Citation
Generalizing from Purchase Outcomes

Material Information

Title:
Generalizing from Purchase Outcomes
Creator:
VANHOUCHE WOUTER ( Author, Primary )
Copyright Date:
2008

Subjects

Subjects / Keywords:
Anticipation ( jstor )
Causal theory ( jstor )
Cognitive psychology ( jstor )
Customers ( jstor )
Generalization ( jstor )
Marketing ( jstor )
Rate bases ( jstor )
Restaurants ( jstor )
Social psychology ( jstor )
Waiters ( jstor )

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright Wouter Vanhouche. Permission granted to University of Florida to digitize and display this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Embargo Date:
7/30/2007
Resource Identifier:
71638298 ( OCLC )

Downloads

This item has the following downloads:

vanhouche_w ( .pdf )

vanhouche_w_Page_76.txt

vanhouche_w_Page_89.txt

vanhouche_w_Page_86.txt

vanhouche_w_Page_54.txt

vanhouche_w_Page_03.txt

vanhouche_w_Page_66.txt

vanhouche_w_Page_39.txt

vanhouche_w_Page_75.txt

vanhouche_w_Page_61.txt

vanhouche_w_Page_09.txt

vanhouche_w_Page_58.txt

vanhouche_w_Page_70.txt

vanhouche_w_Page_37.txt

vanhouche_w_Page_29.txt

vanhouche_w_Page_53.txt

vanhouche_w_Page_83.txt

vanhouche_w_Page_73.txt

vanhouche_w_Page_48.txt

vanhouche_w_Page_25.txt

vanhouche_w_Page_65.txt

vanhouche_w_pdf.txt

vanhouche_w_Page_01.txt

vanhouche_w_Page_32.txt

vanhouche_w_Page_94.txt

vanhouche_w_Page_31.txt

vanhouche_w_Page_26.txt

vanhouche_w_Page_22.txt

vanhouche_w_Page_19.txt

vanhouche_w_Page_51.txt

vanhouche_w_Page_88.txt

vanhouche_w_Page_08.txt

vanhouche_w_Page_45.txt

vanhouche_w_Page_90.txt

vanhouche_w_Page_14.txt

vanhouche_w_Page_23.txt

vanhouche_w_Page_63.txt

vanhouche_w_Page_36.txt

vanhouche_w_Page_20.txt

vanhouche_w_Page_91.txt

vanhouche_w_Page_67.txt

vanhouche_w_Page_12.txt

vanhouche_w_Page_74.txt

vanhouche_w_Page_56.txt

vanhouche_w_Page_46.txt

vanhouche_w_Page_92.txt

vanhouche_w_Page_02.txt

vanhouche_w_Page_42.txt

vanhouche_w_Page_79.txt

vanhouche_w_Page_10.txt

vanhouche_w_Page_57.txt

vanhouche_w_Page_35.txt

vanhouche_w_Page_97.txt

vanhouche_w_Page_43.txt

vanhouche_w_Page_59.txt

vanhouche_w_Page_78.txt

vanhouche_w_Page_38.txt

vanhouche_w_Page_49.txt

vanhouche_w_Page_95.txt

vanhouche_w_Page_81.txt

vanhouche_w_Page_64.txt

vanhouche_w_Page_17.txt

vanhouche_w_Page_18.txt

vanhouche_w_Page_82.txt

vanhouche_w_Page_11.txt

vanhouche_w_Page_84.txt

vanhouche_w_Page_05.txt

vanhouche_w_Page_62.txt

vanhouche_w_Page_06.txt

vanhouche_w_Page_77.txt

vanhouche_w_Page_13.txt

vanhouche_w_Page_68.txt

vanhouche_w_Page_69.txt

vanhouche_w_Page_27.txt

vanhouche_w_Page_96.txt

vanhouche_w_Page_60.txt

vanhouche_w_Page_21.txt

vanhouche_w_Page_24.txt

vanhouche_w_Page_34.txt

vanhouche_w_Page_55.txt

vanhouche_w_Page_47.txt

vanhouche_w_Page_50.txt

vanhouche_w_Page_33.txt

vanhouche_w_Page_30.txt

vanhouche_w_Page_93.txt

vanhouche_w_Page_15.txt

vanhouche_w_Page_52.txt

vanhouche_w_Page_44.txt

vanhouche_w_Page_72.txt

vanhouche_w_Page_41.txt

vanhouche_w_Page_71.txt

vanhouche_w_Page_85.txt

vanhouche_w_Page_04.txt

vanhouche_w_Page_98.txt

vanhouche_w_Page_40.txt

vanhouche_w_Page_80.txt

vanhouche_w_Page_16.txt

vanhouche_w_Page_28.txt

vanhouche_w_Page_87.txt

vanhouche_w_Page_07.txt


Full Text












GENERALIZING FROM PURCHASE OUTCOMES


By

WOUTER VANHOUCHE

















A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA


2005





























Copyright 2005

by

Wouter Vanhouche
































I dedicate this dissertation to my wife and the little miracle she is carrying with her.















ACKNOWLEDGMENTS

Initiation, development and completion of this dissertation would not have been

possible without the contribution of a number of people. I thank Joe Alba, my dissertation

chair, for making me a better researcher and for his dedicated guidance throughout the

dissertation process. I have also benefited from interactions with Chris Janiszewski and

thank my other committee members--Lyle Brenner, Rich Lutz and Gary McGill--for their

valuable comments on the dissertation.

I am grateful to JefNuttin, Hans Vertommen, and Frank Baeyens. Each in his own

way has played a significant role in my development from a student to a researcher.

Special thanks go to Luk Warlop, without whom I might never have pursued a Ph.D.

abroad.

My parents have provided invaluable emotional support, and the same is true for

my wife, Maddy, who has made our time in Gainesville so much fun. I thank her for

enabling me to initiate this Ph.D. and for her support in completing it. I look forward to

the next stage in our life as a family of three instead of two.
















TABLE OF CONTENTS

page

A C K N O W L E D G M E N T S ................................................................................................. iv

L IST O F FIG U R E S .............. ............................ ............. ........... ... ........ viii

ABSTRACT .............. ......................................... ix

CHAPTER

1 IN TRODU CTION ................................................. ...... .................

2 LITERATURE REVIEW AND THEORY DEVELOPMENT................................

Generalization and Induction Research ............................. ................ 3
Causal Lay Theories as Alternative Framework ........................ ..............7
Do People Hold Causal Lay Theories about Events that Have Occurred Once
or Twice? .................... .................... ............ ... ............... 9
Is the human cognitive system designed to draw causal inferences based
on very sm all sam ples? ............. ....................................... .... ...........9
How many theories do people typically generate?.................. ............10
Are Causal Lay Theories Used as Input into Generalization Judgments? ..........12

3 RESEARCH QUESTIONS AND INITIAL EVIDENCE IN MARKETING ............16

R research Q questions and H ypotheses ............................................... ..................... 16
Initial Evidence in M marketing Context ............................................ ............... 19

4 TEST OF EXPLANATORY POWER OF COMPETING ACCOUNTS ................22

E x p e rim e n t 1 ......................................................................................................... 2 2
M e th o d ........................................................................................................... 2 3
P reduction s ..................................................................................................23
R e su lts ........................................................................................................... 2 4
D isc u ssio n .....................................................................................2 5
E x p e rim e n t 2 ......................................................................................................... 2 7
M e th o d ........................................................................................................... 2 8
P reductions ..................................................................................................29
R e su lts ........................................................................................................... 3 0
D isc u ssio n .....................................................................................3 2









E x p e rim e n t 3 .......................................................................................................... 3 4
M e th o d ..................................................................................................... 3 5
Predictions ..................................... .......... ........ ................. 36
R e su lts ............................................................................................................ 3 7
D iscu ssion .............................................................................................. ....... 38

5 IN SEARCH OF LAY THEORIES AND THEIR IMPACT ON
G E N E R A L IZ A T IO N ............................................................................................ 40

E x p e rim e n t 4 .......................................................................................................... 4 0
M ethod ......................................................................................................40
R e su lts ............................................................................................................ 4 1
D iscu ssion .............................................................................................. ....... 4 1
E x p e rim e n t 5 .......................................................................................................... 4 2
M ethod ......................................................................................................42
R e su lts ............................................................................................................ 4 3
D iscu ssion .............................................................................................. ....... 43
E x p e rim e n t 6 .......................................................................................................... 4 4
M ethod ......................................................................................................44
R e su lts ............................................................................................................ 4 4
D iscu ssion .............................................................................................. ....... 45
E x p e rim e n t 7 .......................................................................................................... 4 7
M ethod ...................................................................... ......... 47
R e su lts ............................................................................................................ 4 8
D iscu ssion .............................................................................................. ....... 49
E x p e rim e n t 8 .......................................................................................................... 5 0
M ethod and results ............................................................50
D iscu ssion .............................................................................................. ....... 5 1

6 MULTIPLE "UNSTABLE" OBSERVATIONS ........................ .................. 53

E x p e rim e n t 9 .......................................................................................................... 5 3
Method ............... .............................. ...............53
R e su lts ............................................................................................................ 5 4
Discussion ............. ..... ......... ... ...............54
E xperim ent 10............................................................ 55
M eth o d an d R esu lts ........................................................................................ 56
Discussion ............. ..... ......... ... ...............57

7 GENERAL DISCUSSION AND CONCLUSION........................ ...............58

T he L ay T heory A account ....................................................................................... 59
Specific Lay Theories ......................... ..................63
C on clu sion .................................................................................................6 6











APPENDIX

A STIMULI FOR EXPERIMENT 1 ........................................ .......................... 67

B STIMULI FOR EXPERIMENT 2.................. ....... ................... 69

C STIMULI FOR EXPERIMENT 3 ........................................ .......................... 70

D STIMULI FOR EXPERIMENT 4.................. ....... ................... 71

E STIMULI FOR EXPERIMENT 5 ........................................ .......................... 72

F STIMULI FOR EXPERIMENT 6...........................................................................73

G STIMULI FOR EXPERIMENT 7.................. ...... .................... 75

H STIMULI FOR EXPERIMENT 8 ........................................ .......................... 78

I STIM ULI FOR EXPERIM ENT 9..............................................................................79

J STIM ULI FOR EXPERIM EN T 10........................................ ........................ 80

L IST O F R E F E R E N C E S ......... ..... ............ ................. ...............................................8 1

B IO G R A PH IC A L SK E T C H ...................................................................... ..................88
















LIST OF FIGURES


Figure page

4-1. Patterns of results as anticipated by various accounts (A,B and C) and as observed
in Experiment 1 (D).................... ................. ...............24

4-2. Possible outcome es in Experim ent 2................................................ ........ ....... 29

4-3. The observed pattern of results in Experiment 2 ....................................................31

4-4. Pattern of results as anticipated by several accounts (A, B and C) and as observed
in Experim ent 3. ........................................................................ ............................. 36

5-1. Classification of causes as either stable or unstable in Experiment 4 ......................41

5-2. Average generalization as a function of number of experiences and purchase
context (product service) in Experiment 5. ................................... ............... 43

5-3. Mean generalization per replicate in the product and in the service condition in
E xperim ent 6 ...................................................... ................. 4 5

5-4. Mean generalization as a function of replicate and dimension in Experiment 7........48

5-5. Mean generalization as a function of purchase context and behavior in
E xperim ent 8 .........................................................................5 1

6-1. Mean generalization as a function of replicate and number of experiences in
E xperim ent 9 ...................................................... ................. 54

6-2. Mean generalization per condition in Experiment 10. .............................................56















Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy

GENERALIZING FROM PURCHASE OUTCOMES

By

Wouter Vanhouche

August 2005

Chair: Joseph W. Alba
Major Department: Marketing

Induction is a ubiquitous but rarely investigated process in marketing contexts.

Consumers frequently interact with a vendor, usually through purchase, and then must

assess the likelihood that future interactions will produce the same outcome. Research in

decision science suggests that generalizing from small samples is common but

ill-advised. Research in cognitive psychology suggests that generalization varies as a

function of the perceived typicality of the episode or exemplar. In line with a hypothesis

about conceptual coherence in the categorization literature, I argue that induction will be

driven primarily by the theories consumers have regarding the reason for an outcome. I

refer to this perspective as the "causal lay theory" view of induction. Moreover, I

hypothesize that generalization will be greater when the driving mechanism is perceived

as stable rather than unstable.

I find systematic support for this hypothesis in a series of experiments. Specifically,

consumers generalize more quickly from a positive outcome than from a negative one,

and more quickly in a product context than in a service context. Surprisingly, a strong









typicality manipulation failed to impact generalization. With multiple inconsistent

outcomes, generalization is determined by the valenced order of the outcomes and the

time lag between those outcomes, which is consistent with the lay theory perspective.

With multiple consistent outcomes, generalization quickly reaches asymptote, even when

an unstable mechanism had initially been assumed.

Results suggest that some major findings from the induction literature are not

transferable to the non-taxonomic stimuli encountered in consumer contexts. The Law of

Small Numbers, too, cannot easily be generalized to a consumer context. Instead, the

results are more in line with a hypothesis in the category-formation literature and with

findings in the social realm.














CHAPTER 1
INTRODUCTION

The question of how people generalize from a limited sample of observations to a

larger population of observations has received considerable attention in decision science

and cognitive psychology. Philosophy, too, has recognized the fundamental importance

of generalization, as well as its ubiquitous nature. Reichenbach (1951) contended that

generalization is the essence of knowledge, and long before him Aristotle (1963)

maintained that inductive reasoning, rather than divine revelation, is the prime source of

human beliefs. Psychologists have added that generalization is perhaps the simplest and

most pervasive of everyday inductive tasks through which people come to know their

physical and social world (Krueger and Clement 1996; Nisbett et al. 1983).

Generalization from instances is surely ubiquitous and important in a marketing

context and has obvious implications for repeat purchase and customer loyalty.

Surprisingly though, behavioral consumer researchers have raised hardly any explicit

questions on the topic. Is one purchase experience with a vendor enough to induce strong

expectations about future quality with that vendor? If not, does a typical purchase

experience induce stronger generalization? Does a negative experience lead to stronger

generalization than a positive one? Does the nature of the purchase determine the level of

generalization, that is, does a product experience lead to different generalization than a

service experience? Is the objective base rate of the occurrence of a certain outcome the

sole driver of generalization? Although insights from a variety of literatures are









suggestive, systematic behavioral research is missing in marketing (Folkes and Patrick

2003 for an interesting exception).

The goal of this dissertation is to directly address these unanswered questions by

systematically testing the validity of pre-existing generalization frameworks in a

consumer context and ultimately proposing a framework that has not been used

systematically in the generalization or induction literature. Results suggest that

pre-existing frameworks do not hold up as well as might have been anticipated. The

newly proposed framework fits the overall data pattern better.

The focus is on generalization from one or two purchase experiences with a

vendor. The dependent measure probes the degree to which consumers find this small

sample representative for a large population of experiences, often expressed in

percentages. For instance, in a restaurant setting, the question pertains to the percentage

of all meals on the menu that respondents believe will have the same quality as the meal

chosen.

Presentation and discussion of empirical research make up the bigger part of this

dissertation (Chapters 4-6), but a theoretical overview is outlined first (Chapter 2). Three

pre-existing frameworks are introduced before an alternative approach is outlined, one

that is a virtual stranger to classical generalization or induction research. Chapter 3

explicates the research questions and evaluates initial evidence in the marketing domain.














CHAPTER 2
LITERATURE REVIEW AND THEORY DEVELOPMENT

Generalization and Induction Research

Three distinct frameworks emerge from research closely and directly related to

the question of how and when consumers generalize: (1) the Law of Small Numbers,

(2) the heterogeneity account, and (3) the typicality account. A seminal marketing paper

relies on the first of these accounts and is the starting point of this literature review.

