived to be stronger in argument quality and is evaluated more favorably than the atypical me ssage. One needs to be sure that what is defined as an atypical argument is conceptually different from a silly or a childish argument. Previous research suggested more positive evaluations about the typical items than about the atypical ones (e.g., Ne dungadi & Hutchinson, 1985; Loken & Ward, 1990). Loken and Ward (1990) argued that t ypical items are more positively evaluated because: (1) they have more valued attributes and, (2) familiarity increases likeability. Particularly, the second reasoning regarding the familiarity effect could be buttressed by the findings of ZajoncÂ’s ( 1980) seminal study in which he demonstrated that mere repetitious exposure by Americans to some Chinese characters resulted in positive attitudes toward the objects. From a processing fluency point of view, typical arguments are also expected to induce more positive affective judgments. For example, Tvesky and Kahneman (1973) suggested that individuals often make judgmen ts based on Â“the ease with which instances or associations come to mindÂ” (p.208). A typi cal argument can be considered a stimulus that comes to mind readily. Lee and Labr oo (2004) called it conceptual fluency. Whittlesea (1993) reported a study in which he found that participants liked the stimuli (common words) as a result of enhanced conceptual fluenc y (presented in a predictive context rather than a neutral context) . Based on WhittleseaÂ’s study, Lee and Labroo
65 (2004) also examined the effects of concep tual fluency on affective judgments. They assumed that the meanings of stimuli are mo re easily grasped when the stimuli are more readily processed. To test their assump tion, they predicted that both perceptual1 and conceptual2 fluency led to more positive affectiv e evaluations than neutral ones in a replication of WhittleseaÂ’s (1993) experime nt. Results of their study showed more favorable attitudes for those words in a predic tive rather than a neutral context. Since a typical argument is more read ily and fluently processed th an an atypical argument, the former will be likely to lead to more favorable attitudes. If a typical message induces more positive affective evaluations than atypical one, how come so many advertising messages are using atypical concepts ? Indeed, such an assumption that a typical argument brings a bout more favorable evaluations than an atypical may be controversial, because atyp ical arguments in some cases could be evaluated positively. In some cases, an atypical argument results in more favorable evaluations, when: (1) the audience is so familiar with typical arguments that they consider those typical argument s as banal, and (2) the atypi cal argument is both creative and informative. In addition, as discussed ear lier, atypical arguments are at least believed to leave a more distinctive memory representation than typical arguments. When Madison AvenueÂ’s primary goal is to leav e any segment of a persuasive message on consumersÂ’ minds, then it makes sense that atyp ical concepts are used more than typical 1 To manipulate perceptual fluency, participants were first presented with a sentence, the last word of which was either the same or different from the target word that was presented immediat ely after the sentence for evaluation (e.g., Â”They spent three hours look ing for the houseÂ”: Â“houseÂ” or Â“drinkÂ”). 2 Conceptual fluency was manipulated by presenting a target word in a predictive context of the sentence using semantic association between the last word and preceding sentence. Fo r example, the preceding sentence in a predictive context was Â“The woman soaked the white sweater in some cold waterÂ” when the target word was Â“water.Â” For this target word, a neutral context was Â“The woman looked out of her window and saw the waterÂ” (Lee & Labroo, 2004, p.153)
66 ones in the advertising industry. To clarify this issue, two advertising campaign messages were provided as examples of either a t ypical or an atypical message. One of the following messages was adapted from Microsof tÂ’s so called Â“Get the FactsÂ” campaign. Microsoft launched the campaign against Linux, an open-source operating system, in an effort to check the spread of the public perceptions that Linux is more secure and more reliable. [Example 1] Â“Customers say that reliability is very important to them and that they are hearing that Linux and Unix are more reliable than Windows. But more and more independent analysts and leadin g companies find that Windows Server System outperforms Linux on reliability a nd security. Get the facts, and make the decision thatÂ’s right for youÂ” [Microsoft Get the Facts Home] With conceptual coherence and appropri ate logical sequences, the above message in [Example 1] is presumably a typical one that advocates the be tter performance of MicrosoftÂ’s server system. In terms of logi cal sequences, it first refuted the opponentÂ’s argument. Then this message reinforced Mi crosoftÂ’s position with seemingly the third partyÂ’s endorsement followed by some remarkab le cases that benefited from MicrosoftÂ’s operating system. Compare it w ith the he following message: [Example 2] Â“People told Columbus the world was flat. He didnÂ’t insist it was round. He got a boat. People will tell you Linux and Unix are more reliable than Windows. WeÂ’re not telling you to insist they are wrong. Get the boat, and make the decision thatÂ’s right for you.Â” This is a message adapted from an ad campaign for a network solution company ( The New York times , Oct. 4, 2005, C3). This message lacks conceptual coherence and logical sequences given that it is a message advocating a sever system. In other words, Â“ColumbusÂ” and Â“boatÂ” are least expected wo rds from oneÂ’s Â“Microsoft server systemÂ” schema. Presumably, the message shown in the second example will be rated as less
67 typical or atypical. However, there are severa l sophisticated rhetorical tactics in this message such as figurative comparisons, an innuendo3, and an indirect conclusion. Although some hypotheses regarding memory performance4 can be posited, it is hard to predict which one of those two messages will receive more positive message evaluations and subsequently lead to more persuasion. Typicality and Argument Quality One of the ways to compare a messageÂ’s pe rsuasiveness by typicality is to consider argument quality in the persuasive message. Ar gument quality refers to Â“a recipientÂ’s perception that a messageÂ’s arguments are st rong and cogent as opposed to weal and speciousÂ” (Eagly and Chaiken, 1993, p.311). Argu ment quality was a construct proposed by Petty and Cacioppo and their associates (Petty, Cacioppo, & Hessacker, 1981; Petty, Cacioppo, & Goldman, 1981) in the developm ent of Elaboration Likelihood Model (ELM), but has rarely been elaborated in li terature. An earlier distinction of a strong argument was "...logically sound, defensible and compelling," whereas a weak argument was described as being "...open to skepticis m and easy refutation" (Petty, Cacioppo, and Heesacker, 1981, p. 435). Another earlier work in the conceptualization of this term was the lack of scientific valid ity: "..the strong version of th e message provided persuasive evidence (statistics, data, etc.) in support of the exam [...] the weak version of the message relied more on quotations, persona l opinion and examples to support its 3 According to the Oxford Dictionary, an innuendo is defined as Â“an allusive or oblique remark of hint, typically a suggestive or disparaging one.Â” Bell (1997) defined it as Â“a non-overt intentional negative ascription, whether true or false, usually in the form of an implicature, which is understood as a charge or accusation against what is, for the most part, a non-pr esent partyÂ” (p.36). According to Wegner and his colleagues, an innuendo is consisted of a statemen t and a qualifier about th e statement (Wegner, 1984; Wegner et al., 1981). In the example, the sentence Â“W eÂ’re not telling you to insist they are wrongÂ” is an innuendo that is composed of a st atement and a qualifier for denial. 4 For example, recognition accuracy or response late ncy (Perfetti Beverly, Bell, Rodgers & Faux, 1987).
68 position...the strong arguments were selected from a pool that elicited primarily favorable thoughts in a pretest, and the weak counter arguments in a pretes t" (Petty, Cacioppo, & Goldman, p. 850). A few consumer researcher s (Areni & Lutz, 1988; Boller, Swash, & Much, 1990) challenged the lack of construct va lidity in defining this term, but there is still a tendency that this construct is us ed with a common-sense approach without identifying the underlying dimensions on whic h arguments may differ in "quality." Based on those definitions, Petty, Cacioppo, and Schum ann (1983) reported that they pre-tested various arguments for disposal razors with rega rd to their Â“potency.Â” Then they reported, Â“In the strong arguments ad, the razor was characterized as Â“scientifically designed,Â” while the same razor was characterized as Â“d esigned for beautyÂ” in the weak arguments adÂ” (p. 139). Five arguments from each version are as follows: The strong version of the Edge razor ad New advanced honing method creates unsurpassed sharpness. Special chemically formulated coating e liminates nicks and cuts and prevents rusting. Handle is tapered and ribbe d to prevent slipping. In direct comparison tests, the Edge blad e gave twice as many close shaves as its nearest competitors. Unique angle placement of the blade pr ovides the smoothest shave possible. The weak version of the Edge ad Floats in water with a minimum of rust. Comes in various sizes, shapes and colors. Designed with the bathroom in mind. In direct comparison tests, the Edge blad e gave no more nicks or cuts than its competition. Can only be used once but will be memorable. (Petty et al., 1983, p.139). Areni and Lutz (1988) criticized Pe tty, Cacioppo, and SchumannÂ’s way of conceptualization and manipula tion of argument quality:
69 Examination of these arguments as well as the arguments used in the editorials in Petty, et al. (1981) leads to the conclusi on that argument quality, as manipulated, may not be reflecting solely the logical as pects of a persuasive message. Rather, a considerable difference appears across conditions in the desirability of the arguments (i.e., attributes of the adver tised razor). For instance, getting the "smoothest shave possible" is almost certa inly seen as being quite desirable. It probably represents a basic objective of shaving. On the other hand, a razor that "can only be used once" is rather undesirab le for most individuals. In fact, the two sets of attributes appeari ng in the two ads are so di sparate in terms of their desirability that a subject seeing both ads might easily imagine that the ads are for two completely different products, one whic h is very good and one which is not. As a result, interpretation of past results associated with the argument quality construct must be tempered somewhat due to the nebulous nature of the construct, as conceptualized and operationali zed. (Areni & Lutz, 1988, p.198). As seen in this observation, Petty an d Cacioppo and their associatesÂ’ strong arguments were often associated with positive attributes of the objects while their weak arguments were associated with less positive or negative attributes (Eagly & Chaiken, 1993). Thus, Areni and Lutz raised a possibility that this construct in previous studies of the ELM might be a measure of argument vale nce rather than argum ent strength. In fact, when Areni and Lutz tested this possibility in an experiment with previously tested arguments in Petty et al. (1983) and Pe tty, Cacioppo, and Goldman (1981), they found that the set of Â“strongÂ” arguments ascribed more desirable attribut es to the attitude objects, but there was no difference in subj ectsÂ’ rating of perceived strength (i.e., likelihood) between strong and weak arguments. Nonetheless, the construct of argument quality has widely been used in persua sion research according to Petty, Cacioppo and their associatesÂ’ definition. For example, Zuwerink and Devine (1996) defined argument quality primarily in terms of "how worthy th e arguments were of serious consideration and how difficult the arguments were to c ounterargue." They also revealed that, to maximize differences in message quality, th ey used "better in itial arguments and
70 references to scientific studies in th e strong message but weaker arguments and references to opinion in the weak one." (p.938). In spite of the difficulty in judging argume nt quality for either typical or atypical arguments, there are several reasons that typi cal arguments are believed to be stronger and to be more convincing than atypical ones in terms of argument quality: (1) A typical argument is less refutable than an atypical argument. Since a typica l argument is more frequently used for the purpose of persuasion, it is likely that the ty pical arguments have been challenged more than the atypical one s, making the typical argument resistant to counterarguments; (2) A typical argument can be comprehended more easily and faster5 than an atypical argument (Loken & Ward, 1990), since it is logically sequenced and conceptually coherent with a generic schema; (3) Typical arguments sound more logical and valid than atypical arguments since th e former complies with common sense or ordinary peopleÂ’s expectancy. For example, in a debate about gun control, a typical argument that reads Â“Gun cont rol decreases the av ailability of guns to criminalsÂ” may sound more valid than an atypical argument th at reads Â“Gun control decreases the killing of wildlifeÂ” (Schmidt & Sherman, 1984). To elucidate further, compare the following examples of typical and atypi cal arguments that were used in the previous studies. [Example 3: Typical] Capital punishment is irreversible if new evidence later shows the person to be innocent, it is too late (Schmidt & Sherman, 1984). [Example 4: Atypical] Capital punishment increases the stress on judges and jurors since they are deciding on someoneÂ’s life (Schmidt & Sherman, 1984). [Example 5: Typical] Marijuana could threaten your teenÂ’s success (PARENTS campaign, the New York Times , October 10, 2005). 5 Both conceptual and perceptual fluency lead to more positive attitude change (Lee & Labro, 2004).
71 [Example 6: Atypical] Marijuana impairs your teenÂ’s vision. (Schmidt & Sherman, 1984) Presumably, typical arguments in both cas es are stronger than atypical arguments in argument quality. An atypical argument al so might be perceive d as less strong in argument quality than a typical argument in that the former often employs some inappropriate rhetorical tactics such as Â“exa ggeration,Â” Â“incorrect an alogy,Â” Â“the slippery slope,Â” Â“inappropriate simile or metaphor,Â” a nd so forth. For example, this argument is proposed to ridicule the Supr eme CourtÂ’s ruling that peer-t o-peer (P2P) companies like Grokster could be held responsible for copyright infringement. [Example 7] Â“Should all gun, bullet, and kni fe manufacturers be held responsible for all law breaking activities perp etrated by their products?Â” What if that argument were used to defe nd GroksterÂ’s copyright infringement case in the Supreme Court? Such an argument is probably the least lik ely to be heard in the Supreme Court. Compare the atypi cal argument with a typical one: [Example 8] Â“P2P technology should be protected under copyri ght law as long as they are capable of substa ntial noninfringing uses.Â” The atypical arguments above are likely to be evaluated as weaker arguments than the typical ones, because it lacks conceptual coherence and/or is not in a logical sequence. However, whether the typical argument has a stronger effect on attitude change than the atypical argument is hard to tell. At least, two other variables may influence the affective evaluations due to message typica lity: (1) Who said th at? Â– that is, an interaction of source credibility and me ssage typicality; and (2) Will the message typicality effect on persuasion be persistent over time? For instance, think about the aforem entioned [Example 7] -Â“Should all gun, bullet, and knife manufacturers be held responsible for all la w breaking activities perpetrated by their products?Â” If that argument was made by a leading authority in
72 Internet copyright law who was a professor at Stanford University then the atypical argument may be viewed as a stronger ar gument than if it was said by a high-school student who enjoys downloading music fr om a peer-to-peer (P2P) service. Despite the importance of source-message congruence, it is not known whether the effects of message typicality on attitude s will be moderated by source credibility perceptions. In an effort to e xplore this issue, the following questions are raised in this present study. (1) How will message typicality influence th e effect size of the sleeper effect? (2) If an interaction occurs, will the inte raction of source credibility and message typicality on persuasion be stable over time or will it show a different pattern over time? To answer these research questions, rele vant literature was reviewed in the following chapters. The expectation that message typicality will interact with source credibility perception elicited a question of how the interaction will affect an individualÂ’s attitude toward the object endorsed. A few resear chers (e.g., Bohner & Insko, 1966; Sternthal, Phillips, & Dholakia, 1978; Ku mkale & AlbarracÃn, 2004) consid ered the joint effects of source credibility and messages on persuasion. Among various types of the joint effects are source credibility and message discrepancy from a recipientÂ’s initial opinions or attitudes (Bohner & Insko, 1966; Kumkale & AlbarracÃn, 2004) and source-message (in)congruity. While the effect of the messageÂ’s discrepancy from a recipientÂ’s prior attitudes has often been studied, little re search has attempted to explain how the messageÂ’s discrepancy from a recipientÂ’ s source schemas interacts with source credibility. Those two areas ar e fundamentally different, because the former studies the
73 messageÂ’s congruency with a recipientÂ’s prio r attitudes or opinions. On the other hand, the latter investigates the me ssageÂ’s congruency with the r ecipientÂ’s prior source schemas in which the message does not violate oneÂ’s kno wledge or experiences about the source. In exploring this issue, an emphasis is gi ven to whether the inte raction effect will persist over time or whether it will show a di fferent pattern over a passage of time. How the source credibility -message congruency6 plays a role in the persistence of attitude change is hard to answer because there is no single study available in U.S. refereed journals to date that has di rectly addressed this questi on. To answer this research question, it is necessary to review the sleeper effect as well as the source-message congruency effect on the persistence of attitude change. The Effect of Source Credib ility on Attitude Change Ethos and Source Credibility The importance of source credibility in pe rsuasion has been emphasized as early as during the time of Aristo tleÂ’s (trans. 2003) essay on Rhetoric . Strictly, the concept described in Aristotle's Rhetoric was not source credibility but ethos , but both concepts were often equated in (speech) communica tion research (Logue & Miller, 2005). In Aristotle's exposition, ethos referred principally to a speak er's character exhibited in speech (Logue & Miller, 1995). Aristotle (trans . 2003) identified a speakerÂ’s character as one of three7 means of effective persuasion. Of th e source character, Aristotle said, Persuasion is achieved by the speaker's pe rsonal character when the speech is so spoken as to make us think him credible. We believe good men more fully and more readily than others: this is true generally whatever the question is, and 6 A high-credibility source is expected to make a cogent or a typical message, while a low-credibility source is expected to a specious or a less typical message. 7 Other two means are logical reasoning, and a consideration of the audienceÂ’s emotions that may put them in a certain mind.
74 absolutely true where exact certainty is impossible and opinions are divided. This kind of persuasion, like othe rs, should be achieved by wh at the speaker says, not by what people think of his character be fore he begins to speak (pp.10-11). Inch and Warnick (2002) thus argued that a speakerÂ’s credibility does not result from intrinsic characteristics of the speaker, but it results from the recipientÂ’s impressions and beliefs about the source when the speech is spoken so as to make us think the source is credible. In this view of context-depende nt source credibility, it is assumed that the perceptions about the credibility of an argue r may increase or decrease even within a speech or during the reception of a message, depending on how recipients perceive and react to the points made by the arguer (Inch and Warnic k, 2002). Brooks and Scheidel (1968) empirically demonstrated this cont ext-driven credibil ity judgment in the comparison of an experimental group whose attitudes toward the videotaped speaker (Malcolm X) were measured in a few break sessions and two control groups whose attitudes toward Malcolm X were measured e ither without a message or after presenting an uninterrupted spee ch (see Figure 7).
75 2.4 2.5 2.6 2.7 2.8 2.9 3 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 4 4.1 4.2 4.3 4.4 4.5 4.6Pretest Time1 Time2 Time3 Time4 Time5 Time6 Time7 Time8 Posttes t Exprimental Control A Control B Linear (Control A) Linear (Control B) Figure 7. Evaluative responses (on a multiple 8point semantic differential scale) toward the speaker over time intervals Analogous to AristotleÂ’s view that credibi lity Â“achieved by what the speaker saysÂ” might be attributed to Â“what people think of his character before he begins to speak,Â” a message recipientÂ’s credibil ity judgment about a speaker could change with the interaction of an audience (Inch & Warnic k, 2002). Inch and Warn ick (2002) called the latter form initial credibility, and called the former derived credibility . According to Inch and Warnick (2002), Â“initial cr edibility is based on an argue rÂ’s credentials, status, and reputation as known to recipients before they hear or read the messageÂ” (p.82). On the other hand, derived credibility Â“results from what is said in the message Â– the quality of
76 the claims and evidence used and the ways arguers employ their own expertise to get their claims acceptedÂ” (p.82). Factors Influencing Message RecipientsÂ’ Credibility Judgment About a Source Compared to ethos, source credibility is a relatively newer concept that largely was formulated after World War II by Carl Hovl and and associates. However, nowadays it seems that both concepts undoubtedly overl ap (Logue & Miller, 1995). While arguing that the two concepts cannot simply be equa ted, Logue and Miller (1995) attributed the merger trend to the Hovland school's cont ribution of giving operational meaning and measurable content to what Aristotle and others called ethos . Indeed, several researchers (e.g., Anderson & Clevenger, 1963; Baudhuin & Davis, 1972; Schweitzer, 1970) in speech communication measured the ethos in the same way that Hovland and his associates did. Hovland and his colleagues (Hovland & We iss, 1951; Hovland, Janis, & Kelley, 1953; Kelman & Hovland, 1953) made a las ting contribution to source credibility research by uncovering the factor s that made a communicator credible to the audience. Generally, a communicatorÂ’s credibility refers to the communicatorÂ’s perceived ability or motivation to provide accurate and truthful information (Kelman & Hovland, 1953). In order to be evaluated as hi ghly credible, a persuasive communicator should not only possess comprehensive knowledge or a skill in a particular ar ea but also (s)he has to be trustworthy. There have been quite a few factor-analytic studies (see OÂ’Keefe, 2002 for the list ) to examine what dimensions constitute a personÂ’s credibility judgments since Hovland and his colleagues initia lly conceptualized the concept with two factors Â– that is, a communicatorÂ’s expertise a nd trustworthiness. Even t hough findings obtained from those factor-analytic reviews have different names for dimensions, most of them can be
77 converged into two broad dimensions that Hovland and associates earlier defined ; a source of a persuasive message can be perceived as credible if he or she is expert and/or trustworthy (Lashbrook et al., 1977; Sternthal, Phillips, & Dholakia, 1978; Tormala & Petty, 2004; OÂ’Keefe, 2002; Ols on et al., 1984l; Ohanian, 1990). The expertise factor consists of scale items such as e xpert, qualified, experienced (Applbaum & Anatol, 1973; Wu & Sh affer, 1987; Harmon & Coney,1982) and knowledgeable (Wu & Shaffer, 1987). Source of the persuasive communication is considered as being trustworthy if he or she is unselfish, honest, fair or unbiased, believable and trustworthy (Applbaum & Anat ol,1973; Dean & Bisw as, 2001; Dholakia & Sternthal, 1977; Harmon & Coney,1982; La fferty & Goldsmith, 1999; Lashbrook et al., 1977; McGinnies et al., 1974; McGuire, 1969; Ohanian, 1990; Sternthal; Dholakia, & Leavitt, 1978; Wu & Shaffer, 1987). However, researchers often face two dilemmas in conceptualizing source credibility. The first dilemma is that these tw o factors of perceived credibility are often independent of each other. In other words, it is possible that a sour ce can be perceived to be an expert without creating the perception of trustworthiness, or , vice versa. Another dilemma is that it is not easy to experiment ally manipulate the construct considering both factors together. Since it is di fficult to do so, researchers ofte n consider either one of the other (e.g., Priester & Petty, 2003; Szybillo & Heslin; Ziegle r, Diehl, & Ruther, 2002; Petty, Cacioppo, & Goldman, 1981), depending on their research questions. Some researchers (e.g., Priester & Petty, 2003; Szybi llo & Heslin) prefer trustworthiness to expertise because it is the trustworthiness fact or that really matters in human relations. In the field of public relations, most problems in terms of source credibility are related to a
78 situation in which a source viol ates some ethical rules that regulate the prof ession Â– that is source trustworthiness (e.g., Elliott, 2005; Kirkpatrick, 2005). This does not mean that the expertise factor is less important than the trustworthiness factor. When it comes to a decision-making situation such as the purch ase of a highly-invol ved product (e.g., car, computer), expertise is more important than trustworthiness. When a researcher managed to manipulate either one of th e two factors, he/she should be wary of using the term of source credibility. For instance, some resear chers referred to Â“source trustworthinessÂ” (Priester and Petty, 2003) or Â“source expertiseÂ” (DeB ono & Harnish, 1998; Wood & Kallgreeen, 1988; Maddux & Rogers, 1980), when they only manipulated one dimension of source credibility. In persuasion literature, it is a well -accepted premise that a highly-credible source/endorser will induce more positive attitu dinal and behavioral change toward the position advocated than will a less-credibl e source (Tripp, Jensen, & Carlson, 1994; Wang, 2005). For instance, Wang (2005) compar ed the average ratings about movies endorsed by institutional sources and by We b users. Results from WangÂ’s study showed that participants perceived institutionsÂ’ averag e ratings as more credible than consumersÂ’ average ratings. There has been considerable su pport in the literature for the notion that a highly-credible source will have a positive message effect on peopleÂ’s evaluations about the endorsed product/brand (Atkin & Bl ock, 1983; Craig & McCann, 1978; Goldsmith, Lafferty, & Newell, 2000; Lafferty & Goldsmit h, 1999; Priester & Petty, 2003) compared with a low-credibility source. Thus, the following hypotheses were proposed. H5a. A highly credible endorser will indu ce more attitude change than a less credible endorser instantaneously.
79 H5b. A highly credible endorser will i nduce more attitude change than a less credible endorser after a two-week delay. However, literature (Eagly & Chaike n, 1993; Inch & Warnick, 2002; OÂ’Keefe, 2002; Pratkanis & Aronson, 1992) in persuasion also discusse s the strategy that less credible sources could take to enhance thei r persuasiveness. Referring to the importance of derived credibility , Inch and Warnick (2002) articulate that the challenge to enhance credibility of a less credible source coul d be achieved Â“through and by means of the message itselfÂ” (p.85). That is, to increase persuasion, a low-credib ility source should make either cogent arguments or speak agains t his/her self-interests. This approach to enhancing persuasiveness from a low-credibility source has received sporadic research attention (e.g., Walster, Aronson, & Abra hams, 1966; Eagly, Wood, & Chaiken, 1978; Ziegler, Diehl, Zigon, & Fett, 2004). An emerging question is whether or not the endorser credibility effect will prevail over the message typicality effect, with more attitude change from a highly-credibile source regardless of message typicality. Otherwise, will it make a different pattern in an interaction with message typica lity? In the previous chapter about the effect of message typicality on persuasion, it wa s argued that a typical messa ge would have a persuasive advantage over an atypical message, but it would be hard to conclude the message typicality effect without cons iderations for source credibil ity and the passage of time. Little scholarship has examined the role of me ssage typicality in attitude change let alone the interaction between typi cality and source credibility. Based on an assumption that match or mismatch8 between message typicality and source credibility may cause schema 8 A highly credible source is often expected to make a cogent or a typical argument. On the other hand, a recipient may be less surprised when he or she finds a less cogent message or an atypical message from a low-credibility source.
80 unexpectedness/incongruence in information processing, a schema-congruency effect was reviewed. The Schema-Congruency Effect Since we live in a media-saturated envi ronment in which we are constantly encountering ads, news, gossip and rumors from sources that vary widely in credibility, the issues regarding source memory in genera l and source credibility in particular have important social implications for ever yday life (Schacter, 1996). For instance, by remembering that the source of a story was a supermarket tabloid such as the National Inquirer and not the New York Times , one can evaluate the veridicality of the purported rumor (M. K. Johnson et al., 1993). Thus, a personÂ’s attitudes or judgments about an object are often rated or decide d by the match/mismatch of the object with the category in which the object is evaluated. A fundamental question in social cognition is how messages that are congruent or incongruent with the recipientÂ’s prior know ledge or experiences are represented in memory and how that cognitive representation affects attitudes toward the object (Devine & Ostrom, 1985; Judd, & Kulik,1980). Research about the role of s ource perceptions in accepting a message can be traced back to as early as Lorge (1936) and Asch (1948). In the classic study of source prestige, Lorge (1936) demonstrated that peopleÂ’s ratings of agreement with the quote Â“Those who hold and those who are without property have ever formed two distinct classes" changed as th e source changed from Karl Marx to Thomas Jefferson. Similarly, a statement Â“Service to th e people is more valuable than itselfÂ” might be represented differently if attributed to Mother Theresa versus Saddam Hussein (Slater & Rouner, 1997, 975).
81 As such, a quote or an argument can l ead to entirely different evaluations, depending on the source to which it is attrib uted. In fact, each side advocating different views about a controversial issue often adopts a quote th at is tightly tied to the opponentÂ’s message. For instance, in the 1990s, the famous sentence from Martin Luther King, Jr.Â’s speech -Â“I have a dream, that my four little children will one day live in a nation where they will not be judged by the colo r of their skin, but by the content of their character.Â” -was adopted by the group least expected to us e it: opponents of affirmative action (McGlone, 2005). There is a similar example. The followi ng sentence comes from Dante AlighieriÂ’s The Divine Comedy . [Example 9] Â“The hottest places in hell ar e reserved for those who, in time of great moral crisis, maintain their neutrality.Â” The famous sentence was previously quot ed by JFK and Martin Luther King, Jr. However, it will generate entirely diffe rent cognitive responses and attitudinal evaluations if it is addressed in a religious right groupÂ’s message to buttress the following argument: [Example 10] Â“Our society should not s how contempt for those who draw political arguments from their faith and refuse to settle for a Godless middle ground in morality.Â” As seen in those examples, the same message could generate different cognitive responses and different evalua tions depending upon to whom the message is attributed. The schema-driven expectancy during oneÂ’s processing of an incoming message is dependent on his or her previous knowledge or experience about the messageÂ’s source or a source schema. A source schema refers to the conflux of semantic knowledge and episodic experiences that help identify the sour ce with the content of their messages. This
82 schematic judgment about the source of in formation often regulates how people process information from each source. The audience of any type of communication distinguishes the sources of information, such as speakers or media, based on source schemas, so long as the sources in general make meaningful st atements that fit their role and the social situation (Bayen et al., 2000). In this process, a message r ecipient continuously engages in a source-monitoring9 task so as to extract the real meaning from the message source. For instance, assume that a person read th e following atypical argument in a paper: Â“I would say same-sex couples should be happy that they are not allowed to get married. It is not a disadvantageÂ—itÂ’s a benefit.Â” Exposed to this atypical argument, one may try to figure out who is the source of this message . Whether the source of that argument is presented as being a top Vatican Cardinal or a radical feminist makes a significant difference in an audienceÂ’s message processi ng. In a similar context, a statement that Â“What this country needs is another Ronald Reag anÂ” will be heard as more ironic if it is made by a liberal Democrat than if it is made by a conservative Republican (Wyer, 2004). Indeed, this real-world example of sour ce-schema driven information processing was tested by Mather et al . (1999) who examined how individualsÂ’ schematic knowledge or experiences influenced their source-s chema congruent message processing when source-specifying features were less available. They provide d participants with several statements that fit the particular stereotypes for either a Republican or a Democrat. In two 9 Johnson and her colleagues (M. K. Johnson et al., 1993) have proposed a theoretical framework called source-monitoring framework (SMF) to integrate di verse source-memory-misattribution phenomena. Source monitoring is defined as Â“the set of processe s involved in making attributions about the origins of memories, knowledge, and beliefs (M. K. Johnson et al., 1993, p.3). In this theoretical framework, general knowledge about how particular sour ces may relate to a remembered event plays a critical part in the attribution of a source to the matched event (Mather et al., 1999).