Hoch and Deighton (1989) propose that generalization will occur even when only

very small samples are available as input to the estimate. Although not focusing on

generalization, they suggest that consumers do not generate many hypotheses when

confronted with a certain purchase outcome; for instance, a bad meal in a restaurant.

Instead, consumers jump to conclusions fairly quickly and generalize, even when

statistical criteria do not support such conclusions.

Hoch and Deighton (1989) based their view on two streams of literature, the first

being concerned with the Law of Small Numbers, which states that people generalize

quickly because they believe that even small samples are representative (Tversky and

Kahneman 1971). An example in a marketing context is the consumer who labels a

restaurant as bad on the basis of one negative experience (e.g., a bad meal). The second,

arguably less well-known literature on hypothesis generation seems to confirm that

people often generate no more than one hypothesis (Gettys and Fisher 1979; Gettys,

Mehle, and Fisher 1986). Moreover, hypothesis formation may happen in a rather passive

way so that salient aspects of the problem drive the content of the actual hypothesis









(Hoch and Deighton 1989, p. 4). Thus one bad meal may easily be taken as sufficient

evidence of a bad restaurant. The Law of Small Numbers and the work by Hoch and

Deighton are clearly in line with the claim that people perceive less variability than is

actually present (Kareev, Amon, and Horwitz-Zeliger 2002).

Other researchers have argued that people may not always generalize from small

samples. In a reply to Tversky and Kahneman (1971), Nisbett et al. (1983) show that

heterogeneity is perceived, in some situations at least, and thus generalization is low. This

view will henceforth be referred to as the "heterogeneity account." Empirical evidence is

provided in a scenario requiring participants to imagine they arrive at a previously

undiscovered island. They have to estimate the percentage of tribe members, Barratos,

who are obese. The only input available for the estimate is the knowledge that a given

sample is obese. The size of the sample is varied: 1, 3 or 20 obese tribe members.

Clearly, generalization is low when the sample consists of 1 member (-38%), higher with

a sample of 3 (-56%), and highest with a sample of 20 (-75%).

In stark contrast, generalization about an object that is found on this imaginary

island, floridium, which is said to burn with a green flame when heated, is extremely

high, even with an n=l sample. Nisbett et al. (1983) argue that obesity is perceived as

containing more heterogeneity in a tribe population than does a population of floridium

elements. This difference in perceived heterogeneity allows Nisbett et al. to make their

point that, at least in some situations, people perceive heterogeneity even though only a

small and arguably homogenous sample is overtly available. In those situations, people

are notably less likely to generalize and will instead reason statistically. This

"generalization" literature, concerned with the Law of Small Numbers and perceived









heterogeneity, focuses on statistical reasoning and whether or not people engage in it.

However, at least two important questions remain largely unanswered. When are people

likely to perceive heterogeneity and when are they not likely to do so? Also, and in more

general terms, what drives perceived heterogeneity?

One seemingly obvious factor that may affect perceived heterogeneity is the

typicality of the sample. When a given sample is atypical for its population, perceived

heterogeneity may be high and thus generalization low. Conversely, when the sample is

typical for its population, perceived heterogeneity may be low and generalization high

(Rothbart and Lewis 1988). This pattern of results is anticipated by what is usually

referred to as the "induction literature" in cognitive psychology, which will be referred to

here as the typicalityy account" (Heit 2000; Osherson et al. 1990; Rips 1975; Sloman

1993).

The typicality effect is perhaps the single most-demonstrated effect in the

induction literature, and typicality as a factor may account for more explained variance

than any other factor. A priori, the induction literature may be considered as the single

most-important source that can help answer questions raised in this dissertation because

of its rich history, its significant volume, and the force of one of its main arguments, that

typicality drives generalization.

Although the typicality effect is compatible with the notion of perceived

heterogeneity and the shared focus on generalization in the induction and generalization

literature, the induction literature does not build on the generalization literature. Reasons

for this lack of cross-referencing are unclear, but focus in the two streams of research

differs slightly. Induction research focuses on "how people project information from









known cases to the unknown" (Heit 2000, p. 569), rather than on the (non)normativeness

of any such generalization, which concerned Tversky and Kahneman (1971). Also, the

induction literature relies almost exclusively on natural or biological categories (e.g.,

animal categories) for its stimuli, while Tversky and Kahneman and Nisbett et al. (1983)

rely more on "social" stimuli. And although there are a variety of induction models, most

assume some kind of typicality calculation underlying the induction judgment. The

output of the typicality judgment almost automatically leads to the induction judgment.

That is, very basic processes, sometimes called "bottom-up" processes, drive

generalization. In contrast, the framework developed in Chapter 3 assumes that higher

level processes--"top down" rather than "bottom up"--drive generalization.

Induction research focuses on the typicality of the sample in relation to the

population, but explicitly ignores the dimension or characteristic on which the sample is

considered (Heit and Rubinstein 1994, for an exception). Like Nisbett et al., the

framework developed here focuses less on the typicality of the sample than on the

importance of the dimension or characteristic. Finally, generalization in induction

research is usually based on category knowledge rather than object knowledge: an entire

category such as all sparrows rather than just one or a few sparrows. Despite the

differences between the two lines of research, a 25-year review of induction research by

Heit (2000) concluded that most induction models apply widely. Typical samples lead to

stronger generalization than do atypical samples (e.g., a "sparrow" sample leads to

stronger inferences concerning "all birds" than does a "penguin" sample). An equivalent

example in a marketing context would be generalization to all meals on a steakhouse

menu from a steak rather than a pasta dish. In sum, given the dominant status of the









induction literature and the extensively documented typicality effect, typicality may well

be the most important factor that drives perceived heterogeneity and thus generalization.

Available insights in the generalization and induction literature suggest that

consumers are highly likely to generalize, even with very small samples. If they do not,

the perception of heterogeneity of the characteristic in its population is the inhibiting

force. Despite the lack of cross-referencing between the induction and generalization

literature, typicality of the sample appears to be an extremely important factor in the

perception of heterogeneity and thus generalization. However, a fourth account is

introduced to explicitly consider consumer contexts.

Causal Lay Theories as an Alternative Framework

Although insights from the literature review seem compellingly documented and

comfortably intuitive, easy generalization to a consumer context is not guaranteed. Do

consumers always generalize from small samples, as the Law of Small Numbers

suggests? The induction literature, as well as research by Nisbett et al. (1983), suggests

they may not. If they do not, will typicality be a major determining factor? Or will other

factors also drive perceptions of heterogeneity and thus generalization?

In an effort to understand generalization in a consumer context, I introduce the

concept of causal lay theories as a determinant of generalization, an alternative option to

the frameworks already reviewed. Unlike the Law of Small Numbers, the causal lay

theory framework assumes that consumers will not always generalize. It is also broader

than the typicality account in allowing for low generalization when the sample is typical

and for high generalization when the sample is atypical. It is more specific than the

perceived heterogeneity account by assuming a definite source for perceived









heterogeneity: the causal lay theory that is invoked to understand the occurrence of a

particular outcome.

Lay theories are causal knowledge structures about how the world functions or is

organized. Depending on the specific causal lay theory generated, generalization will

vary. When the mechanism underlying the theory assumes stability or systematicity,

generalization tends to be high, even when based on very small samples. When the

assumed mechanism reflects instability, generalization is low. An example of a stable

mechanism may be the expertise of the chef in a restaurant scenario, whereas an example

of an unstable mechanism may be an "off day" for the chef. If the chef scores low on

expertise today, he probably did so yesterday and is likely to continue to do so tomorrow.

However, a chef having a bad day may have had a good day yesterday and may be

expected to have another good one tomorrow. Depending on which mechanism is

invoked, generalization from a single experience will vary.

Note that lay theories--the central concept of the framework--often do not reach

the rigor and consistency expected of a scientific theory (Nisbett and Ross 1980; Tversky

and Kahneman 1980). While lay theories have been shown to improve judgment in some

situations in the social realm (Wright and Murphy 1984), they may hurt judgment in

other situations (Chapman and Chapman 1969). However interesting, the appropriateness

of the impact of those lay theories is not the focus of this dissertation. Instead, the

emphasis is on whether consumers hold such causal lay theories and if so, whether they

drive generalization. The induction literature implicitly suggests that they do not and that

typicality calculations drive generalization. A major goal of this dissertation is to address

these questions explicitly, empirically (Chapters 4-6) and theoretically.









The remainder of this chapter reviews a variety of literatures that may be

theoretically relevant but that has not been relied on by induction or by generalization

researchers. Insights from the literature on causal reasoning and that on hypothesis

generation address the question of whether people hold causal lay theories. Evidence

from the social psychology literature on stereotyping, the correspondence bias, the

negativity effect, and person versus group perception is discussed to provide initial

evidence for the second question of whether causal lay theories drive generalization. The

suggestion is that people can and do hold causal lay theories, even about events that have

occurred only once, and that such theories impact generalization, at least in the social

realm.

Do People Hold Causal Lay Theories about Events that Have Occurred Once or
Twice?

This question implies two more subquestions. First, is the human cognitive

system able to draw causal inferences from very small samples? Second, if so, how many

of those inferences (theories) are typically made per observation?

Is the human cognitive system designed to draw causal inferences from very small
samples?

If people have a causal lay theory--if they draw a causal inference--about a

one-time event, then they must feel they know why something happened after having

experienced it only once. Such a claim may run counter to common wisdom as well as to

insights from some of the most influential thinkers in psychology (Heider 1958, Kelley

1967) and philosophy (Mill 1973). For instance, at the heart of Kelley's ANOVA model

lies the necessity to observe covariation between presumed cause and effect before a

causal inference can be made. According to the model, causal inferences will be drawn to

the extent that cause and effect coincide in a series of observations. One observation is,









by definition, not a series; so according to this account, it seems unlikely for people to

draw causal inferences based on only one observation.

When two observations are presented in this dissertation, only outcome

information without information about a cause was provided. As such, covaration

information is again not overtly available and causal inferences seem unlikely. The

ANOVA and similar accounts may make one wonder whether people actually draw

causal inferences in the type of situations studied in this dissertation. However, it has

been recognized (Hilton and Slugoski 1986) that causal attributions are not always made

as prescribed by the normative ANOVA or any other covariational model (Cheng 1997;

White 2002).

Different approaches have been proposed (Ahn et al. 1995; Ahn and Kalish 2000;

Einhorn and Hogarth 1982, 1986; Johnson, Long and Robinson 2001; Mandel 2003;

White 1990). The approach advanced by Ahn and her colleagues may reveal relevant

insights. The basic argument is that people seek out causal mechanisms in their

knowledge base to develop an explanation for a specific event, rather than relying solely

on covariation information to identify a causal relationship between sometimes arbitrary

factors. Somewhat simplified, this means that people confronted with a certain event will

search for an explanatory mechanism or theory. Clearly, the suggestion is that it is not

impossible for the human cognitive system to draw causal inferences from extremely

small samples. A remaining question then is how many reasons people typically generate.

How many theories do people typically generate?

As previously suggested, people seem able to draw causal inferences based on

only one observation, at least in some situations. A following question pertains to the

number of hypotheses generated for a single event or observation. Granted that any given









outcome is likely to be caused by multiple factors, it is not unreasonable to expect that

people may generate multiple reasons or theories. If so, uncertainty about exactly which

theory holds may increase and the degree of generalization may consequently decrease.

The literature on hypothesis generation may reveal valuable insights inasmuch as it deals

with the number of hypotheses (theories) people tend to generate for a particular problem

(event or instance in our analysis). (The terms "hypotheses" and "theories" are used

interchangeably in this section.)

The possible number of hypotheses or theories one can consider for a given

problem can theoretically be large, and people are clearly able to take into account or

generate multiple theories for a given problem (Alba, van Osselaer and Vanhouche 2003;

Arocha, Patel and Patel 1993; Bockenholt and Weber 1993; Fisher et al. 1983; Gettys and

Fisher 1979; Koehler 1994; Kruglanski 1990; Liberman et al. 2001; McClure, Jaspars and

Lalljee 1993; Mehle 1982; Trope and Liberman 1996; Weber et al. 1993). For instance, it

is likely that multiple theories will be generated by a physician faced with a patient's

problem or by a detective dealing with a crime.

However, an extensive review on hypothesis (theory) generation and testing

(Sanbomatsu et al. 1998) concludes that hypothesis testing is often a first-come,

first-confirmed process. The first viable theory generated has an enormous comparative

advantage (p.202). This research suggests that people often generate only one explanation

for a given problem. This is consistent with work by Simon (1956, 1982): People search

for a satisfying but not necessarily optimal solution (Garst et al. 2002). This tendency is

consistent with research that suggests that people tend to pursue confirmatory strategies

(Arocha et al. 1993; Hoch 1984; Hodgins and Zuckerman 1993; Klayman 1995; Schaklee









and Fischoff 1982; Tschirgi 1980; Zuckerman et al. 1995): Most of us look for the

presence of what we expect, not for what we would not expect. In addition, it has been

suggested that once a hypothesis becomes focal (i.e., has been generated) its strength is

overestimated (Sanbomatsu, Akimoto and Biggs 1993). This renders a causal inference

more plausible and makes the "hypothesizer" more confident (Gettys, Mehle and Fisher

1986; Koehler 1994).

If the foregoing analysis of the literature on causal reasoning literature and

hypothesis generation is valid, it makes sense to expect that people draw causal

inferences, even from samples with only one observation. This is exactly what Read

(1983, 1984) has empirically demonstrated. In line with the work of Ahn et al. (1995) he

has shown that causal inferences based on one observation are more likely when a

plausible theory (an analogy) is available (Read 1984; Weber et al. 1993).

It seems fair to conclude that people can and do draw causal inferences based on

samples with as few as one or two observations. We tend to do this when we have a lay

theory that explains the datum in a satisfactory way. As soon as one lay theory has been

activated, the search ceases, and uncertainty originating from the plausibility of multiple

theories is not experienced. A subsequent question is whether a person's causal theories

are used as input in generalization judgments. I address this issue empirically in a

marketing context (Chapters 4-6), but first outline evidence from social psychology.

Are Causal Lay Theories Used as Input in Generalization Judgments?

Evidence supporting the impact of lay theories comes from research streams that

are not necessarily focusing on generalization or induction per se, but instead deal with

lay theories in one way or another. Four significant bodies of research, all from the social

psychology literature, are briefly discussed.









Research on person versus group perception indicates that people perceive more

heterogeneity when a certain behavior is performed by a group rather than by an

individual (Hamilton and Sherman 1996). Consequently, generalization is stronger

regarding a person than a group. That is to say, people seem to factor in the origin of a

certain behavior and thus they seem to have a certain theory about its causes.

Research on stereotyping is rather explicit about the impact of lay theories. It no

longer questions whether people's lay theories--in this case called "stereotypes"--affect

judgment, but instead accepts that they do and asks when they are activated and applied

and when they are not (Kunda and Spencer 2003). For instance, if people hold the causal

lay theory that advanced age slows down certain responses, the question in this literature

is no longer 1i heile'r the causal belief affects future judgment about speed-related

behavior but when it does. The impact of lay theories is found to be so pervasive that

research is even focusing on situations in which judgment is not impacted. For the

purpose of this dissertation, the message is that research other than the core induction or

generalization work suggests that lay theories impact judgment.

Research on the correspondence bias also seems relevant (Gilbert and Malone

1995). Participants in a representative experiment draw conclusions about the personality

characteristics of an actor, even on the basis of one isolated observation of behavior that

is explicitly said to be caused by situational factors rather than personality traits (e.g,.

Jones and Harris 1967). The social psychology research is interested in the fact that

people perceive personality traits to be stable predictors of behavior, whereas situational

characteristics often are (set up to be) the actual predictors. The point of interest here is

that people seem to bring a specific causal lay theory to the scene, one that attributes









behavior to personality traits, not to certain cues in the environment. It is this stable lay

theory that determines the level of generalization: People generalize extensively

regarding the future behavior of the actor, even on the basis of a single observation. The

implication is that they would have generalized much less--or at least much differently--if

the lay theory concerned situational cues rather than personality traits. That is, not only is

a lay theory driving generalization, but impact of the stability of the assumed mechanism

is implicitly recognized as well.

A fourth significant body of research finds that negative information receives

more weight than positive information ( Folkes and Kamins 1999; Herr, Kardes and Kim

1991; Peeters and Czapinski 1990; Reeder and Brewer 1979; Skowronski and Carlston

1987). In a social context, Ybarra (2002) translates this observation into the claim that

people believe that negative behavior is caused by stable personality traits while positive

behavior is caused by variable situational cues. The implication is that stronger

generalization occurs based on negative rather than positive behavior, which suggests that

consumers will generalize more strongly from a negative rather than from a positive

purchase experience. Further, theories featuring a "personality" mechanism lead to strong

generalization while those underlying a "situational" mechanism lead to weak

generalization. Ybarra goes even as far as claiming that the perception of social behaviors

is driven by the goal of inferring underlying causes in the person. That is, not only do

people apply lay theories when available, they actively search for them, at least in the

social realm.

To summarize, research using social stimuli tends to support the notion that lay

theories impact generalization judgment and that hypothesized stable mechanisms lead to






15


high generalization while unstable mechanisms lead to comparatively low generalization.

In addition, it is argued that people actively search for causal lay theories. Before

considering whether lay theories impact generalization in a marketing context, I compare

the four theoretical frameworks, advance hypotheses to test them, and review the scarce

evidence available in the marketing literature in light of the four accounts.














CHAPTER 3
RESEARCH QUESTIONS AND INITIAL EVIDENCE IN MARKETING

Research Questions and Hypotheses

When and how do consumers generalize? This general but central question is

approached by testing the validity of the three frameworks that have been extensively

investigated in the generalization or induction literature and by examining a fourth

approach that has remained virtually untested as a generalization framework. This chapter

explicates how each of the accounts differ and advances hypotheses to test their validity.

One stunningly simple yet powerful answer to our general question comes from

the Law of Small Numbers (Tversky and Kahneman 1971) in the generalization

literature: Consumers always generalize because they find even the smallest sample

representative. Several experiments allow testing of this hypothesis. The first includes

two manipulations--typicality and outcome valence--that can be expected to induce lower

generalization.

The perceived heterogeneity view is non-compatible and posits that consumers

will not always generalize. Generalization will be low when perceived heterogeneity is

high (Nisbett et al. 1983). Confirmation of this hypothesis implies disconfirmation of the

previous one. Experiment 1 is again applicable in this regard, and Experiment 2 tests

whether generalization is driven only by objective heterogeneity (in the sample and/or the

population).

The typicality account (Heit 2000) is also incompatible with the Law of Small

Numbers but is more specific than the heterogeneity framework in explicitly anticipating









lower levels of generalization when the sample is atypical rather than typical. It further

distances itself from other accounts by suggesting that generalization occurs virtually

automatically as a result of low-level typicality calculations. If valid, the implication is

that generalization necessarily covaries with typicality. This hypothesis is also tested in

Experiment 1.

The fourth account originates from research other than strict generalization and

induction work. Like the typicality and heterogeneity accounts, the lay theory framework

is incompatible with the Law of Small Numbers in that it assumes low generalization in

some situations. It is broader than the typicality account in allowing for high

generalization when typicality is low, and low generalization when typicality is high, and

it is more specific than the heterogeneity account in assuming a specific source of

perceived heterogeneity: the lay theory invoked to account for the outcome of a specific

purchase situation. Although the exact content of lay theories can vary greatly, the

stability of the underlying mechanism is one-dimensional. This stability supposedly

drives generalization. The lay theory framework, contrary to the other three accounts, has

remained virtually untested as an explicit generalization account. To make up for this

lapse, this research project started testing specific marketing-relevant hypotheses, as

outlined below. Most are derived from the extant literature.

If the causal lay theory perspective is to be a valid generalization account, it must

(1) show that consumers hold causal lay theories and (2) that the underlying stability of

these theories drives generalization. For this approach to differentiate itself from the

typicality account, it is further desirable (3) that it shows stability and instability in

contexts independent of typicality. Experiment 4 explicitly addresses the first and third









requirements, testing whether negative outcomes elicit stable theories while positive

outcomes elicit unstable mechanisms (Ybarra 2002) or whether the opposite occurs, as

claimed by Folkes and Patrick (2003).

Experiments 3, 5 and 6 address the second requirement. The question is whether

consumers' lay theories about product versus service experiences affect generalization,

and whether their theories are similar to those held by academics. For instance, Zeithaml

et al. (1985) recognized that heterogeneity is higher in a service than in a product context.

If consumers' lay theories conform, generalization is expected to be higher in a product

than in a service context because more stability is assumed in the former.

Causal lay theories about single outcomes are not the only theories investigated.

Experiment 2 tests the presence of lay theories about a sequence of two inconsistent

outcomes. This allows for a test of the widely held belief that "the first impression

matters most" (Nisbett and Ross 1980) and thus that more stability is inferred from a

"first" rather than a "recent" experience.

If a lay theory perspective turns out to be defendable, another question pertains to

the speed of belief (theory) updating. For instance, consumers may hold a positive belief

about a certain company but encounter a negative experience that is attributed to an

unstable cause. In such a situation they may generalize only to a limited degree. It is

imperative to investigate how many more negative experiences are needed before the

negative outcomes are perceived to be caused by a stable mechanism. Experiments 3, 4, 9

and 10 test the hypothesis that consumers update beliefs slowly (Boulding, Kara and

Staelin 1999) but also consider the possibility that consumers update quickly. Fast









updating would not be inconsistent with a lay theory perspective that assumes an unstable

mechanism after one experience but a stable mechanism after two experiences.

Initial Evidence in Marketing Context

In evaluating previous research in marketing, I identified three articles that deal

explicitly with the generalization question. These give supportive evidence for the

typicality and the perceived heterogeneity accounts. An unmodified Law of Small

Numbers does not seem to fit the scant available data and evidence for the lay theory

framework is mixed.

Using a paradigm similar to that used in standard induction research, Joiner and

Loken (1998) studied the typicality of the "population" rather than of the sample. For

instance, employing the traditional induction paradigm whereby participants are exposed

to the following format, they found that (2) is rated as a stronger argument than (1).

(1) Sony TVs have attribute X; therefore, Sony bicycles have attribute "X".

(2) Sony TVs have attribute X; therefore, Sony cameras have attribute "X"

In other words, generalization from one specific category (e.g., TVs) to another is

stronger when the latter is more typical (e.g., camera rather than bicycle). Although this is

not the "typical" typicality effect, the results clearly show an effect of typicality on

generalization. It is important to note, however, that as in most induction experiments,

typicality information is made salient by explicit comparison of the typical and atypical

information.

Boulding et al. (1999) do not investigate the impact of typicality but advance

suggestions about the pace with which evaluative beliefs about a service experience are

updated as a function of the number of experiences with the service. Specifically, the

authors investigated generalization from one hotel experience (sample) to the overall









population of experiences with the hotel. They conclude that "any time a firm wants to

change consumers' beliefs of its perceptual positioning through the delivery of goods or

services, managers must recognize that these beliefs may be slow to adjust" (p. 481). It

may be informative to note that some of the data reported by Nisbett et al. (1983) seem in

line with this claim. Even with a sample of 20 observations generalization remains short

of asymptote in estimates of tribal obesity. Boulding et al. argue that multiple

observations are needed to update belief. This seems counter to a lay theory perspective,

which could conceivably allow for fast updating as a function of an increasing number of

observations. For instance, while one negative experience in a restaurant may be

perceived as being caused by an unstable mechanism, a second negative experience may

induce the perception of a stable cause and thus high levels of generalization.

The third and most recent study (Folkes and Patrick 2003) is less pessimistic

about the impact of lay theories on generalization. Although these authors did not

manipulate typicality either, they observed differential generalization as a function of

valence of the outcome. Generalization to the population of colleague service providers

was more pronounced when the sample consisted of a friendly, rather than unfriendly,

service employee. That is, a positivity effect is observed in a service context. The mere

observation that generalization is rather low, based on negative outcomes, may point to

the limited generalizability of the Law of Small Numbers in a marketing context. The

combination of low generalization in the negative outcome condition with high

generalization in the positive outcome condition is taken by Folkes and Patrick as

evidence for the existence of lay theories. Such lay theories would imply that a good

outcome is perceived as caused by a stable mechanism (e.g., the policy of the firm),









whereas a negative outcome is perceived as caused by an unstable mechanism (e.g., the

personality of one person who decided not to follow the firm's policy). Direct evidence

for these theories is not provided, however.

To summarize, the Law of Small Numbers is not always consistent with the

available evidence. For example, generalization is lower in the negative outcome

condition in Folkes and Patrick (2003). Both the typicality (Joiner and Loken 1998) and

lay theory accounts (Folkes and Patrick 2003), and thus also the perceived heterogeneity

perspective, seem to receive at least some support. However, Boulding et al. (1999)

anticipate slow updating of beliefs, which is not necessarily anticipated by a lay theory

perspective.














CHAPTER 4
TEST OF EXPLANATORY POWER OF COMPETING ACCOUNTS

The empirical section is organized around three sets of experiments, each

presented in a separate chapter. Chapter 4 explores the explanatory power of the four

theoretical accounts. Chapter 5 targets the lay theory account and examines the

characteristics and impact of specific lay theories on generalization, especially when the

sample is n=l. Chapter 6 focuses on those situations when two observations are available

as input, the first of which is perceived as caused by an unstable mechanism.

Experiment 1

The main goal of Experiment 1 is to investigate the effect of typicality on

generalization in a marketing context. At the same time the valence of the purchase

outcome is manipulated to investigate whether the negativity effect widely observed in

the psychology literature can be replicated.

The typicality effect has been demonstrated so extensively in the induction

literature that it may seem redundant to aim for replication. However, as noted before, a

specific set of stimuli and a specific paradigm have been used in the induction literature,

and both may have favored the impact of typicality in induction research in an artificial

way.

The induction literature does not make differential predictions regarding

generalization about positive versus negative purchase outcomes. However, this

distinction is obviously relevant to marketers, especially since most available evidence









suggests that generalization is more extreme when based on negative rather than positive

experiences. Interestingly, some consumer researchers have reported a weak negativity

effect (Ahluwalia 2002) or even a positivity effect (Folkes and Patrick 2003).

Method

Seventy business majors at the University of Florida participated in the

experiment for partial class credit. Participants imagined they went to a new restaurant

with a group of friends. For at least 30 seconds, they read a menu with seven meat dishes

(e.g., New York steak of black angus beef, prime rib of beef) and one pasta dish

(Appendix A for the full stimuli). Half of the participants were told they had selected a

beef item (typical), while the others learned they had selected the pasta (atypical item).

Contrary to the standard induction experiment, typicality in this experiment needs to be

inferred. Typicality of the selected item is crossed with the outcome of the meal. After

having selected their meal, half of the participants learned the quality of their meal was

low while the other half learned it was high. Participants were next asked (1) what

percentage of all meals in this restaurant did they think would be of the same quality as

the meal they had selected and (2) what was the rationale behind their generalization

estimate. This experiment, as well as all the others, was conducted entirely on computer.

Predictions

Various predictions can be made. First, the induction literature anticipates a main

effect of typicality in that generalization is more extreme in the typical than atypical

condition (Figure 4-1A). Second, a large body of evidence anticipates higher

generalization with a negative rather than positive outcome (Figure 4-1B). Third, the Law











of Small Numbers proposes that generalization will be high across all conditions (Figure

4-1C).


m positive negative


Positive negative


typical atypcal


*positive Enegative


typical atypcal


*positive Enegative


typical atypical typical atypical
C D
Figure 4-1. Patterns of results as anticipated by various accounts (A, B and C) and as
observed in Experiment 1 (D). A) Predicted pattern by typicality account, B)
Anticipated pattern of negativity effect is observed, C) Pattern anticipated by
Law of Small Numbers, and D) Observed pattern in Experiment 1.

Results

Manipulation check. In an independent study, 32 undergraduate students judged

the degree to which they found their item (steak or pasta) typical, given the menu.

Despite the relatively low cell sizes (n=16), the difference between the typical and

atypical conditions was highly significant [t(30)=5.4; p<.001], thereby demonstrating the

strength of the typicality manipulation.









Numerical data of main experiment. Figure 4-1D clearly shows that neither of

the three predicted results is actually obtained. A 2x2 typicalityy x outcome valence)

between-subject ANOVA revealed no main effect of typicality (F<1); instead there is an

outcome effect of lower generalization in the negative condition [F(1,66)=4.3; p<.05]; no

evidence for an interaction is obtained (F<1).

Cognitive responses. Two independent judges analyzed the cognitive responses

that were recorded after the generalization estimate had been made. In all experiments

except Experiment 2 the coding categories were "stable," "unstable," and "ambiguous."

"Stable" indicates a stable mechanism, such as the expertise of the chef, while "unstable"

refers to an unstable mechanism such as the chef having a bad day. Responses that did

not fit either category were coded "ambiguous." The relation between the level of

generalization and the assumed underlying mechanism is established by calculating the

correlation between the two, with the "mechanism" variable being a dichotomous

variable (unstable=0, stable=l). The goal of this measure is to provide directional

evidence for the effect of lay theories, even when a specific manipulation does not reveal

the anticipated impact of lay theories. In Experiment 1, the judges agreed in 91% of the

cases and came to a joint conclusion in the other 9%. The resulting correlation between

generalization and stability was r=.56 (p<.001).

Discussion

At least three observations are surprising in that they do not correspond to

predictions of prominent literatures: (1) contrary to predictions in the induction literature,

typicality failed to impact generalization; (2) a positivity instead of a negativity effect is

observed since generalization is higher in the positive condition than in the negative









condition; (3) contrary to predictions of the Law of Small Numbers, one condition

showed significantly lower generalization than another.

Do the results imply that typicality does not impact generalization "naturally"? A

claim based on this one marketing experiment would be far-fetched, but the results

suggest that generalization in the marketplace is not driven by mere typicality

calculations as suggested by the induction literature. Moreover, the results may indicate

that, if lay theories drive generalization, typicality may not always be the primary input

for such theories. Perhaps typicality does not "jump out" as much as previously believed.

The second surprising observation refers to the non-occurrence of a negativity

effect. Instead, generalization is higher in the positive than in the negative condition. This

result, too, is interesting for several reasons. First, it validates the "null effect" of the

typicality manipulation in that the latter cannot be attributed to negligent or inattentive

participants who slid the scale without thinking about or even reading the scenario shown

to them. Second, the mere observation that generalization varies between any two

conditions rules out an unmodified Law of Small Numbers as a framework to account for

the results. Third, whereas large bodies of literature anticipate a negativity effect,

Experiment 1 shows a positivity effect.

One may wonder why a positivity effect is observed in this experiment (and in

Folkes and Patrick's 2003 study) while a negativity effect is found in so many others.

Also, what drives the positivity effect? In line with the positive correlation between

generalization and stability of the underlying mechanism, one could argue that lay

theories drive the results and thus the positivity effect. If so, consumers believe that

positive outcomes are caused by stable mechanisms, such as the expertise of the chef,









while negative outcomes are caused by unstable mechanisms, such as the chef having a

bad day. This would suggest that people have different lay theories about positive and

negative outcomes in the commercial world--where a positivity effect is observed--than

in the social world, where a negativity effect has been reported (Ybarra 2002).