83 separate videotapes, those statements were read by a woman who was labeled as either a Republican or a Democrat. Half of the stat ements read by the woman corresponded to a political label schema, while the other half did not. Two factors (i.e., age and focus of attention) were the main inte rests of their study. Results show ed that older adults were more likely than younger adults to make sche ma-consistent source misattributions when assigning spoken statements to sp eakers, i.e., to misattribute a liberal political statement to a Democrat rather than the Republican w ho actually said it. Aside from this agingrelated issue, their study demonstrated that relying on source schemas could lead to problems when specific qualitative informa tion describing the source did not match the source schemas. Researchers have argued that the extent to which a person cognitively responds to different levels of source-message incongruity can affect the personÂ’s attitudes toward the object discussed. In general, source and message are incongruent in two cases. First, a source can disconfirm an audien ceÂ’s expectation by addressing a message that is against his/her self-interest. This case has been scrutinized in terms of source attribution and source consistency (Walst er, Aronson, & Abrahams, 1966; Eagly, Wood, & Chaiken, 1978; Ziegler, Diehl, Zigon, & Fe tt, 2004). Some researchers inspired by attribution theory assumed that a message re cipient might infer different reasons why a communicator presents a certain position (Bohner & Wanke, 2002). It was known that more persuasion occurred when the advocat ed position disconfirmed the audienceÂ’s expectations even when it was proposed by a low-credibility source (O'Keefe, 2002; Pratkanis & Aronson, 1992, Walster et al., 1966). Eagly, Wood, and Chaiken (1978) found that a communicator who advocated a pr o-environmental policy was more likely to
84 affect recipientsÂ’ attitudes when the source was believed to be a pr o-business affiliation than when the source was depicted to be someone with pro-environment affiliations. A study by Walster et al. (1966) varied the persuasive messages with source credibility variations. The evid ence from Walster et al.Â’s ( 1966) research suggests that communicators could make themselves perceived as more credible by delivering messages that disconfirm audiencesÂ’ schematic expectation of co mmunicatorsÂ’ views. They presented participants in their expe riments with a newspaper clipping of an interview with Joe Â“the ShoulderÂ” Napolita no, who was identified as a serious criminal. In one experimental condition, Joe Â“the Shoulde rÂ” argued that criminal justice should be tougher. In another condition, participants heard his critic ism of the harshness of a judicial system and his suggestion for more le niency and less severe sentences. They ran a parallel set of experimental c onditions in which the same ar guments were attributed to a respected public official. Results demonstrated that when Joe Â“the ShoulderÂ” was arguing for stricter and severe court sentences, the low-credibility source was as effective as was the high credible source (i.e., the respected pub lic official). Walste r et al. (1966) thus argued that a communicatorÂ’s persuasiveness does not come entirely from source credibility, but rather comes from the inte raction between the source credibility and source schemas. They focused especially on recipientsÂ’ schema-expectancy of communicatorÂ’s self-interest in the effect of persuasive communications. That is, when the audience perceives the persuasive communications to be contrary to a communicatorÂ’s self-interest, it may override th e effect of low-credibility, resulting in attitude change even in the discounting cue setting. For instance, a female gossip columnist who was presented as a low-credib ility source in Hovland and WeissÂ’s (1951)
85 study may have been far more effective than the highly-credible c ounterpart if she was arguing against her own best in terest (Walster et al, 1966). The crucial point is that audiencesÂ’ knowledge of the background and se lf-interest of source (or source schemas) lead them to expect schema-congruent messages before they listen to an actual message. When the audience receives the contradictory communication that disconfirms their initial expectations of a low-cr edibility source, they may perceive the message as credible unless they are suspicious that some kind of pressure may have induced the lowcredibility source to ex press their message (Pratkanis & Aronson, 1992). Another case in which a lowcredibility source may genera te a positive response to his or her position is when the low-credibility source makes a cogent or a typical message with which many people would agree. This case has rarely been investigated in persuasion research. Source credibility refers to a message sourceÂ’s perceived ability or motivation to provide accurate and truthful information (e.g., Kelman & Hovland, 1953). As noted, source credibility is a sort of source schema in th at people generally expect to hear common, familiar, and frequently heard arguments from a highly-credible source. For instance, if the President of United States made uncommon, unfamiliar, and rare arguments to sell his/her policy positions, pe ople might be embarrassed. Therefore, the more credible a source is, the more he or sh e should reflect typical ideas and thoughts in a society. However, it does not mean that a hi ghly-credible source al ways makes a typical argument. Highly credible sources often come up with atypical arguments or uncommon ideas in the expectation of making their messa ge more memorable or in the consideration of audience levels of know ledge or understanding.
86 Although a typical message is more likely to induce attitude change, it is also likely that that an audienceÂ’s awareness of sour ce credibility would mode rate the effect of typicality on evaluations about an attitudinal object. For ex ample, an argument, Â“Should all gun, bullet, and knife manufact urers be held responsible fo r all law breaking activities perpetrated by their products?Â” could be less effective in pe rsuasion than a typical one, Â“P2P technology should be prot ected under copyright law as l ong as they are capable of substantial noninfringing us es.Â” However, to which source those arguments were attributed also makes a difference in pers uasion. If the atypical argument Â“should all gunÂ…Â” is attributed to a Stanford University professor and authority in cyber law, then it could receive a more positive evaluation than if it is attributed to a less credible source (e.g., a high-school student). A logical jump or weird comparison made by a Stanford law professor could be perceived as being funny or sarcastic, while the same logical fallacy could generate unfavorable evaluations when it is attributed to an anonymous high-school student on a Web bulletin. Despite many implications that the sourcemessage (in)congruency effect may give, its effect on persuasion has rarely been inve stigated. Devine and Ostrom (1985) observed that a jurorÂ’s source credibility perceptions provided a vital ke y to interpreting the validity of a witness's testimony. In their experiments that simulated the criminal court system, the evaluation of source credibility (no relationship with the defendant vs. the defendantÂ’s sister) and its a ssociation with testimony conten ts (inconsistent testimony) were critical components of the participantsÂ’ (i.e., juror) predeliberation task. There have been a few attempts in marketing research that examined the sour ce-message congruity
87 effect on persuasion (Kirmani & Shiv, 1998; Mandler, 1982; Meye rs-Levey & Tybout, 1989; Hunt & Kernan, 1984; Sujan, Bettman, & Sujan, 1986). Kirmani & Shiv (1989) examined the c onditions under which source-congruity influenced brand attitudes. Defining source congruity as the degree of match between accessible endorser associations and brand association, particip ants in one of Kirmani and ShivÂ’s (1989) experiments watched a rugged wa tch ad endorsed by a celebrity (i.e., Clint Eastwood). In the high-congruity condition, th e watch was positioned as Â“a watch for the rugged,Â” while in the low-congruity conditi on, the same watch was positioned as Â“a watch for the romantic.Â” To reinforce partic ipantsÂ’ perceptions about Clint EastwoodÂ’s characteristics, a photo in which he wa s pointing a gun was also shown to the participants. There was a significant interacti on effect of issue-rele vant elaboration and congruity: When participan tsÂ’ issue-relevant elabor ation was high (i.e., brand processing), brand attitude was more fa vorable under high than it was under low congruity. Under the low issue-relevant el aboration condition (i.e., ad processing), however, brand attitude was not significantly different acr oss the level of congruity. Similar results were found when the source-message congruity was manipulated by only varying the endorser (i.e., eith er Tom Hanks or Clint East wood) while the ad message (i.e., Â“a watch for the ruggedÂ”) was the same across the subjects. A slightly different version of source-messa ge congruity has been tested in a few experiments called Â“salesperson schemaÂ” (Sujan, Bettman, & Sujan, 1986; Hunt & Kernan, 1984; T. Stafford & M. Stafford, 2002) . In this salesperson schema research, Hunt and Kernan (1984) directly examined th e effects of expectancy disconfirmation on attitudes in the intera ction of source credibil ity and message typicality. Three groups of
88 subjects in Hunt and Kernan Â’s (1984) study were informed that they would see a clock radio ad in which the companyÂ’s spokesperson would talk about his experience with the new clock radio. Then, two groups of participan ts were given a booklet that contained the test ad for the clock radio. S ubjects in an expectancy-confirme d condition read an ad that made superior claims on all salient feat ures of the clock ra dio. On the other hand, counterparts in an expectancy-disconfirmed c ondition were given an ad that limited such claims to all but one product feature. To the subjects in the thir d group (an expectancy group), no ad was given, but they were aske d to respond to items designed to measure their pre-message expectancies. Results s howed that an ad in the expectancydisconfirmed condition resulted in more pe rsuasion than an ad in the expectancyconfirmed condition. A post-hoc analysis was do ne to examine whether or not enhanced credibility perceptions are positively re lated to persuasion. As expected, source credibility ratings influenced persuasion onl y in the expectancy-d isconfirmed group. In Sujan et al.Â’s (1986) study, participants re ad a description of either a computer or a clothing salesperson, the opening statement used by each salesperson, and the product arguments used by the salesperson to persua de the customers. In a schema-incongruent version, a product-oriented opening statement (i.e ., There is a particular computer IÂ’d like you to see) was presented to the computer salesperson, whereas a customer-oriented statement was (i.e., We have several suits that you can try on) presented to the clothing salesperson. In the schema-congruent versi on, a product-oriented statement was matched to a clothing salesperson, with a custom er-oriented statement being matched to a computer salesperson. Then, they had subj ects read six arguments presented by the salesperson advocating the product. All six of the arguments were either strong or weak
89 arguments. In other words, their research was deigned in a 2 (positive/negative salesperson schema) X 2 (an opening statem entÂ’s match/mismatch to the salesperson schema) X 2 (strong/weak arguments) between -subject design. They predicted that because more elaborative processing occu rs in the source-message incongruent conditions, product evaluations would be affected by argument quality only when the message was discrepant from the salespers on schema. In other words, in the schemaincongruent conditions, strong ar guments resulted in higher product evaluations than did weak arguments. Their assumption that any kind of schema-incongruency will make a recipient elaborate a message is reasonable . Can such an assumption be applied to a situation where oneÂ’s schematic expectation is violated due to an incongruency between source credibility and message typicality? I ndeed, people tend to expect a message to contain more valid arguments when it stems fr om a highly-credible source than from a less credible source (Chaiken & Maheswaran, 1994). Hence, a highlycredible source is expected to make more typical arguments in a well-organized message structure than a less-credible source. It would not be surpri sing if an atypical argument were related to a low-credibility source. However, the initial amount of attitude change induced by an atypical message from a highly-credible source will be smalle r than that by a typical message from the same source. When a highly-credible sour ce makes a typical (or strong) message, it should lead to the greatest attitude ch ange among all possible conditions. But the incongruency caused by an atypical message from a highly-credible message still motivates a reader to reduce the cognitive conflict, and thus to elaborate the message, which in turn induces attitude change that may persist over time. Then, one can expect
90 less decrease in the amount of attitude change with the passa ge of time when an atypical message is presented than when a typical message is presented by a highly-credible source10. Consider the aforementioned [Example 711] and [Example 812]. If both arguments were made by a a profe ssor at Stanford University who was a leading authority in Internet copyright law to defend the P2P service in the online music copyright infringement case, [Example 8] may induce more positive opinion change immediately after the message exposure. The ar gument in [Example 7], if it was told by the highly-credible source, could also induce a positive attitude change. In addition, as the schema-incongruency effect assumed, th e audience will elaborate this atypical message to figure out why the credible source argued in such a way. Thus, the audience of the message can better remember the source of the atypical message even after a lapse of time than he or she remembers the source of the typical message. In this case, an atypical message from a highly-credible sour ce is less effective in persuasion than a typical message from the same source in an immediate test, but attitude change induced by the former may be more persistent or less declined over time than attitude change by the latter. Thus presenting some atypical arguments could be an effective message strategy for a high-credibility source. A low-credibility source also could be used for strategic communication. Discussing the influence of spokesperson tr ustworthiness on message elaboration and 10 According to a sleeper effect, persuasion from a high credibility source decrease rapidly because of source-message dissociation or a more rapid d ecay of source information than a message. 11 Â“Should all gun, bullet, and knife manufacturers be held responsible for all law breaking activities perpetrated by their products?Â” 12 Â“P2P technology should be protected under copyrigh t law as long as they are capable of substantial noninfringing uses.Â”
91 attitude, Priester and Petty (2003) argued th e following with the s ub-heading Â“why to use untrustworthy endorsersÂ”: Given that the goal of advert ising is to influence a cons umerÂ’s behavior over a long period of time and in the face of counter-persuasion attempts by other advertisements, it is sensible for market ers to be interested in establishing thoughtful (i.e., elaborated) attitudes toward the product that are able to persist, resist, and guide behavior. And the use of an untrustworthy endorser might be a useful strategic tool in accomp lishing this very goal (p.419). The rationale for their argument is that messages delivered from a low-credibility source could generate more cognitive thoughts probably because people will be skeptical of messages from the low-credibility sour ce. But the use of an untrustworthy source without the considerat ion of any argumentation stra tegy may not produce persuasion because the low-credibility source may bias the processing of persuasive arguments. However, if an untrustworthy source ma kes some cogent arguments or if the untrustworthy source speaks against his/her self-interest, it may induce more positive cognitive responses, which, in turn, leads to persuasion. A less-credible source who makes typical arguments may induce attitude change in certain cases. First, a recipient may di sregard the discounting information when arguments in the typical message have mer it despite the expectations induced by the discounting cue (Kumkale and AlbarracÃn, 2004). The higher oneÂ’s motivation to process the message, the more likely the cogent message is to induce attitude change after a lapse of time. However, even in this case, the in itial attitude change could be suppressed or reduced on the awareness of the discounting cu e. But what will happen if the recipient forgets the discounting cue or if he/she cannot match persuasive communication with the source cue over time? Then, the cogent message will be awakened and may influence oneÂ’s opinion change after a delay.
92 Thus, another consideration to develop a strategic persuasive message for a lowcredibility source is a Â“timeÂ” factor. There is a possibility that the negative effect of low credibility can be minimized over time as a result of source-memory decay or desensitization to the impact of the discounting cue. An interaction between message typicality and source credibility in terms of attitude change and its persiste nce over time is important because in previous research it is known that a persuasive message projecte d by a low-credibility source has a delayed effect on attitude, which is generally known as the sleeper effect. The assumption of the sleeper effect is that information in a message and a source cue are dissociated or decayed differently with the passage of time. However, it is not known whether the interaction between time and source credibili ty could be affected in the presence of different levels of message typicality. To exp licate this process, related literature in the sleeper effect is reviewed in the subsequent chapter. The Sleeper Effect Theoretical Definition of Sleeper Effect The sleeper effect is a phenomenon in which a persuasive communication from a low-credibility source has great er persuasive impact after a lapse of time than it did immediately following exposure (Cialdini & Petty, 1981). The term Â“sleeperÂ” stems from an expectation that the long-term effect is larger than the short-term effect in some manner Â– that is, the effect is asleep but awak es to be effective later (Mike Allen & Stiff, 1998). Original findings. More than a half-century a go, Hovland and his colleagues (Hovland et al., 1949; Hovland & Weiss, 1951) demonstrated that i ndividuals who were suspicious of the source of persuasive me ssages tended to be less responsive to such
93 communications, resulting in an initial gap in the amount of opinion changes, depending on the credibility of sources. However, th e gap in the amount of opinion change was narrowed with a four-week time interval. In their studying of the effectiveness of Army propaganda films, Hovland, Lumsdaine, and Sheffield (1949) reported that opinion chan ge induced by the films was sometimes greater after a lapse of time than immediately after exposure. In a subsequent study, Holvand and Weiss (1951) varied the cr edibility of the message source with an identical persuasive communication message. Examining the amount of attitude change over time, they found that the initial opini on change was greater for a message from a high-credibility source than a message from a low-credibility source. However, contrary to conventional wisdom, there was no signi ficant difference in opinion change induced by varied credibility at three to four weeks after message exposure. Specifically, opinion change in the high credibility source condition showed a significant drop between time intervals, but there was a non-significant increase for the message from the low credibility source (Schulma n and Worrall. 1970). Forgetting hypothesis. The initial account for this effect was the source forgetting hypothesis in which Â“forgetting is the rule, bu t the source of information is more quickly forgotten than the material presentedÂ” (Hovland et al., 1949, p.196). The forgetting hypothesis (or the Yale model) pr edicts that Â“Attitude change by participants measured at the delayed post-test will be greater for the low credible message source than the high credible message sourceÂ” (Mike Allen & Stiff, 1989, p.414) Dissociation hypothesis. However, this initial h ypothesis was challenged by a contradictory finding (Hovland & Weiss, 1951) that even individuals who showed the
94 sleeper effect were able to recall the source, albeit with lower credibility, thereby failing to support their original in terpretation. Thus, Hovland and Weiss (1951) proposed an alternative explanation calle d the dissociation hypothesis. The dissociation hypothesis explains that attitude is initially influenced by both the source and the contents, but over time the contents are dissociated from their or iginal source, resulting in attitude change only by the original messages (Priester et al., 1999). In other words, in the long term, the credibility of the source is not relevant to th e final attitude, and the effect of credibility will diminish over time (Mike Allen & Stiff, 1989). Thus, in general, it is predicted that Â“Attitude change by participants measured at the delayed posttest will not be significantly different between a highly credible message source and a less credible message sourceÂ” (Mike Allen & Stiff, 1989, p.41 8). The dissociation model assumes that persuasion is a function of both the source a nd the content of the message. However, in this model, source credibility is considered to function only in the short term to overstate or understate the eventual impact of the message. Conditions for the sleeper effect. Nevertheless, the sleeper effect has often been criticized for the difficulty in replicating its original effect as Hovland and his colleagues showed (Cook et al., 1979; Gillig & Green wald, 1974). Amidst the confusion and controversy, researchers have turned their attention from the existence of the sleeper effect to the theoretical c onditions under which a sleeper effect may or may not be observed (Lariscy & Tinkham, 1999). Attempts to demonstrate the sleeper effect suggest that it is obtained only in a limited set of circumstances (Hannah & Sternthal, 1984). Gruder et al.Â’s (1978) study suggested that an absolute sleeper effect could be found when: 1) a persuasive message has a substantia l initial impact on att itude; 2) this change
95 is totally inhibited by a discounting cue; 3) the cue and messages are dissociated over time; and 4) the cue and message are dissociated so quickly that the message by itself still has some impact when dissociation occurs. Deferential decay hypothesis. However, even when these conditions were satisfied, Gruder and his colleagues (1978) suggest ed that the effect only occurred in the participants group that receive d the discounting cue after th e persuasive message. This finding that the sleeper effect failed rep eatedly when a low-credibility source was available as a discounting cue at the time of the message presentation was also confirmed in Pratkanis et al.Â’s (1988) elaborate study. When it comes to the effect of the timing-ofsource-information variation, thei r replication of 16 computeriz ed experiments resulted in no sleeper effect with the discounting-cue-be fore manipulation. The failure to obtain a sleeper effect using the source-cue-before procedure prodded researchers to propose alternative explanations for th e sleeper effect. Indeed, Pratka nis et al. (1988) proposed the differential decay hypothesis based on the pr imacy-recency analysis. They state, Â“a sleeper effect results when message and cue have near initial impact, but the impact of the cue decays more rapidly and independently than that of the message. According to the differential decay interpretation, a sleeper effect is not like ly to occur when discounting information precedes the message. In such cases, subjects may be more disposed to counterargue with the message as they read it. Thus, the persuasive impact of the message is attenuated, and the message and s ource are more likely to form a unit in memoryÂ” (Pratkanis et al., 1988, p.215). In other words, a perceiver in this condition may wonder, Â“why bother reading an article alrea dy proven to be false?Â” (Priester et al., 1999). Based on that finding, Petty and Caci oppo (1996) also suppor ted the differential
96 decay hypothesis arguing that Â“This result suggests that the assumption regarding the separate lives of the cue and message is , in many instances, incorrectÂ” (p.90). Sitton and Griffin (1980) provided a simila r explanation based on the interference effect and recency vs. primacy effect. When two types of information (e.g., a message and a source) enter a recipientÂ’s cognitive processes, two sources compete with each other, and new material is more likely to be accepted as accurate at the time it is presented, thus, a recency effect occurs. As time passes, interference theory says that a primacy effect will occur, with the original message recovering strength and becoming the dominant memory. Pratkanis et al. (1988) detailed what distinguishes the differential decay interpretation from previous dissociation hypotheses. They argued that the two accounts are similar in that both hypotheses claim the effect is based on forgetting and retention processes as opposed to other processes. However, the tw o accounts differ as to how information is organized during the encoding pha se and what information is retained over time. The dissociation hypothe sis assumes that message, discounting cue, and topic category are integrated or associated in memory during the encoding phase. As time passes, the source cue dissociates (or separate s) from the topic/message representation. Thus, it is expected in the dissociation hypothe sis that a delayed opinion change is based on the retrieval of message arguments acco mpanied by the failure to retrieve a dissociated discounting cue. In contrast, th e differential decay hypothesis assumes that the messages and source cues are poorly integr ated at the time of encoding (Pratkanis et al., 1988).
97 A distinct point proposed by the differential decay hypothesis compared to the traditional sleeper effect models (i.e., the dissociation hypothesis or the forgetting hypothesis) is that attitude changes between two groups will converge over time because the highly-credible message source group e xperiences greater decay in the amount of attitude change than does the low credible source group (Mike Allen & Stiff, 1989). Availability-valence hypothesis . In a slightly different theoretical assumption, Hannah and Sternthal (1984) also attempted to provide a reason for the failure of the dissociation hypothesis to predict the effect of timing-of-discountin g-cue presentation. They proposed the availability-valence hypot hesis to explain the increased message influence over time. According to this hypothe sis, Â“individualÂ’s attitudinal judgments in response to persuasive communi cations are determined by fa vorableness (or valence) of the issue-relevant information available in memory at the time of judgmentÂ” (Hannah & Sternthal, 1984, p.633). What is noteworthy in Hannah and SternthalÂ’s (1984) study is that it attempted to consider both the role of an active audience and the limited capacity of memory during the in formation processing. Thus, they a ssumed that an attitude change could occur when prior information held by a recipient is augmente d or supplanted by a currently processed message that is more or less favorable. In their research, both the persuasive communication and the discounting cue were considered as two separate pieces of information, as was the case in Si tton and GriffinÂ’s (1980) study. Thus, it is expected that the sleeper effect occurs wh en, with the passage of time, unfavorable associations evoked in responses to the di scounting cue are suppre ssed or supplanted by associations that are more favorable to message advocacy.
98 Some researchers, however, warn that this source-cue-b efore-presentation procedure needs to be considered as Â“additive13Â” instead of Â“necessary.Â” For example, Gruder et al. (1978) who proposed the order of discounting-cue-presentation stated that Â“a discounting cue following a message is not a sufficient condition for an absolute sleeper effect to occur, although this order of presentation may facilitate obtaining the effect when all the necessar y theoretical conditions are metÂ” (p.1073). In an empirical study, Lariscy and Tinkham (1999) found that a sleeper effect occurred in both the discounting-cue-before and af ter-the-message presentation, while the strongest effect occurred for subjects who were exposed to the discounting cue af ter the message (i.e., political attack ads). Thus, they concluded that the order of exposure could be additive in its effect, arguing that the sleeper effect was independent of message order: Â…low perceived credibility is not an esse ntial condition for the sleeper effect to occur, nor is a discounting cue followi ng the persuasive message a necessary condition for its occurrence. Â….By dismissi ng the order requirement as a necessary prerequisite the sleeper eff ect may be observed in more naturalistic settings and is not dependent on highly struct ured experimental protoc ols. (Lariscy & Tinkham, 1999, p.28). According to Kumkale and AlbarracÃn (2004) , 74 percent of studies that were analyzed in their meta-analytic review of th e sleeper effect presen ted the discounting cue before the message, while only 9 percent presented the cue after the message. The decision to present the discounting cue before or after the message is in part dependent upon research design and external validity. For in stance, in political advertising, it is not uncommon to present the sponsor after presenting the message. 13 The prediction that the sleeper effect will be most likely when the discounting cue is presented after the persuasive communication does not exclude the possib ility of the same effect in the source-before-themessage condition.
99 In summary, findings in prev ious studies identified two s lightly different types of sleeper effects and explicated several experimental conditi ons to observe a relatively stable sleeper effect. The Absolute Sleeper Effect An absolute sleeper effect (see B1 & B2 in Figure 8) refers to a significant increase in attitude change over time induced by a lowcredibility source, whic h is not observed in a control group (baseline) (Kumkale & AlbarracÃn, 2004; Gruder et al., 1978). Any attitudinal change in the contro l group must be smaller than that of participants exposed to a message from a low-credibility source (Kumkale & AlbarracÃn, 2004). The Relative Sleeper Effect The relative sleeper effect (see C1, C2, & C3 in Figure 8) refers to non-significant increase or less decrease in attitude over time among a low-credibility source compared to those who received messages from a hi ghly-credible source. It results from Â“significantly slower decay of the impact of a message accompanied by a discounting cue than of one accompanied by an accepting cue or no cue.Â” (Greenwald, Pratkanis, Leippe, & Baumgardner, 1986, p.217).
100 Figure 8. Persistence of persuasion. A: nonpersisting boomerang effect. B: absolute sleeper effect. C: relative sleeper effect. A Necessary Condition to Observe a Sleeper Effect To diagnose a sleeper effect, post-attit ude change induced by a low-credibility source should not fall below the post-attitude change among particip ants in a control group (baseline). If the post-attitude change induced by a low-credibility source regresses to the baseline, it is called the boomerang effect14 (see A in Figure 8 ) that is conceptually distinct from the sleeper effect (Kumka le & AlbarracÃn, 2004). Thus, hypothesis 6 was proposed to diagnose (1) if the treatment of a discounting cue i nduced post-attitude change, and (2) if the change induced by the discounting cue did not fall below the change observed in a control group. 14 A nonpersisting boomerang effect occurs when a me ssage initially backfires but later loses the adverse reaction, which is independent of message effect (Kumkale & AlbarracÃn, 2004)
101 H6. The post-attitudes observed in the lo w-credibility conditions will be greater than those in the control condi tion in the immediate tests. Some Experimental (But Not Necessary) Conditions to Observe the Sleeper Effect 1. The message itself must be persuasive e nough to make a significant initial impact on attitudes (Cook et al., 1979; Lariscy & Tinkham, 1999). 2. The discounting cue must be strong enough in its immediate impa ct to suppress the attitude change the original message w ould have otherwise produced (Cook et al., 1979; Lariscy & Tinkham, 1999). 3. There must be sufficient time prior to the delayed measure for the discounting cue and the original message to become dissociated (Cook et al., 1979; Lariscy & Tinkham, 1999). 4. The dissociation of the discounting cue and the original message must occur quickly enough that forgetting will not have occurred yet (Cook et al., 1979; Lariscy & Tinkham, 1999). 5. Subjects must rate the credibility of th e communicator immediately after receiving the discounting cue (Pratkanis, Greenwald & Leippe, 1988; Lariscy & Tinkham, 1999). 6. Participants should pay a ttention to the important arguments in the message (Pratkanis, Greenwald & Leippe, 1988; Lariscy & Tinkham, 1999). 7. The discounting cue should follow the message processing (Pratkanis, Greenwald & Leippe, 1988; Lariscy & Tinkham, 1999). Despite the relatively robust findings of the sleeper effect when some theoretical conditions were satisfied, assumptions found in the previous studies of the sleeper effect have a critical limitation in terms of ecological validity (Gru der et al., 1978; Mazursky & Schul, 1988). In other words, all three theoretic al accounts for the sleeper effect that were proposed in previous research assumed that the source cue would be either forgotten or dissociated or decayed. But the discounting cue may not be forgotte n, dissociated with the message, and/or decayed faster than the message cue. That is , the source cue can be remembered with the message even after a lapse of time if the sour ce spoke about somethi ng that violated the
102 recipientÂ’s expectancy, and thus left a very unique memory cu e. In this regard, Petty and Cacioppo (1996) criticized the assumption of the cue-message separation in previous studies of the sleeper effect: It was thought that the cues and message arguments have separate (noninteracting) effects on attitude change Â– that is, that cues did not affect the reception, retention, or yielding induced by the message argumen ts and the arguments did not alter the effects of the cues. Thus, attitude change could be determined at any point simply by adding the separate effects of the cue( s) and the message arguments (Petty & Cacioppo, 1996, p.90). If a less credible source delivers a cogent me ssage, this also can result in a schemacongruency effect. If the low-credibility source makes a typical message, the message will be elaborated. Nonetheless, the immedi ate opinion change can be suppressed in the awareness of the discounting cue. However, as time passes, the initial source information may decay or be dissociated from the typical message. If a source cue decays faster than a message, and thus only cogency of the message matters in the passage of time, then a typical message having been presented by a low-credibility source may induce persuasion over time. Thus, the sleeper effect could be observed when the message is typical. This hypothesis can be tested in terms of an inte raction effect between credibility and time under typical message conditions. On the other hand, an atypical message pr ojected by a low-credibility source will be discounted during the encoding stage, wh ereas the atypical me ssage from a highlycredible source will receive more attention a nd be elaborated. As discussed earlier, if a highly-credible source makes an atypical argument, the incongruency will motivate an audience to make an effort to figure out why the speaker ma de such an atypical argument, which results in message elaboration. For ex ample, an audience may think, Â“the speaker must have expressed his disappointment with the Court decision and ridiculed the logic
103 behind the decision using sarcasm.Â” Thus, th e audience of the message may experience less decay of source memory for an atypical me ssage from a highly-credible source than a typical message from a high-credibility s ource. Since a sleeper effect is maximized when the source cue and message are dissoci ated or poorly integrated, an atypical message by a highly-credible source will reduce the sleeper effect. Thus, under the atypical message condition, there will be no interaction between time and credibility. Instead, there will be main effects of time and credibility on attitude change. When the credibility of a source is high and the message is typical (or strong), it initially induces the greatest immediate at titude change among all possible conditions. When the source cue is available in the s hort-term memory, the cogent message may augment the persuasiveness of the high-credibility source. In summary, the sleeper effect will be mo re likely to occur when the message is typical than when it is atypical . Based on the review of theori es in attitudes research, the following hypotheses were proposed that descri be the roles of endor ser credibility and message typicality in attitude change at different times of measurement. H7a. There will be an interaction between credibility and time on attitudes toward Firefox under the typical message conditi on. More specifically, a typical message presented by a highly-credible endorser will lead to a significant decrease of attitude change over time, while a typical message presented by a low credible source will lead to an increase of attitude change over time. H7b. There will be a main effect of time on attitudes toward Firefox under the atypical message condition. H7c. There will be a main effect of credib ility on attitudes toward Firefox under the atypical message condition. In the hypotheses above, both H6a and H6b are necessary conditions for a sleeper effect. If attitude change observed in the base line is greater than that observed in a low-
104 credibility condition, it is de fined as a boomerang effect. The following Figure 9 is a pictorial summary of those hypotheses. Immediate-HighDelayed-HighImmediate-Low Delayed-Low Typical Atypical Figure 9. A pictorial summary of H5-H7
105 CHAPTER 4 METHODS Overview of the Experiment To explore this uncharted territory, a time-delayed experiment was conducted on the Web by manipulating message typicality and source credibility. In a 2 (endorser credibility: high v. low) X 2 (message t ypicality: typical v. atypical) X 2 (time: immediate v. delayed) between-subject desi gn, participants read a short biographical sketch about an endorser and his suppor ting message about a Web browser called Firefox, an open-source Web browser de veloped and produced by a non-profit organization called Mozilla Foundation. Th e supporting message for Firefox was presented by either a high-credibility e ndorser or a low-credib ility endorser with perceived typicality varied. In recruiting participants for this study a nd in the consent form, the purpose of this study was described as Â“to examine the di ffusion of innovation in the technological adoption of Weblogs (a.k.a, blogs) and Web br owsers.Â” Both the experimental and the control groups received a pre-test regard ing their attitudes toward Firefox and their interests in and concerns about the Internet security issue. Th en, half of th e participants who were randomly assigned to an immediate test group evaluated the attitudes toward Firefox and showed their behavioral intenti ons to use the endorsed product immediately after their reading the endor sement. This was followed by an unexpected recognition memory test for arguments that either had been presented or not had been presented.