It is also possible, however, that generalization is not driven by typicality

calculations or by lay theories. Instead, people may know and use the objective

percentage of positive or negative experiences in restaurants. That is, base rates may

drive the positivity effect and thus generalization. Although we do not know the exact

base rates of positive and negative experiences in restaurants, it seems reasonable to

expect more positive than negative experiences, rather than the opposite.

In sum, exactly which factors are driving the results in Experiment 1 is impossible

to determine at this point. It is clear, however, that typicality does not drive

generalization. Furthermore, the overall data pattern is inconsistent with the Law of Small

Numbers, but not with the perceived heterogeneity framework, the lay theory account, or

a base rate explanation. The plausibility of a base rate explanation is addressed more

explicitly in Experiment 2.

Experiment 2

Experiment 2 allows further investigation of the impact of base rates on

generalization by presenting participants with two experiences, one positive and one

negative. The order in which they appear and the time interval between them are

manipulated. Neither factor should impact generalization when objective base rates about

the occurrence of good and bad outcomes drive generalization. However, manipulation of

both factors may induce differential lay theories about why the specific set of outcomes









occurred. If lay theories drive generalization, differences may be expected as a function

of those factors. A third goal is to investigate whether consumers reason statistically

when confronted with overt heterogeneity: the positive and the negative purchase

experience. Nisbett. et al. (1983) argued that statistical reasoning is more likely to occur

when people have reason to believe that there is heterogeneity on the considered

dimension. This experiment explores whether overt heterogeneity is a sufficient source to

fuel perceived heterogeneity, induce statistical reasoning and thus inhibit generalization.

Method

In anticipation of a manipulation in Experiment 3, the restaurant context used in

Experiment 1 was changed to an explicit product or service context. Participants from the

University of Florida (n=182) and Erasmus University (n=l 10) read a scenario in which a

consumer experience--either a product or service experience--was depicted with either a

positive or a negative outcome. Before any assessment was recorded, participants were

told they experienced the same brand again--either one year or one week later, this time

with the opposite outcome as result (Appendix B for the full stimuli). The dependent

variable pertained to expectations about a third experience being positive, negative, or

unpredictable. A 201-point scale was presented, ranging from very negative (-100) over

fifty-fifty (0) to very positive (100). As in Experiment 1, participants were allowed to

express the rationale behind their response in an open-ended question.

A 2x2x2 between-subject design was employed, crossing the order of outcomes

(bad-good vs. good-bad) with delay between outcomes (one week/one year) and purchase

context (product/service).










Predictions

If people reason statistically--that is, they treat their two outcomes in the same way

they would treat two coin flip outcomes--the average estimate in all conditions should not

deviate from zero (Figure 4-2A). However, it is also possible that consumers plug in base

rates when generalizing. As Experiment 1 suggests, people may believe that positive

purchase experiences outnumber negative experiences. If such base rates are the basis for

generalization, a pattern similar to the one shown in Figure 4-2B can be expected.

*good-bad ,bad-good Egood-bad ,bad-good











A B
goodbad badgood good-bad Ebad-good










C D
Figure 4-2. These are some possible outcomes in experiment 2. A) Anticipated pattern of
people reason statistically, B) Expected pattern of base rates dominate, C)
Pattern of results indicating a primacy effect and D) Pattern of results
indicating a negativity effect.

A third possibility is that consumers have a theory as to why their specific set of

outcomes occurred. A variety of theories can be adopted, two of which are considered

here. First, consumers may emphasize either the more recent or the first experience









(Figure 4-2C). This would result in either a recency or a primacy effect. Both phenomena

have been reported extensively by a variety of researchers. For instance, in the social

realm, Nisbett and Ross (1980) stated that "primacy effects are overwhelmingly more

probably" (p.172). Research on cognitive ability (Jones et al. 1968) as well as a review

paper on order effects suggest that, indeed, the first impression is often considered the

most important one, at least in a short and easy task such as that used in Experiment 2

(Hogarth and Einhorn 1992). However, research in other domains suggests recent

information may be more dominant (Cuccia and McGill 2000 regarding accounting,

Davis 1984 regarding jury decision-making).

Alternatively, but not necessarily mutually exclusively, consumers may hold

theories in which either the negative or positive experience dominates while the other is

perceived as an outlier or the beginning or end of a trend. In line with the result of

Experiment 1, it may be that the positive outcome will dominate. In that case, a recency

effect would be observed in the bad-good and a primacy effect in the good-bad condition.

Note that such a result would be very much in line with a base rate explanation.

Alternatively, as suggested by a large body of other evidence, it is possible that negative

information will dominate the estimation responses (Figure 4-2C).

Results

Brief inspection of the means in Figure 4-3 indicates that none of the predicted

patterns is obtained. Instead, the means show a combined emphasis on recent and positive

information. In addition, generalization is stronger when there is a one-year, as opposed

to a one-week, delay between the two outcomes. This pattern is confirmed by a 2x2x2

[delay (one week/one year) x order of outcome (good-bad/bad-good) x purchase context

(product-service)] ANOVA on the absolute values, with all three factors considered as









between-subject factors. Absolute values are used rather than the raw data because the

raw data may fail to pick up a time lag effect even if it is present. The main effects of

time lag [F(1,284)=9.3; p< .01] and order of outcome [F(1,284)=18.2; p<.001] confirm,

respectively, that generalization is more extreme after one year rather than one week, and

more extreme in the bad-good condition than the good-bad condition. No evidence is

provided for a main effect of purchase context [F(1,284)=1.8; p<.18)] or for any

interaction (highest F=1.9; p<.17).

*good-bad mbad-good










Figure 4-3. The observed pattern of results as a function of time delay and order in which
the outcomes are presented.

Because this is the only experiment with inconsistent outcomes, the coding scheme

for the cognitive responses is different in this experiment. Judges classified the responses

either as evidence for statistical reasoning or for a lay theory (other than statistical

reasoning). "Statistical reasoning" responses refer to a 50/50 chance of the next

experience being positive/negative. A lay theory response was defined as a reference to

one experience dominating the other, such as 'the company is improving', or 'the

negative experience was a fluke'. The judges agreed in 94% of the cases. References to a

lay theory tend to dominate as generalization became more extreme. For instance, in the

bad-good condition with a one-year time interval, 85% of the responses were classified as

referring to a trend or an outlier, while 14% were classified as statistical reasoning (1%









was not classified in either category). References to statistical reasoning were relatively

more dominant when generalization was less extreme. For instance, in the good-bad

condition with a one-week interval, 57% of the responses were classified as a lay theory,

and 36% as statistical reasoning (7% were not classified in either category).

Discussion

Experiment 1 failed to support the typicality account, but could not rule out a base

rate account as an explanation for its findings. Indeed, consumers may have had

knowledge about the overall percentage of positive and negative purchase outcomes and

used that knowledge exclusively to make their generalization estimate. In a similar way,

participants in Experiment 2 could have plugged base rate knowledge into their estimate.

For instance, if 60% of purchases can be objectively classified as having a positive

outcome, at least a positive number should have appeared in each of the four conditions

without significant differences across the conditions. However, the observed pattern of

results is very different.

In addition, the pattern of results deviates significantly from what could be

expected if consumers reason statistically. Aggregate generalization levels were strongly

different from zero in all four conditions, despite the strong overt heterogeneity. Clearly,

the combination of one positive and one negative purchase experience is not perceived in

the same way as a head-and-tail result from two coin flips. In other words, the results in

Experiment 2 cannot be fully explained by the objective heterogeneity of the sample or

by the objective heterogeneity of the larger population of positive and negative purchase

experiences; that is, the base rates.

Still, considerable systematic variance is left: A combined recency and positivity

effect is even more pronounced as the time interval increases. What is causing this









variance? One possible interpretation is that consumers try to give meaning to--come up

with a lay theory about--the sequence of events encountered and that this meaning is

reflected in their generalization estimate. For instance, and in line with the results of the

cognitive responses, participants in the bad-good condition may believe that this

sequence of outcomes occurred because the company is improving. Understandably,

there is more room for improvement in a time span of one year as opposed to one week.

The result is therefore more extreme in the one-year condition. The opposite seems true

for the good-bad condition, but to a lesser degree.

These results, then, may provide the first piece of evidence, albeit indirect, in favor

of a lay theory perspective, while at the same time making a simple base rate

interpretation less likely. The Law of Small Numbers is supported in that people do not

tend to reason statistically, even in the face of strong heterogeneity cues, but it is

contradicted in that consumers do not find a small sample representative.

The specific content of the lay theories is not a first concern, but it is certainly not

inappropriate to ask why these theories and not others? One reason positive rather than

negative information may dominate theories is that consumers may assume the company

they interacted with is in business. Given this assumption, it is more logical to expect that

a company is improving or performing well (bad-good condition), rather than

deteriorating or performing badly (good-bad condition).

The other question pertains to why lay theories are dominated by recent rather than

primacy information. Research on cognitive--rather than commercial--skills has shown a

primacy effect (Jones et al. 1968). The same pattern dominates in the social realm

(Nisbett and Ross 1980) and was anticipated in an impressive review on recency and









primacy effects (Einhorn and Hogarth 1992). Einhorn and Hogarth expect a recency

effect when more complex and cognitively demanding tasks are involved or when two

estimates are made (one after each outcome) instead of one. The task in Experiment 2 is

classified as short and easy, and participants make only one estimate, as soon as they

received information about the valence of both outcomes. The only way to align our

findings with Einhorn and Hogarth's framework is to assume that participants made two

estimates--one implicitly after the first experience and the other explicitly after the

second experience. Future research may want to explore this possibility, for instance, by

explicitly having participants make two estimates--one after each outcome. If the result

does not change, Einhorn and Hogarth's framework is supported. However, it is not

inconceivable that the procedural change might induce a primacy--instead of a recency--

effect, given that consumers may make up their mind quickly after a first experience.

Taken together, the results of Experiment 2 suggest (1) that consumers do not

reason statistically in the face of overt heterogeneity, (2) that simple base rates do not

(always) drive generalization, and (3) that, instead, lay theories may impact

generalization. A recurring observation is that generalization is more extreme on the

positive than on the negative side. Experiment 3 tests the boundaries of this positivity

effect.

Experiment 3

Experiment 3 explores the boundaries of the positivity effect. It also tests the

impact of lay theories about purchase experiences in a product versus a service context.

Experiments 1 and 2 showed that generalization is more extreme on the positive

than on the negative side. An emerging question is how persistent this positivity effect is.









Are consumers incurably positive? In other words, is even a set of multiple negative

experiences perceived as an exception and is the updating process consequently slow, as

suggested by Boulding et al. (1999)? Or does a second negative experience lead rather

quickly to the same level of extreme generalization as does one positive experience?

Such a finding would not be inconsistent with a lay theory perspective. It would suggest

that consumers perceive one negative experience as an exception, but two negative

experiences as an indication of stable negative performance.

The second goal of Experiment 3 is to explore whether consumers hold different

theories about various sets of purchase experiences and, more importantly, whether those

lay theories drive generalization. For instance, many scholars may agree with Zeithaml et

al. (1985) that heterogeneity is higher in a service context than in product context. If so,

consumers should generalize to a larger degree in a product than in a service context. The

specific theories underlying generalization may recognize that products are the output of

an automated production process whereas services are generated by a much more variable

mechanism. A final question is whether the positivity effect observed in a restaurant

scenario in Experiment 1 can be replicated in a different context.

Method

In this experiment, 93 students at the University of Florida were asked to imagine

they had had one or two experiences with "a product" or "a service," with the description

left vague. If two experiences occurred, the outcomes were consistent. In both the product

and the service conditions, the scenario showed four examples of a product/service in

parentheses (Appendix C for the full stimuli).









This resulted in a 2x2x2 design with purchase context (product vs. service),

number of experiences (one vs. two) and outcome valence (positive vs. negative) as three

between-subject factors. As in Experiment 2, the task was to estimate the likelihood that

the next experience would be positive or negative, and cognitive responses were recorded

after the generalization estimate had been submitted.


Predictions

A persistent positivity effect and thus slow updating, as suggested by Boulding et

al. (1999), would be evidenced by a pattern similar to the one depicted in Figure 4-4A.

.lexp *2exp UPosltve *Negatve










A B

*positive *negative *lep 2e'p











C D
Figure 4-4. Pattern of results as anticipated by several accounts (A, B and C) and as
observed in Experiment 3. A) Expected pattern if the positivity effect is to
persist when n=2, B) Pattern as anticipated by work of Zeithaml et al. (1990),
C) Expected pattern if a positivity effect occurs in a service context and a
negativity effect in a product context. D) The actually observed means as a
function of number of experiences and outcome valence.









Figure 4-4B shows the possible impact of a product-service effect: The direction

of generalization is identical in both conditions, but the magnitude is higher in the

product condition (Zeithaml et al. 1990). If, however, a positivity effect occurs in the

service condition, and a negative effect in the product condition (Folkes and Patrick

2003), the pattern should look like the one depicted in Figure 4-4C.

Results

Descriptively, Figure 4-4D shows a replication of the positivity effect when one

experience is involved, but no evidence for a positivity effect when two consistent

episodes are experienced. A 2x2x2 [valence (positive-negative) x number of exposures

(one-two) x purchase context (product-service)] ANOVA confirmed this pattern. The

number of exposures by valence of outcome interaction is significant [F(1,85)=7.8;

p<.01], thereby confirming the annihilation of the positivity effect after two exposures

[t(44)=1.2; p<.24 for the difference between positive and negative conditions after two

episodes on absolute values]. Although the main effects of number of exposures

[F(1,85)=4.7; p<.05] and valence [F(1,85)=540.7; p<.001] are significant, the main effect

of purchase context is not (F<1). Overall, generalization is more extreme after two

experiences than after one, and more extreme on the positive than negative side. Whether

the experience involves a product or service, makes no difference. In analyzing the

cognitive responses, the two judges agreed in 96% of the cases. The overall correlation

between generalization and stability is r=.39 (p<.053).









Discussion

Experiment 3 was designed to test boundaries of the positivity effect and thus the

speed with which consumers update their beliefs. At the same time, a first test probed the

impact of product- and service-related lay theories on generalization.

Despite the changed context and the use of a slightly different dependent measure,

the positivity effect observed in Experiment 1 is compellingly replicated in Experiment 3,

thus supporting the robustness of the phenomenon.

In addition, the results show that consumers confronted with two experiences

generalize equally extreme on the negative as on the positive side. That is, the positivism

of consumers has clear limits. One implication is that consumers may switch easily from

an unstable to a stable mechanism when generalizing. In other words, the increased

generalization in the negative outcome condition may be suggestive of a fast updating

process, which is inconsistent with Boulding et al.'s (1999) predictions but not with a lay

theory perspective.

The other interesting observation is the lack of differentiation between the product

and the service condition. Experiment 3 suggests that consumers do not hold the same

theories as academic scholars. Another possibility is that consumers do hold the same lay

theories, but that these do not impact generalization. One could argue that this is evidence

against the lay theory perspective. Still another interpretation, however, is that these lay

theories are held by consumers, but that the manipulation in Experiment 3 was too weak

to elicit them.

The full pattern of results in Experiment 3, then, seems at least partially

inconsistent with each of the three frameworks that reasonably could make predictions.









The Law of Small Numbers has difficulties with the low level of generalization after one

negative experience, while both the heterogeneity and the lay theory framework would

have anticipated differences as a function of purchase context.

When taken together, the full pattern of results observed in Experiments 1-3

cannot easily be accounted for by any of the reviewed frameworks. An unmodified Law

of Small Numbers and the typicality account cannot be retained, while the heterogeneity

and lay theory perspectives receive mixed support. Because the heterogeneity account

does not add much to the lay theory perspective, the next set of experiments targets the

validity of the lay theory perspective and further explores the degree to which consumers

hold theories about positive and negative experiences. They also further examine

purchase experiences in a product versus a service context.














CHAPTER 5
IN SEARCH OF LAY THEORIES AND THEIR IMPACT ON GENERALIZATION

The major goals of experiments reported in this chapter are to obtain direct

evidence for the existence of lay theories in a consumer context and for their impact on

generalization.

Experiment 4

Given that the evidence in favor of lay theories has been indirect and even then

inconclusive at best, the main goal of Experiment 4 is to provide direct evidence. Do

consumers believe that positive outcomes are caused by stable mechanisms, such as the

expertise of the chef, and negative outcomes by unstable mechanisms, such as the chef

having a bad day? Such beliefs are necessary to interpret the positivity effect in

Experiments 1 and 3 in terms of a lay theory perspective. A variant of the restaurant

scenario employed in Experiment 1 is used in Experiment 4.

Method

This experiment included 55 students, 27 of whom were randomly assigned to the

positive outcome condition. All participants were presented with a scenario in which a

restaurant critic goes to a particular restaurant to write a review. The critic's meal is said

to be "very good" for half of the participants and "not very good" for the other half. As

part of the review, the critic wants to explain why the restaurant produced a good/bad

meal. The task of each participant is to provide reasons that could be used in the critic's

review (Appendix D for full scenario). In a second phase, two independent judges who










were unfamiliar with the hypothesis of the experiment coded each reason as either a

"stable" or "unstable" mechanism.


Results

There were 120 separately codable units (causes) in the negative condition and

109 in the positive condition, all coded as "stable," "unstable" or "non-codable." The two

judges agreed in 74% of the cases. Inconsistencies were resolved through deliberation by

the two judges.


*stable Hunstable











positive negative

Figure 5-1. Classification of causes as either stable or unstable.

Of the codable responses, 94% was classified as "stable" (6% as "unstable") in

the positive outcome, compared to 51% in the negative outcome condition, where 49%

were coded "unstable" (Figure 5-1). The difference in the number of stable/unstable

causes between the positive and negative outcome condition is statistically significant

[X2=23.7 ; p<.001]. Non-codable responses comprised 52% of all responses in the

positive outcome condition and 61% in the negative outcome condition.


Discussion

Experiment 4 provides the first direct evidence for lay theories. Consumers are

more likely to believe that good outcomes (e.g., a good meal) are caused by a stable









mechanism and bad outcomes (e.g., a bad meal) by an unstable mechanism. This result is

consistent with the positivity effect observed in Experiments 1 and 3. However, as

interesting and important as this observation is, it is not helpful unless consumers use

such theories when generalizing. Experiments 5 and 6 further investigate the impact of

lay theories on generalization by probing the product-service distinction employed in

Experiments 2 and 3.

Experiment 5

Experiment 5 takes both Experiments 3 and 4 one step further. Having provided

evidence for the existence of lay theories (Experiment 4), the next step is to show the

impact of those theories on generalization. The context of product versus service

(Experiment 3) is selected again, with the manipulation slightly strengthened. The

generic product and service description in Experiment 3 is made more specific in

Experiment 5. The rationale is that the more specific the description, the more likely that

specific theories will be triggered, and therefore the more likely that differences in

generalization will occur between the product and the service condition.

Method

The goal in the stimuli selection was to include products that typically show little

variance in the production process and to choose service stimuli in which more variability

can be perceived. Toothpaste, a printer cartridge, and a battery were used as product

stimuli; financial advice, a hotel stay, and delivery service were used as service stimuli

(Appendix E for full stimuli).

In this experiment, all the purchase experiences were negative. The 127

participants imagined they had had either one or two negative experiences. The outcome









information was still rather general, just as in Experiment 3. A 2x2 [number of exposures

(one vs. two) x purchase context (product vs. service)] was employed.

Results

*products Mservlce












Figure 5-2. Average generalization as a function of number of experiences (one/two) and
purchase context (product/service).

The results are shown in Figure 5-2. Once again, despite the stronger

manipulation, no significant difference is observed between the product and the service

condition (F<1). Instead, the main effect of "number of exposures" is highly significant

[F(1,123)=50.5; p<.001]. The interaction is not significant (F<1) and neither is the single

main effect of replicate in the product [F(2,62)=1.6; p<.21] or service condition (F<1).

Agreement in coding the cognitive responses was reached in 91% of the cases and the

correlation was r=.39 (p<.05).

Discussion

Given the intuitively plausible difference in heterogeneity between a product and

a service context, the lack of any significant difference is striking. Again, one possible

explanation is that consumers do not hold the same theories about products and services

that most academics do. Or, consumers may hold those theories but not apply them when

generalizing. Experiment 6 further probes a third possibility: Consumers hold lay theories









about products and services similar to those that academics hold, but a stronger

manipulation is needed to elicit them.

Experiment 6

The rationale behind Experiment 5 was taken one step further in Experiment 6.

The purchase scenario was made even more specific. This time participants were not only

provided with specific product or service information, they were also informed about the

specific dimension on which the product or service did not perform appropriately.

Method

A pool of 95 students evaluated either four service or four product replicates.

Replicates in each condition were selected as intuitive representatives of exemplars in

their respective categories and to fall short in a specific characteristic. Replicates in the

product condition are "a razor blade that provides a somewhat rough shave," "a bad

tasting chocolate bar," "a wristwatch that is running behind," and "a pen that does not

distribute the ink evenly." "An undercooked meal," "rude service in a coffee shop," "your

lawn is not carefully mowed," and "being placed on hold for a long time at a call center"

are the replicates in the service condition (Appendix F for the full scenario). The question

to the participant was, "If you purchased 100 'replicate,' what percentage would perform

equally poorly?"

Results

Figure 5-3 shows the means in the product and service condition for each of the

replicates. Descriptively, average generalization seems higher in the product condition

than in the service condition. Statistical analyses confirm the main effect of purchase









context when the replicate is treated as a repeated measure, [F(1,93)=11.5; p<.001] and

almost when treated as a between-subject factor [F(1,87)=3.1; p<.09].













prod serv

Figure 5-3. Mean generalization per replicate in the product and service condition.

The main effect of replicate in the product [F(3,141)=20.3; p<.001], but also in

the service condition [F(3,138)=19.5; p<.001] was highly significant when the replicate

was treated as a repeated measure in two one-way ANOVAs, which indicates high

variability even within the product and service contexts. Agreement among judges in the

coding task was 88%. The correlation was r=.57 (p<.001).


Discussion

The results of Experiment 6 extend those observed in Experiments 3, 4, and 5 in

various ways. Relative to Experiments 3 and 5, overall generalization in the negative

condition of Experiment 6 increased in the product, but not in the service, condition. That

is, the specific description of the purchase situation in Experiment 6 led to the pattern

inferred from Zeithaml et al. (1990): higher generalization in a product context than in a

service context. A product-service distinction in Experiment 6 that is absent in

Experiments 3 and 5 suggests that, even if consumers hold generic lay theories about

purchase experiences in a product versus service context (e.g., the "assembly line"









theory), more particular lay theories about specific purchase situations, rather than broad

product-service differences, may be elicited more naturally.

By showing evidence for the product-service distinction, Experiment 6 extends

the results of Experiment 4--consumers hold lay theories--in multiple ways. First, it

provides evidence for the existence of another set of lay theories, a set reflecting the work

ofZeithaml et al. (1985). More importantly, it reinforces the lay theory perspective by

showing that those lay theories do affect generalization.

However much Experiment 6 provides important support for the lay theory

perspective and answers crucial questions, it certainly sparks new ones. It is informative

about the specific level of product- or service-related lay theories that consumers may

apply when generalizing. Although the observed main effect of product-service is

interesting and in line with the proposition inferred from Zeithaml et al. (1985), one

should not disregard the considerable systematic variance within both the product and

service conditions. This variance suggests that the specific dimension and/or the specific

product or service category may be as crucial as the broad product-service distinction.

Indeed, if just one of the replicates had been selected for each of the two conditions, it

would have been possible to observe a main effect of purchase context that is opposite to

the observed pattern: higher generalization in a service than in a product context. Such an

outcome is not anticipated by the classical heterogeneity distinction between products

and services (Zeithaml et al. 1990) and may add an important insight to the existing

theoretical literature, as well as to the knowledge base of the manager. Whether the

specific dimension or the specific category or a combination of both determines the









specific lay theory is impossible to infer from the design employed in Experiment 6.

Experiment 7 aims at disentangling the impact of both factors.

Experiment 7

Although Experiment 6 shows differences in generalization between a product

and a service context, in line with Zeithaml et al. (1990), the considerable variance within

the product and service conditions suggests that this distinction does not suffice to predict

an appropriate level of generalization. Instead, the specific product/service category may

be crucial just like the dimension on which the product/service performs inadequately.

Experiment 7 disentangles the impact of the two factors by keeping the category constant

and varying the dimension. Because more overall variance is expected in a service than in

a product context, only service replicates are included in Experiment 7. Dimensions that

had been considered relevant by previous researchers (Zeithaml et al. 1990; Coulter and

Coulter 2003) were selected and crossed with three service categories.

Method

Four dimensions of service quality were selected from Zeithaml et al. (1990):

competence, reliability, courtesy, and credibility. Each dimension was crossed with each

of three replicates: a restaurant scenario, a car repair scenario, and a painter scenario.

Operationalization for the four dimensions in the restaurant scenario was as follows

(Appendix H for the full stimuli): "The waiter mixed up the orders and therefore no one

gets exactly what he ordered" (competence); "The waiter is slow to take your order"

(reliability); "The waiter seems to be abrupt and unfriendly" (courtesy); "The waiter has

overcharged you" (credibility). The dependent measure was worded as follows: "For each










set of 100 customers, what percentage do you feel would have the same experience with

this waiter/mechanic/painter?" The study participants were 259 students.


Results

Descriptively, the results in Figure 5-4 show that generalization is a function of a

combination of the replicate (category) and the dimension. This observation is confirmed

by the statistical analyses in a 3x4 (replicate x dimension) ANOVA, with both factors

treated as between-subject factors. The interaction between dimension and replicate is

significant [F(6,247)=7.6; p<.001] and so are the main effects of dimension

[F(3,247)=16.7; p<.001] and replicate [F(2,247)=14.6; p<.001].


Compp *reliab Ocourt Ocred










restaurant car painter

Figure 5-4. Average generalization as a function of replicate and dimension.

Agreement among judges in coding the cognitive responses was 89%. The overall

correlation between the numeric level of generalization and the dichotomous level of

stability was r=.43 (p<.001). To illustrate, the extremely low generalization in the

credibility-restaurant scenario is backed up by an "unstable" rating for 14 out of 15

codable responses.









Discussion

No specific directional predictions were advanced at the outset of this experiment.

Instead, the main goal was to explore whether generalization varies as a function of the

specific dimension and/or category. The interaction between dimension and category

confirms that both the category and the dimension matter. Together with the results of

Experiment 6, the data in Experiment 7 indicate that generalization can vary as a function

of the broader category (product vs. service: Experiment 6) but also as a function of the

specific category and dimension within a service context. The suggestion is that

consumers come up with a different theory, depending on the very specific purchase

situation in which they find themselves. For instance, overcharging seems to be perceived

more of an exception when performed by a waiter than when performed by a mechanic or

painter. Overcharging by a waiter also seems to be perceived as more of an exception

than unfriendliness of a waiter. That is, not only is the product-service distinction too

broad to determine an accurate generalization level, so is the specific product or service

category. The performance dimension is also needed to predict generalization.

An even more extreme example of how specific lay theories can be is suggested

by a comparison across experiments. Experiment 7 includes a restaurant/unfriendly

(courtesy) scenario while Experiment 6 describes a coffee shop/rude scenario. Although

both scenarios seem highly similar and even interchangeable--especially against a

background of broad product-service differences--generalization is considerably higher in

the restaurant/unfriendly scenario (mean = 59) than in the coffee shop/rude scenario

(mean = 38). It is entirely possible that this difference is not systematic and is attributable









to error. However, it would be interesting to see whether such subtle differences spark

differential theories. Experiment 8 pursues this possibility.

Experiment 8

Context (restaurant/coffee shop) and behavior (unfriendly/rude) are orthogonally

manipulated. The suggestion is that consumers may perceive negative outcomes to be

caused by more stable mechanisms in a coffee shop than in a restaurant and that

unfriendly behavior is seen as more stable than rude behavior. It would be interesting if

consumers are sensitive to these subtle differences, but not to such seemingly obvious

marketing variables as "product" and "service" at the broad generic level (Experiments 3

and 5).

Method and results

Sixty-five undergraduate students participated in this 2x2 experiment that crosses

context (restaurant/coffee shop) with behavior (rude/unfriendly) as between-subject

factors. The students were told to imagine an encounter with a rude/unfriendly waiter in a

restaurant/coffee shop and were asked what percentage of customers they feel would

have the same experience in this place (Appendix I for the full stimuli). Confirming the

across-experiment differences twice (Figure 5-6), the two main effects are significant.

Generalization is higher in a coffee shop than in a restaurant [F(1,61)=4.7; p<.04], and

higher with the unfriendly than the rude behavior [F(1,61)=4.1; p<.05]. There is no

evidence for an interaction (F<1). Coding of the cognitive responses resulted in 89%

agreement and a correlation of r=.66 (p<.001).










Erude *unfnendly










restaurant coffee

Figure 5-6. Mean generalization as a function of purchase context and behavior.

Discussion

The differences observed across experiments between rudeness and unfriendliness

on the one hand and a restaurant and coffee shop context on the other hand are replicated

in a well-controlled environment. This suggests that consumers perceive true differences

between seemingly interchangeable contexts or behaviors. This finding, together with the

positive correlation between generalization and stability, is important from a marketer's

perspective and is interesting because it suggests that rudeness/unfriendliness are

perceived as being caused by a more unstable mechanism in a restaurant than in a coffee

shop. At the same time, rudeness is perceived as being caused by a more unstable

mechanism than unfriendliness. However, one cannot rule out a simple base rate

explanation: Rude behavior may be less likely in a restaurant than in a coffee shop and

rude behavior may occur less frequently than unfriendly behavior. If so, future research

may want to investigate the interplay between causal lay theories and base rates in

determining generalization.


Before turning to Chapter 6, I summarize the evidence collected in Experiments

4-8. The second set of experiments supports the lay theory account where the first set did

not. Experiment 4 shows that consumers hold lay theories consistent with the positivity









effect observed in the first set. Experiment 6 indicates that consumers not only hold lay

theories but also apply them when generalizing. Experiments 7 and 8 suggest that those

lay theories tend to be more specific than could reasonably be anticipated. The

suggestion--when n=l--is that generalization is dependent on the very specific lay theory

that is applied, and not easily predictable. To what degree the same is true as the sample

size increases is not clear. The question is especially pertinent in those situations when

n=l and where an unstable mechanism is assumed. The third set of experiments explores

this issue more systematically.














CHAPTER 6
MULTIPLE "UNSTABLE" OBSERVATIONS

Experiment 9

Given that consumers seem to be unwilling to generalize in some situations, it

becomes important to understand the boundaries of this reluctance as the number of

observations increases. It is possible that Boulding et al.'s (1999) prediction holds and

that an unstable mechanism dominates even after multiple negative outcomes have been

experienced. Alternatively, consumers may quickly assume a stable mechanism when

n=2 after initially having perceived an unstable mechanism. In other words, the

assumption of an unstable mechanism may be short-lived. If true, the implication may

well be that the Law of Small Numbers holds when n=2, if not when n=1.

Method

Ninety-five students participated in the study. The two scenarios producing the

lowest mean generalization in Experiment 7 were included in Experiment 9: the

wristwatch running behind and the undercooked meal. In addition, two scenarios from

Nisbett et al. (1983) were included for comparison: the element floridium burning with a

green flame on an imaginary island and obese members of a tribe (Barratos) on this

island. Extreme generalization has been observed with floridium, even after one trial,

while moderate to low generalization has been observed for obesity (Appendix I for full

stimuli).