106 The other half of subjects who were ra ndomly assigned to the delayed test group were not given any test right after their e xposure to the stimuli, and thanked for their participation. In the consent form, however, they were informed that some would be contacted for supplementary questions. After a two-week delay from their exposure to the endorsement message, those who were assigne d to the delayed group were invited to a follow-up survey by e-mail. Then, they were aske d to complete the same test that subjects in an immediate group had received. Design This study used a 2 (source credibility: hi gh versus low) X 2 (message typicality: typical versus atypical) X 2 (test: immediate versus delayed) betw een-subjects factorial design with two control groups in a differential test time (immediate and delayed). The study employed a true experimental desi gn, performed under Campbell and StanleyÂ’s (1963) suggestion for Â“testing for effects ex tended in timeÂ” (p. 31). Based on Hovland and his colleaguesÂ’ studies, Campbell and Stanley (1963) extended the pretest-posttest control group design, suggesting separate expe rimental and control groups for each time delay for the posttest as follows: Figure 10. The pretest-posttest control group design extended in time In accordance with the model in Figure 10, approximately the same number of participants were assigned to each control group in the immediate and the delayed test without receiving any manipula tion of source credibility an d message typicality. Subjects
107 in both control groups (immediate and dela yed control group) were also randomly selected from the same subject pool of other experimental groups. Stimuli: Typical versus Atypical Message In traditional research testing the SC + T model, subjects in a separate pool were requested to list either typical or atypical items or actions for specific places (e.g., room) or for particular contexts (e.g., lecture scen e). This process is called a generation phase. While this method has been widely employe d in cognitive psychology, it has a generic limitation; that is, it is difficult to construc t a message that requires a certain syntax structure with a story or logical flow of ideas. Some researchers (Shapiro & Fox, 2 002; Schmidt & Sherman, 1984), however, used this generation method to construct either typical or atypical stories or arguments. For example, participants in Shapiro and Fox (2002) were asked to list both typical and atypical sentences they might expect in nine different situations that could be seen in either a real or a fictional mass media presentation. One of the situations was Â“A man is held up at gunpoint on a city street.Â” One of the generated typical sentences was Â“Man pleads with gunman not to shoot him.Â” An ex ample of the generated atypical items was Â“Gunman gives man the gun.Â” Subjects in Schmidt and ShermanÂ’s (1984) st udy were instructed to list pro and con arguments concerning four social issues: gun co ntrol, abortion, legali zation of marijuana, and capital punishment. Another instruction given to subjects was to Â“include both arguments that might typically be found in an essay or speech (pro or con) concerning each topic and arguments they would expect to see or hear less frequentlyÂ” (Schmidt & Sherman, p.20). Arguments produced in Schmid t and Sherman (1984) then were used to construct an interview transcript for a fic titious political candidate. There are two other
108 limitations of using the generation method in developing a persuasive message: (1) An argument or a script generated by students in a laboratory often lack s external validity; (2) It is difficult to contrast a typical argument with an atypic al argument. In other words, to claim the effect of typicality, an atypical argument must be contrasted with a typical argument in the same context. In an af orementioned example of Shapiro and FoxÂ’s (2002) study, a typical description for the Â“ gunman on the streetÂ” sc ene was Â“Man tells gunman to take anything he wants,Â” while the atypical one was Â“Man congratulates gunman.Â” In this present study, the pool of argu ments for constructing the endorsement message for the Firefox Web browser were firs t collected from the previous endorsement messages that were found in a variety of s ources such as newspaper columns, magazine reviews, Firefox campaign Web sites, Web bu lletin boards and discussion forums, blogs, and so on. Firefox was selected because it occupi ed a relatively small size of pie in the Web browser market when this study was c onducted. MicrosoftÂ’s Internet Explorer (IE) had a monopoly position in the Web browser ma rket, claiming more than 95 percent of it before Firefox (FF) made inroads into the market. At the time when this study was conducted, the Firefox browser received much attention from the technology industry and among Web-savvy users, and furthermore from gr assroots public relations campaigns as well as numerous endorsements from technol ogy columnists/reviewers and ordinary people such as bloggers. Unlike its popularity among early adopters, however, most Web users were either unaware or indifferent to Firefox. Ind eed, a market data report by Jupiter Media in June, 2005 showed that Fi refox only succeeded in reducing Microsoft
109 IEÂ’s market share from 95 percent before Firefox was launched in the market to 90 percent. Numerous endorsement messages from a vari ety of sources were collected in this present study, to construct either typical or atypical arguments. T ypical arguments were mostly obtained from reliable or credible sources such as The New York Times , The Wall Street Journal , The Washington Post , a couple of magazines (e.g., Wired , Forbes , PC World , etc.) or an online technology news sour ce (CNET News.com). For each argument that was considered typical , a corresponding atypical message was derived from a discussion thread left in an online news site (e.g., CNET News.com) or from an individualÂ’s endorsement poste d to his or her Web log (blog). Most comments in the discussion forum were left anonymously, a nd those comments were sometimes very atypical, illogical, or extreme. For example, in a piece of an endorsement for Firefox by a technology columnist, there was the following sentence: Â“Keep in mind that Microsoft hasn't made any important functional improve ments in Internet E xplorer for years --aside from opening additional security hol es.Â” An anonymous reader left a comment saying, Â“NASA hasn't made any major advancem ents since the fall of the Soviet Union.Â” This sentence was later adopted and slightly changed to construct an atypical argument. Two pilot tests were conducted to measur e the typicality of each argument drawn. Each pilot test was followed by a discu ssion about the clarity, plausibility, and comprehensibility of each version of messages. In this way of generating arguments in the variation of typicality, a total of 28 arguments that ha d the equal number of typical and atypical arguments were constructed. Ne xt, a pretest was conducted to check the typicality of each of those 28 arguments. Twenty six student s majoring in public relations
110 read the 28 statements and rated the typicality of each statement. Participants in the pretest were asked to rate whether each of th e sentences would be typical or atypical on a 7-interval typicality scale1 if it were found in an endorsement for a Web browser. The presentation order of each of 28 arguments was randomized to avoid a possible order effect. The mean scores between typical and atypical arguments were compared using a ttest. Then, 10 arguments that did not make significance differences to mean scores were omitted in constructing a message for an experimental stimulus. Nine out of those ten arguments were later used as foil items in a recognition memory test in the experiment. Two different versions of an endorsement message with each having 9 typical or 9 atypical arguments were created. Each versi on shared some basic sentences that were necessary to describe background information or to make a transition between each sentence. In this way, each version of the endorsement message contained either mostly typical or mostly atypical arguments for claiming FirefoxÂ’s superior features over Internet Explorer. Stimuli: Endorser Credibility Endorser credibility was manipulated by presenting subjects with a brief biographical sketch for a person called Wa lt Morse before their reading of his endorsement message. The biographical sketch for the higher credibility source was adopted from two sources: The Â“About Walt MossbergÂ” Web page on the Walt MossbergÂ’s Web site and Wired MagazineÂ’s article (D eutschman, 2004) about Walt Mossberg. Biographical information for the low credibility source was also adopted from 1 This manipulation-check question was derived from Schmidt and Sherman (1984) and from Shapiro and Fox (2002). Schmidt and Sherman (1984) determined typicality by obtaining normative ratings for arguments on a scale from 0 (very atypical) to 9 (very typical). Shapiro and Fox (2002) also asked participants to rate each of items on a 6-point typicality scale anchored by Â“1: very typicalÂ” and Â“6: very atypical.Â”
111 an ordinary bloggerÂ’s Â“About MeÂ” page with a slight revision to make him less credible. The scripts for each endorserÂ’s biography were created after two pilot tests and subsequent discussions with th e participants in the pilot studies. After obtaining reliable results for manipulation checks from the two pilo t tests, a pretest was conducted to ensure that the final scripts for an endorserÂ’s biography effectivel y separated two levels of endorser credibility groups. CronbachÂ’s alpha for the six items measured in the pretest was .95. The manipulation check of endorser credib ility in the pretest showed that two versions of credibility information distinct ively separated high and low-credibility groups. With 24 participants in a pretest, those w ho were assigned to the high-credibility groups rated significantly higher on six evaluation items than counterparts in the low-credibility groups (see Table 1). Table 1. Manipulation check of endor ser credibility in a pretest Low Credibility Group (N=13) High Credibility Group (N=13) t df Qualified 4.4 (1.4) 5.8 (1.0) 2.91 ** 24 Expert 3.8 (1.1) 5.3 (1.3) 3.24 ** 24 Professional 3.6 (1.3) 5.5 (1.3) 3.78 ** 24 Trustworthy 3.6 (1.3) 5.6 (1.0) 4.41 *** 24 Believable 3.8 (1.3) 5.7 (1.1) 4.10 *** 24 Reliable 3.9 (1.4) 5.6 (1.2) 3.26 ** 24 Note. Measures are based on a 7-point seman tic differential scale, with 4 indicating neutrality of impression evaluations. ** p < .01; *** p < .001 Following the generation of the typical a nd atypical messages and the biographical sketch for a low and a high-credibility endor ser, each version of the stimuli received manipulation checks using a separate sample of 26 volunteer students from the same university that subsequently provided experimental subj ects. Students were mostly
112 juniors who were enrolled in a publ ic relations research class at the University of Florida. All the stimuli and questionnaire used in the pr etest were the same as those used in the experiment except that there were no recogni tion test items. Instead participants in the pretest rated typicality for each recogniti on item. The 26 students who were randomly assigned into 4 groups by message typical ity (typical vs. atypical) and endorser credibility (low vs. high) r eceived a booklet that contai ned stimuli and a questionnaire. After reading the source information and endor sement message, particip ants in the pretest were asked to rate their impressions about th e source in six 7-point semantic differential scales and to rate the message typicality on one bipolar 7-poin t semantic differential scale (very atypical-very typical)2. To differentiate the levels of endorser cr edibility, two versions of an endorserÂ’s biographical sketch were created including the endorserÂ’s identity, credentials, and his affiliation with an organization or an instituti on. Participants in the main experiment were led to believe that the information containe d in each endorsement had been written by either an expert/trustworthy or an inexpert /untrustworthy source. In the high-credibility (expert/trustworthy) condition, participants were given a brief biography of Walt Morse who was depicted as The New York Times ' tech columnist and one of the most influential tech reviewers in United States. In th e low-credibility c ondition, Walt Morse was portrayed as a paid blogger w ho is expected to endorse Firefox. Key sentences from each biography for different vers ions of source credibility were as follows: 2 The question used to ask typicality rating was Â“Please evaluate whether each of the following arguments might typically be found in an endorsement message fo r a Web browser. If an argument is so typical that you feel familiar to it or could think of it easily on your own, please circle 7. On the other hand, if an argument is so rare and atypical that if feels very unfamiliar or something you might never think of your own, please circle 1. If you feel that it is in between the two extremes in typicality, please circle 4 on the scales below.
113 High-Credibility Condition Walt S. Morse is the author and creator of the weekly technology review column "What's Next" in The New York Times , which has appeared every Thursday since 1991. Mr. Morse has been known as a fiery cr usader since the opening line of his column "What's Next": "Personal computers are just too hard to use, and it isn't your fault." His quest has ear ned him legions of fans. [Â…] Yahoo! Cofounder Jerry Yang says, "He's one of the most trusted and influential voices in technology." [Â…] The Wall Street Journal calls him "the most respected computer journalist." Mr. Morse was awarded the 1999 Loeb award for Commentary, the only technol ogy writer to be so honored. Low-Credibility Condition Walt S. Morse is one of the Petereus blogge rs. Petereus is a company that is paying bloggers $800 a month to blog their sponsors' products or services. The content that runs on these web pages won't be marked as advertising, it will be editorial comment, right in the body of the blog itsel f. Bloggers are paid to write and report freely about their own views on any product that sponsors Petereus. Mr. Morse uses his blog, Junk Drawer, to put up links to music videos, pop culture, video games, and Japanese manga. [Â…] When he's not goofing off online, Wa lt Morse writes a book whose upcoming title is Book of the Dumb (a book on stupid ity). He also reviews stuff for Official US PlayStation Magazine . Now, Walt Morse lives in a te eny tiny little town in Ohio called Bradford, where a traffic jam is defined as three cars caught behind an Amish buggy. Walt Morse lives there with his wife Kristine, dog Kodi and cats Rex, Lopsided Cat, and Ghlaghghee (pr onounced "fluffy"). He enjoys pie. [Disclosure: Mr. Morse is paid by Petereus to comment on Firefox.] In the first version of this biography, the low-credibility endorser was depicted as a professional gambler with average knowledge about Internet security and Web browsers. While this scenario was successful in lowering scores in the expertis e factor, it was not effective in manipulating the trustworthiness fa ctor. Suggested by participants in the first pretest, the new version of biography for a low-credibility source was developed.
114 After reading these two versi ons of source credibility ma nipulation, participants in the second pretest were asked to evaluate th eir feelings or impressions about Walt Morse on six 7-point semantic differential scales (unqualified-qualified, not expert-expert, unprofessional-professional, untrustwort hy-trustworthy, unbelieva ble-believable). Subjects Participants were 532 students at three large public universit ies [University of Florida (452), Washington State University ( 20), and University of Maryland (28)] and a large private university [Syracuse University (32)] in the United States. All students participated in the experiment voluntarily, but they received extra credit for completing a series of questionnaires. Courses varied fr om mass communication research methods to public relations principles/campaigns to American politics & mass media. Among the 532 students who registered for the study, 40 st udents dropped out during the delayed test and those responses were excluded from the analysis of this study. Procedure The experiment was conducted exclusivel y on the Web. All experimental materials including source information, both typical and atypical messages, and a questionnaire were posted on the Web. Although the experime nt was conducted on the Web, it was the extension of a laboratory experiment as well as a computer-based laboratory experiment. Based on the experimental design of this study, subjects who participated in this experiment were randomly assigned to either eight different experi mental groups or two control groups. Subjects who agr eed to voluntarily participate for this study were asked to sign up for this experiment by registering their email address with the investigator of the study. Each participant was then randomly a ssigned to one of ten conditions, and was sent an invitational email message that in cluded a Web link to a designated condition. All
115 other procedures were very cl ose to those of laboratory expe riments except that reading time for experimental stimuli was strictly c ontrolled by a JAVA-scri pt program. In other words, subjects were expected to read the experimental stimuli during the enforced time intervals by making a next button appear on the screen after the minimum time passage. All subjects, including those assigned to control groups, completed a pretest in which they were asked to answer several questions regarding their involvement in Internet ExplorerÂ’s (IE) security issu e and were then asked to eval uate how they felt about the Firefox browser on six 7-point semantic differential scales. Then, participants in the experimental conditions read the biography for Walt Morse whose information varied in two levels of credibility. After the exposur e to these manipulations, subjects in an atypical group read an atypical endorsemen t message, while counterparts in a typical group read a typical endorsement message. Afte r reading these messages, subjects in the immediate test group again evaluated their attitudes toward Firefox. Then, they were asked to rate the messageÂ’s persuasiveness. Finally, a recogniti on memory test was administered. The recognition memory test incl uded nine target sentences and the equal number of foil sentences3 that had not been presented in the endorsement. The instructions4 indicated to participants that some of the arguments had not, in fact, been presented in the endorsement. The 18 stat ements used in the recognition test were randomly ordered. On each Web page, only one statement was presented followed by a confidence rating for the recognition task. Each recognition test requir ed participants to 3 The nine foil statements had been produced during the message development process, but they were not adopted as experimental stimuli because they received mixed scores in typicality testing in pilot tests. 4 Among the following sentences, some appeared in the article you read about Firefox and some did not. For each statement below, please check Yes if you rememb er that the statement was in the article or check No if it was not. For each YES or NO judgment, please indicate how sure or unsure you are of your answer.
116 respond in order for them to go to the next pa ge of questionnaire. Th e entire experimental session took approximately 20 minutes. Subjects assigned to a delayed group were thanked and were informed that they might be contacted a few weeks later for some supplementary questions. Two weeks later, they were contacted by email and asked to answer some supplementary questions in a follow-up survey. Every possible method was employed to reduce the dropout reduction rate (Reips, 2002). Several bar graphs displaying each participantsÂ’ percentagecomplete were shown to notify the participan ts of their progress during the course of experiment5. Subjects were informed of extra credit in return for completion of the follow-up survey. To decrease the variance in time passage6, participants were asked to complete the follow-up survey as soon as po ssible, and were notified that the survey would be closed in three days. To redu ce possible communication between delayed and immediate-test groups regardi ng the recognition test, particip ants assigned to the delayed group were given the pre-test two weeks befo re those assigned to the immediate test condition. Independent Variables Typicality Typicality was defined in accordance w ith Loken and WardÂ’s (1990) definition. That is, typicality in a persuasi ve argument refers to Â“the degree to which an argument is logically sequenced and conceptually coherent with a generic schema.Â” 5 It is known that participants will be more likely to complete the experiment, if they know how long it will take approximately (Reips, 2000). 6 Without time constraints, participants tend to forget about responding to follow-up surveys. In addition, if a participantÂ’s answer is made in a time long after other subjectsÂ’ answers were collected, their answer could be considered as to be measured in a different time interval. It is a shortcoming of a Web-based experiment.
117 Endorser credibility In accordance with traditional definitions of source credibility, some scholars have discussed source credibility on the Web in terms of a s ource's trustworthiness and competence (Warnick, 2004). Fritch and Crom well (2001) argued that several factors made it difficult to determine credibility (or authority) of the source of Web-based information such as "no filtering of informa tion," "no entry barrier," Â“absence of author qualifications and cred entials," and so on. They suggest ed that perceived credibility (author competence and trustworthiness) of Web source can be increased by revealing author identity, author credentials (degree, titles, biographical information, experiences) and author's affiliation with an organization or an institution. Thei r evaluation criteria were well reflected in the ma nipulation of endorser credibil ity in this present study. Dependent Variables Post-attitude Measures This studyÂ’s primary dependent measure we re post-attitudes to ward Firefox, which was measured with six 7-point semantic di fferentials, anchored by the following bipolar adjectives adapted from MacKenzie and Lutz (1989) and Priester and Petty (2003): goodbad, useful-not useful, securenot secure, easy-difficult, lik able-dislikable, and harmfulbeneficial. The six itemsÂ’ scores were av eraged to form a composite measure of participantsÂ’ attitudes. Each participantÂ’s at titude index could range from 1 to 7, with higher scores indicating more pos itive attitudes toward Firefox. Recognition Performance The most widely used dependent variable in previous studies of the SC + T model is recognition performance (Graesser, 1981; Graesser et al., 1979, 1980, 1998; Hess, 1985; Macrae et al., 2002; Nakamura et al., 1985; Rice & Okun, 1994; Schmidt &
118 Sherman, 1984; Smith & Graesser; Vakil et al., 2003). The recognition performance in memory research is generally7 measured by hit rates (HR), false alarm rates (FAR), and d scores in terms of signal de tection theory. Sentences that had been explicitly named in the acquisition phase and that were correctly recognized we re classified as Â“hits.Â” Incorrect recognition of sentences that were not presented during acquisition but were recognized as having been seen was cl assified as Â“false alarms.Â” The d index is most widely used in memory research to assess subjectsÂ’ ability to discriminate between presented and non-presented items in r ecognition memory tasks (see Murdock, 1974; 1982). Thus, when it comes to d scores for recognition performance, it is often called memory discrimination (Smith and Graesser, 1981) or memory accuracy (Neuschatz et al., 2002; Schmidt & Sherman, 1984). It is also called recognition Â“sensi tivityÂ”: the larger the d score, the more sensitive the judge is at discriminating between target and foil items (Fox, 2004; Macmillan & Creelman, 1991; Shapiro, 1994). In this study, recognition accuracy ( d) for typical and atypical arguments was assessed by computing d indices for each subject. Each d index was computed using a participantÂ’s hit rate (i.e., the rate at which one correctly identified statements that were really presented during their r eading of the message) and oneÂ’s false alarm rate (i.e., the rate at which he or she inco rrectly claimed that new stat ements had been made during their reading of the message). The d statistics is the differe nce between these two rates divided by the standard deviation of responses to the new statements. The d scores provide a measure of recognition performan ce that corrects for guessing (Smith & 7 Some researchers measure the experiential cont ent of subjectsÂ’ memories using Remember/Know distinction introduced by Tulving (1985) and developed by Gardiner (2002; Gardiner & Java, 1990, 1993) and Rajaram (1993; Rajaram & Roediger, 1997).
119 Graesser, 1981). The highest possible hit rate and the lowest possible false alarm rate indicate the most accurate recognition. (Hol brook et al., 2005; Cradit, Tashchian, & Hofacker, 1994; Tavassoli, 1998; Shapiro & Fox, 2002; Fox, 1999, 2004). In the signal detection theo ry, participantsÂ’ judgments about the stimuli are based on familiarity values which are assumed to be normally distributed (Macmillan & Creelman, 1991; Shapiro, 1994; Fox 2004). When requested to judge the familiarity of stimuli, participants will have a theoretical decision space with two probability distributions along a familiar ity continuum, one for old items and one for new items (Macmillan & Creelman, 1991; Fox, 2004). These pr obability distributions may overlap in parts depending on how sensitive oneÂ’s reco gnition memory is in distinguishing old items from new items. The more sensitive it is, the less probability distributions will overlap and the greater the distance will be between the means of those distributions (Macmillan & Creelman, 1991; Fox, 2004). According to Rotello, Macmillan, and Reeder (2004), this distance between the means of the two distributions is theoretically a measure of memory accuracy that is inde pendent upon the decision criterion when the variances of the two distribu tions are equal (p. 589). Confidence for Memory Discrimination The confidence rating task is often request ed in the memory test, because it is unrealistic to ask participants to respond to a block of stimulus items several times each time a participant will become either more or less cautious8 in deciding whether he or she remembers the stimulus items (Tashchian, White, & Pak, 1988). 8 Respondents may be asked to use a number of decisional criteria along the line of extremely conservative responses (in which the respondent must have an unusually high level of certainty before reporting recognition of the stimuli) and of an extremely liber al response (in which even the slightest feeling of familiarity is sufficient to trigger a recognition response).
120 In the experiment from this present study, respondents were asked to report their confidence9 in their yes/no recognition judgment for an endorsement along a 3-point scale: Â“Sure,Â” Â“Not SureÂ” and Â“Unknown or Guess.Â” 9 In Schmidt and ShermanÂ’s (1984) study, the confidence level for each statement was rated on the following 6-point scale: 1 = definitely not presented in the interview; 2 = fairly sure was not presented; 3 = uncertain, but think it was not presented; 4 = uncertain , but think it was presented; 5 = fairly sure was presented; 6 = definitely presented.
121 CHAPTER 5 RESULTS Demographics Group Size A total of 532 participants were randomly assigned to one of either 8 experimental groups or 2 control groups (immediate vs. delayed). Among them, responses of 40 participants who did not complete the quest ionnaire in the delayed test were excluded from the analyses. The number of participants in the low and high-credibility groups was 199 and 207 respectively (see Table 2).. Table 2. Number of particip ants by experimental groups Immediate Delayed Total Low Credibility 49 46 95 Typical High Credibility 54 53 107 Low Credibility 49 55 104 Atypical High Credibility 51 49 100 Subtotal 203 203 406 Control 48 38 86 Total 251 241 492 An equal number (N = 203) of participants completed either an immediate or a delayed test for the experimental groups. Two-hundred-and-two participants who were assigned to the typical message condition return ed their responses, and 204 in the atypical message condition group completed the test. Forty-eight participants returned their responses to an immediate control group test , while 38 students completed their answers for a delayed control group test.
122 RespondentsÂ’ Demographics Most participants were female (71.0%). The mean age was 21.3 with a standard deviation of 3.2. Participants were diverse in majors (see Table 3). About 25% of participants (N = 124) majore d in public relations, 15.5% (N = 76) in business, 10.6% (N = 52) in telecommunication, 10.2 % (N = 50) in sports & recreation management, 6.5% (N = 32) in political science, 5.9% (N = 29) in journalism, 5.1% (N = 25) in mass communication, 4.1% (N = 20) in sociology & criminology, 3.5% (N =17) in literature, 2.9% (N = 14) in psychology, and so forth (see Table 3). Table 3. ParticipantsÂ’ majors Major Participants Percent Public Relations 124 25.3 Business 76 15.5 Telecommunication 52 10.6 Sport & Recreation Management 50 10.2 Political Science 32 6.5 Journalism 29 5.9 Advertising 26 5.3 Mass Communication 25 5.1 Sociology & Criminology 20 4.1 Literature (English, French, etc.) 17 3.5 Psychology 14 2.9 Science 8 1.6 Engineering 7 1.4 History 5 1.0 Economics 2 0.4 Others (Nursing, Architecture, WomenÂ’s Studies) 3 1.0 Missing 2 Total 492 100.0 About 38% of participants reported their family income below $50,000 (see Table 4). The median household income range was from $75,000 to $99,999. Participants predominantly used Windows as their main desktop operating system (96.5%).
123 Table 4. Participants Â’ household income Household Income Participants Percent Less than $30,000 119 25.5 $30,000 ~ $39,999 29 6.2 $40,000 ~ $49,999 29 6.2 $50,000 ~ $59,999 22 4.7 $60,000 ~ $74,999 22 4.7 $75,000 ~ $99,999 54 11.6 $100,000 ~ $149,999 83 17.8 $150,000 ~ $249,000 65 13.9 $250,000 ~ $490,999 25 5.4 $500,000 ~ $999,999 15 3.2 $1 million or more 4 0.9 missing 25 Total 492 100.0 Asked whether they used Internet Explor er, 87.1% (N = 427) of participants answered that they were curren tly using the Internet Explorer (IE). About thirteen percent (N = 63) of participants, however, answered th ey were not using Inte rnet Explorer (see Table 5). Table 5. Most frequently used Web browser among participants Manipulation Checks Endorser credibility A manipulation check on cred ibility was performed with data obtained from the main experiment of this study. As a check on the measure of endorser credibility, participants in the experimental conditions (N = 404) were asked to indicate their impressions about the endorser with the sa me six items used in the pretest. The biographical sketch used in the experiment was the same as the biography used in the Browsers Participants Percent Internet Explorer (IE) 427 87.1 Non-IE 63 12.9 missing 2 Total 492 100.0
124 pretest. Again, those who read a high-credib ility version scored significantly higher for all six items of impression measures than counterparts who received a low-credibility version (see Table 6). Table 6. Manipulation check of endorser credibility in the main experiment Low Credibility Group (N=199) High Credibility Group (N=207) t df Qualified 4.3 (1.6) 5.3 (1.3) 6.90 *** 404 Expert 4.3 (1.5) 5.0 (1.4) 5.09 *** 404 Professional 4.0 (1.6) 5.4 (1.3) 9.38 *** 404 Trustworthy 4.1 (1.6) 5.3 (1.4) 8.65 *** 404 Believable 4.1 (1.5) 5.3 (1.4) 8.26 *** 404 Reliable 4.0 (1.6) 5.3 (1.4) 8.78 *** 404 Note. Measures are based on a 7-point seman tic differential scale, with 4 indicating neutrality for impression evaluations. *** p < .001. Argument typicality Twenty-six participants who voluntarily part icipated in a pretest rated typicality for nine typical and nine atypical arguments that were presente d in the endorsement message. The mean scores for the typical and the at ypical arguments were 4.94 (SD = .78) and 3.17 (SD = .56) respectively on a 7-poi nt scale. Each typical argu ment that contrasted with each atypical argument was reported in Tabl e 7 with mean and standard deviation. Typicality rating was based on a 7-point sema ntic differential scale with 4 indicating neither typical nor atypical. Paired-sample ttests were performed to compare the mean differences for each version of argument ty picality. Every paired item was significantly different, providing evidence that the manipulation of typicality may have been effective.
125 Table 7. Manipulation check of t ypicality in a pretest (N=26) Note. Typicality was rated on a 7-point semantic differential scale (1: very atypical, 7: very typical) with 4 indicating neither typical nor atyp ical. T and AT indicate Â“typicalÂ” and Â“atypicalÂ” respectively. *** p <.001, ** p <.01, * p <.05 Typicality Arguments M SD t df T1 Seeing Microsoft piling on supp osedly "cool" features without regards to the consequences gives me the impression that their programmers are completely clueless about how things work in the "real world." 4.2 1.2 At1 Watching Microsoft pile on "cool" features without regards to the consequences gives me the impression that their average programmer is about 7 years old. 3.2 1.3 2.42 * 12 T2 Keep in mind that Microsoft hasn't made any important functional improvements in Internet Explorer for years. 4.9 1.4 At2 Like the US space program in which NASA hasn't made any major advancements since the fall of the Soviet Union, Microsoft has not released any major upgrades for Internet Explorer since its defeat of Netscape. 2.7 1.8 3.43 ** 12 T3 However, you are advised to surf with discretion because it may endanger your privacy and security. 5.9 .8 At3 However, to be blunt, he or she should be fired for stupidity because they are endangering the company. 3.8 1.7 4.27 ** 12 T4 Firefox is built and distributed free by the Mozilla Foundation, a small nonprofit corporation spun off last year from the fastfading remnants of Netscape. 4.5 1.4 At4 The name Firefox is a common name for the red panda which eats large amounts of bamboo. 2.4 1.3 3.67 ** 12 T5 Wouldn't it be great if you could read papers, do searches, and check the weather altogether? 5.6 1.0 At5 Wouldn't it be great if you could see multiple pictures of dating partners in a single browser window? 3.0 1.7 6.20 *** 11 T6 Since it is open source, the software is guaranteed to not have intended malicious code or spyware included. 4.5 1.3 At6 Since it is open source, the software can be protected by international hackers who are sympathetic to the open source community. 3.1 1.6 2.29 * 12 T7 Getting the same kind of protection in your Explorer requires expensive add-on software created by third parties. 5.0 1.3 At7 If you want the same thing in your Explorer, Microsoft and its toadies will charge you extra money, even for their buggy and inadequate add-on software. 3.8 1.3 3.05 * 11 T8 While interviewing Bill Gate s for "ABC World News Tonight" in 2004, Peter Jennings asked Gates whether it is fair to say that Firefox's perceived ability is better than IE. 4.8 1.5 At8 While interviewing Bill Gates for "Eye to Eye" in 2004, Connie Chung likened Gates' business tactics to knife-fighting. 3.1 1.2 2.69 * 11 T9 A new battle has begun and the browser war is on again. 5.5 1.2 At9 I declare a war against th e Internet Explorer. 3.5 2.2 3.02 * 11
126 Reliability Source credibility . Reliability for the source credibility measures was high ( = .95). Following the reliability te st, a principal axis factorin g analysis was performed for construct validation. One factor accounted for a bout 76.8% of total vari ance (see Table 8). Table 8.. Factor loadings for items in the source credibility measures Items Loading Reliable .93 Believable .91 Trustworthy .90 Professional .85 Qualified .83 Expert .80 Post-attitudes . To asses the post-attitudes toward Firefox, an attitude index was developed using a six-item sema ntic differential scale. To ex amine the reliability of the attitude index, Cronbach's alpha was calculated. The reliabilit y for this index was high ( = .96). A principal axis factoring analysis yi elded one factor that accounted for 83.1% of the variance. All items loaded highly on th is component (ranging from .81 to .96). Table 9.. Factor loadings for items in the attitude measures Items Loading Bad-Good .96 Useful-Not useful .96 Likable-Not likable .94 Harmful-Beneficial .91 Difficult-Easy .90 Not secure-Secure .81 In the following section, results for hypothesis tests were reported.