Results

Descriptively, the low levels of generalization observed after one episode are

replicated for the marketing stimuli (Figure 6-1). Similarly, the high level of

generalization for the floridium element in Nisbett et al. (1983) is replicated, as is the

moderate level of generalization for the obese Barratos. Generalization after two episodes

increases dramatically for the two marketing stimuli, but much less or not at all for the

other replicates.


exposs m2expos









watch restaurant obese element

Figure 6-1. Mean generalization as a function of replicate and number of exposes.

These observations are confirmed by a replicate by number of exposures

interaction [F(3,87)=6.1; p<.01]. The single main effect for number of exposures is

significant for the marketing stimuli [F(1,46)=43; p<.001], but not for the two other

replicates [F<1]. Overall, the main effect of number of observations is significant

[(F(1,87)=28.7; p<.001)], as is the main effect of replicate [F(3,87)=9.2; p<.001]. Coding

of the cognitive responses led to agreement in 93% of the responses and an overall r=.63

correlation (p<.01).


Discussion

All four cells with a sample of n=l replicate generalization previously observed

either in Experiment 6 or in Nisbett et al. (1983). Generalization is low in the wristwatch,









restaurant, and obese scenario, while extremely high in the floridium condition. In

addition, generalization based on two observations in the obesity and floridium cases

seems in line with the results that Nisbett et al. reported. The observed pattern in the

obese condition seems to correspond to the slow updating process anticipated by

Boulding et al (1999). However, the contrast with the marketing stimuli is stark. Whereas

generalization was extremely low after only one observation, the addition of a second

observation induced a huge increase. In this case, the updating process does not seem to

be slow but extremely fast. The suggestion is that even though consumers may initially

surmise an unstable mechanism, a stable mechanism is assumed as soon as the same

episode is experienced twice. The lesson for the marketer may not be that consumers

learn and update slowly, as argued by Boulding et al., but that they infer consistent low

quality easily, even when high quality is initially expected. However, it is also possible

that the high generalization when n=2 is the result of participants' compliance with what

they think is the hypothesis pursued by the experimenter. Experiment 10 pursues this

possibility.

Experiment 10

Thus far all experiments that included marketing stimuli showed high

generalization in the two-experience conditions. Experiment 10 explores whether the

high generalization when n=2 should be attributed to "demand" or "blind generalization."

To test this contention, one condition is included in which the target service behavior for

the second negative experience is explicitly said to be performed by a different rather

than the same waiter. The actual outcome is identical to that when the second experience

is caused by the same waiter. If the previous results are attributable to "blind










generalization" when n=2, high generalization may reasonably be expected in

Experiment 10 when n=2, even when the waiter in the second episode is different.

Alternatively, if lay theories--instead of "demand"--drive generalization, one might

expect low generalization when the waiter is different in the second episode.


Method and Results

Sixty participants were asked to imagine they went to a coffee shop and had one

or two negative service experiences with a rude waiter. The second experience either

involved the same waiter or a different waiter. When the waiter was the same,

participants were asked, "For each set of 100 customers, what percentage do you feel

would have the same experience with this waiter?" When the waiter of the second

experience was a different one, the question was, "For each set of 100 customers, what

percentage do you feel would have the same experience in this coffee shop?" (Appendix J

for the full stimuli) This resulted in a 2x2 design that crosses the number of experiences

with the target of generalization (this waiter/this coffee shop). Because the dependent

measure is different in the two "target of generalization" conditions, the data are analyzed

separately. For purposes of presentation, the means are presented in one figure.


*1 expos 2 expos











Indiv shop

Figure 6-2. Mean generalization in each of the four conditions in Experiment 10.









The mean levels of generalization per condition indicate that generalization is

considerably higher after two experiences than after one in the "individual" condition but

much less in the "coffee shop" condition (Figure 6-2). T-tests confirm that the former is

highly significant [t(28)=-4.7; p<.001] while the second does not even approach

significance [t(28)=-1.1=p<.3]. The correlation between level of generalization and

stability is r=.47 (p<.01). The judges agreed in 97% of the responses.

Discussion

When rude behavior was displayed in the second experience by the same waiter,

generalization increased dramatically, just as in previous experiments. However, when

the target behavior was associated with a different waiter, generalization did not increase

relative to the one-experience condition. This result suggests that participants did not

generalize "blindly" in previous experiments when n=2 and that the Law of Small

Numbers does not always hold, even when n=2. Instead, the suggestion is that

generalization can be low when n=2 if the appropriate theory is cued, e.g., a different

waiter is responsible for the negative outcome.














CHAPTER 7
GENERAL DISCUSSION AND CONCLUSION

"When do consumers (not) generalize?" and "How should one understand

generalization in the marketplace theoretically?" are the central questions in this

dissertation. The experiments show that generalization is high in a product context rather

than a service context and when the outcome is positive rather than negative, but not

when the sample is typical rather than atypical. Even within a product or service context,

considerable variance indicates that generalization can be high in a service context and

low in a product context. Although two consistent outcomes tend to be perceived as

representative of the larger population, two inconsistent outcomes tend not to be.

The results are interpreted in line with a causal lay theory perspective.

Generalization is high when the assumed mechanism is stable, low when the mechanism

is unstable. Direct evidence for the existence of such lay theories is provided by

Experiment 4, which shows that consumers tend to believe that stable mechanisms cause

a positive outcome while unstable mechanisms cause a negative outcome. Evidence for

the impact of lay theories on generalization is provided by Experiment 6, where

generalization is higher in a product context than in a service context. Across

experiments, correlational evidence shows that generalization tends to be higher when a

stable rather than unstable mechanism is assumed.

The level of correspondence between these results and previous research differs

depending on the domain selected for comparison. Correspondence is lowest with the









majority of research in the generalization and induction domain, but surprisingly high

with research in the social psychology literature and with category formation research.

Relevant literatures are discussed and suggestions for future research are incorporated in

the following discussion that focuses first on the claim that lay theories drive

generalization and then on the actual lay theories.

The Lay Theory Account

A major inconsistency occurs when the focus of comparison is the "calculation"

account suggested by the induction literature (Osherson et al. 1990; Sloman 1993; Heit

2000). In many induction studies, typicality exerts a pervasive and reliable impact, yet it

failed to influence generalization in this research. Does this discrepancy suggest that two

qualitatively different mechanisms drive generalization in a biological (induction)

context compared to a marketing or social context? Do lay theories drive generalization

in marketing and social contexts, while calculations predominate in a biological context?

Or is the lay theory account a special case of the calculation account or perhaps the other

way around?

Another look across the borders of strict generalization, induction, and even social

psychology research might be instructive. A seminal paper on category formation by

Murphy and Medin (1985) suggests a way to resolve the apparent discrepancy. Murphy

and Medin rejected the long-held belief that similarity calculations determine which

objects are grouped together to form a category (Medin and Schaffer 1978; Posner and

Keele 1968; Rosch and Mervis 1975). They argued that the concept of similarity is too

unconstrained (Goodman 1972). Instead, they proposed that "concepts are coherent to the

extent they fit people's background knowledge and naive theories about the world"









(Murphy and Medin 1985, p.289; also Rehder and Hastie 2001; Rips and Collins 1993).

Simply put, even if similarity calculations drive category formation, lay theories drive

and constrain similarity judgments and thus category formation. Transposed to the

domain of generalization, Murphy and Medin's claim suggests that even typicality

findings in the induction literature are the result of a specific set of theories about

biological categories rather than the output of simple and "objective" calculations. The

suggestion is that generalization in the induction literature is therefore not qualitatively

different from generalization in a consumer context or in a social context. The paradox is

resolved by considering the typicality effect as induced by subjective lay theories rather

than by objective calculations. Future research may want to investigate this empirically.

Still, even if valid, Murphy and Medin's (1985) suggestion does not explain why

generalization seems hypersensitive to typicality information in the induction literature,

but not at all in our context. Are there fundamental differences between lay theories about

the biological world and lay theories in the consumer and social world? This is certainly

an interesting hypothesis to be investigated in future research. Alternatively, there may be

differences between the paradigms that are responsible for the discrepancy. Perhaps the

induction paradigm favors typicality more than does the paradigm used in this research

project. Induction scenarios tend to include both a typical and atypical sample. As such,

one can argue that typicality information is made salient and a typicality effect is more

likely. Should one go as far as implying that the induction paradigm induces artificial

effects that do not exist in the real world? Probably not. Instead, the induction paradigm

may simply highlight certain aspects of a situation, and historically this has often been

typicality information. As such, the paradigm can be considered a valuable tool for









detecting specific sets of theories. In fact, an interesting challenge for future research

would be to employ the induction paradigm as a tool to test the impact of theories not

based on typicality. For instance, managers may want to know if consumers hold

different lay theories according to whether a company owns or franchises its stores

(Agrawal and Lal 1995). Vertically integrated systems (owning) may induce more

generalization across stores than does the franchising format because more independence,

and potentially instability or heterogeneity, is allowed in the latter. Similar questions can

be addressed regarding "make" or "buy" decisions (Anderson and Weitz 1986). Do

consumers infer more stability when something is made by the company rather than

bought from another company? Also, consumers may hold different theories about

"direct" salespeople versus "representatives" (Anderson 1985). Perhaps theories about

representatives imply lower levels of generalization than do theories about "direct"

salespeople. Systematically including both options in a stimuli set, may give a sense of

the degree to which consumers hold different theories in each of these situations.

My results are not necessarily inconsistent with the hypothesis generation

literature and with Hoch and Deighton's (1989) suggestion that consumers generate very

few hypotheses--often only one--about why a certain outcome occurred. However, my

research extends theirs by positing that the stability of the underlying mechanism is

important in determining the level of generalization, even if only one hypothesis is

generated.

The results are at least partially inconsistent with the Law of Small Numbers.

Although generalization tends to be high when the outcome is positive, and in a product

context even when the outcome is negative, it is often low in a service context when the









outcome is negative. Even an overtly heterogeneous sample is not perceived as

representative.

Interestingly, many of the findings that are inconsistent with the Law of Small

Numbers are in line with Nisbett et al.'s (1983) view that people, in at least some

contexts, perceive heterogeneity and thus refrain from generalizing. Our results extend

Nisbett et al.'s research by identifying factors that do and do not seem sufficient to

explain the variance in generalization. Neither the objective heterogeneity of the

population (base rates) nor the objective heterogeneity of the sample proved sufficient to

understand variance in generalization. That even typicality calculations failed to impact

perceived heterogeneity is perhaps even more perplexing. Instead, causal lay theories

seem to determine the level of perceived heterogeneity and thus generalization.

My results seem most consistent with findings in social psychology that people

hold and apply causal lay theories in a variety of judgments. Research on stereotypes has

long agreed that people hold beliefs--mostly stable--that impact judgment dramatically

(Kunda and Spencer 2003). Research on the correspondence bias (Gilbert and Malone

1995), on person versus group perception (Hamilton and Sherman 1996) and on valenced

behaviors (Ybarra 2002) implies that people perceive more stability in some situations

than in others.

The results are also consistent with research on causal attribution, which states

that people seek out causal mechanisms in developing an explanation for a specific event

and that they do not necessarily need covariation information (Ahn et al. 1995).

Thus far, the discussion has concentrated on consistency between relevant

literatures and my claim that lay theories drive generalization. At a more specific level, it









may be instructive to compare the actual lay theories that have been observed with what

could reasonably be expected on the basis of previous research.

Specific Lay Theories

The number of lay theories that can be applied is virtually countless. We have

started to document a few of them--some about individual purchase experiences, others

about a sequence of two experiences.

First, consumers tend to believe that a positive outcome is generated by a stable

mechanism, while a negative outcome is believed to be caused by an unstable

mechanism. Consequently, positive outcomes lead to higher generalization than do

negative outcomes. This finding is consistent with what Folkes and Patrick (2003)

observed in a service context, but is opposite to what an entire body of research in the

psychology literature anticipated. For instance, Ybarra (2002) argued that negative

human behavior tends to be seen as the consequence of a stable mechanism--a personality

trait--whereas positive behavior is perceived as the consequence of an unstable

mechanism, e.g., situational factors. Together, these results suggest a perplexing

dissociation between the social and the commercial world regarding the valence-stability

relationship: Whereas people seem to believe that their own peer is inherently bad, a

commercial entity seems to be perceived as inherently positive. Although this

dissociation sounds like a scary thought in a world of ever-present and sometimes

aggressive marketers, it is comforting to know that such positive beliefs about a

commercial entity are not long-lasting, as suggested by Boulding et al. (1999), but change

rapidly as multiple negative outcomes are experienced.









Second, the results suggest that consumers tend to believe that a product is

generated by more stable mechanisms than is a service experience. Consequently,

generalization tends to be higher in a product than in a service context, as suggested by

the work of Zeithaml et al (1990). However, this seems true only when the outcome is

negative. When the outcome is positive, stable mechanisms are assumed across the

board--in line with the positivity effect--and generalization is high.

Another deviation pertains to the specificity of product- and service-related lay

theories. Contrary to marketing scholars, consumers do not seem to hold differential

theories regarding heterogeneity in a product versus service context at the generic level.

Indeed, the product-service difference did not occur until very specific scenario

descriptions were introduced. The implication is that, even though the product-service

distinction may be an interesting guideline, generalization can vary greatly even within a

product or service context, possibly to such a degree that generalization can be higher in a

service context than in a product context. Equally important in determining the specific

level of generalization is the specific product/service category and even the specific

dimension on which the product/service is performing inadequately.

Still, the product-service distinction is an interesting parallel to findings in the

social psychology literature about person versus group perception (Hamilton and

Sherman 1996). Group behavior and a service experience seem to induce perceived

heterogeneity (instability) much more than individual behavior and a product experience.

My research suggests that this is true only when the outcome is negative. It would be

interesting for future research to investigate whether valence moderates the effect in the

social realm as well. If it does, one may wonder about the direction of the moderation









since a negativity rather than a positivity effect has been reported in a social context

(Ybarra 2002).

While theories about individual experiences seem to vary as a function of the

valence and the nature of the purchase (product-service), theories about multiple

inconsistent experiences tend to vary as a function of the delay between the outcomes and

the order in which they appear. In addition, in line with theories about individual

experiences, generalization tends to be more stable as function of a positive rather than

negative experience. The order effect is interesting because it indicates that the most

recent--rather than the first--outcome is more likely to be considered as a stable indicator

of true quality. This is inconsistent with the majority of findings in the social psychology

literature where a primacy-- instead of a recency--effect has more often been observed

(Nisbett and Ross 1980). Whether the discrepancy between my findings in a marketing

context and the bulk of evidence in the social psychology literature is attributable to

procedural or more fundamental differences may be the target of future research. One

possibility is that my paradigm induced recency by probing only one response right after

information about the recent outcome had been processed, rather than two responses--

each of them administered after each outcome was experienced. More fundamental

differences between the two domains cannot be ruled out, and neither can a third

possibility.

In the social realm, a primacy effect has been observed in person perception but a

recency effect in group perception (Hamilton and Sherman 1996). Given the earlier

analogy between person perception and generalization in a product context on the one

hand and group perception and generalization in a service context on the other hand, it is









possible that a primacy effect would be observed in a product context and a recency

effect in a service context. One reason this pattern did not occur in Experiment 2 may be

the lack of specificity in the scenario.