127 Overall Review of Findings Two separate three-way analyses of variance (ANOVAs) were conducted to examine the overall association between de pendent and independent variables. The dependent variables were recognition memory performance which was measured by sensitivity ( d) and postattitudes toward Firefox. For each analysis, the independent variables were retention inte rval (time), message typicali ty, and source credibility. In Table 10, a 3-way ANOVA testing the asso ciation between retention interval, message typicality, and source cr edibility showed a significant main effect of retention interval and an interaction effect of retention interval by typicality. Table 10.The effects of typicality, time, and credibility on recognition performance Source MS df F p 2 Power. Constant .03 1 .02 .00 .05 Typicality 5.31 1 2.95 .01 .40 Time 93.49 1 51.96*** .12 1.00 Credibility 1.29 1 .72 .00 .14 Typicality X Time 16.82 1 9.35** .02 .86 Typicality X Credibility 1.91 1 1.06 .00 .18 Time X Credibility 1.26 1 .70 .00 .18 Typicality X Time X Credibility 2.48 1 1.38 .00 .22 Error 1.22 398 Note . R squared = .15 (Adj. R squared = .13). *** p < .001, ** p < .01 Retention interval (time) s howed the strongest effect ( F = 51.96, p < .001, partial eta 2 = .12), followed by a time-bytypicality interaction (F = 9.35, p < .01, partial eta 2 = .02). Table 11 shows the means ( SD s) for recognition performa nce of eight different
128 experimental groups by typicality, credibility, an d retention interval. As can be seen in Table 11, the lowest mean d score was observed among partic ipants who had received an atypical endorsement message from a highl y credible cue but whose memory was measured after a two-week delay. Â… Table 11. Means (SDs in parenthesis)for d by time, typicality and credibility Typical Atypical Low credibility High CredibilityLow credibility High Credibility Immediate .28 cd (1.21) N =49 .49 d (1.21) N =54 .44 d (1.55) N =48 .69 d (1.99) N =51 Delayed -.32 abc (.87) N =44 -.02 bcd (.98) N =53 -.67 ab (1.41) N =52 -.96 a (1.11) N =48 F = 9.75, df = 7/391, p < .001, Means with different subscripts within measure differ at p < .05. (post-hoc TukeyÂ’s B test). In a next step a 3-way ANOV A was run for attitudes toward Firefox, testing the association between time, message typical ity, and source credibility. There was a significant main effect for time ( F = 20.39, p < .000) and a significant three-way interaction between time, typicality, and credibility ( F = 5.53, p < .05). Together these variables explain about 6% of the variation in attitude change. Table 12.The effects of typicality, time, a nd credibility on attitudes toward Firefox Source MS df F p 2 Power. Constant 11977.85 1 9823.72*** .96 1.000 Typicality .10 1 .08 .00 .06 Time 24.87 1 20.39*** .05 .96 Credibility .26 1 .21 .00 .06 Typicality X Time 1.27 1 1.04 .00 .18 Typicality X Credibility .14 1 .12 .00 .06 Time X Credibility .51 1 .42 .00 .10 Typicality X Time X Credibility 6.74 1 5.53* .01 .65 Error 1.22 398 Note . R squared = .06 (Adj. R squared = .05). *** p < .001, * p < .05
129 Table 13 shows means and standard deviations for attitude scores by eight experiment groups. A one-way analysis of vari ance (ANOVA) revealed that treatments in this study made a difference of m eans for attitudes between groups, F (7, 398) = 3.86, p <.001. Table 13. Means ( SD s in parenthesis) for atti tude toward Firefox by group Typical Atypical Low credibility High CredibilityLow credibility High Credibility Immediate 5.8 c (.9) N =49 5.7 c (.1) N =54 5.4 b (1.1) N =49 5.8 c (1.1) N =51 Delayed 5.0 a (1.2) N =46 5.3 ab (1.2) N =53 5.4 (1.2) ab N =55 5.1 a (1.2) N =49 F (7, 398) = 3.86, p < .001, Means with different subscripts within measure differ at p < .05. (post-hoc TukeyÂ’s B test). A post-hoc analysis using TukeyÂ’s B test showed that the means of the immediatetypical groups were signifi cantly different from means fo r the delayed-typical groups, p < .05. Means for the atypical-immediate a nd the atypical-delaye d groups were only different when the source credibility was high. Tests of Hypotheses Test of Hypothesis 1. H1a. ParticipantsÂ’ recognition performan ce will be greater for atypical than for typical arguments in an immediate test. H1b. ParticipantsÂ’ recognition performa nce will be greater for atypical than for typical arguments in a delayed test. It was predicted that par ticipantsÂ’ recognition performa nce would be greater for atypical than for typical arguments in both an immediate and a delayed test (Hypothesis 1a and 1b). To test these hypot hesis, t-tests were performe d on the mean sensitivity ( d)
130 scores over different retenti on intervals (i.e., immediate vs . delayed). Table 14 and Table 15 show recognition scores from the discrete yes/no recognition test s in different time intervals. In the proposed hypothesis 1a, it was pred icted that the recognition performance would be higher for atypical arguments than for typical arguments in an immediate memory test. Although the d scores were about 20% better for at ypical sentences ( M = 0.6 SD = 1.8) than for typical ones ( M = 0.4, SD = 1.2) in an immediate test, no statistically significant difference was found between typical and atypical groups, t = .8, df = 200 (see Table 14). Thus, hypot hesis 1a was not supported. Table 14. Recognition performance by typi cality at an immediate test Typical (N = 103) Atypical (N = 99) t df p Mean d .4 (1.2) .6 (1.8) .84 200 .404 Note. d= Z (HR) Â– Z(FAR). To test hypothesis 1b for delayed memory performance by typicality, a t-test was conducted with d as the dependent variable. The re sult produced a significant mean difference between the two groups ( t = 4.05, df = 195, p < .001), but the direction was reversed from that of the proposed hypothesis. The mean d score for atypical arguments was -0.8 (SD = 0.9) while it was -0.2 (SD = 1.3) for typical arguments (see Table 15). This finding contradicted the proposed hypothesis 1b; thus, hypothesis 1b was not supported. Table 15.Recognition performance by t ypicality at a delayed test Typical (N=97) Atypical (N=101) t df p Mean d -.2 (.9) .8 (1.3) 4.05 *** 195 < .001 Note. d= Z (HR) Â– Z(FAR), ** p < .01, *** p < .001
131 Test of Hypothesis 2. H2. There will be a main effect of time of measurement on recognition performance. In other words, partic ipantsÂ’ recognition performance will decrease over time. In Hypothesis 2, it was predicted that part icipantsÂ’ recognitio n performance would decrease over time. Table 16 presen ts the results of a t-test with d scores as a dependent variable and time of measurement as inde pendent variables. Consistent with the Hypothesis 2, the results of the t-test indicated a significant e ffect of retention interval on d scores, with mean d of 0.5 (SD = 1.5) without a dela y and -0.5 (SD=1.2) with a twoweek delay, t = 7.11, df = 397, p < .001 (see Table 12). Thus, Hypothesis 2 was supported. Table 16. Recognition performan ce by retention intervals Immediate (N=202) Delayed (N=198) t df Mean d .5 (1.5) .5 (1.2) 7.11 *** 397 Note. d= Z (HR) Â– Z(FAR), *** p < .001 Test of Hypothesis 3. H3. The decrease of recognition performance over time will be greater for atypical arguments than for typical arguments. In Hypothesis 3, it was predicted that r ecognition memory for atypical arguments would decay faster than for typical argume nts after a two-seek delay. The interaction effect of typicality by reten tion interval on rec ognition performance was supported, F (1, 395) = 9.54, MS = 17.2, p < .01, p 2 =.02 (see Table 17).
132 Table 17. The effect of reten tion interval and typicality on d (sensitivity) Source MS df F p 2 Constant .0 1 .00 .00 Retention interval 92.7 1 51.58*** .12 Typicality 5.5 1 3.08 .01 Typicality X Retention interval 17.2 1 9.54** .02 Error 1.80 395 R Squared= .14 (Adj. R Squared=.13). *** p < .001, ** p < .01. Findings reported in Table 18 showed that memories for atypical arguments decayed more rapidly than for t ypical ones after a delay. The mean d for atypical arguments dropped from .6 to -.8 between imme diate and delayed tests, while the mean d dropped from .4 to -.2. The recognition sensitivity ( d) was about 20% better for atypical than for typical items in an immediate test, but it turned out to be negatively higher for atypical than for typical items at the delay. Table 18. Interaction of typicality and time interval on recognition performance Typicality Typical Atypical Retention intervals Immediate (N=103) Delayed (N=97) Immediate (N=99) Delayed (N=101) Mean d M (SD) .4 (1.2) -.2 (.9) .6 (1.8) -.8 (1.3) In summary, subjectsÂ’ memory performance was initially better for atypical than for typical arguments, but it became poorer as their memory was measured after a twoweek delay. Thus, Hypothe sis 3 was corroborated.
133 Test of Hypothesis 4. H4. There will be an interaction effect of typicality by retention interval on the confidence of recognition judgment. More sp ecifically, there will be more Â“SureÂ” responses for atypical hits than for typical hits at the immedi ate test, while there will be less Â“SureÂ” responses for atypical hits than for typical hits after a two-week delay. Shapior and Fox (2002) found that particip ants tested immediately were more confident in their judgment about atypical ite ms than typical items. However, it was found that as memory becomes less certain over time, people tend to be more conservative about their memories of atypi cal items, but more liberal about their memories of typical items (Shapiro & Fox, 2002) . Thus, it was predicte d that participants tested immediately would be more confid ent in their recognition judgments about atypical hits than typical hits, while those tested after a dela y would be less confident in their judgments about atypical hits than typical hits. To test this interaction effect, the tota l number of Â“SureÂ” responses for both typical hits and atypical hits was computed. A possi ble range of the total number is 0 (none of Â“SureÂ” responses) to 9 (all Â“SureÂ” responses). Once the total number of Â“SureÂ” responses for hit items was calculated, data were submitted to a two-way ANOVA for the interaction effect proposed in H4. Results (see Table 19) of a two-way ANO VA showed a significant typicality-bydelay interaction for confidence ab out correct recognition judgments, F (1, 402) = 12.02, MS = 43.2, p <. 01 with partial eta 2 of .03. Not surprisingly, part icipantsÂ’ confidence for their correct judgments dropped significantly over time, explaining 17% of variance for confidence about hits [ F (1, 402) = 81.6, MS = 293.4, p < .001].
134 Table 19. Interaction of typical ity and retention interval on confidence of recognition judgments for hits Source MS df F p 2 Constant 4346.3 1 1209.09*** .75 Retention interval 293.4 1 81.62*** .17 Typicality .1 1 .04 .00 Typicality X Retention interval 43.2 1 12.02** .03 Error 1.80 402 R Squared= .19 (Adj. R Squared=.18). *** p < .001, ** p < .01. Table 20 shows the mean Â“SureÂ” responses fo r participantsÂ’ correct recognitions for either typical or atypical presented sentences . As predicted, more Â“SureÂ” responses were attained for atypical items ( M = 4.4) than for typical items ( M = 3.8) when recognition was tested immediately. However, participants tested at a two-week delay scored more Â“SureÂ” responses for typical items ( M = 2.8) than for atypical items ( M = 2.1). A post-hoc LSD test revealed that the m ean differences between these f our groups were significant at p < .05 level. In summary, findings presented above supported Hypothesis 4 that predicted a typicality-b y-delay interaction. Table 20. Mean Â“SureÂ” responses for rec ognition judgments about typical/atypical arguments with retention intervals 95% CI for M Time Typicality N M SD Lower Upper Typical 103 3.8 1.7 3.5 4.1 Immediate Atypical 100 4.4 2.5 3.9 4.9 Typical 99 2.8 1.6 2.4 3.1 Delayed Atypical 104 2.1 1.6 1.8 2.4 Note : Numbers are the mean respons es of Â“SureÂ” for hit items. F (3, 402) = 31.36, p < .001. Mean differences between groups were significant at p < .05 (LSD test).
135 Figure 11. Interaction of Typicality X Time on confidence for recognition judgments. Confidence was measured by the number of Â“SureÂ” responses for either typical or atypical hits. Tests of H5a and H5b H5a. A highly-credible endorser will indu ce more attitude ch ange than a lesscredible endorser instantaneously. H5b. A highly-credible endor ser will induce more attit ude change than a lesscredible endorser after a delay. In hypothesis 5a and 5b, it was predicted that a highly-credible endorser would make more positive attitude change than a less-credible endorser both instantaneously and after a delay. However, the credibility vari able did not make any statistical difference in post-attitude scores in either immediate a nd delayed tests (see Table 21). Thus, neither H5a nor H5b was supported.
136 Table 21. Attitudes by credibility in an immediate and a delayed test group Low (N=98) High (N=105) t df p Immediate 4.9 (1.2) 4.9 (1.1) .28 201 .781 Low (N= 101) High (N= 102) t df p Delayed 5.0 (1.2) 4.9 (1.2) .63 199 .527 Note. The dependent measure of attitudes toward Firefox shows the average of six measures on a 7-point semantic differential scale, with higher values indicating a more positive attitude. Test of Hypothesis 6 H6. The post-attitudes observed in th e low-credibility conditions will be greater than those in the immediate contro l condition at the immediate tests. H6 was proposed to diagnose (1) if the tr eatment of a discounting cue induced postattitude change, and (2) if the change induced by the di scounting cue did not fall below the change observed in a control group. To test H6, a t-test was pe rformed with the dependent va riable of attitudes toward Firefox. As predicted, participants in the low credibility groups ( M = 5.86, SD = 1.0) showed significantly higher post-attitudes th an counterparts in a control group (M = 4.8, SD = 1.1) at an immediate test. These finings confirmed a treatment effect in which a discounting-cue-effect on attitudes would be greater than the ba seline attitude. Table 22. Comparison of means between low cred ibility groups and a control group at the immediate test Credibility N Mean SD df T Low credibility 98 5.6 1.0 Immediate control 48 4.8 1.1 144 4.19 *** Note. The dependent measure of attitudes towa rd the company also shows the average of six measures on a 7-point semantic differenti al scale, with higher values indicating a more positive attitude. *** p < .001. Test of Hypothesis 7a H7a. There will be an interaction between credibility and time on attitudes toward Firefox under the typical message conditi on. More specifically, a typical message presented by a highly-credible endorser will lead to a significant decrease of
137 attitude change over time, while a typical message presented by a low credible source will lead to an increase of attitude change over time. In H7a, it was predicted that under the typical message condition, attitudes induced by a high-credibility endorser would decrease rapidly over time, while attitudes by a lowcredibility endorser would increase. To test this hypothesis, a tw o-way analysis of variance was performed. A two-way ANOVA was performed to examine the interaction effect of credibility by time on attitudes toward Firefox. Tabl e 23 shows the summary of the two-way ANOVA under the typical message condition. There was no interaction of Time and Credibility under the typical message condition, F (1, 202) = 1.54, p =.216. Thus, H7a was not supported. Table 23. The effects of time and credibility on attitudes under typical message condition Source Mean Square df F p 2 Power Constant 5988.05 1 5225.41*** .96 1.000 Time 18.57 1 16.21*** .08 .98 Credibility .38 1 .34 .00 .09 Time X Credibility 1.76 1 1.54 .01 .24 Error 1.15 202 R 2 = .08 (Adj. R 2 = .07). *** p < .001 Table 24 presents the means and standard de viations for post-attitudes under typical message condition. Means for the low credib ility groups were 5.8 (SD = 0.9) and 5.0 (SD = 1.2) at immediate and delayed tests resp ectively. The mean for the high-credibility group was 5.7 (Sd = 1.0) during an immediate test, while it was 5.3 (SD = 1.2) during a delayed test. A post-hoc analys is (Scheffe test) revealed that means for both low and high-credibility groups at an immediate test were significantly higher than the mean for the low-credibility group at a delayed test.
138 Table 24. Comparison of mean post-attit udes by credibility and time under typical message condition 95% CI for M Time Credibility N M SD Lower Upper Low 49 5.8 b 0.9 5.6 6.1 Immediate High 54 5.7 b 1.0 5.4 6.0 Low 46 5.0 a 1.2 4.7 5.4 Delayed High 53 5.3 ab 1.2 5.0 5.6 Note : F (3, 198) = 5.83, p < .01. Means that do not share th e same subscript significantly differ at p < .05 (Scheffe test). Figure 12 illustrates the effects of time a nd credibility on attitudes under typical message condition. As reported in Table 23, there was only a main effect of time on attitudes. 4.2 4.4 4.6 4.8 5 5.2 5.4 5.6 5.8 6 Immediate DelayedImmediate control Delayed control Low Credibility High Credibility Figure 12. Effects of Time and Credibility on attitudes toward Firefox under typical message condition
139 Tests of H7b and H7c H7b. There will be a main effect of time on attitudes toward Firefox under the atypical message condition. H7c. There will be a main effect of credib ility on attitudes toward Firefox under the atypical message condition. Table 25. The effects of time and credibi lity on attitudes under atypical message condition Source MS df F p 2 Power. Constant 5989.90 1 4636.60*** .96 1.000 Time 7.49 1 5.80* .03 .67 Credibility .01 1 .01 .00 .05 Time X Credibility 5.51 1 4.27* .02 .54 Error 1.29 200 R 2 = .05 (Adj. R 2 = .03). *** p < .001, * p < .05 Under the atypical message condition, ther e was a main effect of time on postattitudes toward Firefox, F (1, 200) = 5.80, MS = 7.49, p < .05, p 2 = .03 (see Table 25). Thus, H7b was supported. However, there was no main effect of cr edibility. Thus, H7c was rejected. Table 26 shows the means and standard de viations for post-attitude scores under the atypical message condition. There was no difference of means for low-credibility groups between immediate and delayed tests (M = 5.4, SD = 1.1 during an immediate test; M = 5.4, SD = 1.2 during a delayed test). On the other hand, th ere was a significant difference for means for high-credibility gr oups between immediate and delayed tests. Tested immediately, the mean for the high-credibility group was 5.8 ( SD = 1.1), whereas the mean score decreased significantly when it was measured after a delay ( M = 5.1, SD = 1.1).
140 Table 26. Comparison of mean post-attitude s by credibility and time under atypical message condition 95% CI for M Time Credibility N M SD Lower Upper Low 49 5.4 ab 1.1 5.1 5.8 Immediate High 51 5.8 b 1.1 5.5 6.1 Low 55 5.4 ab 1.2 5.1 5.7 Delayed High 49 5.1 a 1.1 4.7 5.4 Note : F (3, 200) = 3.30, p < .05. Means that do not share th e same subscript significantly differ at p < .05 (Scheffe test). The following Figure 13 illustrates the interaction between time and credibility under the atypical message condition. 4.2 4.4 4.6 4.8 5 5.2 5.4 5.6 5.8 6 Immediate DelayedImmediate control Delayed control Low Credibility High Credibility Figure 13. Interaction of Time X Credibility on attitudes toward Firefox under atypical message condition
141 Summary of Tests of Hypotheses H1a. ParticipantsÂ’ recognition performance will be greater for atypical than for typical arguments in an immediate test. (Not supported) H1b. ParticipantsÂ’ recognition performance will be greater for atypical than for typical arguments in a delayed test. (Not supported) H2. There will be a main effect of time of measurement on recognition performance. In other words, participantsÂ’ recognition performance will decrease over time. (Supported) H3. The decrease of recognition performance over time will be greater for atypical arguments than for typical arguments. (Supported) H4. There will be an interaction effect of typicality by retention interval. More specifically, there will be more Â“SureÂ” responses for atypi cal hits than for typical hits at the immediate test, wh ile there will be less Â“SureÂ” responses for atypical hits than for typical hits after a two-week delay. (Supported) H5a. A highly-credible endorser will indu ce more attitude ch ange than a lesscredible endorser instantaneously. (Not supported) H5b. A highly-credible endor ser will induce more attit ude change than a lesscredible endorser after a delay. (Not supported) H6. The post-attitudes observed in the lo w-credibility conditions will be greater than those in the immediate contro l condition at the immediate tests. (Supported) H7a. There will be an interaction between credibility and time on attitudes toward Firefox under the typical message conditi on. More specifically, a typical message presented by a highly-credible endorser will lead to a significant decrease of attitude change over time, while a typical message presented by a low credible source will lead to an increase of attitude change over time. (Not supported) H7b. There will be a main effect of time on attitudes toward Firefox under the atypical message condition. (Supported) H7c. There will be a main effect of credib ility on attitudes toward Firefox under the atypical message condition. (Not supported)
142 CHAPTER 6 DISCUSSIONS AND LIMITATIONS Post-Hoc Analyses Post-Hoc Analyses for H1a, H1b & H3 In the test of H1a and H1b, an interaction effect of typicality by time of delay on memory sensitivity was found. To examine wh ether this interaction resulted from decreased hit rate (HR) or increased false alarm rate (FAR), data were submitted to twoway ANOVAs with dependent measures of HR and FAR. Results shown in Table 27 revealed main effects for retention interval [ F (1, 395) = 20.71, MS = .93, p < .001] and for typicality [ F (1, 395) = 5.06, MS = .23, p <. 05] on HR. An interaction between retention interval and ty picality was also observed, F (1, 395) = 8.45, MS = .38, p < .01. The comparison of the effect sizes indicated th at retention interval had the largest effect on HR explaining five percent of the total variance, followed by th e interaction between typicality and retention interval (p 2 =.02). Table 27. The effect of retention interval and typicality on HR Source MS df F p 2 Constant 135.1 1 3017.7*** .88 Retention interval .9 1 20.7*** .05 Typicality .2 1 5.1* .01 Typicality X Retention interval .4 1 8.5** .02 Error .1 395 R Squared= .08 (Adj. R Squared=.07). *** p < .001, ** p < .01, * p < .05.
143 However, FAR was explained only by retention interval, F (1, 395) = 29.66, MS =1.49, p < .001, p 2 = .07 (Table 28). There was no main effect for typicality, F (1, 395) = .04; nor was there an interaction betw een typicality and retention interval, F (1, 395) = 1.96, p 2 = .01. Table 28. The effect of retention interval and typicality on FAR Source MS df F p 2 Constant 97.9 1 1951.6*** .83 Retention interval 1.5 1 29.7*** .07 Typicality .0 1 .0 .00 Typicality X Retention interval .1 1 2.0 .01 Error .1 395 R Squared= .07 (Adj. R Squared=.07). *** p < .001. A closer look at HR and FAR by typicalit y and retention interval showed that poorer d scores for atypical items in a delaye d test was due to bot h increased FAR for atypical arguments and rapid drops in HR (see Table 29). Table 29. Interaction of typicality and time interval on recognition performance Typicality Typical Atypical Retention intervals Immediate (N=103) Delayed (N=97) Immediate (N=99) Delayed (N=101) M (SD) .62 (.16) .59 (.20) .63 (.24) .47 (.24) Mean HR t (df), p 1.37 (198), p = .172 4.66 (198), p < .001 M (SD) .45 (.24) .54 (.18) .42 (.25) .57 (.21) Mean FAR t (df), p 3.00 (198), p < .01 4.63 (197), p < .001 M (SD) .39 (1.21) -.16 (.94) .57 (1.79) -.80 (1.28) Mean d t (df), p 3.57 (198), p <.001 6.26 (197), p < .001 Note . HR= Hit Rate; FAR= False Alarm Rate.
144 The mean FAR for atypical arguments at the immediate test was 0.4 ( SD = 0.3), but it became higher at a delayed test ( M = 0.6, SD = 0.2), t = 4.63, df = 197, p < .001. The mean HR for atypical arguments al so dropped after a two-week delay ( t = 4.66, df = 198, p <.001). Under the atypical message condition s, the mean HR for immediate groups was 0.6 (SD = 0.2), while the mean HR for delaye d groups was 0.5 (SD = 0.2). On the other hand, there was no change in HR for typica l arguments between immediate and delayed tests with the mean for immediate groups = 0.6 and the mean for delayed groups = 0.6. As noted, it is an important finding that the d score was higher for an atypical argument ( M = .6) than for typical arguments ( M = .4) instantaneously, but the d for atypical arguments ( M = -.8) was lower than that for typical arguments ( M = -.2) after two weeksÂ’ delay. Figure 14 illustrates the changes in subjectsÂ’ recognition performance ( d) with mean HR and FAR as an inter action of typicality and time of measurement. At a glance, it is noteworthy that the mean FAR for bot h typical and atypical arguments increased after a delay (this increase wa s statistically significant as reported in Table 29). Another important finding was that the mean HR fo r atypical arguments decreased after a twoweek delay (this result was statistically signifi cant as reported in Table 29), while the HR for typical arguments did not change. Intere stingly, discriminative accuracy in a delayed test generated negative d scores for both typical and at ypical sentences. Figure 14 also demonstrates that recognition accuracy for atypi cal arguments declined at a faster rate over time than for typical ar guments, supporting Hypothesis 3.
145 Figure 14. Recognition performance as a f unction of typicality and time interval Supplementary Analysis for H4 As a supplement to the ANOVA for the te st of H4, chi-square analyses were performed on observed frequencies for conf idence ratings of sure, not sure, and guess/donÂ’t know (see Appendix F) fo r each recognition judgment. Post-Hoc Analyses for H5 In the test of H5a and H5b, no effect of credibility on attitudes was found. Although the manipulation check for credibility showed mean differences between low and high-credibility groups, it was found that a ll six items for the low-credibility groups were rated moderately rather than low. That is, all six items for cred ibility measures were rated around the median point in six 7-point semantic differential scales (see Table 6 in Chapter 5).
146 As a post-hoc analysis, post-a ttitude scores were compared across three levels of perceived credibility (low, moderate, and high) based on participantsÂ’ ratings of six items of credibility measures. An ANOVA revealed that there was an effect of perceived credibility on post-attitudes, F (2/143) = 9.40, p <.001. A subsequent post-hoc analysis (Scheffe test) specifically indicated that the attitude scores among the low-perceivedcredibility group were significantly highe r than those among the high-perceivedcredibility group. Table 30. Comparison of means between differe nt levels of perceived credibility 95% CI for means Group N Mean SD Lower Upper df F Low perceived credibility 49 5.2 1.1 5.0 5.4 Moderate perceived credibility 49 5.4 1.1 5.2 5.6 High perceived credibility 48 5.7 1.1 5.6 5.9 2/143 9.40 *** Note . Three levels of perceived credibility were created based on the frequency distribution of participantsÂ’ ra tings of six 7-point semantic differentials. The dependent measure of attitudes toward the company also shows the average of six 7-point semantic differential scales, with higher values indi cating a more positive attitude. CI = 95% confidence interval for group means. The mean for low perceived credibility is significantly different from the mean for hi gh perceived credibility at p < .05 (Scheff post-hoc test). *** p < .001. Post-Hoc Analyses for H6 H6 was proposed to determine whether or not attitude ch ange induced by a discounting-cue is greater than the change without any tr eatment. If the former is statistically smaller than the latter, a possi ble finding regarding the increased attitude change induced by the discounting cue cannot be attributed to a sleep er effect. Strictly speaking, it was not a test for a theoretical assumption, but was for checking a necessary condition for the subsequent theory test. One pr oblem of the test of H6 is that the low
147 credibility condition contained two groups: a low credibility w ith a typical message and a low credibility with an atypical message. Th atÂ’s why the cell size reported in Table 22 (see Chapter 5) is unequal. As a supplement ary analysis, an ANOVA was performed with breaking down the low-credibility groups by typicality. Table 31. Comparison of means for low credib ility groups and for a control group at the immediate test 95% CI for means Group N Mean SD Lower Upper df F Low Credibility & Atypical Message 49 5.4 1.1 5.1 5.8 Low Credibility & Typical Message 49 5.8 0.8 5.6 6.1 Immediate Control 48 4.8 1.1 5.2 5.6 2/143 10.37 *** Note. The dependent measure of attitudes towa rd the company also shows the average of six 7-point semantic differenti al scales, with higher valu es indicating a more positive attitude. CI = 95% confidence interval for gr oup means. Means for low credibility groups are significantly different from the mean for the immediate control group at p < .05 (Scheff post-hoc test). *** p < .001. Table 31 presents the group means, st andard deviations, and 95% confidence intervals for the means for lowcredibility groups with either typical or atypical message condition and for a control group in the immedi ate test. The difference in post-attitudes was significant, F (2, 143) = 10.4, p < 001, and a post hoc analys is revealed that postattitude scores among the participants in th e immediate low-credibility conditions were significantly higher than those in the im mediate control group (ScheffÃ©, p <.05). A three-way ANOVA was run with typicality, time of measur e, and credibility as a between-groups factor. Table 32 shows the results from the 3-way ANOVA.
148 Table 32.The effects of typicality, time , and credibility on attitude change Source MS df F p 2 Power. Constant 11977.85 1 9823.72*** .96 1.000 Typicality .10 1 .08 .00 .06 Time 24.87 1 20.39*** .05 .96 Credibility .26 1 .21 .00 .06 Typicality X Time 1.27 1 1.04 .00 .18 Typicality X Credibility .14 1 .12 .00 .06 Time X Credibility .51 1 .42 .00 .10 Typicality X Time X Credibility 6.74 1 5.53* .01 .65 Error 1.22 398 Note . R squared = .06 (Adj. R squared = .05). *** p < .001, * p < .05 Most of all, a significant main effect of time emerged from the three-way analysis of variance. The time factor accounted for 5% of total variability of the total effect, F (1, 398) = 20.39, p < .001. The three-way interaction between time, typicality, and credibility was significant [ F (1, 398) = 5.53, p <. 05], but the effect size was small, p 2 = .01. The findings reported in Table 32 also s uggest that interact ion between time and credibility may be observed only when the messa ge typicality is considered. In fact, the interaction by time and credibility was obser ved only when the message typicality was atypical in tests of H7a, H7b, and H7c (see Table 25 & Figure 13 in Chapter 5). Post-Hoc Analyses for H7 To examine if participantsÂ’ involvement and credibility judgments influenced the results of H7a, H7b, and H7c, an analysis of covariance (ANCOVA) was performed with
149 the involvement index1 and the perceived credibility index2 as covariates. Table 33 displays results from the analysis of covariance. Table 33.Analysis of covariance on attitude ch ange with the involvement and perceived credibility index as covariates Source MS df F p 2 Power. Constant 311.92 1 275.23*** .41 1.00 Typicality .00 1 .00 .00 .01 Time 25.43 1 22.44*** .06 1.00 Credibility 2.85 1 2.52 .01 .35 Typicality X Time .88 1 .77 .00 .14 Typicality X Credibility .16 1 .14 .00 .07 Time X Credibility .72 1 .64 .00 .13 Typicality X Time X Credibility 6.04 1 5.33* .01 .63 Involvement 6.12 1 5.40* .01 .64 Perceived Credibility index 23.10 1 20.38*** .05 1.00 Error 1.13 389 Note . The credibility index was the mean score for each participantÂ’s credibility ratings on six 7-point semantic differential scales. The involvement index was the means for two involvement measures: Â“how important is IE Â’s security to you personally?Â” (valuerelevant importance) and Â“how concerned are you about IEÂ’s s ecurity?Â” (outcomerelevant involvement). R squared = .13 (Adj. R squared = .10). *** p < .001, * p < .05 Even when the involvement and the perceived credibility were controlled, a significant main effect of time [(F (1,389) = 22.44, p <.001) and a 3-way interaction 1 The involvement index was the means for two involvement measures: Â“how important is IEÂ’s security to you personally?Â” (value-relevant importance) and Â“how concerned are you about IEÂ’s security?Â” (outcomerelevant involvement): see Eagly & Chaiken, 1993; Kiene, Barta, Zelenski, & Cothran, 2005. 2 The credibility index was the mean score for each par ticipantÂ’s credibility ratings on six 7-point semantic differential scales.