Conclusion

Analogous to a hypothesis on concept coherence (Murphy and Medin 1985) and

to work in the social realm (Hamilton and Sherman 1996; Ybarra 2002), but not

anticipated by induction (Heit 2000) or generalization research (Tversky and Kahneman

1971), this dissertation claims that generalization is primarily driven by people's lay

theories about how the world functions--rather than by typicality calculations. To

understand how and when consumers generalize, this account suggests, one should know

which theory the consumer invokes in the specific purchase situation rather than (1)

probe the typicality of the purchase situation or (2) assume that each first experience with

a vendor is representative of all future experiences with the vendor. More specifically,

one wants to know the stability of the underlying mechanism. A virtually endless set of

theories can be applied, but it is the one-dimensional stability of the underlying

mechanism that determines the level of generalization.















APPENDIX A
STIMULI FOR EXPERIMENT 1

Imagine it's Friday evening and you're going out for dinner with a group of friends.

You decide to go the new restaurant downtown. You've arrived in the restaurant and

you're about to explore the menu.

Look at it carefully.

--- New screen ---

New York steak of Black Angus Beef with Crushed Black Peppercorn

Tenderloin Filet of Beef

SBaby Back Ribs slow cooked with soy, beer and garlic, glazed with the

house made barbecue sauce

Beef Kebab In Teriyake Marinade

Flat Iron Steak of Black Angus Beef Topped with Fresh Herb Butter

Pasta with your choice of Tomato, Pesto or Clam Sauce

Prime Rib of Beef

Medaillons of Buffalo Tenderloin wrapped in Apple-Wood Smoked Bacon,

served on a Grilled Portobello Mushroom

All entrees come with a salad, bread and one side dish

--- New screen ---

You decided to have the Pasta / Flat Iron Steak of Black Angus Beef

It turns out you liked the meal a lot. / It turns out you didn't think it was a very

good meal compared to other restaurants where you've eaten at.










--- New screen ---

What percentage of all meals in this restaurant do you expect to be of the same

quality as the meal you had?

Please move the sliding scale below to give your answer.

--- New screen ---

Why did you guess this percentage of other meals that would have the same

quality, rather than a lower or higher number?

Write down your answer in the box below. Don't press Continue before you have

finished writing.















APPENDIX B
STIMULI FOR EXPERIMENT 2

Imagine that you purchased a product/service (e.g. coffee, toothpaste, battery,

cartridge for printer, disposable lenses, financial advice, hairdresser, hotel stay, delivery

service, babysitting) from a particular firm for the first time. Your assessment of the

product/service was positive/negative. A week/year later you decide to try the brand

again. This time, however, your experience is negative/positive.

--- Next screen ---

Now, imagine that you have the opportunity to purchase a third time from the same

firm. Using the scale below, please express your belief regarding the probability that the

third experience will be good versus bad.

--- Next screen ---

Why did you give the answer you gave? That is, please describe the rationale

behind your response.

Write down your answer in the box below. Don't press Continue before you have

finished writing.















APPENDIX C
STIMULI FOR EXPERIMENT 3

Imagine that you purchased a product/service (e.g. coffee, toothpaste, battery,

cartridge for printer, disposable lenses, financial advice, hairdresser, hotel stay, delivery

service, babysitting) from a particular firm for the first time. Your assessment of the

product/service was positive/negative. [Sometime later you decide to try the brand again.

Your experience is positive/negative again.]

--- Next screen ---

Now, imagine that you have the opportunity to purchase once more from the same

firm. Using the scale below, please express your belief regarding the probability that your

next experience will be good versus bad.

--- Next screen ---

Why did you give exactly the rating you gave, instead of a lower or higher number?

That is, please describe the rationale behind your response.

Write down your answer in the box below. Don't press Continue before you have

finished writing.














APPENDIX D
STIMULI FOR EXPERIMENT 4

Restaurants try to provide good meals to their customers. Imagine that you just

went to a particular restaurant and you had a meal that was (not) very good. Please list as

many reasons as possible as to why this could have happened.

Write down your answer in the box below. Don't press Continue before you have

finished writing.

Please separate different reasons with a star (*).















APPENDIX E
STIMULI FOR EXPERIMENT 5

Imagine that you purchased a tube of toothpaste / battery / cartridge for a printer //

financial advice / hotel stay / delivery service from a particular firm for the first time.

Your assessment of toothpaste / battery / cartridge for a printer // financial advice / hotel

stay / delivery service was negative. [Sometime later you decide to try the brand again.

Your experience is negative again].

--- Next screen ---

Now, imagine that you have the opportunity to purchase once more from the same

firm. Using the scale below, please express your belief regarding the probability that your

next experience will be good versus bad.

--- Next screen ---

Why did you give exactly the rating you gave, instead of a lower or higher number?

That is, please describe the rationale behind your response.

Write down your answer in the box below. Don't press Continue before you have

finished writing.














APPENDIX F
STIMULI FOR EXPERIMENT 6

Imagine that you purchased a brand of razor blade for the first time. You try it and

find that it provides a rough shave. If you purchased this brand 100 times, what

percentage of the blades do you think would perform at the same level as the first one

you purchased?

Imagine you bought a brand of chocolate candy bar for the first time. You try it and

find that it doesn't taste very good. If you purchased 100 of these candy bars, what

percentage do you think would taste the same as the first one you tried?

Imagine that you purchased a new wristwatch. After a week you find out that the

clock is running behind. If you tried 100 watches of this model, what percentage do you

think would give an inaccurate time?

Imagine that you purchased a brand of ballpoint pen that you have never purchased

previously. You discover that the pen does not distribute the ink evenly, leaving the page

with blobs and smears. If you purchased 100 pens of this model, what percentage do you

think would perform equally poorly?

Imagine that you try a restaurant for the first time. You order your meal and find

that it is undercooked. If you visited this restaurant 100 times, what percentage of all the

meals you have do you think would be improperly cooked?

Imagine you go to a coffee shop and find the service to be rude. If you visited this

coffee shop 100 times, what percentage of the service encounters do you think would be

rude?









Imagine you just bought a new home and hired a company to take care of the

garden. The company will send someone to your house once a week to mow the lawn.

After the first week you find that the lawn is not mowed carefully. If the company

mowed your lawn 100 times, what percentage of times do you think the lawn would be

mowed poorly?

Imagine that you just bought a cell phone with a new service provider. You have

some questions about your plan and decide to call the help desk. It turns out that you are

placed on hold for 15 minutes before someone gets to your call. If you were to call the

help desk 100 times, for what percentage of your calls do you think the waiting time will

be (at least) as long as during your first call?














APPENDIX G
STIMULI FOR EXPERIMENT 7

Imagine that you go to a restaurant with a few friends. Each person orders a meal.

When the meals are delivered, you find that the waiter has mixed up the orders and that

no one gets exactly what they ordered in the way they wanted it prepared. For each 100

set of customers, what percentage do you feel would have the same experience with this

waiter?

Imagine that you go to a restaurant with a few friends. The waiter is slow to take

your order and to deliver the check at the end. For each 100 set of customers, what

percentage do you feel would have the same experience with this waiter?

Imagine that you go to a restaurant with a few friends. The waiter takes your order

but seems abrupt and not very friendly. For each 100 set of customers, what percentage

do you feel would have the same experience with this waiter?

Imagine that you go to a restaurant with a few friends. At the end of the meal you

receive your check and find that the waiter has overcharged you. For each 100 set of

customers, what percentage do you feel would have the same experience with this waiter?

Imagine that you go to a mechanic for an oil change and some other standard

maintenance. Afterward, you pick up your car and when you get home you find that your

car is leaking oil. For each 100 set of customers, what percentage do you feel would have

the same experience with this mechanic?

Imagine that you go to a mechanic for an oil change and some other standard

maintenance. The mechanic promises that the car will be ready in 2 hours but it actually









takes much longer. For each 100 set of customers, what percentage do you feel would

have the same experience with this mechanic?

Imagine that you go to a mechanic for an oil change and some other standard

maintenance. The mechanic agrees to fix it but you find him to be somewhat unfriendly.

For each 100 set of customers, what percentage do you feel would have the same

experience with this mechanic?

Imagine that you go to a mechanic for an oil change and some other standard

maintenance. The mechanic quotes you a price. When the job is finished, you pick up

the car and find that the mechanic charged you more than he initially quoted. For each

100 set of customers, what percentage do you feel would have the same experience with

this mechanic?

Imagine that you own a home and would like to have some rooms painted. You

hire a painter who does the job. Afterward, you notice that the painter had been sloppy in

places and that the end result is not as nice as you expected. For each 100 set of

customers, what percentage do you feel would have the same experience with this

painter?

Imagine that you own a home and would like to have some rooms painted. You

hire a painter who does the job. The painter says that he can get to your job in one week

but it actually takes significantly longer. For each 100 set of customers, what percentage

do you feel would have the same experience with this painter?

Imagine that you own a home and would like to have some rooms painted. You

talk to a painter who is willing to do the job. You find him to be somewhat unfriendly.






77


For each 100 set of customers, what percentage do you feel would have the same

experience with this painter?

Imagine that you own a home and would like to have some rooms painted. You

hire a painter who does the job. When the job is finished, you get a bill that is higher than

the price that the painter originally quoted. For each 100 set of customers, what

percentage do you feel would have the same experience with this painter?














APPENDIX H
STIMULI FOR EXPERIMENT 8

Imagine that you go to a restaurant / coffee shop. The restaurant / coffee shop

employs a lot of waiters and you are being served by one of them. You find him/her to be

somewhat rude / unfriendly. For each set of 100 customers, what percentage do you feel

would have the same experience in this place?














APPENDIX I
STIMULI FOR EXPERIMENT 9

Imagine that you purchased a new wristwatch. After a week you find out that the

clock is running behind. [You decide to test a second watch of the same model. The

performance of the second watch is identical to the first.] If you tried 100 watches of this

model, what percentage do you think would give an inaccurate time?

Imagine that you try a restaurant for the first time. You order your meal and find

that it is undercooked. [You decide to try the restaurant again. You order a meal and get

the same result.] If you visited this restaurant 100 times, what percentage of all your

meals do you think would be improperly cooked?

Imagine that you are an explorer who has landed on a previously unknown island in

the Southeastern Pacific. You encounter several new animals, people and objects.

Suppose you encounter a native who is a member of a tribe called the Barratos. He is

obese. [You encounter a second member of the tribe and he is also obese.] If you

encountered 100 male Barratos, what percentage do you think would be obese?

Imagine that you are an explorer who has landed on a previously unknown island in

the Southeastern Pacific. You encounter several new animals, people and objects.

Suppose you encounter a sample of a new element you call floridium. Upon being heated

to a very high temperature, it burns with a green flame. [You encounter a second sample

and it also burns with a green flame.] If you encountered 100 samples of floridium, what

percentage do you think would bum with a green flame?















APPENDIX J
STIMULI FOR EXPERIMENT 10

Imagine that you go to a restaurant with a few friends. The restaurant employs a lot

of waiters and you are being served by one of them. S/he takes your order but seems

abrupt and not very friendly. [Sometime later you visit the restaurant again. It turns out

that you are served by the same/a different waiter]. For each set of 100 customers, what

percentage do you feel would have the same experience with this waiter/in this

restaurant?















LIST OF REFERENCES


Agrawal, Deepak and Rajiv Lal (1995) "Contractual Arrangements in Franchising: An
Empirical Investigation," Journal ofMarketing Research, 32 (May), 213-221.

Ahluwalia, Rohini (2002) "How prevalent is the negativity effect in consumer
environments?" Journal of Consumer Research, 29 (September), 270-279.

Ahn, Woo-Kyoung and Charles W. Kalish (2000) "The Role of Mechanism Beliefs in
Causal Reasoning," In F.C. Keil and R.A. Wilson (Eds.) Explanation and
Cognition, Cambridge, MA: MIT Press, (199-227).

Charles W. Kalish, Douglas L. Medin, and Susan A. Gelman (1995) "The Role of
Covariation versus Mechanism Information in Causal Attribution," Cognition, 54,
299-352.

Alba, Joseph W., Stijn M.J van Osselaer and Wouter Vanhouche (2003) "Multicausal
thinking and Free Will," Manuscript under preparation.

Anderson, Erin (1985) "The Salesperson as Outside Agent or Employee: A Transaction
Cost Analysis," Marketing Science, 4 (3), 234-254.

and Bart A. Weitz (1986) "Make-or-Buy Decisions: Vertical Integration and
Marketing Productivity," Sloan Management Review, Spring, 3-19.

Aristotle (1963), Organon graece, (vol. 2) Dubuque, IA: W.C. Brown Reprint Library.
(Original translation published 1846).

Arocha, Jose F., Vimla L. Patel and Yogesh C. Patel (1993) "Hypothesis generation and
the coordination of theory and evidence in novice diagnostic reasoning," Medical
Decision Making, 13, 198-211.

Bockenholt, Ulf and Elke U. Weber (1993) "Toward a theory of hypothesis generation in
diagnostic decision making," Investigative Radiology, 76-80.

Boulding, William, A. Kalra, and Richard Staelin (1999) "The Quality Double
Whammy," Marketing Science, 18(4), 463-484.

Chapman Loren J. and Jean P. Chapman (1969), "Illusory Correlation as an Obstacle to
the Use of Valid Psychodiagnostic Signs," Journal ofAbnormal Psychology, 71
(3), 271-280.









Cheng, Patricia (1997) "From Covariation to Causation: A Causal Power Theory,"
Psychological Review, 104, 367-405.

Coulter, Keith S. and Robin A. Coulter (2003) "The Effects of Industry Knowledge on
the Development of Trust in Service Relationships," International Journal of
Research in Marketing, 20, 3 1-43.

Cuccia, Andrew D. and Gary A. McGill (2000) "The Role of Decision Strategies in
Understanding Professionals' Susceptibility to Judgment Biases," Journal of
Accounting Research, 38 (2), 419-435.

Davis, J. H. (1984) "Order in the courtroom," In D.J. Miller, D.G. Blackman, & A.J.
Chapman (Eds.) Perspectives in Psychology and Law. New York: Wiley.

Einhom, Hillel J. and Robin M. Hogarth (1982) "Prediction, Diagnosis, and Causal
Thinking in Forecasting," Journal ofForecasting, 1, 23-36.

and Robin M. Hogarth (1986) "Judging Probable Cause," Psychological Bulletin,
99 (1), 3-19.

Fisher, Stanly D, Charles F. Gettys, Carol Manning, Tom Mehle, & Suzanne Baca (1983)
"Consistency checking in hypothesis generation," Organizational Behavior and
Human Performance, 31, 233-254.

Folkes, Valerie S. and Michael A. Kamins (1999) "Effects of Information about Firms'
Ethical and Unethical Actions on Consumers' Attitudes," Journal of Consumer
Psychology, 8 (3), 243-259.

and Vanessa M. Patrick (2003) "The Positivity Effect in Perceptions of Services:
Seen One, Seen them all?" Journal of Consumer Research, 30 (June), 125-137.

Garst, Jennifer, Norbert L. Kerr, Susan E. Harris and Lori A. Sheppard (2002)
"Satisficing in Hypothesis Generation," American Journal ofPsychology, 115 (4),
475-500.

Gettys, Charles F and Stanely D. Fisher (1979) "Hypothesis Plausibilty and Hypothesis
Generation," Organizational Behavior and Human Performance," 24, 93-110.

Thomas Mehle and Stanly Fisher (1986) "Plausibility assessments in hypothesis
generation," Organizational Behavior and Human Decision Processes, 37, 14-33.

Gilbert, Daniel T. and Patrick S. Malone (1995) "The Correspondence Bias,"
Psychological Bulletin," 117 (1), 21-38.

Goodman, N (1972) "Seven Strictures on Similarity," In N. Goodman, Problems and
Projects (pp. 437-447). Indianapolis: Bobbs-Merrill.









Hamilton, D.L. and Steven J. Sherman (1996) "Perceiving Persons and Groups,"
Psychological Review," 103 (2), 336-355.

Heider, Fritz (1958) The Psychology of Interpersonal Relations. New York: Wiley.