150 effect on [(F (1,389) = 5.33, p <.05] attitudes remained, demonstrating a robust treatment effect of time and a time-credibility-t ypicality interaction on attitudes. A hierarchical regression analysis was run to compare the relative effects of involvement and perceived credibility after c ontrolling the treatments of time, credibility, and typicality. In a first ste p, three treatments were entere d using dummy-variable coding. In a second step, involvement and perceived credibility were adde d to predict relative contributions of those variables to th e attitude change (see Table 34). Table 34.Hierarchical regression predic ting attitudes toward Firefox Variable B SE Step 1 Typicality a -.03 .11 -.01 Time b -.49 .11 -.22 *** Credibility c .04 .11 .02 Step 2 Typicality -.02 .11 -.01 Time -.50 .11 -.22 *** Credibility -.17 .12 -.08 Involvement .09 .04 .12 * Perceived Credibility .18 .04 .23 *** Note . Step 1, R 2 = .05 *** ; Step 2, R 2 = .06 *** . a Typical = 0, Atypical = 1; b Immediate = 0, Delayed = 1; c Low-credibility = 0, High credib ility = 1. The credibility index was the mean score for each participantÂ’s credibility ratings on six 7-point semantic differential scales. The involvement index wa s the means for two i nvolvement measures: Â“how important is IEÂ’s secu rity to you personally?Â” (value-relevant importance) and Â“how concerned are you about IEÂ’s securi ty?Â” (outcome-releva nt involvement). R squared = .13 (Adj. R squared = .10). *** p < .001, * p < .05 Not surprisingly, time was negative ly associated with attitudes, = -.22, p < .001. Results of the final model indicated th at both covariates of involvement ( = .12, p < .05) and perceived credibility ( = .23, p < .001) were positively associated with attitudes. Although involvement and perceive d credibility were significant predictors of the attitude
151 change, the effects of experimental factors ha d little change in size and direction in the step 2 of the hierarchical model. These results are consistent with fi ndings in the previous analysis of covariance reported in Table 33. Discussions Discussions about the SC + T Model Cognitive psychologists have long argued i ndividuals form abstract but general knowledge about everything they experience including actions, places, situations, and issues. This general knowledge is calle d Â“schema,Â” whereby lay people make an automatic judgment about what is Â“typical Â” and Â“atypicalÂ” of a specific object. The typicality about an object in fluences not only what we re member and forget over time with regard to the object, but it also influe nces how we interpret, evaluate, and behave observing each of either typical or atypical things. To invest igate the role of schema in the audienceÂ’s remembering and forming of attitudes toward an endorsement message, this study tested the SC + T model of a sc hema theory and the sleeper effect of longitudinal attitude change. The research reported in this study contri buted to the SC + T model by examining peopleÂ’s recognition performance for persuasive messages with typical ity variation and a time delay in recognition tests. Most studies that tested the SC + T model seldom allowed for a time delay longer than a week. An important prediction of Graesser and his colleaguesÂ’ SC + T model lies in the reve rsed recall performan ce over time Â– that is, clearly after a one-week delay. However, the memory accuracy was uniformly higher for atypical than for typical items even with a three-week delay (Smith & Graesser, 1981). While there is a great deal of research that tested the SC + T model, there has been relatively little research that has tested the model after a two-week delay. Thus,
152 researchers who only allowed for a short-term delay may not give a correct conclusion regarding recognition memory performance w ith the passage of time. For instance, Shapiro and Fox (2002) reported that they f ound participantsÂ’ recognition sensitivity was still quite good for atypical information from media stories, but their findings were based on a two-day delay. In the test of H1a and H1b, it was found that participantsÂ’ immediate recognition performance for typical and atypical arguments was not statistically different. However, the memory sensitivity was less decayed for typical than for atypical arguments after a two-week delay, and this result was stat istically significant. The reversed higher d scores for atypical items after a two-week delay indicated that recognition performance of presented atypical items dete riorated rapidly. Although the hypothesis that predicted better recognition accuracy fo r atypical arguments in a delayed test over typical arguments was rejected, this finding is consis tent with the general assumption of the SC + T model; that is, memory for atypical items te nds to decay faster than for typical ones. Then it might be controversial that the main ef fect for typicality was projected in the first hypothesis relying on previous findings of ear lier research. Indeed, as mentioned in Chapter 2, previous studies that found bette r recognition performance for atypical items in a delayed test were mostly based on a s hort delay of two days or at best one week except that of Smith and Graesser (1981) w ho tested memory after three weeksÂ’ delay. Due to this lack of research on the time-dela yed effect of a persuasive message on a longterm memory, Hunt et al. (1992) once summarized the state of art in this area of research as follows: A stimulus is thought to be encoded, st ored, and retrieved in terms of some activated schema in the personÂ’s memory: For example, whether stimulus
153 information is schema-congruent or sc hema-incongruent is highly significant, because incongruency conveys distinctiven ess and salience and therefore greater memorability at least immediately. A lthough schematic-processing models agree on this much, they differ on the long-term memorability of unusual (incongruent) information. One school of thought argue s that incongruent information is particularly forgettable over time; anot her contends that it is particularly memorable (Hunt et al., 1992). Findings in this present study thus may impl y the interaction effect of typicality and retention interval in both reca ll and recognition of persuasive messages. In other words, the audience is likely to forget the atypical arguments more than they forget the typical argument over time, although they recall and recognize those atypica l arguments better than typical arguments instantaneously. Howeve r, these findings need to be interpreted with caution, because d scores were negative in the de layed test. Singh and Churchill, (1986) once argued that nega tivity in sensitivity ( d' < 0) might be due to either measurement error or subjects giving contra ry responses on purposes. In other words, some participants might perform the disc rimination task giving a contrary response intentionally, that is, saying Â“noÂ” when they should have said Â“yesÂ” (and vice versa) on the basis of discrimination ability (Singh & Churchill, 1986). This could happen when participants were able to perform the discrimination task, but they were malingering. However, considering that par ticipants were randomly assi gned to different cells of experimental groups, this possi bility should be opted out. If it had resulted from data contamination by some subjects on purposes, such negative d' scores also would have been observed in the immediate tests other than in the delayed tests. Rather, these negative d' scores imply different possibilities. First, as seen in the post-hoc analyses, the negative mean d' score for atypical sentences was due to a rapid drop of HR for atypical sentences with increased FAR at a
154 delay. These finding may imply that memory for atypical items decayed faster than typical ones, supporting the main assumption of the SC + T model. In earlier studies of the SC + T, this assumption about the faster decay for atypical it ems than for typical items was supported only with a recall performance. Now, findings in this study show that this assumption could be applied to the recognition memory performance. Second, findings in the test of H2b were cons istent with those in the test of H3. In other words, the decrease of recognition perf ormance over time was greater for atypical arguments than for typical arguments. Nonetheless, due to the negativity of d' scores, it is hard to conclude that typical arguments were better recognized over ti me, although HR for typical sentences was significantly higher than fo r atypical ones. Even though there existed significant difference for sensitivity, the mean d' score for typical items was close to zero in the delayed test, implying a possibility of the fl oor effect whereby participants had not accurately recognized the presented items from non-presented items. As mentioned earlier, however, partic ipantsÂ’ recognition memory for atypical arguments decayed faster than for typical ones, showing significant difference in HR between typical and atypical arguments over time. In conclusion, the results reported in the test of H1 and H3 suggest there may be an interaction effect of retention intervals and typicality on recognition memory. A strong effect observed in th is study was the main effect of retention interval on sensitivity. In the test of hypothesis 2, ther e was a significant main effect of time of measure on recognition accuracy, F (1, 392) = 50.6, p < .001. When excluding the possibility of participants Â’ giving contrary responses on purpose, the negative mean d
155 score (-.49) after a delay indi cates that particip ants answered more Â“yesÂ” (i.e., having seen) responses to Â“non-presentedÂ” arguments, while they did more Â“noÂ” (i.e., not having seen) responses to Â“presentedÂ” persuasive arguments. Although there is a possibility that the foils (non-presented arguments in the lear ning phase) were too typical to discriminate, the significant difference [ F (1, 397) = 29.67, p < .001] of the false alarm rates between immediate and delayed tests may be a counterargument for such claim. The result of testing H3 showed the in teraction between retention interval and typicality. A test of two-way ANOVA produ ced a significant interaction effect [ F (1, 395) = 9.54, p < .01] along with the main effect of re tention interval. Ho wever, the effect size was significantly higher for the main effect: that is, using p 2 as the measure of association with the dependent variable, the interaction between typicality and retention interval accounted for only 2% of the total variability in th e recognition sensitivity, while the effect size of the main effect acco unted for 12% of the total variance. While presenting the memory performance over time, it was also proposed that participantsÂ’ confidence ratings for typical ar guments would be more liberal over time, allowing for more Â“SureÂ” answers influenced by the generic schema they had about the issue in discussion. Findings obtained in test s of H4a and H4b are consistent with those found in a test of H3. In the mean comparis ons, atypical arguments instantaneously had a slight advantage over typical arguments in recognition perfor mance, but this pattern was reversed after a delay. Findings reported in tests of H4a indicate that this memory advantage for atypical arguments at an immedi ate test also comes with more strength in confidence for correct recogni tion judgments for the presen ted atypical arguments. On the other hand, the confidence for recogniti on judgments about atypical arguments
156 dropped over time more quickly than did conf idence about typical recognition judgments. In this process, confidence of recognition judgments about typical arguments was less reduced than for confidence of recognition judgments about atypical arguments. The findings support the assumption of criterion bias in peopleÂ’ s recognition judgments. In other words, people become more liberal in making thei r recognition judgments about typical items over time, while they tend to be more conservative in their judgments about atypical items. What could be explained from the results obtained in the tests of hypotheses 1 to 4? First, recognition performance for atypical ar guments tends to decay faster than for typical sentences. It is the main assump tion of the SC + T model that memory performance for atypical items drops more quick ly over time than for typical ones. This principle was formerly applied only to recall pe rformance. In the lack of replications of the original SC + T model in a long-term delay, it is still an issue as to whether initial advantage for atypical items will be reversed after at least two weeksÂ’ delay. Results reported in this study can be used for future replications or meta-analyses on the SC + T model. Second, retention interval is the most influential vari able in determining the memory performance. That is, recognition performance was much influenced by the delay of the recognition measure. The interac tion effect between typicality and retention interval was also signif icant, although its effect size was very low. The confidence ratings were most influen ced by the time of measure, showing significantly more Â“SureÂ” responses in imme diate than in delayed tests. However, confidence for recognition judgments also reve aled a significant inte raction effect. There
157 were more Â“sureÂ” responses for atypical than for typical argument s when recognition was measured immediately. However, Â“sureÂ” res ponses for atypical arguments were less than for typical arguments after a two-week delay. Discussions about the Sleeper Effect Schemas are fundamental to human comprehe nsion of any type of story. If a story does not conform to schematic expectations, it could cause some degree of cognitive disturbance (Plowman, 1996).The degree of schematic congruency between a persuasive message and source information could play a ro le in the dissociation between source cue and a message (dissociation hypothesis) or in the differential decay of source-message information. In this study, it was assumed that the sl eeper effect would be found only when a message conformed to the audienceÂ’s expecta tions about the source of the message. If the message violated the audienceÂ’s expectati on about the source of the message, he/she could remember the source of such incongruent information as a result of their efforts to solve the incongruency during the inform ation processing. A schema-incongruent situation in which atypical argum ents were told by a highly-credible source was expected to reduce the possibili ty of source-memory dissociati on or differential decay. Since a sleeper effect is maximized when the sour ce cue and message is dissociated or poorly integrated, it was hypothesized that the sleeper effect would be only found when the endorsement message was mainly typical in which people would have less motivation to think over why the source made such arguments. Results in the tests of H7a and H7b reveal ed that contrary to this hypothesis, the sleeper effect was found only when the message was atypical. The mean attitude change induced by an atypical endorsement message from a highly-credible endorser dropped
158 significantly over time, while there was no ch ange of attitudes among participants who received the same atypical message in th e less-credible endorse r group. In a post-hoc analysis, a three-way analysis of variance also indicated that a credibility-by-time interaction would not occur unless typicality was considered. Neither credibility nor typicality had an impact on attitude cha nge in the 3-way analysis of variance. It is hard to explain why the sleeper effect was only observed in the atypical message condition. In this study, it was assume d that people would remember better for an atypical message told by a highly-credible source even after a laps e of time. If they remember the atypical message with the relia ble source, the atypical message should lead to more positive attitude change than was projected by a less credible source. This was not the case in the tests of H7a and H7b. A possible reason for this contradictory finding is that the credibility was not important, par ticularly when attitudes were measured after a delay. In other words, participantsÂ’ attitudes toward Firefox after a delay may have been influenced by message typicality rather than endorser credibility when the message was atypical. One interesting question is why differe nt credibility cues did not produce a statistically significant difference in attit ude change. The failur e to find a significant difference in attitude change by credibility does not appear to be a result of the failure of credibility manipulation. As s hown in Chapter 5, the credibili ty manipulation resulted in significant differences in impression evaluations for all six 7-point semantic differential scales. In addition, the CronbachÂ’s alpha for reliability measure was high. Nonetheless, the means for evaluations about a low-credibility endor ser showed that pa rticipants in the low-credibility condition tended to rate the endorser more as neutral than as low in
159 credibility. Thus, a perceive d credibility index was created based on participantsÂ’ evaluations about the endorser. Importantly, a post-hoc anal ysis with this perceived credibility index showed that it was a strong predictor for the attitude ch ange in the hierarchical regression model. In the post-hoc analysis of H 5, it was also found that three levels of perceived credibility made a si gnificant difference in forming postattitudes toward Firefox. Therefore, findings regarding cr edibility effect in this dissertation should be interpreted with caution. Another plausible explanation for not obs erving credibilityÂ’s treatment effect comes from the ElaborationLikelihood Model (ELM). The ELM assumes that a message recipient will be more likely to be influe nced by peripheral cues when his/her issue involvement is low. On the other hand, when a person is highly involved in the issue, he or she will likely scrutinize the message itself rather than the source. In other words, the perceived credibility will exert more influen ce on attitude change in a low-involvement group than in a high-involvement group. In this regard, it could be worthwhile to discuss a relevance override , the reduced impact of cues under high processing motiv ation of persuasion experiments (Pierro, Mannetti, Kruglanski, & Sleeth-Keppler, 2004) . This relevance override hypothesis is a cousin of the cue-weighting hypot hesis proposed by Petty (1994). The cue weighting hypothesis [italics added] assumes that the peripheral cues have relatively little impact on attitudes under high elaboration conditions because when people are highly motivated and able to process all relevant information, although they are aware of the cue, they do not c onsider it particularly relevant in making their evaluative judgments. That is Â… wh en motivation and ability are high Â… the cues are in essence discounted as irrelevant at the time of attitude judgment. (Petty, 1994, p. 234). Then, it could be a limitation of this st udy that no hypothesis about the involvement factor was proposed. From the findings of this study, it can be assumed that the sleeper
160 effect may not be observed unless individual differences (e.g., motivation, involvement, prior experience) are considered. In this regar d, it is an important finding that a relative sleeper effect was observed under the atypica l message condition. This finding may be interpreted in the assumption of the ELM. If the credibility of the source was not so important to the message recipients forming their attitude toward Firefox, then it was possible that the typicality of message wa s a critical criterion to make evaluative judgment especially after a delay. When the information about the credibility was quite available in working memory, then, attitude change was subject to perceived credibility. However, when the source cue was less availabl e after a lapse of time, message recipients were more likely to rate thei r evaluations based on their me ssage typicality. This may be the reason that highly -credible endorsers w ho made atypical arguments induced more attitude change in an immediate test, while their arguments were negatively evaluated in a delayed test group (a 3-way interaction). In earlier research, it was noted that even a highly-credible source could have no added effect on persuasion if the message was fully accepted on its mer its (Hovland et al., 1953). If this is true, findings reported in this study could be e xplained by Hovland and his colleaguesÂ’ hypothesis: Under conditions where positive attitudes to ward the source are very strong (e.g., he is considered as highly credible) and where the tendenc y to reject the conclusions and arguments in the communica tion is very strong, there is a tendency to dissociate the source and c ontent (Hovland et al., 1953, p.41). Thus, findings in this dissertation highlight the need for furthe r investigations of how different contexts of message typicality a nd source credibility affe ct attitude change instantaneously or after a twoweek delay. As a rudimentary analysis this study examined whether there was a significan t impact of involvement on pa rticipants' attitudes over the
161 course of the experimental manipulation. To do so, an analysis of covariance was conducted by entering involvement as one of covariates, followed by a hierarchical multiple regression analysis. Findings in th e ANCOVA and the hierarchical regression showed that the involvement was a signifi cant contributor to participantsÂ’ forming attitudes toward Firefox. Futu re research should investigate how different levels of involvement have an impact on attitude change when message typicality, source credibility, and time were varied. Limitations Recall/Recognition A methodological limitation is re lated to the memory test. In experiments reported in this study, recall was excluded for seve ral reasons. First, a recall test was not considered because a recall test before a r ecognition test might influence subsequent recognition test scores. Second, it was thought that correct recall might not be obtained without direct intervie ws with participants. If participants saw a political ad with few arguments, it would be meaningful to request pa rticipants to recall what they had seen in the 30-second ad. However, in this study, participants read a lengthy endorsement message that was controlled to be shown for at least two minutes. In a laboratory setting, requesting participants to write down what th ey recalled from message reading is not an easy job unless it is related to an official test. Third, the addition of an experimental condition would increase the number of between-groups from eight to 16. As mentioned earlier, previous studies have shown that recall performance was better for atypical items than fo r typical after a shor t delay, but this pattern was reversed after a two-week delay. In this study, it was reported that recognition accuracy was
162 initially better for atypical than for typi cal, but it deteriorated in a two-week delay. However, no comparison could be done without recall data. External Validity This study was conducted in a laboratory se tting, which offered high levels of experimental control and internal validity that are difficult to achieve in the field. However, the use of an experimental simu lation to test hypotheses raises the obvious question as to whether or not these results can be generalized to other, more natural settings and other sources in which the specifi city of the persuasive message will interact with other features of the source and the situation. Untested Underlying Assumptions Hypotheses to test the sleeper effect in this dissertation were based on three theoretical assumptions or hypotheses such as a source-memory differential decay hypothesis, an expectation-violation hypothesi s, and source-message congruency effect. Though these assumptions/hypotheses underlie the main theoretical reasoning for the sleeper effect hypotheses (H7a H7c), they we re not directly tested. Future research should measure the degree of source-message disso ciation or decay. It is also up to future research to test how different levels of source-message cong ruency affect persuasion over time, especially in the varia tions of message typicality. Typicality of Foils The memory test revealed that partic ipants in the delayed group had poorer recognition accuracy than counterparts in an immediate test group. This poorer sensitivity was due to increased false alarms and decr eased hits at the delayed test. Although foil items were selected from sentences that were cl ose to neutral scores from the pretest, it is possible that participants in a delayed test may have recognized those foil items as having
163 been seen before because they looked familiar. Indeed, there was one case in which a non-presented sentence received much more Â“yesÂ” responses than Â“noÂ” responses in both immediate and delayed tests. The senten ce of Â“With tabbed browsing, you can open many Web pages at once in the same browse r windowÂ” was indeed presented as a foil item, but it resulted in very high false al arms both in typical (161 false alarm vs. 37 correct recognition) and in atyp ical (167 FA vs. 37 CR) tests. The above foil is indeed a bad inappropriate choice as a foil because both the typical and atypical version of the endorsement includes a sentence3 that is very close to it in meaning. This may contaminate the results of signal detection task. Individual Differences (Motivat ions and Prior Attitudes) With regard to attitude change, current research did not consider the motivational factors (e.g., involvement, need for cognition) of message recipients in the analysis. This study only considered the message and the source factors in the persuasion process. As mentioned in discussion, it turned out that credibility did not make any difference in attitude change regard less of the time of measure. This failure in obtaining the main effect of credibility indicates that the premise of the current research may be revised in terms of ELM. In other words, examining th e sleeper effect without consideration of individual motivation (or involvement) is grounded in a learning-paradigm of attitude change that presumes attitude change occu rs purely as a consequence of learning or remembering of a message. However, in the cognitive response paradigm that penetrates the contemporary attitude theories, learning or memory has been considered less critical 3 Â“Wouldn't it be great if you could read papers, do se arches, and check the weather altogether?Â” (typical version); Â“Wouldn't it be great if you could see multiple pictures of dating partners in a single browser window?Â” (atypical version)
164 than an individualÂ’s cognitive responses to the message (Greenwald,1968; Petty, Ostrom, & Brock, 1981; Zuwerlink & Devine, 1996). The Order of Source-Message Presentation Several researchers have ar gued that the discounting cue should be presented after the message has been presented to an audien ce in order to find a sleeper effect. They believe that this procedure would elim inate any forewarning about the possible unreliability of the message source. When th e communicatorÂ’s identity is delayed until after the audience has received the message, the message is apparently heard more nearly in its own terms, without the influence of the communicatorÂ’s cr edibility (O'Keefe, 2002). However, this order of source-message presentation was not considered in this study in two reasons. First, many endorsement messages, particularly those posted on the Web, often present the source information be fore people click on each message so that the audience can verify source information. In this experiment, the source information was presented before the message since it was presented on the Web as a form of endorsement. Second, the addition of an experi mental group also increases the number of experimental groups by double. However, due to the failure in finding a sleeper effect in this present study, not consider ing the order of source-messa ge presentation may limit the results reported here.
165 CHAPTER 7 IMPLICATIONS AND SUGGESTIONS FOR FUTURE RESEARCH The SC + T Model The SC + T model is contrary to the commonsense expectation about the human memory function, because it is difficult for in dividuals to presume that initially-better remembered items will be less available than initially-less remembered items over time. One may argue, Â“If this is true, why do so many advertisers attempt to position their product/service in an advertisement message that is atypical of conventional thoughts related to the particular produc t/service?Â” If atypical info rmation decays faster than typical information, then, there is no reason for persuasive communicators to develop an atypical message in an attempt to make a memorable message. Indeed, this study showed that the memory for persuasive arguments d ecays over time, and the patterns of the decay could be dependent upon the typicality of each argument. With a two-week delay, this study showed that recognition memory for atypi cal arguments was better than for typical arguments right after the message exposure, but this pattern was re versed during a longterm delay (i.e., two weeks). Findings in this study could add empirical data for future research to fill the void left due to the lack of research on memory for typicality of persuasive messages in long-te rm retention intervals. Examining memory for persuasive message requires researchers to consider the effect of source credibility. No research has ever examined the interaction of credibility and typicality on memory performance. In the absence of previous re search, this current study did not propose any hypothesis that relate d to the interaction of typicality and
166 credibility on memory for a persuasive me ssage over time. A supplementary analysis on the mean difference of HR, FAR, and d scores showed that those who were given the high-credibility endorser information with an atypical endorsement message had the highest mean HR and sensitivity during an immediate test, however, counterparts in a delayed group marked the lowest mean HR and sensitivity among all experimental groups. These observations may provide a venue for future research. Future researchers may wish to investigate a thre e-way interaction between typica lity, credibility and time. The Illusion of The Truth Effect Typicality effect on memory has much to do with the illusion of truth effect. The illusion of truth effect refers to peopleÂ’s tendency to rate familiar items truer than new ones (Begg, Anas, & Farinacci, 1992; Skur nik, Yoon, Park, & Schwarz, 2005; Hasher, Goldstein, & Toppino, 1977; Hawkins & Stephen, 1992). It is a serious problem in this information-overloaded age in which consumer s are often bombarded with a barrage of information from diverse sources. At an ini tial learning or encoding stage of information, consumers may be vigilant in determining who their source of information is. When consumers are exposed to an important claim Â– for instance, a medical claim Â– they try to determine the veracity of the claim. They might trust the claim less if they remember reading it from a suspicious Web site than if they remember hearing it as advice from their physician. However, once the contextual details are lost, the remaining feeling of familiarity may foster the acceptance of false cl aims as true, which is the case in Skurnik et al.Â’s (2005) experiment. This offers an important implication to health communication researchers as well as consumer research ers who examine the harmful effects of misleading information on consumersÂ’ memory and attitudes. Previously, research on the
167 illusion of truth effect was conducted on message s with repetitions without any typicality variations. Future researchers may wish to apply the SC + T model to the illusion of truth effect, whereby the context of credibility and typicality combination increases or decreases the effect of illusion of truth. Although it is not an illusion of truth effect, this familiarity-driven confusion also could occur in consumersÂ’ remembering of advertising messages. Consumers may falsely believe they had seen an ad for the established brand when they in fact had seen an ad for a new brand, forgetting they saw the new bra nd ad when both the established and the new brand ad are delivered in a similar way. Or they perhaps may remember seeing the established brandÂ’s ad when no such ad ha d appeared but they were exposed to the advertisement in another context (Braun-LaTour and LaTour, 2004). Sleeper Effect While previous studies on persuasive co mmunications often s ought the effect of attitude on memory, this study examines the oppos ite direction of an effect; that is, the effects of episodic memory on attitude. In Hovland's paradigm of learning-based attitude change, the persuasive effect of campaigns starts becausee of the message exposure and message attention to persuasive messages. In this process, a message recipient is assumed to try to know who the source of messages is. Naturally, researchers in the early stage of persuasion research paid special attention to the effect of communicators or message sources. In this line of paradigm, researcher s often considered source credibility as the most important quality that influences the e ffectiveness of persuasi on (Newsom & Carrell, 1990). The decay of source memory was often assumed to influence the attitude change over time independently of me ssage characteristics. Findings of this study raise two issues, however. Credibility of a source may not be an important factor in attitude change.
168 Second, persuasive effects may vary over time not due to one's source memory decay but due to the decay of memory for a message. As for the first issue, a contemporary dual processing model of attitude change assumed that the source cue plays a role in persuasion when the audience's motivation to process the message is low. People with a lo w level of need for cognition (NC) tend not to elaborate the message as much as counter parts with a high level of NC. When the NC is low, recipients are more likely to use the source credibility cue as a basis for attitudinal evaluations. Then the ELM may explain why s ource cue was not an important factor that determined participants' attit ude change in this present st udy. In this regard, O'Keefe's summary offers an important discussion for future research on the sleeper effect: The Elaboration Likelihood Model pr ovides a plausible framework for understanding at least some of the resear ch evidence concerning the persistence of persuasion. Broadly put, the ELM sugge sts that persuasion achieved under conditions of high elaborati on (central processing) is more likely to be enduring than that obtained under conditions of low elaboration (periphera l route processing). That is, persistent effects are more lik ely when persuasion is the result of thoughtful consideration of issues and ar guments as opposed to the results of reliance on heuristics (O'Keefe, 2002, p.258). Therefore, future researchers could a dvance current research by considering different levels of involveme nt or motivation to process the message. Involvement may moderate the effect of source credibility a nd message typicality on persuasion. For issues of little personal relevance, receivers may be content to let their opinions be shaped by the communicatorÂ’s apparent credibility; for su ch an issue, it is not worth the effort to follow the details of the arguments. For highl y relevant topics, re ceivers may be more likely to attend closely to the details of the message, to scrutinize the communicatorÂ’s arguments and evidence, and to invest the e ffort involved in thi nking closely about the contents of the message Â– and that compar atively greater importance of the message
169 contents means that the communicatorÂ’s credib ility will play a smaller role than it otherwise might have. Practical Implications How to prepare persuasive messages and how to convey them effectively to the media and other publics who have a stake on the raised issue is the threshold challenge of effective public relations. The success or fa ilure of a public relations campaign often hinges on how the source of communication is perceived by the intended audience. Although a contemporary theory of attitude change (e.g., Elaboration Likelihood Model) considers source credibility as a Â“periphe ral cueÂ” in the persuasion process, source credibility has been treated as more than a simple peripheral cue in contemporary communication research and the communication industry. Especially, the field of public relations, whose communication goal accomplishment is very dependent upon achieving credibility, has often grounded its existence on the general theo ry of source credibility sources which are perceived as more credible should be more influential than lesscredible sources. (Broom & Dozier, 1990; Reardon, 1981; Cutlip, Center, & Broom, 2000). However, Callison and Zillman (2002) argue d that persuasive messages from even the low-credibility source could be perceived to be effective after the passage of time. The sleeper effect offers a great implication for public relations pr actitioner who want to launch a low-budget public relations campaign that does not feature a high-profile and expensive celebrity like Oprah Winfrey or Mi chael Jordan. The imp act of highly-visible and/or credible endorsers in achieving communication goals does not say too much. Especially, at a time when traditional media were the only outlet for delivering organizational messages, any campaign c ould not succeed without featuring a high-
170 credibility source/endorser. Now that most or ganizations have reali zed that they could reach their targeted publics/consumers w ith a relatively lower budget through the Web, credibility may not be as critical an issue as it used to be. An organization can build an outlet on the Web for delivering its message, a nd it can elicit site visits from people through diverse public relations practices. The message can be told by ordinary people as a testimonial or in a narrativ e story. It does not need to be a formal type of public relations message like a press release. An i ssue is whether or not those ordinary voices can have the same effect in achieving co mmunication goals and obj ectives as credible sources do. This study shows that it is not a matter of credibility that determ ines the audienceÂ’s attitude change, but a matter of time intervals and typica lity. Immediately after the message exposure, people were more influen ced by a highly-credible endorser, but the highly-credible endorser was less effective th an a low credible counterpart when the message was atypical. It implies that the me ssage typicality should be an important consideration for a campaign f eaturing ordinary voices. Regardless of credibility, people tend to be more influenced by a typical message during an immediate test. However, an atypi cal message leads a slight advantage in persuasion over a typical message. If both t ypical and atypical messages could make an impact instantaneously and after a delay, publ ic relations practitioners might be more inclined to send an atypical message for cons ideration of a long-term effect. One notable finding from the current research is that the atypical message from a highly-credible endorser not only marked the lowest score in memory recognition after a delay but it also resulted in less attitude change than the atyp ical message from a low-credibility endorser.
171 No conclusion can be drawn from the findings as to whether or not th e lowest recognition performance for the atypical endorsement by a highly-credible endorser had something to do with the lower attitude change in the delayed test. However, these results may encourage communication profe ssionals to consider featur ing ordinary voices with information as atypical as possible. Indee d, current Web-based campaigns often deliver nontraditional public relations campaigns in terms of messages and sources. From the message point of view, a nontraditional campai gn delivers arguments that are not usual, not frequently heard, and that are irrelevant from or in consistent with laypersonÂ’s established schemas about the issue or topi c. Hence this study offers a practical implication for communicati on professionals to develop memorable atypical messages and to be confident in featuring ordinary voices. For instance, in a CSR campaign recently appeared in Business Week . BP, a global energy grou p, advocated the global warming agenda with denigrating anti-g lobal warming groups, followed by a message about BPÂ’s CSR activities. A brief quote featuring a gra duate student named Paul Lakosky reads: Â“I think anyone that says global warming is not real is kind of burying their head in the sand.Â” (see Business Week , October 24, 2005, p.21). Judging from the findings of the current research, BPÂ’s public relations campaign could be effective in the long run because an atypical argument was presented by an ordinary voice (i.e., a graduate student).
172 APPENDIX A STIMULUS: LOW CREDIBILITY Walt S. Morse is one of the Petereus blogge rs. Petereus is a company that is paying bloggers $800 a month to blog their sponsors' pr oducts or services. The content that runs on these web pages won't be marked as adver tising, it will be editori al comment, right in the body of the blog itself. Bloggers are paid to write and report fr eely about their own views on any product that spons ors Petereus. Mr. Morse uses his blog, Junk Drawer, to put up links to music videos, pop culture , video games, and Japanese manga. He grew up in the suburbs of Los Angele s. After graduating from the Webb School of California, he went to the University of Portland, located in Portland, OR. After graduating from UP with a degree in philos ophy (dude!) he went back to California to watch movies for a living (no, really: He was the movie critic for the Fresno Bee newspaper). He stayed at the job long enough to propose to Laura Dern and to get married. In 1996, he left California to go to work for AOL. Many networked games of Quake followed. In 1998 he went freelance a nd now spends his time writing for anyone who pays him. He started blogging three year s ago, mostly to kill time after he quit the job at AOL. When he's not goofing off online, Wa lt Morse writes a book whose upcoming title is Book of the Dumb (a book on stupidity). He also reviews stu ff for Official US PlayStation Magazine. Now, Walt Morse lives in a teeny tiny little town in Ohio called Bradford, where a traffic jam is defined as three cars caught behind an Amish buggy.
173 Walt Morse lives there with his wife Kristine, dog Kodi and cats Rex, Lopsided Cat, and Ghlaghghee (pronounced "fluffy"). He enjoys pie. Disclosure: Mr. Morse is paid by Petereus to comment on Firefox.
174 APPENDIX B STIMULUS: HIGH CREDIBILITY Walt S. Morse is the author and creator of the weekly technology review column "What's Next" in The New York Times, whic h has appeared every Thursday since 1991. Mr. Morse has been known as a fiery crus ader since the opening line of his column "What's Next": "Personal computers are just t oo hard to use, and it isn't your fault." His quest has earned him legions of fans. His success comes from demystifying the digital realm for readers who aren't regular Slashdot contributors. He's on a mission to remake the tech world according to his own fetish for simplicity, reliability, effectiv eness, and great design. From its launch, Morse knew exactly what he wanted: "What's Next would be utterl y different from the reviews already out there. They were by geeks for geeks, filled with jargon, condescending to nontechies, and reverentia l about the computer and the computer industry," he says. "I wanted to write for the nontech user and be critical of the industry." Naturally, some companies try to use their meetings with him to glean intelligence they can incorporate into products that he'll later review. As a response, he sent a mass email to the publicists who pitch him: "I am not a consultant or a dviser for you. I am not part of your development process. Even when I receive prerelease software or hardware or an advance look at a site or service, I am not a tester. If I make opinionated comments or ask pointed questions during a meeting or by em ail or phone while tryi ng out products, it is strictly to elicit information from you for my own journalistic purposes . I typically try to ask the questions and offer the reac tions I imagine my readers would.
175 Yahoo! Cofounder Jerry Yang says, "He's one of the most trusted and influential voices in technology." Newsweek magazine calls Mr. Morse "the most powerful arbiter of consumer tastes in the computer world to day," TheWall Street J ournal calls him "the most respected computer journalist." Mr. Morse was awarded the 1999 Loeb award for Commentary, the only technol ogy writer to be so honored. Mr. Morse is based in the Time's Washi ngton, D.C., office, where he spent 18 years covering national and internati onal affairs before turning his attention to technology. A native of Warwick, Rhode Island, he gradua ted from Princeton University and the Columbia University Graduate School of Journalism.
177 APPENDIX C STIMULUS: TYPICAL MESSAGE Security, Cool Features Of Firef ox Web Browser Beat Microsoft's IE By Walt S. Morse Netscape's removal from the center stage of the Web browser competition turned out to be a bad thing for Microsoft's Internet Explorer (IE). With a nearly $8 billion R&D budget, why hasn't Microsoft's IE substantia lly advanced since the end of the browser wars? Keep in mind that Microsoft hasn't made any important functional improvements in Internet Explorer for years --asid e from opening additional security holes. In the meantime, other people have been building much better browsers. The most significant of these challengers is Firefox. Firefox is built and distributed free by the Mozilla Organization, a small nonprofit corporat ion spun off last year from the fastfading remnants of Netscape. But in the two years since Firefox debuted as a minimal, browser-only offshoot of those sprawling suites , it has grown into a remarkable product. There is nothing wrong in keeping IE as a regular backup, since it is entrenched in our normal internet usage. However, you are ad vised to surf with discretion because it may endanger your privacy and security. Over time, people have come to agree with me, now more than ever before.There's an important reason why you should dump Microsoft's Internet Explorer Web browser. Th e problem is that hackers continue to find and exploit security holes in Explorer. Ma ny of them take advantage of Explorer's ActiveX system. That's why the U.S. Computer Emergency Readiness Team, a
178 partnership between the tech industry a nd Homeland Security, recently took the unusual step of advising people to consider switching browsers. Just after Firefox version 0.9 became avai lable for download, yet another Trojan horse began harassing Microsoft Windows user s en masse. The Trojan horse exploited a close tie between Internet E xplorer and Microsoft Windows. The Web site of The Kelley Blue Book, an online automobile pricing guide , had to remove the malicious code from its site and now encourages users to switch to Firefox to avoid such a risk. Firefox eschews ActiveX and other well-known in fection paths. You can configure it to automatically download most files when you click on them, but not .exe files, which are runnable programs. Firefox's security successes can be credited to Mozilla's open-source development model. Since it is open source, the software is guaranteed to not have intended malicious code or spyware included. On the ohther hand, all Microsoft is able to offer its Internet Explorer users are incremental security improvements; new patc hes to fix holes in the old patches. We shouldn't have too much confidence in Gates and Co. Seeing Microsoft piling on supposedly "cool" features without regards to the consequences gives me the impression that their programmers are completely clue less about how things work in the "real world." While interviewing Bill Gates fo r "ABC World News Tonight" in 2004, Peter Jennings asked Gates whether it is fair to say th at Firefox's perceived ab ility is better than IE. Instead of directly referring to Firef ox by name, Gates acknowledged that competition is pushing IE. He said "We have got some fa ntastic competitors and it keeps us on our toes."
179 I shouldn't give you the impression that Fire fox is good just because it is safe for my computer and created by a bunch of good pe ople. User experience is what makes or breaks a browser. In fact, the best feature of Firefox is the tabbed browsing, which will become a quick favorite if you've never used it before. Wouldn't it be great if you could read papers, do searches, and check the weat her altogether? Web pa ges are loaded in "tabs" within the same browser window, ma king it easy to switch back and forth among multiple web pages. Each window is marked by named tabs that you can click in and out of easily, allowing you to neatly keep track of them. You probably also like it that Firefox take s care of one of your major annoyances in life. On Firefox, pop-up ads are easily blocked. I've noticed that the program tells me when it has blocked an ad on a particular s ite, though now and then one still gets through. It's imperfect, but better than Explorer. Getting the same kind of protection in your Explorer requires expensive add-on software created by third parties. Microsoft's Explorer may indeed still command the vast majority of the Web browser market, but a new battle has begun a nd the browser war is on again. Microsoft should be watching Mozilla carefu lly. Firefox will not be defe ated as swiftly as Netscape was 10 years ago by Microsoft's IE. The game is now much different. Microsoft may have amassed millions upon millions of dollars but Mozilla has the vast and vibrant opensource army on its side. Give Firefox a da y's worth of Web surfing, and you will never look back on IE.
181 APPENDIX D STIMULUS: ATYPICAL MESSAGE Security, Cool Features Of Firef ox Web Browser Beat Microsoft's IE By Walt S. Morse Netscape's removal from the center stage of the Web browser competition turned out to be a bad thing for Microsoft's Internet Explorer (IE). Like the US space program in which NASA hasn't made any major advancemen ts since the fall of the Soviet Union, Microsoft has not released any major upgrades for Internet Explorer since its defeat of Netscape. In spite of a near ly $8 billion R & D budget, Micr osoft's slow innovation could get listed in the Guinness Book of World Records. In the meantime, other people have been building much fancier browsers. The most significant of these challengers is Fi refox, a free product from an open-source organization called Mozilla. The name Fire fox is a common name for the red panda which eats large amounts of bamboo. In th e two years since Firefox debuted as a minimal, browser-only offshoot of those sp rawling suites like bamboo, it has grown into a remarkable product. There is certainly nothing wrong with ha ving IE as a backup. We know your boss might be stubborn about sticking to IE. But, to be blunt, he or she should be fired for stupidity if they do that because they are endangering the company. It's purely unthinkable for anyone in the business world to want to stick with IE. There's an important reason why I finally dumped Internet Explorer in favor of the Firefox Web browser. The problem is that the IE began spontaneously opening up to a dozen windows
182 displaying triple-X-rated ads, just for misspel ling a URL. History show s there are lots of security bugs in IE that allo w scripts to edit system file s without explicit permission. That's why The Christian Hackers' Asso ciation, a missionary organization for evangelizing the hacker underground, recently took the unusual step of advising people to dump the Internet Explorer. Just after the Firefox 0.9 version became available for download, a Trojan horse began to harass Frank Scheelen, manager of th e porn-specific search engine Ask Jolene. The Trojan horse exploited a close tie betw een porn galleries and user curiosity. To minimize users' headaches, caused by porn sites th at try to slip a Trojan horse into users' home computers, the site ha s taken to endorsing Firefox, o ffering a direct link to the Mozilla Foundation download page. Firefox's security successes can be cred ited to Mozilla's open-source development model. Since it is open source, the software can be protec ted by international hackers who are sympathetic to the open source community. On the other hand, all Microsoft can offer In ternet Explorer users is a promise that users will enjoy a wonderfully rich Web experience with Longhorn, Microsoft's codename for the next version of its Windows operating system, if only they'll wait just a little longer. We shouldn't have too much confidence in Gates and Co. Watching Microsoft pile on "cool" feat ures without regards to the consequences gives me the impression that their average programmer is about 7 years old. While interviewing Bill Gates for "Eye to Eye" in 2004, Connie Chung likened Gates' business tactics to knifefighting. Gates was famously intolerant of in terviewers who asked questions he deemed irrelevant or just plain stupid. Gates t ook off his mike and walked off the set.
183 I shouldn't give you the impression that Fire fox is good just because it is safe for my computer and created by a bunch of good pe ople. User experience is what makes or breaks a browser. In fact, the best feature of Firefox is the tabbed browsing, which will become a quick favorite if you've never used it before. Wouldn't it be great if you could see multiple pictures of dating partners in a single browser window? Recently, former Presient Clinton was shown using the feature to organize and ogle his different webpages on crossword puzzles, saxophone, and golf all within a single window. If Bill Clinton could figure this stuff out, easil y clicking from tab to tab to check out his favorite things, surely no one else will have any problems! Bill Clinton probably also likes it that Firefox takes care of one of his major annoyances in life. On Firefox, pop-up ads are easily blocked. And to protect the kids, I have noticed that this program totally elimin ates the spyware that is sometimes infects your computer through sexually explicit pop-up ads. If you want the same thing in your Explorer, Microsoft and its to adies will charge you extra money, even for their "buggy" and inadequate add-on software. Microsoft's Explorer may indeed still command the vast majority of the Web browser market. But I declare a war against the Internet Explorer and I warn Microsoft to keep an eye on Firefox. But, as Netscape pa infully discovered 10 years ago, Microsoft knows how to fight. Now, the game is differe nt. While the Microsof t Empire has built up billions of dollars, Mozilla has created a spirited community of millions of true believers. The time has come to resist the Empire; break free with Firefox!.
185 APPENDIX E QUESTIONNAIRE
186 2. What is your computer's operating system? If you use two or more operating systems, please choose the one you use most frequently for your personal purposes and check the box. 1) windows 2) Mac OS 3) Linux 4) Others ______________ 3. How many years/months of experience do you have on the internet? _____ years ______ months 4. How many different Web browsers do you usually use? 5. Currently, do you use the Internet Explorer Web browser? 1) yes 2) no 6. Currently, how often do you use each of the following Web browsers for regular Web surfing? Please check a box that indicates your degree of frequency. (1: not at all --: very often) 6.1.Internet Explorer 6.2. AOL/ Netscape 6.3. Firefox 6.4. Safari/Camino 6.5. Opera 7. A blog (a weblog) refers to a Web site that features a personal, distinctive voice, links to other sources and regular postings displayed in reverse chronological order with the newest entry first. To blog is also used as a verb which means to write entries in, add material to, or maintain a weblog. Do you blog? 1) yes 2) no 8. In an average week, roughly how many hours/minutes do you spend reading blogs? ____ hours _______ minutes 9. How often do you think about IE (Internet Explorer)'s security? (1; not at all, 7: very often) 10. How important is IE's security to you personally? (1: not important at all, 7: very important) 11. How concerned are you about IE's security? (1: not concerned at all, 7: very concerned) 12. How confident are you that you can do something to solve IE's security problems for yourself? (1: not confident at all, 7: very confident) 13. Firefox is a Web browser developed by the Mozilla Foundation, a nonprofit group supporting open-source software. Have you ever heard of Firefox? 1) yes 2) not 14. Regardless of whether you have used it or not, please evaluate how you feel about the Firefox browser by checking a box next to the number on each of the lines below. If you have no reaction or if you have never heard of it before, please check the number 0 to indicate your neutrality. How good or bad is Firefox? (-3: bad, 3: good) 15. How useful or not useful is Firefox? (-3: not useful, 3: useful)
187 16. How secure or not secure is Firefox? (-3: not secure, 3: secure) 17. How easy or difficult is Firefox? (-3: difficult, 3: easy) 18. How likable or dislikable is Firefox? (-3: dislikable, 3: likable) 19. How harmful or beneficial is Fire fox? (-3: harmful, 3: beneficial) 20. How old are you? 21. What is your gender? 1) male 2) female 22. What is your undergraduate major? Or, what is your profession? 23. What is your total annual household income ? (Include income for all family members and include all sources). 1) Less than $30,000 2) $30,000 $39,999 3) $40,000 $49,999 4) $50,000 $59,999 5) $60,000 $74,999 6) $75,000 $99,999 7) $100,000 $149,999 8) $150,000 $249,999 9) $250,000 $499,999 10) $500,000 $999,999 11) $1 million or more Now please read the brief bio of Walt Morse on th e next page. You will be asked to give your impressions or feelings about the person later. Once you finish reading the following page, you will see the Next button at the bottom of the page. As a reminder, you will not see the Next button until 30 seconds after the page is loaded. Please feel free to scroll up and down, if needed. [Walt Mossberg Bio] [New Session] 24. Please enter your email address to have secure access to your survey. 25. Based on the background information you just read from Walt Morse's bio on the Web page, please give your first impression or feelings about him on each of the scales below. If you have no reaction to Walt Morse any on a scale, please check the number 0 to indicate your neutrality. First, how qualified or unqualified do you thin k Walt Morse is to inform you about Internet security? (-3: unqualif ied, 3: qualified) 26. How much of an expert or non-expert do you think Walt Morse is to inform you about Internet security? (-3: not expert, 3: expert) 27. How professional or unprofessional do you think Walt Morse is to inform you about Internet security? (-3: unprofessional, 3: professional)
188 28. How trustworthy or untrustworthy do you think Walt Morse is to inform you about Internet security? (-3: untrustwor thy, 3: trustworthy) 29. How believable or unbelievable do you think Walt Morse is to inform you about Internet security? (-3: unbelievable, 3: believable) 30. How reliable or non-reliable do you think Walt Morse is to inform you about Internet security? (-3: not reliable, reliable) Now please read carefully and t horoughly the following technology review written by Walt S. Morse whose brief bio you just read. Please attend to his arguments so that you could deliver his message to your best friend. Once you finish reading the follo wing page, you will see the Next Button at the bottom of the page. As a reminder, you will not see the Next button until one minute after that page is loaded. Please feel free to scroll up and down, if needed. [Typical/Atypical Message] [Post Session] 31. Please enter your email address to have secure access to your survey. 32. Firefox is a Web browser developed by the Mozilla Foundation, a nonprofit group supporting open-source software. Regardless of whether you have used it or not, please evaluate how you feel about the Firefox browser by checking a box next to the number on each of the lines below. If you have no reaction or if you have never heard of it before, please check the number 0 to indicate your neutrality. How good or bad is Firefox? (-3: bad, 3: good) 33. How useful or not useful is Firefox? (-3: not useful, 3: useful) 34. How secure or not secure is Firefox? (-3: not secure, 3: secure) 35. How easy or difficult is Firefox? (-3: difficult, 3: easy) 36. How likable or dislikable is Firefox? (-3: dislikable, 3: likable) 37. How harmful or beneficial is Fire fox? (-3: harmful, 3: beneficial) 38. Based on both the arguments about Firefox that Mr. Morse addressed on the Web page and the background information you read from his bio, please react to Mr. Morse by noting your feelings or impressions toward the author on the scales below. If you have no reaction to the author any on a scale, please check the number 0 to indicate your neutrality. How qualified or unqualified do you think Walt Morse is to inform you about Internet security? (-3: unqualified, 3: qualified) 39. How much of an expert or non-expert do you think Walt Morse is to inform you about Internet security? (-3: not expert, 3: expert) 40. How professional or unprofessional do you think Walt Morse is to inform you about Internet security? (-3: unprofessional, 3: professional) 41. How trustworthy or untrustworthy do you think Walt Morse is to inform you about Internet security? (-3: untrustwor thy, 3: trustworthy)
189 42. How believable or unbelievable do you think Walt Morse is to inform you about Internet security? (-3: unbelievable, 3: believable) 43. How reliable or non-reliable do you think Walt Morse is to inform you about Internet security? (-3: not reliable, reliable) Please evaluate how you feel ab out the messages regarding Firefox Web browswer you read by checking a number on each of the lin es below. If you have no reac tion, please check the number 4 to indicate your neutrality. 44. To what extent did you like the message presented? 1. not at all Â… 7 very much 45. To what extent did you feel th at the arguments were convincing? 1. not at all Â… 7 very much 46. Currently, do you use Firefox browser? 1) yes ( go to 47) 2) no ( go to 49) 47. How likely will you be to try Firefox Web browser in near future? 1) very unlikely Â… 7) very likely 48. How likely will you be to switch to Firefox Web browser in the near future? 1) very unlikely Â… 7) very likely (go to 51) 49. How likely are you to continue usin g the Firefox browser in the future? 1) very unlikely Â… 7) very likely 50. How likely are you to recommend your family or friends to switch to Firefox? 1) very unlikely Â… 7) very likely Among the following sentences, some appeared in the article you read about Firefox and some did not. For each statement below, please check Yes if you re member that the statement was in the article or check No if it was not. For each YES or NO judgment, please indicate how sure or unsure you are of your answer. [Recognition question items for typical Version] 51. Seeing Microsoft piling on supposedly "cool" feat ures without regards to the consequences gives me the impression that their programmers are completely clueless about how things work in the "real world." 1) yes 2) no 52. How confident are you for your answer yes or no? 1) Sure 2) Less Sure 3) Unsure or Guessing 53. Wouldn't it be great if you could see multiple pictures of dating partners in a single browser window?
192 77. Those concerns were validated by the Department of Homeland Security's computer-security team. 79. Clicking it gives you a new bookmark that changes as the content of that site changes. 81. However, to be blunt, he or she should be fired for stupidity because they are endangering the company. 83. With tabbed browsing, you can open many Web pages at once in the same browser window. 85. If you want the same thing in your Explorer, Microsoft and its toadies will charge you extra money, even for their buggy an d inadequate add-on software.
193 APPENDIX F CROSSTABULATION FOR CONFIDENCE RATINGS Immediate-Typical: Hit Miss Sure 35 (76.1 %) 29 (50.9%) Less Sure 9 (19.6 %) 20 (35.1%) Unsure/Guess 2 (4.3%) 8 (14.0 %) Total 46 (100.0%) 57 (100.0%) Recognition test 1 2 =7.243, df = 2, p < .05 Delayed-Typical Sure 19 (26.8%) 9 (34.6%) Less Sure 38 (53.5%) 15 (57.7%) Unsure/Guess 14 (19.7%) 2 (7.7%) Total 71 (100.0%) 26 (100.0%) Recognition test 1 2 =2.14, df = 2, p = .34 Immediate-Typical: Hit Miss Sure 37 (55.2 %) 12 (33.3%) Less Sure 27 (40.3 %) 21 (58.3%) Unsure/Guess 3 (4.5%) 3 (8.3 %) Total 67 (100.0%) 36 (100.0%) Recognition test 4 2 =4.59, df = 2, p = .101 Delayed-Typical Sure 27 (42.2%) 8 (24.2%) Less Sure 28 (43.8%) 13 (39.4%) Unsure/Guess 9 (14.1%) 12 (36.4%) Total 64 (100.0%) 33 (100.0%) Recognition test 4 2 =7.04, df = 2, p < .03
194 Immediate-Typical: Hit Miss Sure 71 (80.7 %) 5 (33.3%) Less Sure 13 (14.8 %) 6 (40.0%) Unsure/Guess 4 (4.5%) 4 (26.7 %) Total 88 (100.0%) 15 (100.0%) Recognition test 7 2 =16.39, df = 2, p < .001 Delayed-Typical Sure 54 (65.1%) 1 (7.7%) Less Sure 16 (19.3%) 5 (38.5%) Unsure/Guess 13 (15.7%) 7 (53.8%) Total 83 (100.0%) 13 (100.0%) Recognition test 7 2 =16.21, df = 2, p < .001 Immediate-Typical: Hit Miss Sure 40 (61.5 %) 13 (34.2%) Less Sure 19 (29.2 %) 19 (50.0%) Unsure/Guess 6 (9.2%) 6 (15.8 %) Total 65 (100.0%) 38 (100.0%) Recognition test 9 2 =7.17, df = 2, p < .05 Delayed-Typical Sure 34 (64.2%) 5 (11.4%) Less Sure 16 (30.2%) 24 (54.5%) Unsure/Guess 3 (5.7%) 15 (34.1%) Total 53 (100.0%) 44 (100.0%) Recognition test 9 2 =30.59, df = 2, p < .001 Immediate-Typical: Hit Miss Sure 71 (80.7 %) 5 (33.3%) Less Sure 13 (14.8 %) 6 (40.0%) Unsure/Guess 4 (4.5%) 4 (26.7 %) Total 88 (100.0%) 15 (100.0%) Recognition test 10 2 =16.39, df = 2, p < .001 Delayed-Typical Sure 54 (65.1%) 1 (7.7%) Less Sure 16 (19.3%) 5 (38.5%) Unsure/Guess 13 (15.7%) 7 (53.8%) Total 83 (100.0%) 13 (100.0%) Recognition test 10 2 =16.21, df = 2, p < .001
195 Immediate-Typical: Hit Miss Sure 53 (71.6 %) 9 (31.0%) Less Sure 16 (21.6 %) 7 (24.1%) Unsure/Guess 5 (6.8%) 13 (44.8 %) Total 74 (100.0%) 29 (100.0%) Recognition test 12 2 =23.04, df = 2, p < .001 Delayed-Typical Sure 32 (53.3%) 8 (22.2%) Less Sure 23 (38.3%) 12 (33.3%) Unsure/Guess 5 (8.3%) 16 (44.4%) Total 60 (100.0%) 36 (100.0%) Recognition test 12 2 =18.79, df = 2, p < .001 Immediate-Typical: Hit Miss Sure 69 (86.3 %) 9 (39.1%) Less Sure 9 (11.3 %) 7 (30.4%) Unsure/Guess 2 (2.5%) 7 (30.4 %) Total 80 (100.0%) 23 (100.0%) Recognition test 13 2 =25.42, df = 2, p < .001 Delayed-Typical Sure 56 (71.8%) 6 (31.6%) Less Sure 19 (24.4%) 10 (52.6%) Unsure/Guess 3 (3.8%) 3 (15.8%) Total 78 (100.0%) 19 (100.0%) Recognition test 13 2 =11.47, df = 2, p < .01 Immediate-Typical: Hit Miss Sure 33 (47.8 %) 11 (32.4%) Less Sure 30 (43.5 %) 13 (38.2%) Unsure/Guess 6 (8.7%) 10 (29.4 %) Total 69 (100.0%) 34 (100.0%) Recognition test 16 2 =7.72, df = 2, p < .05 Delayed-Typical Sure 10 (23.3%) 10 (18.5%) Less Sure 26 (60.5%) 19 (35.2%) Unsure/Guess 7 (16.3%) 25 (46.3%) Total 43 (100.0%) 54 (100.0%) Recognition test 16 2 =10.10, df = 2, p < .01
196 Immediate-Typical: Hit Miss Sure 42 (67.7 %) 12 (29.3%) Less Sure 15 (24.2 %) 17 (41.5%) Unsure/Guess 5 (8.1%) 12 (29.3 %) Total 62 (100.0%) 41 (100.0%) Recognition test 18 2 =16.06, df = 2, p < .001 Delayed-Typical Sure 35 (58.3%) 6 (16.2%) Less Sure 18 (30.0%) 18 (48.6%) Unsure/Guess 7 (11.7%) 13 (35.1%) Total 60 (100.0%) 37 (100.0%) Recognition test 18 2 =17.86, df = 2, p < .001 Immediate-Typical: False Alarm Correct Rejection Sure 10 (62.5%) 59 (67.8 %) Less Sure 6 (37.5%) 16 (18.4 %) Unsure/Guess 0 (0.0%) 12 (13.8%) Total 16 (100.0%) 87 (100.0%) Recognition test 2 2 =4.57, df = 2, p = .102 Delayed-Typical Sure 14 (63.6%) 39 (52.0%) Less Sure 6 (27.3%) 27 (36.0%) Unsure/Guess 2 (9.1%) 9 (12.0%) Total 22 (100.0%) 75 (100.0%) Recognition test 2 2 =.93, df = 2, p = .628 Immediate-Typical: False Alarm Correct Rejection Sure 25 (48.1%) 29 (56.9 %) Less Sure 25 (48.1%) 18 (35.3 %) Unsure/Guess 2 (3.8%) 4 (7.8%) Total 52 (100.0%) 51 (100.0%) Recognition test 3 2 =2.09, df = 2, p = .351 Delayed-Typical Sure 35 (45.5%) 5 (25.0%) Less Sure 35 (45.5%) 8 (40.0%) Unsure/Guess 7 (9.1%) 7 (35.0%) Total 77 (100.0%) 20 (100.0%) Recognition test 3 2 =9.10, df = 2, p < .05
197 Immediate-Typical: False Alarm Correct Rejection Sure 15 (42.9%) 42 (61.8%) Less Sure 18 (51.4%) 19 (27.9%) Unsure/Guess 2 (5.7%) 7 (10.3%) Total 35 (100.0%) 68 (100.0%) Recognition test 5 2 =5.60, df = 2, p = .061 Delayed-Typical Sure 18 (37.5%) 11 (22.4%) Less Sure 24 (50.0%) 23 (46.9%) Unsure/Guess 6 (12.5%) 15 (30.6%) Total 48 (100.0%) 49 (100.0%) Recognition test 5 2 =5.56, df = 2, p = .062 Immediate-Typical: False Alarm Correct Rejection Sure 13 (54.2%) 54 (69.2 %) Less Sure 11 (45.8%) 17 (21.8 %) Unsure/Guess 0 (0.0%) 7 (9.0%) Total 24 (100.0%) 78 (100.0%) Recognition test 6 2 =6.65, df = 2, p < .05 Delayed-Typical Sure 26 (63.4%) 24 (42.9%) Less Sure 13 (31.7%) 20 (35.7%) Unsure/Guess 2 (4.9%) 12 (21.4%) Total 41 (100.0%) 56 (100.0%) Recognition test 6 2 =6.55, df = 2, p < .05 Immediate-Typical: False Alarm Correct Rejection Sure 41 (59.4%) 9 (26.5 %) Less Sure 26 (37.7%) 17 (50.0 %) Unsure/Guess 2 (2.9%) 8 (23.5%) Total 69 (100.0%) 34 (100.0%) Recognition test 8 2 =15.91, df = 2, p < .001 Delayed-Typical Sure 37 (55.2%) 8 (26.7%) Less Sure 21 (31.3%) 7 (23.3%) Unsure/Guess 9 (13.4%) 15 (50.0%) Total 67 (100.0%) 30 (100.0%) Recognition test 8 2 =15.30, df = 2, p < .001
198 Immediate-Typical: False Alarm Correct Rejection Sure 14 (34.1%) 29 (46.8 %) Less Sure 22 (53.7%) 20 (32.3 %) Unsure/Guess 5 (12.2%) 13 (21.0%) Total 41 (100.0%) 62 (100.0%) Recognition test 11 2 =4.80, df = 2, p = .091 Delayed-Typical Sure 13 (26.5%) 7 (14.9%) Less Sure 26 (53.1%) 17 (36.2%) Unsure/Guess 10 (20.4%) 23 (48.9%) Total 49 (100.0%) 47 (100.0%) Recognition test 11 2 =8.78, df = 2, p < .05 Immediate-Typical: False Alarm Correct Rejection Sure 23 (46.0%) 20 (38.5%) Less Sure 17 (34.0%) 16 (30.8%) Unsure/Guess 10 (20.0%) 16 (30.8%) Total 50 (100.0%) 52 (100.0%) Recognition test 14 2 =1.59, df = 2, p = .453 Delayed-Typical Sure 14 (33.3%) 6 (10.9%) Less Sure 21 (50.0%) 25 (45.5%) Unsure/Guess 7 (16.7%) 24 (43.6%) Total 42 (100.0%) 55 (100.0%) Recognition test 14 2 =11.33, df = 2, p < .01 Immediate-Typical: False Alarm Correct Rejection Sure 16 (41.0%) 26 (41.3%) Less Sure 11 (28.2%) 17 (27.0%) Unsure/Guess 12 (30.8%) 20 (31.7%) Total 39 (100.0%) 63 (100.0%) Recognition test 15 2 =.021, df = 2, p = .990 Delayed-Typical Sure 13 (27.7%) 6 (12.2%) Less Sure 24 (51.1%) 17 (34.7%) Unsure/Guess 10 (21.3%) 26 (53.1%) Total 47 (100.0%) 49 (100.0%) Recognition test 15 2 =10.85, df = 2, p < .05 \
199 Immediate-Typical: False Alarm Correct Rejection Sure 65 (75.6%) 3 (18.8%) Less Sure 18 (20.9%) 8 (50.0%) Unsure/Guess 3 (3.5%) 5 (31.3%) Total 86 (100.0%) 16 (100.0%) Recognition test 17 2 =24.26, df = 2, p < .001 Delayed-Typical Sure 60 (80.0%) 4 (19.0%) Less Sure 10 (13.3%) 6 (28.6%) Unsure/Guess 5 (6.7%) 11 (52.4%) Total 75 (100.0%) 21 (100.0%) Recognition test 17 2 =32.00, df = 2, p < .001 ImmediateAtypical: Hit Miss Sure 43 (82.7 %) 15 (31.9%) Less Sure 6 (11.5 %) 23 (48.9%) Unsure/Guess 3 (5.8%) 9 (19.1 %) Total 52 (100.0%) 47 (100.0%) Recognition test 1 2 =26.30, df = 2, p < .001 Delayed-atypical Sure 15 (41.7%) 22 (33.3%) Less Sure 18 (50.0%) 29 (43.9%) Unsure/Guess 3 (8.3%) 15 (22.7%) Total 36 (100.0%) 66 (100.0%) Recognition test 1 2 =3.37, df = 2, p = .186 ImmediateAtypical: Hit Miss Sure 44 (80.0 %) 18 (40.9%) Less Sure 7 (12.7 %) 17 (38.6%) Unsure/Guess 4 (7.3%) 9 (20.5 %) Total 55 (100.0%) 44 (100.0%) Recognition test 2 2 =15.97, df = 2, p < .001 Delayed-atypical Sure 28 (43.9%) 25 (43.9%) Less Sure 13 (29.5%) 20 (35.1%) Unsure/Guess 3 (6.8%) 12 (21.1%) Total 44 (100.0%) 57 (100.0%) Recognition test 2 2 =5.47, df = 2, p = .065
200 ImmediateAtypical: Hit Miss Sure 26 (57.8 %) 31 (58.5%) Less Sure 16 (35.6 %) 18 (34.0%) Unsure/Guess 3 (6.7%) 4 (7.5 %) Total 45 (100.0%) 53 (100.0%) Recognition test 4 2 =.05, df = 2, p = .977 Delayed-atypical Sure 16 (33.3%) 27 (50.9%) Less Sure 25 (52.1%) 18 (34.0%) Unsure/Guess 7 (14.6%) 8 (15.1%) Total 48 (100.0%) 53 (100.0%) Recognition test 4 2 =3.78, df = 2, p = .151 ImmediateAtypical: Hit Miss Sure 36 (58.1 %) 9 (24.3%) Less Sure 21 (33.9 %) 17 (45.9%) Unsure/Guess 5 (8.1%) 11 (29.7 %) Total 62 (100.0%) 37 (100.0%) Recognition test 7 2 =13.41, df = 2, p < .01 Delayed-atypical Sure 20 (40.8%) 8 (15.4%) Less Sure 23 (46.9%) 19 (36.5%) Unsure/Guess 6 (12.2%) 25 (48.1%) Total 49 (100.0%) 52 (100.0%) Recognition test 7 2 =17.10, df = 2, p < .001 ImmediateAtypical: Hit Miss Sure 73 (96.1 %) 8 (40.0%) Less Sure 2 (2.6 %) 3 (15.0%) Unsure/Guess 1 (1.3%) 9 (45.0 %) Total 76 (100.0%) 20 (100.0%) Recognition test 9 2 =39.55, df = 2, p < .001 Delayed-atypical Sure 42 (76.4%) 18 (39.1%) Less Sure 6 (10.9%) 8 (17.4%) Unsure/Guess 7 (12.7%) 20 (43.5%) Total 49 (100.0%) 46 (100.0%) Recognition test 9 2 =15.47, df = 2, p < .001
201 ImmediateAtypical: Hit Miss Sure 55 (80.9 %) 8 (27.6%) Less Sure 11 (16.2 %) 8 (27.6%) Unsure/Guess 2 (2.9%) 13 (44.8 %) Total 68 (100.0%) 29 (100.0%) Recognition test 12 2 =33.31, df = 2, p < .001 Delayed-atypical Sure 15 (38.5%) 12 (19.4%) Less Sure 16 (41.0%) 25 (40.3%) Unsure/Guess 8 (20.5%) 25 (40.3%) Total 39 (100.0%) 62 (100.0%) Recognition test 12 2 =6.15, df = 2, p < .05 ImmediateAtypical: Hit Miss Sure 79 (89.8 %) 7 (70.0%) Less Sure 8 (9.1 %) 1 (10.0%) Unsure/Guess 1 (1.1%) 2 (20.0 %) Total 88 (100.0%) 10 (100.0%) Recognition test 13 2 =10.85, df = 2, p < .01 Delayed-atypical Sure 42 (59.2%) 5 (16.7%) Less Sure 25 (35.2%) 9 (30.0%) Unsure/Guess 4 (5.6%) 16 (53.3%) Total 71 (100.0%) 30 (100.0%) Recognition test 13 2 =32.58, df = 2, p < .001 ImmediateAtypical: Hit Miss Sure 52 (82.5 %) 11 (32.4%) Less Sure 7 (11.1 %) 12 (35.3%) Unsure/Guess 4 (6.3%) 11 (32.4 %) Total 63 (100.0%) 34 (100.0%) Recognition test 16 2 =24.81, df = 2, p < .001 Delayed-atypical Sure 17 (41.5%) 6 (10.2%) Less Sure 19 (46.3%) 25 (42.4%) Unsure/Guess 5 (12.2%) 28 (47.5%) Total 41 (100.0%) 59 (100.0%) Recognition test 16 2 =19.50, df = 2, p < .001
202 ImmediateAtypical: Hit Miss Sure 35 (64.8 %) 8 (18.2%) Less Sure 17 (31.5 %) 22 (50.0%) Unsure/Guess 2 (3.7%) 14 (31.8 %) Total 54 (100.0%) 44 (100.0%) Recognition test 18 2 =25.84, df = 2, p < .001 Delayed-atypical Sure 21 (40.4%) 9 (18.4%) Less Sure 22 (42.3%) 17 (34.7%) Unsure/Guess 9 (17.3%) 23 (46.9%) Total 52 (100.0%) 49 (100.0%) Recognition test 18 2 =11.49, df = 2, p < .01 ImmediateAtypical: False Alarm Correct Rejection Sure 24 (55.8 %) 29 (51.8%) Less Sure 16 (37.2 %) 21 (37.5%) Unsure/Guess 3 (7.0%) 6 (10.7 %) Total 43 (100.0%) 56 (100.0%) Recognition test 3 2 =.448, df = 2, p = .799 Delayed-atypical Sure 28 (41.2%) 7 (21.9%) Less Sure 34 (50.0%) 16 (50.0%) Unsure/Guess 6 (8.8%) 9 (28.1%) Total 68 (100.0%) 32 (100.0%) Recognition test 3 2 =7.72, df = 2, p < .05 ImmediateAtypical: False Alarm Correct Rejection Sure 16 (48.5 %) 36 (54.5%) Less Sure 12 (36.4 %) 22 (33.3%) Unsure/Guess 5 (15.2%) 8 (12.1 %) Total 33 (100.0%) 66 (100.0%) Recognition test 5 2 =.367, df = 2, p = .833 Delayed-atypical Sure 12 (24.0%) 19 (38.0%) Less Sure 32 (64.0%) 21 (42.0%) Unsure/Guess 6 (12.0%) 10 (20.0%) Total 50 (100.0%) 50 (100.0%) Recognition test 5 2 =4.86, df = 2, p = .09
203 ImmediateAtypical: False Alarm Correct Rejection Sure 15 (41.7 %) 40 (63.5%) Less Sure 14 (38.9 %) 18 (28.6%) Unsure/Guess 7 (19.4%) 5 (7.9 %) Total 36 (100.0%) 63 (100.0%) Recognition test 6 2 =5.22, df = 2, p = .073 Delayed-atypical Sure 21 (40.4%) 18 (36.7%) Less Sure 22 (42.3%) 18 (36.7%) Unsure/Guess 9 (17.3%) 13 (26.5%) Total 52 (100.0%) 49 (100.0%) Recognition test 6 2 =1.27, df = 2, p = .530 ImmediateAtypical: False Alarm Correct Rejection Sure 30 (48.4 %) 8 (21.6%) Less Sure 23 (37.1 %) 19 (51.4%) Unsure/Guess 9 (14.5%) 10 (27.0 %) Total 62 (100.0%) 37 (100.0%) Recognition test 8 2 =7.324, df = 2, p < .05 Delayed-atypical Sure 41 (56.9%) 2 (6.9%) Less Sure 26 (36.1%) 12 (41.4%) Unsure/Guess 5 (6.9%) 15 (51.7%) Total 72 (100.0%) 29 (100.0%) Recognition test 8 2 =33.25, df = 2, p < .001 ImmediateAtypical: False Alarm Correct Rejection Sure 12 (42.9 %) 36 (51.4%) Less Sure 10 (35.7 %) 29 (41.4%) Unsure/Guess 6 (21.4%) 5 (7.1 %) Total 28 (100.0%) 70 (100.0%) Recognition test 10 2 =4.100, df = 2, p = .129 Delayed-atypical Sure 24 (45.3%) 9 (18.8%) Less Sure 24 (45.3%) 27 (56.3%) Unsure/Guess 5 (9.4%) 12 (25.0%) Total 53 (100.0%) 48 (100.0%) Recognition test 10 2 =9.65, df = 2, p < .01
204 ImmediateAtypical: False Alarm Correct Rejection Sure 11 (29.7 %) 20 (32.8%) Less Sure 21 (56.8 %) 22 (36.1%) Unsure/Guess 5 (13.5%) 19 (31.1 %) Total 37 (100.0%) 61 (100.0%) Recognition test 11 2 =5.240, df = 2, p = .073 Delayed-atypical Sure 20 (31.7%) 4 (10.5%) Less Sure 28 (44.4%) 16 (42.1%) Unsure/Guess 15 (23.8%) 18 (47.4%) Total 63 (100.0%) 38 (100.0%) Recognition test 11 2 =8.55, df = 2, p < .05 ImmediateAtypical: False Alarm Correct Rejection Sure 8 (44.4 %) 24 (30.0%) Less Sure 9 (50.0 %) 29 (36.3%) Unsure/Guess 1 (5.6%) 27 (33.8 %) Total 18 (100.0%) 80 (100.0%) Recognition test 14 2 =5.744, df = 2, p = .057 Delayed-atypical Sure 5 (14.3%) 14 (21.2%) Less Sure 21 (60.0%) 20 (30.3%) Unsure/Guess 9 (25.7%) 32 (48.5%) Total 35 (100.0%) 66 (100.0%) Recognition test 14 2 =8.47, df = 2, p < .05 ImmediateAtypical: False Alarm Correct Rejection Sure 14 (43.8 %) 19 (28.8%) Less Sure 14 (43.8 %) 30 (45.5%) Unsure/Guess 4 (12.5%) 17 (25.8 %) Total 32 (100.0%) 66 (100.0%) Recognition test 15 2 =3.214, df = 2, p = .200 Delayed-atypical Sure 16 (34.0%) 3 (5.7%) Less Sure 25 (53.2%) 22 (41.5%) Unsure/Guess 6 (12.8%) 28 (52.8%) Total 47 (100.0%) 53 (100.0%) Recognition test 15 2 =23.04, df = 2, p < .001
205 ImmediateAtypical: False Alarm Correct Rejection Sure 56 (65.1 %) 0 (0.0%) Less Sure 24 (27.9 %) 8 (66.7%) Unsure/Guess 6 (7.0 %) 4 (33.3 %) Total 86 (100.0%) 12 (100.0%) Recognition test 17 2 =19.83, df = 2, p < .001 Delayed-atypical Sure 56 (69.1%) 2 (10.5%) Less Sure 21 (25.9%) 6 (31.6%) Unsure/Guess 4 (4.9%) 11 (57.9%) Total 81 (100.0%) 19 (100.0%) Recognition test 17 2 =38.07, df = 2, p < .001
206 LIST OF REFERENCES Abbott, V., Black, J. B., & Smith, E. E. (1985) . The representation of scripts in memory. Journal of Memory & Language, 24 , 179-199. Abelson, R. P. (1981). Psychological status of the script concept. American Psychologist, 36 , 715-729. Alba, J. W., & Hasher, L. (1983). Is memory schematic? Psychological Bulletin, 93 , 203231. Allen, M., & Stiff, J. B. (1989). Testing th ree models for the sleeper effect. Western Journal of Speech Communication , 53, 411-426. Allen, M., & Stiff, J. B. (1998). An analysis of the sleeper effect. In M. Allen & R. W. Preiss (Eds.), Persuasion: Advances through meta-analysis (pp. 175-188). Cresskill, NJ: Hampton Press. Anderson, K., & Clevenger Jr., T. (1963). A summ ary of experimental research in ethos. Speech Monographs, 30 , 59-78. Anderson, J. R., & Reder, L. M. (1979). An elaborative processing explanation of depth of processing. In L. S. Cermak & F. I. M. Craik (Eds .), Levels of processing in human memory (pp. 385-404). Hillsdale, NJ: Lawrence Erlbaum Associates. Applbaum, R. L., & Anatol, K. W. (1973). Di mensions of source credibility: A test for reproducibility. Speech Monographs, 40 , 231-237. Areni, C. S. (2003). The effects of structur al and grammatical vari ables on persuasion: An elaboration likelihood model perspective. Psychology & Marketing, 20 , 349375. Areni, C. S. (2002). The proposition-proba bility model of argument structure and message acceptance. Journal of Consumer Research, 29 , 168-187. Areni, C. S., & Cox, K. C. (1995). Assessing the impact of message cues and arguments in persuasion: Conceptual and methodological issues. Advances in Consumer Research, 22 , 198-202. Areni, C. S., & Lutz, R. J. (1988). The role of argument quality in the elaboration likelihood model. Advances in Consumer Research , 15, 197-203.
207 Areni, C. S., & Sparks, J. R. (2005). Language power and persuasion. Psychology & Marketing, 22 , 507-525. Arias-Bolzmann, L., Chakraborty, G., & Mowen, J. C. (2000). Effects of absurdity in advertising: The moderating role of pr oduct category attitude and the mediating role of cognitive responses. Journal of Advertising , 24, 35-49. Aristotle. (2003). Rhetoric (R. Roberts, Trans.). New York: Nuvision. Asch, S. E. (1948). The doctrine of sugge stion, prestige and imitation in social psychology. Psychological Review , 55, 250-276. Ashcraft, M. (1978). Property norms for typica l and atypical items from 17 categories: A description and discussion. Memory & Cognition , 6, 227-232. Atkin, C., & Block, M. (1983). Effectiveness of celebrity endorsers. Journal of Advertising Research, 23 , 57-61. Axsom, D., Yates, S., & Chaiken, S. (1987). Audience response as a heuristic cue in persuasion. Journal of Personality and Social Psychology, 53 , 30-40. Baker, S., & Green, H. (2005, May 2). Blogs will change your business. Businessweek, 3931, 56-67. Bartlett, F. C. (1932/1995). Remembering; a study in experimental and social psychology (2nd ed.). Cambridge, UK: Cambridge University Press. Batchelder, W. H., & Riefer, D. M. (1990) . Multinomial processing models of source monitoring. Psychological Review, 97 , 548-564. Baudhuin, E. S., & Davis, M. K. (1972). Scales for the measurement of ethos: Another attempt. Speech Monographs, 39 , 296-301. Bayen, U. J., Murnane, K., & Erdfelder, E. (1996). Source discrimination, item detection, and multinomial models of source monitoring. Journal of Experimental Psychology: Learning, memory, and cognition , 22(1), 197-215. Bayen, U. J., Nakamura, G. V., Dupuis, S. E., & Yang, C.-L. (2000). The use of schematic knowledge about sources in source monitoring. Memory & Cognition , 28(3), 480-500. Begg, M. I., Anas, A., & Farinacci, S. (1992). Dissociation of processes in beliefs: Source recollection, statement, familiarity, and illusion of truth. Journal of Experimental Psychology: General , 121, 446-458. Bell, D. M. (1997). Innuendo. Journal of Pragmatics, 27 , 35-59.
208 Bellezza, F. S. (1996). Mnemonic methods to enhance storage and retrieval. In E. L. Bjork & R. A. Bjork (Eds.), Memory (pp. 345-380). New York: Academic Press. Berlo, D. K., Lemert, J. B., & Mertz, R. J. (1969-1970). Dimensions for evaluating the acceptability of message sources. Public Opinion Quarterly , 33, 563-576. Bohner, G., & Wanke, M. (2002). Attitude and attitude change . New York: Taylor & Francis Group. Bochner, S., & Insko, C. (1966). Communicat or discrepancy, source credibility and opinion change. Journal of Personality and Social Psychology , 4, 614-621. Boller, G. W., Swasy, J. L., & Munch, J. M. (1990). Conceptualizing argument quality via argument structure. Advances in Consumer Research, 17 , 321-328. Bourne JR., L. E. (1982). Typicality eff ects in logically defined categories. Memory & Cognition, 10 , 3-9. Bower, G. H., Black, J. B., & Turner, T. J. (1979). Scripts in memory for text. Cognitive Psychology, 11 , 177-220. Bransford, J. D., Barclay, J. R., & Franks, J. J. (1972). Sentence memory: A constructive versus interpretive approach. Cognitive Psychology, 3 , 193-209. Braun-LaTour, K. A., & LaTour, M. S. ( 2004). Assessing the long-term impact of a consistent advertising ca mpaign on consumer memory. Journal of Advertising , 33, 49-61. Brewer, W. F. (2000). Bartlett's concept of the schema and its impact on theories of knowledge representation in contemporary cognitive psychology. In A. Saito (Ed.), Bartlett, culture & cognition (pp. 69-89). UK: Psychology Press. Brewer, M. B., & Nakamura, G. V. (1984). The na ture and functions of schemas. In R. S. J. Wyer & T. K. Srull (Eds.), Handbook of social cognition (pp. 119-160). Hillsdale, NJ: LEA. Brewer, W. F., & Treyens, J. C. (1981). Role of schemata in memory for places. Cognitive Psychology , 13(2), 207-230. Brooks, R. D., & Scheidel, T. M. (1 968). Speech as process: A case study. Speech Monographs, 35 , 1-7. Brown, S. C., & Park, D. C. (2002). Roles of age and familiarity in learning health information. Educational Gerontology, 28 , 695-710.
209 Campbell, D. T., & Stanley, J. C. (1963 ). Experimental and quasi-experimental designs for research . Chicago: Rand McNally College Publishing. Capon, N., & Hulbert, J. (1973). Th e sleeper effect: An awakening. Public Opinion Quarterly, 37 , 333-358. Chaiken, S., Liberman, A., & Eagly, A. ( 1989). Heuristic and systematic information processing within and beyond the persuasion context. In J. A. Bargh & J. S. Uleman (Eds.), Unintended thought (pp. 212-252). New York: The Guilford Press. Chaiken, S., & Maheswaran, D. (1994). He uristic processing can bias systematic processing: Effects of sour ce credibility, argument ambiguity, and task importance on attitude judgment. Journal of Personality & Social Psychology , 66, 460-473. Chattopadhyay, A., & Alba, J. W. (1988). The situational impor tance of recall and inference in consumer decision making. Journal of Consumer Research , 15, 1-11. Chattopadhyay, A., & Basu, K. (1990). Humor in advertising: The moderating role of prior brand evaluation. Journal of Marketing Research , 27, 466-476. Cialdini, R. B., & Petty, R. E. (1981). Anticipatory opinion effect s. In R. E. Petty, T. M. Ostrom & T. C. Brock (Eds.), Cognitive responses in persuasion (pp. 217-235). Hillsdale, NJ: Lawrence Erlbaum. Cloutier, J., Mason, M. F., & Macrae, C. N. (2005). The perceptual determinants of person construal: Reopening the social-cognitive toolbox . Journal of Personality & Social Psychology , 88, 885-894. Cook, T. D., & Campbell, D. T. (1979). Design and analysis of quasi-experiments for field settings . Chicago: Rand-McNally. Cook, T. D., & Flay, B. R. (1978). The persis tence of experimentally induced attitude change. In L. Berkowitz (Ed.), Advances in experimental social psychology (pp. 157). New York: Academic Press. Cook, T. D., Gruder, C. L., Hennigan, K. M., & Flay, B. R. (1979). History of the sleeper effect: Some logical pitfalls in accepting the null hypothesis. Psychological Bulletin , 86(4), 662-679. Coulter, K. S., & Punj, G. N. (2004). The effects of cognitive resource requirements, availability, and argument qua lity on brand attitudes. Journal of Advertising, 33 , 53-64. Cradit, J. D., Tashchian, A., & Hofacker, C. F. (1994). Signal detection theory and single observation designs: Methods and indices for advertising recognition testing. Journal of Marketing Research, 31 , 117-127.
210 Craig, S. C., & McCann, J. M. (1978). Assessing communication effects of energy conservation. Journal of Consumer Research, 5 , 82-88. Craik, F. I. M., & Lockhart, R. S. (1972). Le vels of processing: A framework for memory research. Journal of Verbal Learni ng and Verbal Behavior, 11 , 671-684. Craik, F. I. M., Morris, L. W., Morris, R. G., & Loewen, E. R. (1990). Relations between source amnesia and frontal lobe functioning in older adults. Psychology and Aging, 5 , 148-151. David, P. (1996). Role of imagery in recall of deviant news. Journalism Quarterly, 73 (4), 804-820. Davidson, D. (1994). Recognition and recall of irrelevant and interruptive atypical actions in script-based stories. Journal of Memory & Language, 33 , 757-775. Davidson, D., & Hoe, S. (1993). Children's r ecall and recognition me mory for typical and atypical actions in script-based stories. Journal of Experimental Child Psychology , 55(1), 104-126. Davidson, D., Larson, S. L., Luo, Z., & Bu rden, M. J. (2000). Interruption and bizarreness effects in the r ecall of script -based text. Memory, 8 , 217-234. Dean, D. H., & Biswas, A. (2001). Third-part y organization endorsement of products: An advertising cue affecting consumer pre purchase evaluation of goods and services. Journal of Advertising, 30 , 41-57. Debate over anti-gay marriage consti tutional amendment (February 24, 2004 ). CNN Larry King Live . Retrieved November 24, 2005, from http://transcripts.cnn.com/ TRANSCRIPTS/0402/24/lkl.00.html DeBono, K. G., & Harnish, R. J. (1988). Source expertise, source attr activeness, and the processing of persuasive inform ation: A functional approach. Journal of Personality & Social Psychology, 55 , 541-546. Deutschman, A. (2004, May). The kingmaker. Wired, 24-29. Devine, P. G., & Ostrom, T. M. (1985). Cognitive mediation of inconsistency discounting. Journal of Personalit y & Social Psychology , 49, 5-21. Dholakia, R. R., & Sternthal, B. (1977). Highl y credible sources: Persuasive facilitators or persuasive liabilities? Journal of Consumer Research, 3 , 223-232. Dutta-Bergman, M. J. (2004). The impact of completeness and web use motivation on the credibility of e-h ealth information. Journal of Communication, 54 , 253-269.
211 Eagly, A. H., & Chaiken, S. (1993). The psychology of attitudes . Fort Worth, TX: Harcourt Brace Javanovich. Eagly, A. H., Wood, W., & Chaiken, S. (1978). Causal inferences about communicators and their effect on opinion change. Journal of Personality and Social Psychology , 36(4), 424-235. Ebbinghaus, H. (1885). Memory: A contribution to experimental psychology (H. Ruger & C. E. Bussenius, Trans. 1964 ed.) . New York: Dover Publication. Edgett, R. (2002). Toward an ethical fr amework for advocacy in public relations. Journal of Public Relations Research , 14(1), 2002. Elliott, S. (2005, January 19). Public relations industry debates payments to commentator . The New York Times, pp. C1, C3. Erdfelder, E., & Bredenkamp, J. (1998). R ecognition of script-typical versus scriptatypical information: Effect s of cognitive elaboration. Memory & Cognition , 26(5), 922-938. Eveland, W. P., & Dunwoody, S. (2002). An inve stigation of elabor ation and selective scanning as mediators of learni ng from the web versus print. Journal of Broadcasting and Electronic Media, 46 , 34-53. Eysenbach, G., & KÃ¶hler, C. (2002). How do co nsumers search for and appraise health information on the world wide web? Qualitative study using focus groups, usability tests, and in-depth interviews. British Medical Journal, 324 , 573-577. Eysenck, M. W. (1979). Depth, elab oration, and distinctiveness. In L. S. Cermak & F. I. M. Craik (Eds.), Levels of processing in human memory (pp. 89-118). Hillsdale, NJ: Lawrence Erlbaum Associates. Fabrigar, L. R., Priester, J. R., Petty, R. E., & Wegener, D. T. (1998). The impact of attitude accessibility on elabor ation of persuasive message. Personality and Social Psychology Bulletin, 24 , 339-352. Farzad, R., Elgin, B., & Yang, C. ( 2005, December 5). Googling for gold. Businessweek, 60-66. Flanagin, A. J., & Metzger, M. J. (2000). Pe rceptions of internet information credibility. Journalism & Mass Communi cation Quarterly, 77 , 515-540. Ford, L. A., & Smith, S. W. (1991). Memora bility and persuasiveness of organ donation message strategies. American Behavioral Scientist , 34(6), 695-711.
212 Fox, J. R. (1999). Accurate and false memories of te levision news stories: Confidence and experiential differences in recognition judgments. Unpublished Dissertation, Cornell University. Fox, J. R. (2004). A signal detection anal ysis of audio/video redundancy effects in television news video. Communication Research, 31 , 524-536. Fritch, J. W., & Cromwell, R. L. (2001). Evaluating internet re sources: Identity, affiliation, and cognitive aut hority in a networked world. Journal of the American Society for Information Science and Technology, 52 , 499-507. Gardiner, J. M. (2002). Episodic memory a nd autonoetic consciousne ss: A first-person approach. In A. D. Baddeley, M. A. Conway & J. P. Aggleton (Eds.), Episodic memory: New directions in research (pp. 11-30). New York: Oxford University Press. Gardiner, J. M., & Java, R. I. (1990). R ecollective experience in word and nonword recognition. Memory & Cognition, 18 , 23-30. Gardiner, J. M., & Java, R. I. (1993). Recogni zing and remembering. In A. F. Collins, S. E. Gathercole, M. A. Conway & P. E. Morris (Eds.), Theories of memory (pp. 163-187). Hillsdale, NJ: Lawrence Erlbaum. Garrison, B. (2000). Journalists' perceptions of online information-gathering problems. Journalism & Mass Communi cation Quarterly, 77 , 500-514. Gillig, P. M., & Greenwald, A. G. (1974). Is it time to lay the sleeper effect to rest? Journal of Personality and Social Psychology , 29(1), 132-139. Gillund, G., & Shiffrin, R. M. (1984). A retrie val model for both recognition and recall. Psychological Review, 91 , 1-67. Gilsky, E. L., Polster, M. R., & Routhiea ux, B. C. (1995). Double dissociation between item and source memory. Neuropsychology, 9 , 229-235. Gilsky, E. L., Rubin, S. R., & Davidson, P. S. R. (2001). Source memory in older adults: An encoding or retrieval problem? Journal of Experi mental Psychology: Learning, memory, and cognition, 27 , 1131-1146. Gleitman, H., Rozin, P., & Sa bini, J. (1997). Solomon E. Asch (1907-1996): Obituary. American Psychologist , 52, 984-985. Goldman, S. R., Graesser, A. C., & van Den Broek, P. (Eds.). (1999). Narrative comprehension, causality, and coheren ce: Essays in honor of Tom Trabasso . Mahwah, NJ: Lawrence Erlbaum Associates.
213 Goldsmith, R. E., Lafferty, B. A., & Newell, S. J. (2000). The impact of corporate credibility and celebrity credibility on consumer reaction to advertisements and brands. Journal of Advertising, 29 , 43-54. Graesser, A. C. (1981). Prose comprehension beyond the word . New York: SpringerVerlag. Graesser, A. C., Gordon, S. E., & Sawyer, J. D. (1979). Recognition memory for typical and atypical actions in scripted activities: Te sts of a script point er + tag hypothesis. Journal of Verbal Learning and Verbal Behavior , 18, 319-332. Graesser, A. C., Mills, K. K., & Zwaan, R. A. (1997). Discourse comprehension. Annual Review of Psychology, 48 , 163-189. Graesser, A. C., Kassler, M. A., Kreuz, R. J., & McLain-Allen, B. (1998). Verification of statements about story worlds that deviat e from normal conceptions of time: What is true about Einstein's dreams? Cognitive Psychology, 35 , 246-301. Graesser, A. C., Mills, K. K., & Zwaan, R. A. (1997). Discourse comprehension. Annual Review of Psychology, 48 , 163-189. Graesser, A. C., & Nakamura, G. V. (1982). The impact of a schema on comprehension and memory. Psychology of Learning and Memory , 16, 60-109. Graesser, A. C., Singer, M., & Trabasso, T. (1994). Constructing inferences during narrative text comprehension. Psychological Review, 101 , 371-395. Graesser, A. C., Woll, S. B., Kowalski, D. J., & Smith, D. A. (1980). Memory for typical and atypical actions in scripted activities. Journal of Experimental Psychology: Human learning and memory , 6(5), 503-515. Greenberg, B. S., & Miller, G. R. (1966). The effects of low-credible sources on message acceptance. Speech Monographs, 33 , 127-137. Greenwald, A. G. (1968). Cognitive learni ng, cognitive response to persuasion, and attitude change. In A. G. Greenwald, T. C. Brock & T. M. Ostrom (Eds.), Psychological foundations of attitudes (pp. 147-170). New York: Academic Press. Greenwald, A. G., Pratkanis, A. R., Leippe, M. R., & Baumgardner, M. H. (1986). Under what conditions does theory obstruct research progress? Psychological Bulletin, 93 , 216-229. Gruder, C. L., Cook, T. D., Hennigan, K. M., Flay, B. R., Alessis, C., & Halamaj, J. (1978). Empirical tests of the absolute sleep er effect predicted from the discounting cue hypothesis. Journal of Personality and Social Psychology , 36(10), 1061-1074.
214 Hamilton, M. A. (1998). Message variables th at mediate and moderate the effect of equivocal language on source credibility. Journal of Language and Social Psychology , 17(1), 109-143. Hannah, D., & Sternthal, B. (1984). De tecting and explaining the sleeper effect. Journal of Consumer Research , 11, 632-642. Hansen, C. H., & Hansen, R. D. (1991). Sc hematic information processing of heavy metal lyrics. Communication Research, 18 , 373-411. Harmon, R. R., & Coney, K. A. (1982). The pers uasive effects of source credibility in buy and lease situations. Journal of Marketing Research, 19 , 255-260. Hasher, L., Goldstein, D., & Toppino, T. C. (1977). Frequency and the conference of referential validity. Journal of Verbal Learning and Verbal Behavior , 16, 107-112. Haugtvedt, C. P., & Priester , J. (1997). Conceptual a nd methodological issues in advertising effectiveness: Advertising and consumer psychology. In W. D. Wells (Ed.), Measuring advertising effectiveness: Advertising and consumer psychology (pp. 79-94). Mahwah, NJ: Lawr ence Erlbaum Associates. Hess, T. M. (1985). Aging and context influe nces on recognition memory for typical and atypical script actions. Developmental Psychology, 21 , 1139-1151. Hess, T. M., & Slaughter, S. J. (1990). Sche matic knowledge influences on memory for scene information in young and older adults. Developmental Psychology, 26 , 855865. Hicks, J. L., & Cockman, D. W. (2003) . The effect of general knowledge on source memory and decision processes. Journal of Memory and Language , 48, 489-501. Holbrook, A. L., Berent, M. K., Krosnick, J. A., & Visser, P. S. (2005). Attitude importance and accumulation of attitude-relevant knowledge in memory. Journal of Personality & Social Psychology, 88 , 749-769. Hovland, C. I., Janis, I. L., & Kelley, H. H. (1953). Communication and persuasion: Psychological studies of opinion change . New Haven, CT: Yale University Press. Hovland, C. I., Lumsdaine, A. A., & Sheffield, F. D. (1949). Experiments on mass communication . Princeton, NJ: Princeton University Press. Hunt, R. R. (1995). The subtlety of dis tinctiveness: What von Restorff really did. Psychonomic Bulletin & Review, 2 , 105-112. Hovland, C. I., & Weiss, W. (1951). Th e influence of sour ce credibility on communication effectiveness. Public Opinion Quarterly , 15(4), 635-650.
215 Huh, J., DeLorme, D., & Reid, L. (2005). Fact ors affecting trust in on-line prescription drug information and impact of trust on behavior following exposure to DTC advertising. Journal of Health Communication, 10 , 711-731. Hunt, J. M., & Kernan, J. B. (1984). The ro le of disconfirmed e xpectancies in the processing of advertising messages. Journal of Social Psychology , 124, 227-236. Hunt, J. M., Kernan, J. B., & Bonfield, E. H. (1992). Memory structure in the processing of advertising messages: How is unusual information represented? Journal of Psychology , 126(4), 343-356. Inch, E. S., & Warnick, B. (2002). Critical thinking and communication: The use of reason in argument (4th ed.). Boston: Allyn and Bacon. Jacoby, L. L., & Craik, F. I. M. (1979). Effect s of elaboration of pr ocessing at encoding and retrieval: Trace distinctiveness and recovery of initial context. In L. S. Cermak & F. I. M. Craik (Eds.), Levels of processing in human memory (pp. 1-21). Hillsdale, NJ: Lawrence Erlbaum Associates. Jacoby, L. L., Craik, F. I. M., & Begg, I. (1979). Effects of decision difficulty on recognition and recall. Journal of Verbal Learning and Verbal Behavior , 18, 585606. Jacoby, L. L., Kelley, C., & Dywan, J. (1989). Memory attribution. In H. L. Roediger & F. I. M. Craik (Eds.), Varieties of memory and consciousness (pp. 391-422). Hillsdale, NJ: Lawrence Erlbaum. Johnson, M. K. (2005). The relation between source memory and episodic memory: Comment on Siedlesc ki et al. (2005). Psychology and Aging, 20 , 529-531. Johnson, M. K., Bush, J. G., & Mitchell, K. J. (1998). Interpersona l reality monitoring: Judging the sources of other people's memories. Social Cognition , 16(2), 199-224. Johnson, M. K., Hashtroudi, S., & Lindsa y, D. S. (1993). Source monitoring. Psychological Bulletin , 114(1), 3-28. Johnson, T. J., & Kaye, B. K. (2002). Webe lievability: A path model examining how convenience and reliance predict online credibility. Journalism & Mass Communication Quarterly, 79 , 619-642. Johnston, E. B. (2001). The repeated re production of Bartlett's remembering. History of Psychology, 4 , 341-366. Johnston, W. A., & Hawley, K. J. (1994). Per ceptual inhibition of expected inputs: The key that opens closed minds. Psychonomic Bulletin & Review, 1 .
216 Judd, C. M., & Kulik, J. A. (1980). Schematic effects of social attitudes on information processing and recall. Journal of Personality and Social Psychology , 38(4), 569578. Kahn, J. (2005, September 23). China's best frie nd in Taiwan lectur es in Beijing about freedom. The New York Times , A7. Kelly, J. R., & McGrath, J. E. (1988). On time and method . Beverly Hills, CA: Sage. Kelman, H. C., & Hovland, C. I. (1953). "rei nstatement" of the communicator in delayed measurement of opinion change. Journal of Abnormal and Social Psychology, 1953 , 48. Kiene, S. M., Barta, W. D., Zelenski, J. M., & Cothran, D. L. (2005). Why are you bringing up condoms now? The effect of me ssage content on framing effects of condom use messages. Health Psychology , 24, 321-326. Kintsch, W. (1988). The role of knowledge in discourse comprehension: A constructionintegration model. Psychological Review, 95 , 163-182. Kintsch, W., & van Dijk, T. A. (1978). Towa rd a model of text comprehension and production. Psychological Review , 85, 363-394. Kirkpatrick, D. D. (2005, January 8). TV hos t says U.S. paid him to back policy . The New York Times, pp. A1, C3. Kirmani, A., & Shiv, B. (1998). Effects of source congruity on brand attitudes and beliefs: The moderating role of issue-relevant elaboration. Journal of Consumer Research , 7, 25-47. Koriat, A., Goldsmith, M., & Pansky, A. (2000). Toward a psychology of memory accuracy. Annual Review of Psychology, 51 , 481-537. Kumkale, G. T., & AlbarracÃn, D. (2004). Th e sleeper effect in persuasion: A metaanalytic review. Psychological Bulletin, 130 , 143-172. Lafferty, B. A., & Goldsmith, R. E. (1999). Co rporate credibility's role in consumers' attitudes and purchase intentions when a hi gh versus a low credibility endorser is used in the ad. Journal of Business Research, 44 , 109-116. Lampinen, J. M., Copeland, S. M., & Neuschat z, J. S. (2001). Recollections of things schematic: Room schemas revisited . Journal of Experi mental Psychology: Learning, memory, and cognition , 27(5), 1211-1223.
217 Lampinen, J. M., Faries, J. M., Neuschatz, J. S., & Toglia, M. P. (2000). Recollections of things schematic: The influence of scripts on recollective experience. Applied Cognitive Psychology , 14, 543-554. Lariscy, R. A. W., & Tinkham, S. F. (1999). The sleeper eff ect and negative political advertising. Journal of Advertising , 28(4), 13-30. Lashbrook, W. B., Sanvely, W. B., & Sulliv an, D. L. (1977). The effects of source credibility and message information quantit y on the attitude change of apathetics. Communication Monographs, 44 , 252-262. Lasica, J. D. (2005, February 17). The cost of ethics: Influencing peddling in the blogosphere. Online Journalism Review . Retrieved October 12, 2005, from http://www.ojr.org/oj r/stories/050217asica Lee, A. Y., & Labroo, A. A. (2004). The effect of conceptual and perceptual fluency on brand evaluation. Journal of Marketing Research, 41 , 151-165. Lessig, L. (2005, September). A rotten ruling. Wired, 94. Lingle, J. H., & Ostrom, T. M. (1981). Princi ples of memory and cognition in attitude formation. In R. E. Petty, T. M. Ostrom & T. C. Brock (Eds .), Cognitive responses in persuasion (pp. 399-420). Hillsdale, NJ: Lawrence Erlbaum. Logue, C. M., & Miller, E. F. (1995). Rhetorical status: A study of its origins, functions, and consequences. Quarterly Journal of Speech, 81 , 20-47. Loken, B., & Ward, J. (1990). Alternative appr oaches to understanding the determinants of typicality. Journal of Consumer Research , 17, 111-126. Lord, C. G., Desforges, D. M., Pugh, S. F., & Lepper, M. R. (1994). Typicality effects in attitudes toward social policie s: A concept-mapping approach. Journal of Personality and Social Psychology , 66, 658-673. Lorge, I. (1936). Prestige, suggestion, and attitudes. Journal of Social Psychology , 7, 386-402. MacKenzie, S. B., & Lutz, R. J. (1989). An empirical examination of the structural antecedents of attitudes toward the ad in an advertising pretesting context. Journal of Marketing, 53 , 48-65. Macmillan, N. A., & Creelman, C. D. (1991). Detection theory: A user's guide . New York: Cambridge University Press. Macrae, C. N., Schloerscheidt, A. M., B odenhausen, G. V., & Milne, A. B. (2002). Creating memory illusion: Expectancy-based processing and the generation of
218 false memories. Memory , 10(1), 63-80. Maddux, J. E., & Rogers, R. W. (1980). E ffects of source expertness, physical attractiveness and supporting arguments on persuasion: A case of brains over beauty. Journal of Personality & Social Psychology, 39 , 235-244. Maggio, R. (2001). How to say it: Choice words, phr ases, sentences & paragraphs for every situation . New York: Prentice Hall. Malt, B. C., & Smith, E. E. (1982). The role of familiarity in determining typicality. Memory & Cognition , 10, 60-75. Mandler, G. (1982). The structure of value: Accounting for taste. In M. S. Clark & S. T. Fiske (Eds.), Affect and cognition: The 17th annual Carnegie Symposium (pp. 336). Hillsdale, NJ: Lawrence Erlbaum Associates. Mandler, G. (1992). Memory, arousal, and mo od: A theoretical in tegration. In S.-Ã…. Christianson (Ed.), The handbook of emotion and memory (pp. 93-110). Hillsadale, NJ: Lawrence Erlbaum Associates. MÃ¤ntylÃ¤, T. (1997). Recollections of f aces: Remembering differences and knowing similarities. Journal of Experimental Psychol ogy: Learning, Memory, & Cognition , 23, 1203-1216. MÃ¤ntylÃ¤, T., & BÃ¤ckman, L. (1992). Aging and memory for expected and unexpected objects in real-world settings. Journal of Experimental Psychology: Learning, Memory, & Cognition, 18 , 1298-1309. Mantwill, M., KÃ¶hnken, G., & Aschermann, E. (1995). Effects of the cognitive interview on the recall of familiar and unfamiliar events. Journal of Applied Psychology , 80, 68-78. Markham, R., & Lissner, D. (1994). The effect of typicality and re tention intervals on discriminative memory for instrument. Australian Journal of Psychology, 46 , 17176. Markoff, J., & Wyatt, E. (2004, December 14) . Google is adding majo r libraries to its database . The New York Times, p. A1. Markus, H., & Zajonc, R. B. (1985). The cogni tive perspective in social psychology. In G. Lindsey & E. Aronson (Eds.), Handbook of social psychology (pp. 137-230). New York: Random House. Martin, B. A. S., Lang, B., & Wong, S. (2003) . Conclusion explicitnes s in advertising: The moderating role of need for cogn ition (NFC) and argument quality (AQ) on persuasion. Journal of Advertising, 32 , 57-65.
219 Mather, M., Johnson, M. K., & De Leonardis, D. M. (1999). Stereotype reliance in source monitoring: Age differences and neuropsychological test correlates . Cognitive Neuropsychology , 16(3/4/5), 437-458. Mazursky, D., & Schul, Y. (1988). The effects of advertisement encoding on the failure to discount information: Impli cations for the sleeper effect . Journal of Consumer Research , 15(1), 24-36. McDaniel, M. A., & Einstein, G. O. (1986). Bi zarre imagery as an effective memory aid: The importance of distinctiveness. Journal of Experimental Psychology: Learning, memory, and cognition, 12 , 54-65. McGinnies, E., & Ward, C. D. (1974). Persuasibility as a f unction of source credibility and locus of control: Five cross cultural experiments. Journal of Personality, 42 , 360-371. McGlone, M. S. (2005). Quoted out of c ontext: Contextomy and its consequences. Journal of Communication , 330-346. McGuire, W. J. (1969). The nature of attitude s and attitude change. In G. Lindsey & E. Aronson (Eds.), Handbook of social psychology (2nd ed., Vol. 3, pp. 136-314). Reading, MA: Addison-Wesley. McKoon, G., & Ratcliff, R. (1992). Inference during reading. Psychological Review, 99 , 440-466. Means, M. L., & Voss, J. F. (1985). Star wars: A developmental study of expert and novice. Journal of Memory & Language, 24 , 746-757. Mehrley, R. S., & McCroskey, J. C. (1970) . Opinionated statements and attitudes intensity as predictors of attit ude change and source credibility. Speech Monographs, 37 , 47-52. Meyers-Levy, J., Louie, T. A., & Curren, M. T. (1994). How does the congruity of brand names affect evaluations of brand name extensions ? Journal of Applied Psychology , 79, 46-53. Meyers-Levy, J., & Tybout, A. M. (1989). Sc hema congruity as a basis for product evaluation. Journal of Consumer Research, 16 , 39-54. Mick, D. G. (1992). Levels of subjective co mprehension in advertising processing and their relations to ad perceptions, attitudes, and memory. Journal of Consumer Research , 18, 411-424.
220 Miller, M. D., & Burgoon, M. (1979). The relationship between violations of expectations and the inducti on of resistance to persuasion. Human Communication Research , 5, 301-313. Miller, G. R., & Hewgill, M. A. (1964). Th e effect of variations in nonfluency on audience ratings of source credibility. Quarterly Journal of Speech, 50 , 36-44. Munch, J. M., Boller, G. W., & Swasy, J. L. (1993). The effects of argument structure and affective tagging on product attitude formation. Journal of Consumer Psychology, 20 , 294-302. Murdock, B. B., Jr. (1974). Human memory: Theory and data . Potomac, Md.: Lawrence Erlbaum Associates. Murdock, B. B., Jr. (1982). A theory for the storage and retrieval of item and associative information. Psychological Review, 89 , 609-626. Nakamura, G. V., Graesser, A. C., Zimm erman, J. A., & Riha, J. (1985). Script processing in a natural situation. Memory & Cognition , 13(2), 140-144. Nedungadi, P., & Hutchinson, J. W. (1985). The prototypicality of brands: Relationships with brand awareness, preference and usage. In E. C. Hirschman & M. Holbrook (Eds.), Advanced in consumer research (pp. 498-503). Provo, UT: Association for Consumer Research. Neisser, U. (1976). Cognition and reality . San Francisco, CA: Freeman. Neuschatz, J. S., Lampinen, J. M., Preston, E. L., Hawkins, E. R., & Toglia, M. P. (2002). The effect of memory schemata on me mory and the phenomenological experience of naturalistic situations. Applied Cognitive Psychology , 16, 687-708. Norris, F. (2005, November 18). Rapid rise: Google passes $400 a share . The New York Times, p. C1. O'Brien, E., & Myers, J. L. (1985). When co mprehension difficulty improves memory for text. Journal of Experimental Psychol ogy: Learning, memory, and cognition , 11, 12-21. Ohanian, R. (1990). Construction and validation of a scale to measure celebrity endorsers' perceived expertise, trustwor thiness, and attractiveness. Journal of Advertising, 19 , 39-52. O'Keefe, D. J. (2002). Persuasion: Theory & research (2nd ed.). Thousand Oaks, CA: Sage.
221 Oliver, M. B. (1999). Caucasian viewers' memo ry of black and white criminal suspects in the news. Journal of Communication , 49(3), 49-60. Olson, J. M., & Victoria, C. A. (1984). Source cr edibility, attitudes, and the recall of past behaviors. European Journal of Social Psychology, 14 , 203-210. O'Sullivan, C. S., & Durso, F. T. (1984). E ffect of schema-incongruent information on memory for stereotypical attributes. Journal of Personality and Social Psychology , 47(1), 55-70. Owens, J., Bower, G. H., & Black, J. B. (1979) . The "soap opera" effe ct in story recall. Memory & Cognition, 7 , 185-191. Papageorgis, D. (1963). Bartlett effect and the persistence of induced opinion change. Journal of Abnormal and Social Psychology , 67(1), 61-67. Pashupati, K. (2003). "i know this brand, but did i like the ad?" an investigation of the familiarity-based sleeper effect. Psychology & Marketing, 20 , 1017-1043. Peracchio, L., & Tybout, A. M. (1996). The moderating role of prior knowledge in schema-based product evaluation. Journal of Consumer Research , 23, 177-192. Perfetti, C. A., Beverly, S., Bell, L., Rodgers, K., & Faux, R. (1987). Comprehending newspaper headlines. Journal of Memory & Language, 26 , 692-713. Petty, R. E. (1994). Two routes to persuasion: State of the art. In G. d'Ydewalle, P. Eelen, & P. Berteleson (Eds.), International perspect ives on psychological science (Vol. 2, pp. 229Â–247). Hillsdale, NJ: Erlbaum. Petty, R. E., & Cacioppo, J. T. (1996). Attitudes and persuasion: Classic and contemporary approaches . Boulder, CO: Westview Press. Petty, R. E., & Cacioppo, J. T. (1984). The effects of involvement on responses to argument quantity and quality: Central a nd peripheral routes to persuasion. Journal of Personality and Social Psychology, 46 , 69-80. Petty, R. E., Cacioppo, J. T., & Goldman, R. (1981). Personal involvement as a determinant of argument-based persuasion. Journal of Personality and Social Psychology, 41 , 847-855. Petty, R. E., Cacioppo, J. T., & Heesacker, M. (1981). Effects of rhetorical questions on persuasion: A cognitive response analysis. Journal of Personality and Social Psychology, 40 , 432-440.
222 Petty, R. E., Cacioppo, J. T., & Schumann, D. (1983). Central and peri pheral routes to advertising effectiveness: The moderating role of involvement. Journal of Consumer Research, 10 , 135-146. Petty, R. E., Priester, J. R., & Wegener, D. T. (1994). Cognitive processes in attitude change. In R. S. J. Wyer & T. K. Srull (Eds .), Handbook of social cognition: Applications (pp. 69-142). Hillsdale, NJ: LEA. Petty, R. E., & Wegener, D. T. (1991). Thought systems, argument quality, and persuasion. In R. S. J. Wyer & T. K. Srull (Eds.), Advances in social cognition (Vol. 4, pp. 147-162). Hillsdale, NJ: Larwrence Erlbaum. Pezdek, K., Whetstone, T., Reynolds, K., As kari, N., & Dougherty, T. (1989). Memory for real-world scenes: The role of consistency with schema expectation. Journal of Experimental Psychology: Le arning, memory, and cognition , 15, 587-595. Pham, M. T., & Johar, G. V. (1997). Con tingent processes of source identification. Journal of Consumer Research , 24, 249-265. Pierro, A, Mannetti, L, Kruglanski, A. W., & Sleeth-Keppler, D. (2004). Relevance override: On the reduced impact of "c ues" under High-Motivation Conditions of persuasion studies. Journal of Personalit y and Social Psychology , 86, 251-264. Pratkanis, A. R., & Aronson, E. (1992 ). Age of propaganda: The everyday use and abuse of persuasion . New York: W. H. Freeman & Company. Pratkanis, A. R., Greenwald, A. G., Leippe , M. R., & Baumgardner, M. H. (1988). In search of reliable persuasion effects: Iii. The sleeper effect is dead. Long live the sleeper effect. Journal of Personality and Social Psychology , 54(2), 203-218. Priester, J., & Petty, R. E. (2003). The in fluence of spokesperson trustworthiness on message elaboration, attitude strengt h, and advertising effectiveness. Journal of Consumer Psychology, 13 , 408-421. Priester, J., Wegener, D., Petty, R. E ., & Fabrigar, L. (1999). Examining the psychological process underlying the sleep er effect: The el aboration likelihood model explanation. Media Psychology , 1(1), 27-48. Rajaram, S. (1993). Remembering and knowi ng: Two means of access to the personal past. Memory & Cognition, 21 , 89-102. Rajaram, S., & Roediger, H. L. (1997). Remembering and knowing as states of consciousness during retrieval. In J. D. Cohen & J. W. Schooler (Eds.), Scientific approaches to consciousness (pp. 213-240). Mahwah, NJ: Lawrence Erlbaum.
223 Reips, U.-D. (2002). Theory and techniques of conducting web experiments. In B. Batinic, U.-D. Reips & M. Bosnjak (Eds.), Online social sciences (pp. 229-250). Seattle, WA: Hogrefe & Huber Publishing. Riccio, D. C., Rabinowitz, V. C., & Axelr od, S. (1994). Memory: When less is more. American Psychologist , 49(11), 917-926. Rice, G. E., & Okun, M. A. (1994). Older reader s' processing of medical information that contradicts their beliefs. Journal of Gerontology, 49 , 119-128. Rifon, N. J., Choi, S. M., Trimble, C. S., & Li, H. (2004). Congruity effects in sponsorship. Journal of Advertising , 33, 29-42. Roediger, H. L., III, & Guynn, M. (1996). Retrieva l process. In E. L. Bjork & R. A. Bjork (Eds.), Memory (pp. 197-236). New York: Academic Press. Roediger, H. L., III, & McDermott, K. B. (1995). Creating false memories: Remembering words not presented in lists. Journal of Experimental Psychology: Learning, memory, and cognition, 21 , 803-814. Rojahn, K., & Pettigrew, T. F. (1992). Memory for schema-relevant information: A metaanalytic resolution. British Journal of Social Psychology, 31 , 81-109. Rosch, E., & Mervis, C. (1975). Family resembla nces: Studies in the in ternal structure of categories. Cognitive Psychology , 7, 573-605. Rosch, E. (1978). Principles of categoriz ation. In E. Rosch & B. Lloyd (Eds.), Cognition and categorization (pp. 27-48). Hillsdale, NJ: Lawrence Erlbaum. Rosch, E., Simpson, G., & Miller, S. (1976). Structural basis of typicality effects. Journal of Experimental Psychology: Hum an perception and performance, 2 , 491-502. Rotello, C. M., Macmillan, N. A., & Reeder , J. A. (2004). Sum-difference theory of remembering and knowing: A two-dime nsional signal-detection model. Psychological Review , 111, 588-616. Roth, E. M., & Shoben, E. J. (1983). The effect of context on the structure of categories. Cognitive Psychology , 15, 346-378. Roediger, H. L., III, & Guynn, M. (1996). Retrieva l process. In E. L. Bjork & R. A. Bjork (Eds.), Memory (pp. 197-236). New York: Academic Press. Salmaso, P., Baroni, M. R., Job, R., & Pe ron, E. M. (1983). Schematic information, attention, and memory for places . Journal of Experiment al Psychology: Learning, memory, and cognition , 9(2), 263-268.
224 Schacter, D. L. (1996). Searching for memory: The brain, the mind, and the past (1st ed.). New York: BasicBooks. Schacter, D. L., Kaszniak, A. W., Kihlstrom, J. F., & Valdis erri, M. (1991). The relation between source memory and aging. Psychology and Aging, 6 , 559-568. Schacter, D. L., Norman, K. A., & Koutstaa l, W. (1998). The cognitive neuroscience of constructive memory. Annual Review of Psychology , 49, 289-318. Schank, R. C. (1999). Dynamic memory revisited . UK: Cambridge University Press. Schank, R. C., & Abelson, R. P. (1977). Scripts, plans, goals, and understanding: An inquiry into human knowledge structures . Hillsdale, NJ: Erlbaum. Schmidt, S. R. (1991). Can we have a distinctive theory of memory ? Memory & Cognition , 19, 523-542. Schmidt, D. F., & Sherman, R. C. (1984). Memory for persuasive messages: A test of a schema-copy-plus-tag model . Journal of Personality and Social Psychology , 47(1), 17-25. Schmidt, T. L., & Hitchon, J. C. (1999). When advertising and public relations converge: An application of schema theory to th e persuasive impact of alignment ads. Journalism & Mass Communication Quarterly , 76(3), 433-455. Schulman, G., & Worrall, C. (1970) . Salience patterns, source credibility, and the sleeper effect. Public Opinion Quarterly , 34, 371-382. Schweitzer, D. A. (1970). The effect of presentation on source evaluation. Quarterly Journal of Speech, 56 , 33-39. Shapiro, M. A. (1994). Signal detection meas ures of recognition memory. In A. Lang (Ed.), Measuring psychological responses to media messages (pp. 133-148). Hillsdale, N.J.: Erlbaum Associates. Shapiro, M. A., & Fox, J. R. (2002). The role of typical and atypi cal events in story memory. Human Communicati on Research, 28(1), 109-135. Sherman, J. W., & Bessenoff, G. R. (1999). Stereotypes as source-monitoring cues: On the interactions between episodic and se mantic memory. Psychological Science, 10(2), 106-110. Sherman, J., Lee, A. Y., Bessenoff, G. R., & Frost, L. A. (1998). Stereotype efficiency reconsidered: Encoding flexibili ty under cognitive control. Journal of Personality and Social Psychology , 75, 589-606.
225 Shimamura, A. P., & Squire, L. R. (1987). A neuropychological study of fact memory and source amnesia. Journal of Experimental Ps ychology: Learning, memory, and cognition, 13 , 464-473. Siedlecki, K. L., Salthouse, T. A., & Berish, D. E. (2005). Is there anything special about the again of source memory? Psychology and Aging, 20 , 19-32. Singer, M., Graesser, A. C., & Trabasso, T. (1994). Minimal or gl obal inference during reading. Journal of Memory & Language, 33 , 421-441. Sitton, S. C., & Griffin, S. (1980). Th e sleeper effect in reconstructive memory. Journal of General Psychology , 103(1), 21-25. Slater, M. D., & Rouner, D. (1997). How messa ge evaluation and source attributes may influence credibility assessment and belief change. Journalism & Mass Communication Quarterly , 73(4), 974-991. Slotnick, S. D., & Dodson, C. S. (2005). Suppor t for a continuous (single-process) model of recognition memory and source memory. Memory & Cognition , 33, 151-170. Smith, D. A., & Graesser, A. C. (1981). Memory for actions in scripted activities as a function of typicality, retention interval, and retrieval task. Memory & Cognition , 9(6), 550-559. Smith, E. R. (1998). Mental representation and memory. In D. T. Gilbert, S. T. Fiske & G. Lindzey (Eds.), The handbook of social psychology (4th ed., pp. 391-445). Boston, MA: The McGraw-Hill Companies. Smith, E. R. (1990). Content and process specific ity in the effects of prior experiences. In T. K. Srull & R. S. J. Wyer (Eds.), Advances in social cognition (Vol. III, pp. 1-59). Hillsdale, NJ: LEA. Solso, R. L. (1991). Cognitive psychology (3rd ed.). Boston, MA: Allyn and Bacon. Spaniol, J., & Bayen, U. J. (2002). When is schematic knowledge used in source monitoring? Journal of Experimental Psyc hology: Learning, memory, and cognition , 28(4), 631-651. Spencer, W. D., & Raz, N. (1994). Memory fo r facts, source, and context: Can frontal lobe dysfunction explain age-related differences? Psychology and Aging, 9 , 149159. Stafford, T. F., & Stafford, M. R. (2002). Th e advantage of atypical advertisements for stereotyped product categories . Journal of Current Issues and Research in Advertising , 24, 25-37.
226 Stangor, C., & McMillan, D. (1992). Memory for expectancy-congruent and expectancyincongruent information: A review of the so cial and social developmental literature. Psychological Bulletin , 111(1), 42-61. Stayman, D. M., Alden, D. L., & Smith, K. H. (1992). Some effects of schematic processing on consumer expectation and disconfirmation judgments. Journal of Consumer Research, 19, 240-255. Sternthal, B., Dholakia, R., & Leavitt, C. (1978). The persuasive effect of source credibility: Tests of cognitive response. Journal of Consumer Research, 4 , 252260. Sternthal, B., Phillips, L. W ., & Dholakia, R. (1978). The persuasive effect of source credibility: A situational analysis. Public Opinion Quarterly, 42 , 285-314. Story, L. (2005, November 26). As corporat e ad money flows their way, bloggers risk their rebel reputation . The New York Times, p. C4. Stross, R. (2004, December 19). The Fox is in Microsoft's henhouse . The New York Times, p. C5. Sujan, M., & Bettman, J. R. (1989). The effects of brand positioning strategies on customersÂ’ brand and category perceptions : Some insights from schema research. Journal of Marketing Research , 26, 454-467. Sujan, M., Bettman, J. R., & Sujan, H. (1986). Effects of consumer expectations on information processing in selling encounters. Journal of Marketing Research , 23, 346-353. Sundar, S. S. (1998). Effects of source attr ibution on percep tion of online news stories. Journalism & Mass Communication Quarterly , 75(1), 55-68. Sundar, S. S. (1999). Exploring receivers' crit eria for perception of print and online news. Journalism & Mass Communication Quarterly , 76(2), 373-386. Sundar, S. S., & Nass, C. (2001). C onceptualizing sources in online news . Journal of Communication , 51(1), 52-72. Szybillo, G. J., & Heslin, R. (1973). Resistan ce to persuasion: Inoculation theory in a marketing context. Journal of Marketing Research, 10 , 396-403. Tashchian, A., White, J. D., & Pak, S. (1988). Signal detection analys is and advertising recognition: An introduction to measurement and interpretation issues. Journal of Marketing Research, 25 , 397-404.
227 Tavassoli, N. T. (1998). Language in multim edia: Interaction of spoken and written information. Journal of Consumer Research, 25 , 26-37. Tesser, A., & Leone, C. (1977). Cognitive sc hemas and thought as determinants of attitude change. Journal of Experiment al Social Psychology, 13 , 340-356. Thorndyke, P. W. (1977). Cognitive structur es in comprehension and memory of narrative discourse. Cognitive Psychology , 9, 77-110. Thorndyke, P. W., & Yekovich, F. R. (1980). A critique of schema-based theories of human story memory. Poetics, 9 , 23-49. Tormala, Z. L., & Petty, R. E. (2004). Sour ce credibility and attitude certainty: A metacognitive analysis of resistance to persuasion. Journal of Consumer Psychology, 14 , 427-442. Tourangeau, R., & Rasinski, K. A. (1988). C ognitive processes underl ying context effects in attitude measurement. Psychological Bulletin , 103(3), 299-314. Trabasso, T., & van Den Broek, P. (1985). Ca usal thinking and the representation of narrative events. Journal of Memory and Language, 24 , 612-630. Trabasso, T., & Sperry, L. L. (1985). Causal relatedness and importance of story events. Journal of Memory and Language, 24 , 595-611. Trafimow, D., & Wyer, R. S. J. (1993). C ognitive representation of mundane social events. Journal of Personality and Social Psychology, 64 , 365-376. Tripp, C., Jensen, T., & Carlson, L. (1994). Th e effects of multiple product endorsements by celebrities on consumers' attitudes and intentions. Journal of Consumer Research, 20 , 535-547. Tuckey, M. R., & Brewer, N. (2003). The infl uence of schemas, stimulus ambiguity, and interview schedule on eyewitness memory over time. Journal of Experimental Psychology: Applied , 9(2), 101-118. Tulving, E. (1972). Episodic and semantic memory. In E. Tulving & W. Donaldson (Eds.), Organization of memory (pp. 381-403). New York: Academic Press. Tulving, E. (1983). Elements of episodic memory . New York: Oxford University Press. Tulving, E. (1985). Memory and consciousness. Canadian Psychology, 26 , 1-26. Tulving, E. (1993). What is episodic memory? Current Directions in Psychological Science, 2 , 67-70.
228 Tulving, E. (1995). Varieties of consciousness an d levels of awareness in memory. In A. D. Baddeley & L. Weiskrantz (Eds.), Attention: Selection, awareness, and control (pp. 283-299). London: Oxford University Press. Tulving, E. (2002a). Episodic memory and common sense: How far apart? In A. D. Baddeley, M. A. Conway & J. P. Aggleton (Eds.), Episodic memory: New directions in research (pp. 269-287). New York: Oxford University Press. Tulving, E. (2002b). Episodic memory: From mind to brain. Annual Review of Psychology, 53 , 1-25. Tversky, A., & Kahneman, D. (1973). Availabil ity: A heuristic for judging frequency and probability. Cognitive Psychology, 5 , 207-232. Vakil, E., Sharot, T., Markowitz, M., Aber buch, S., & Groswasser, Z. (2003). Script memory for typical and atypi cal actions: Controls vers us patients with severe closed-head injury. Brain Injury, 17 , 825-833. van Dijk, T. A., & Kintsch, W. (1983). Strategies of discourse comprehension . San Diego, CA: Academic Press. Wallace, W. P. (1965). Review of the historic al, empirical, and theoretical status of the von Restorff phenomenon. Psychological Bulletin , 65, 410-424. Walster, E., Aronson, E., & Abrahams, D. (1966) . On increasing the persuasiveness of a low prestige communicator. Journal of Experimental Social Psychology , 2, 325342. Wang, A. (2005). The effects of expert and consumer endorsements on audience response. Journal of Advertising Research, 45 , 402-412. Warlaumont, H. G. (1997). Appropriating r eality: Consumer' per ceptions of schemainconsistent advertising. Journalism & Mass Communication Quarterly , 74(1), 39-54. Warnick, B. (2004). Source credibility in an "authorless" environment. American Behavioral Scientist, 48 , 256-265. Wegner, D. M. (1984). Innuendo and damage to reputations. In T. C. Kinnear (Ed.), Advances in consumer psychology (pp. 691-696). Provo, UT: Association for Consumer Research. Wegner, D. M., Wenzlaff, R., Kerker, R. M ., & Beattie, A. E. (1981). Incrimination through innuendo: Can media questions become public answers? Journal of Personality and Social Psychology, 40 , 822-832.
229 Whittlesea, B. W. A. (1993). Illusions of familiarity. Journal of Experimental Psychology: Learning, Memory, & Cognition, 19 , 1235-1253. Wicks, R. H. (2001). Understanding audiences: Learning to use the media constructively . Mahwah, NJ: Lawrence Erlbaum Associates. Wicks, R. H., & Drew, D. G. (1991). Learning from news: Effects of message consistency and medium on recall and inference making. Journalism Quarterly, 68 (1/2), 155-164. Woll, S. B., & Graesser, A. C. (1982). Memory discrimination for information typical or atypical of person schemata. Social Cognition , 1(4), 287-310. Wood, W., & Kallgren, C. A. (1988). Communicator attr ibutes and persuasion: Recipient's access to attitude-rele vant information in memory. Personality and Social Psychology Bulletin, 14 , 172-182. Wu, C., & Shaffer, D. R. (1987). Susceptibility to persuasive app eals as a function of source credibility and prior experi ence with the attitude object. Journal of Personality & Social Psychology, 52 , 677-688. Wyer, R. S. J. (2004). Social comprehension and judgmen t: The role of situational models, narratives, and implicit theories . Mahwah, NJ: Lawrence Erlbaum Associates. Wyer, R. S. J., & Srull, T. K. (1986). Human cognition in it s social context. Psychological Review , 93, 322-359. Yonelinas, A. P. (1999). The contribution of recollection and familiarity to recognition and source-memory judgments: A formal dual-process model and an analysis of receiver operating characteristics. Journal of Experiment al Psychology: Learning, Memory, & Cognition, 25 , 1415-1434. Zajonc, R. B. (1980). Feeling and thin king: Preferences need no inferences . American Psychologist , 35, 151-175. Zaragoza, M. S., & Lane, S. M. (1994). S ource misattributions a nd suggestibility of eyewitness memory. Journal of Experimental Psychology: Learning, memory, and cognition, 20 , 934-945. Zeller Jr., T. (2005, June 21). Blogs fearlessly forecast Grokster ruling. International Herald Tribune . Ziegler, R., Diehl, M., & Ruther, A. ( 2002). Multiple source characteristics and persuasion: Source inconsistency as a determinant of message scrutiny. Personality and Social Psychology Bulletin, 28 , 496-508.
230 Zuwerlink, J. R., & Devine, P. G. (1996) . Attitude importan ce and resistance to persuasion. Journal of Personality and Social Psychology, 70 , 931-944.
231 BIOGRAPHICAL SKETCH Joon Soo Lim holds a B.A. in journalism and mass communication (1994) and an M.A. in mass communication (1996) from th e Korea University in Seoul, Korea. He earned his Ph.D. in mass communications from the University of Florida and joined the Middle Tennessee State Universi ty (MTSU) faculty in 2006. His research concerns cognitive proce ssing of persuasive communication, with particular attention to the role of schemas and heuristics in memory for the persuasive messages and attitudinal evaluations.