Heit, Evan (2000) "Properties of Inductive Reasoning," Psychonomic Bulletin and
Review, 7(4), 569-592.

and Joshua Rubinstein (1994) "Similarity and Property Effects in Inductive
Reasoning," Journal ofExperimental Psychology: Learnin/,ig. Memory and
Cognition, 20 (2), 411-422.

Herr, Paul M., Frank Kardes, and John Kim (1991) "Effects of Word-of-Mouth and
Product-Attribute Information on Persuasion: An Accessibility-Diagnosticity
Perspective," Journal of Consumer Research, 17 (March), 454-462.

Hilton, Denis J. and Ben R. Slugoski (1986) "Knowledge-Based Causal Attribution: The
Abnormal Conditions Focus Model," Psychological Review, 93 (1), 75-88.

Hoch, Stephen J. (1984) "Hypothesis testing and consumer behavior: If it works, Don't
mess with it," Advances in Consumer Research, Vol. 11 T.C. Kinnear, ed. Ann
Arbor, MI: Association for Consumer Research, 478-483.

and John Deighton (1989) "Managing What Consumers Learn From Experience,"
Journal ofMarketing, 53 (April), 1-20.

Hodgins, Holley S. and Miron Zuckerman (1993) "Beyond Selecting Information: Biases
in Spontaneous Questions and Resultant Conclusions," Journal ofExperimental
Social Psychology, 29, 387-407.

Hogarth, Robin M. and Hillel J. Einhorn (1992) "Order Effects in Belief Updating: The
Belief-Adjustement Model," Cognitive Psychology, 24, 1-55.

Johnson, Joel T., Debra L. Long and Michael D. Robinson (2001) "Is a Cause
Conceptualized as a Generative Force? Evidence from a Recognition Memory
Paradigm," Journal of Experimental Social Psychology, 37, 398-412.

Joiner, Christopher and Barbare Loken (1998) "The Inclusion Effect and Category-Based
Induction: Theory and Application to Brand Categories," Journal of Consumer
Psychology, 7 (2), 101-129.

Jones, Edward E. and V.A. Harris (1967) "The Attribution of Attitudes," Journal of
Experimental Social Psychology, 3, 1-24.

Jones, Edward, E., Leslie Rock, Kelly G. Shaver, George R. Goethals, and Lawrence M.
Ward. (1968) "Pattern of Performance and Ability Attribution: An Unexpected
Primacy Effect," Journal of Personality and Social Psychology, 10 (4), 317-340.









Kareev, Yaakov, Sharon Arnon, and Reut Horwitz-Zeliger (2002) "On the Misperception
of Variability," Journal ofExperimental Psychology: General, 131 (2), 287-297.

Kelley, Harold, H. (1967) "Attribution Theory in Social Psychology," In D. Levine (Ed.),
Nebraska Symposium on Motivation (pp. 192-242). Lincoln: University of
Nebraska Press.

Klayman, Joshua (1995) "Varieties of confirmation bias," In D.L. Medin, J.R.
Busemeyer, & R. Hastie (Eds.), The psychology of learning and motivation:
Decision making from a cognitive perspective. New York: Academic Press.

Koehler, Derek J. (1994) "Hypothesis generation and confidence in judgment," Journal
ofExperimental Psychology: Learniing. Memory and Cognition, 20, 461-469.

Krueger Joachim and Russel W. Clement (1996) "Inferring category characteristics from
sample characteristics: Inductive reasoning and social project," Journal of
Experimental Psychology: General, 125 (1), 52-68.

Kruglanski Arie W. (1990) "Motivations for Judging and Knowing: Implications for
Causal Attribution," in Handbook of Motivation and Cognition: Foundations of
Social Behavior, Vol. 2, ed. E. Torry Higgins and R. M. Sorrentino, NY: Guilford
Press, 333-368.

Kunda, Ziva and Steven J. Spencer (2003) "When Do Stereotypes Come to Mind and
When Do They Color Judgment? A Goal-Based Theoretical Framework for
Stereotype Activation and Application," Psychological Bulletin, 129 (4), 522-544.

Liberman, Nira, Daniel C. Molden, Lorraine C. Idson and E. Tory Higgins (2001)
"Promotion and Prevention Focus on Alternative Hypotheses: Implications for
Attributional Functions," Journal ofPersonality and Social Psychology, 80 (1), 5-
18.

McClure John, Jos Jaspars and Mansur Lalljee (1993) "Discounting Attributions and
Multiple Determinants," The Journal of GeneralPsychology, 12 (2), 99-122.

Medin, Douglas L. and M.M. Shaffer (1978) "Context theory of Classification Learning,"
Psychological Review, 85, 207-238.

Mehle, Thomas (1982) "Hypothesis generation in an automobile malfunction inference
task," Acta Psychologica, 52, 87-106.

Mill, John S (1973) System of Logic (8th ed.) In J.M. Robson (Ed.) Collected Works of
John Stuart Mill (Vols. 7 & 8). Toronto Canada: University of Toronto Press
(Original work published 1872)

Murphy, Gregory L. and Douglas L. Medin (1985) "The Role of Theories in Conceptual
Coherence," Psychological Review, 92 (3), 289-316.









Nisbett, Richard A., David H. Krantz, Christopher Jepson, and Ziva Kunda (1983), "The
Use of Statistical Heuristics in Everyday Inductive Reasoning," Psychological
Review, 90 (4), 339-363.

Nisbett, Richard E. and Lee Ross (1980) Human Inference: Strategies and shortcomings
for social judgment. Englewood Cliffs, NJ: Prentice-Hall.

Osherson, Daniel N., Edward E. Smith, Ormond Wilkie, Alejandro Lopez, and Eldar
Shafir (1990) "Category-Based Induction," Psychological Review, 97 (2), 185-200.

Peeters Guido and Janusz Czapinski (1990) "Positive-Negative Asymmetry in
Evaluations: The Distinction Between Affect and Informational Negativity
Effects," European Review of Social Psychology, 1, 33-60.

Posner, M.I. and S.W. Keele (1968) "On the Genesis of Abstract Ideas," Journal of
Experimental Psychology, 77, 353-363.

Read, Stephen J. (1983) "Once is Enough: Causal Reasoning from a Single Instance,"
Journal ofPersonality and Social Psychology, 45 (2), 323-334.

Read, Stephen J. (1984) "Analogical Reasoning in Social Judgment: The Importance of
Causal Theories," Journal ofPersonality and Social Psychology, 46 (1), 14-25.

Reeder, Glenn D. and Marilynn B. Brewer (1979) "A Schematic Model of Dispositional
Attribution in Interpersonal Perception," Psychological Review, 86 (1), 61-79.

Rehder, Bob and Reid Hastie (2001) "Causal Knowledge and Categories: The Effects of
Causal Beliefs on Categorization, Induction, and Similarity," Journal of
Experimental Psychology: General, 130 (3), 323-360.

Reichenbach, H. (1951) The Rise of Scientific Philosophy. Berkeley: University of
California Press.

Rips, Lance (1975) "Inductive Judgments about Natural Categories," Journal of Verbal
Learning and Verbal Behavior, 14, 665-681.

and Allan Collins (1993) "Categories and Resemblance," Journal ofExperimental
Psychology: General," 122 (4), 468-486.

Rosch, E. and C.B. Mervis (1975) "Family Resemblance: Studies in the Internal Structure
of Categories," Cognitive Psychology, 7, 573-605.

Rothbart, Myron and Scott Lewis (1988) "Inferring Category Attributes from Exemplar
Attributes: Geometric Shapes and Social Categories," Journal ofPersonality and
Social Psychology, 55 (6), 861-872.









Sanbomatsu, David, M. Akimoto, and Earlene Biggs (1993) "Overestimating Causality:
Attributional Effects of Confirmatory Processing," Journal of Personality and
Social Psychology, 65, 892-903.

Steven S. Posavac, Frank Kardes and Suzan P. Mantel (1998) "Selective
Hypothesis Testing," Psychonomic Bulletin andReview, 5 (2), 197-220.

Schaklee, Harriet and Baruch Fischoff (1982) "Strategies in Information Search in Causal
Analysis," Memory and Cognition, 10 (6), 520-530.

Simon, Herbert A. (1956), "Rational choice and the structure of the environment,"
Psychological Review," 63, 129-138.

(1982), Models of bounded rationality, Cambridge, MA: MIT Press.

Skowronski John J. and Donal E. Carlston (1987) "Social Judgment and Social Memory:
The Role of Cue Diagnosticity in Negativity, Positivity and Extremety Biases,"
Journal ofPersonality and Social Psychology, 52 (4), 689-699.

Sloman, Steven A. (1993) "Feature-Based Induction," Cognitive Psychology, 25, 231-
280.

Tschirgi, Judith E. (1980) "Sensible Reasoning: A Hypothesis about Hypotheses," Child
Development, 51, 1-10.

Trope, Yaacov and A. Liberman (1996) "Social Hypothesis Testing: Cognitive and
Motivational Mechanisms," in Social Psychology: Handbook ofBasic Principles,
NY: Guilford Press, 239-270.

Tversky, Amos and Daniel Kahneman (1971), "Belief in the Law of Small Numbers,"
Psychological Bulletin, 76 (2), 105-110.

and Daniel Kahneman (1980), "Causal Schemas in Judgments under Uncertainty,"
In M. Fishbein (Ed.) Progress in Social Psychology (pp. 49-72). Hillsdale, NJ:
Erlbaum

Weber, Elke, UlfBockenholt, Denis J. Hilton, Brian Wallace (1993) "Determinants of
diagnostic hypothesis generation: Effects on information, base rate and
experience," Journal ofExperimental Psychology: Learniuing. Memory and
Cognition, 19, 1151-1164.

White, Peter A. (1990) "Ideas About Causation in Philosophy and Psychology,"
Psychological Bulletin, 108 (1), 3-18.

(2002) "Causal Attribution from Covariation Information: The Evidential
Evaluation Model," European Journal of Social Psychology, 32, 667-684.









Wright, Jack C. and Gregory L. Murhpy (1984) "The Utility of Theories in Intuitive
Statistics: The Robustness of Theory-Based Judgments," Journal ofExperimental
Psychology: General, 113 (2), 301-322.

Ybarra, Oscar (2002) "Naive Causal Understanding of Valenced Behaviors and Its
Implications for Social Information Processing," Psychological Bulletin, 128 (3),
421-441.

Zeithaml, Valarie, A., A. Parasuraman, and Leonard L. Berry (1985). "The Problems and
Strategies in Services Marketing," Journal ofMarketing, 49 (2), 33-46.

A. Parasuraman, and Leonard L. Berry (1990) "Delivering Quality Service.
Balancing Perceptions and E\pxVi, atiii\" The Free Press, NY

Zuckerman, Miron, C. Raymond Knee, Holley S. Hodgins and Kunitate Miyake (1995)
"Hypothesis Confirmation: The Joint Effect of Positive Test Strategy and
Acquiescence Response Set," Journal ofPersonality and Social Psychology, 68,
(1), 52-60.















BIOGRAPHICAL SKETCH

Wouter Vanhouche graduated with a Masters of Science degree in psychology from

the University of Leuven in Belgium in 1995. He held several positions as a research

assistant in his hometown university before he decided to take his academic career one

step further and pursue a doctorate in the United States in 2001. Four more years of

rigorous training at the Marketing Department of the University of Florida culminated in

the defense of his dissertation in the summer of 2005.




Full Text

PAGE 1

GENERALIZING FROM PURCHASE OUTCOMES By WOUTER VANHOUCHE 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 2005

PAGE 2

Copyright 2005 by Wouter Vanhouche

PAGE 3

I dedicate this dissertation to my wife and the little miracle she is carrying with her.

PAGE 4

iv ACKNOWLEDGMENTS Initiation, development and completion of this dissertation would not have been possible without the contribution of a number of people. I thank Joe Alba, my dissertation chair, for making me a better researcher and for his dedicated guidance throughout the dissertation process. I have al so benefited from interactions with Chris Janiszewski and thank my other committee members--Lyle Brenne r, Rich Lutz and Ga ry McGill--for their valuable comments on the dissertation. I am grateful to Jef Nuttin, Hans Vertom men, and Frank Baeyens. Each in his own way has played a significant role in my de velopment from a student to a researcher. Special thanks go to Luk Warlop, without wh om I might never have pursued a Ph.D. abroad. My parents have provided invaluable emo tional support, and the same is true for my wife, Maddy, who has made our time in Gainesville so much fun. I thank her for enabling me to initiate this Ph.D. and for her support in co mpleting it. I look forward to the next stage in our life as a family of three instead of two.

PAGE 5

v TABLE OF CONTENTS page ACKNOWLEDGMENTS.................................................................................................iv LIST OF FIGURES.........................................................................................................viii ABSTRACT....................................................................................................................... ix CHAPTER 1 INTRODUCTION........................................................................................................1 2 LITERATURE REVIEW AND THEORY DEVELOPMENT....................................3 Generalization and Induction Research........................................................................3 Causal Lay Theories as Alternative Framework..........................................................7 Do People Hold Causal Lay Theories a bout Events that Have Occurred Once or Twice?...........................................................................................................9 Is the human cognitive system designe d to draw causal inferences based on very small samples?...............................................................................9 How many theories do people typically generate?.......................................10 Are Causal Lay Theories Used as In put into Generalization Judgments?..........12 3 RESEARCH QUESTIONS AND INITIAL EVIDENCE IN MARKETING............16 Research Questions and Hypotheses..........................................................................16 Initial Evidence in Marketing Context.......................................................................19 4 TEST OF EXPLANATORY POWE R OF COMPETING ACCOUNTS..................22 Experiment 1...............................................................................................................22 Method.................................................................................................................23 Predictions...........................................................................................................23 Results.................................................................................................................24 Discussion............................................................................................................25 Experiment 2...............................................................................................................27 Method.................................................................................................................28 Predictions...........................................................................................................29 Results.................................................................................................................30 Discussion............................................................................................................32

PAGE 6

vi Experiment 3...............................................................................................................34 Method.................................................................................................................35 Predictions...........................................................................................................36 Results.................................................................................................................37 Discussion............................................................................................................38 5 IN SEARCH OF LAY THEORIES AND THEIR IMPACT ON GENERALIZATION.................................................................................................40 Experiment 4...............................................................................................................40 Method.................................................................................................................40 Results.................................................................................................................41 Discussion............................................................................................................41 Experiment 5...............................................................................................................42 Method.................................................................................................................42 Results.................................................................................................................43 Discussion............................................................................................................43 Experiment 6...............................................................................................................44 Method.................................................................................................................44 Results.................................................................................................................44 Discussion............................................................................................................45 Experiment 7...............................................................................................................47 Method.................................................................................................................47 Results.................................................................................................................48 Discussion............................................................................................................49 Experiment 8...............................................................................................................50 Method and results..............................................................................................50 Discussion............................................................................................................51 6 MULTIPLE “UNSTABLE” OBSERVATIONS........................................................53 Experiment 9...............................................................................................................53 Method.................................................................................................................53 Results.................................................................................................................54 Discussion............................................................................................................54 Experiment 10.............................................................................................................55 Method and Results.............................................................................................56 Discussion............................................................................................................57 7 GENERAL DISCUSSI ON AND CONCLUSION.....................................................58 The Lay Theory Account............................................................................................59 Specific Lay Theories.................................................................................................63 Conclusion..................................................................................................................66

PAGE 7

vii APPENDIX A STIMULI FOR EXPERIMENT 1..............................................................................67 B STIMULI FOR EXPERIMENT 2..............................................................................69 C STIMULI FOR EXPERIMENT 3..............................................................................70 D STIMULI FOR EXPERIMENT 4..............................................................................71 E STIMULI FOR EXPERIMENT 5..............................................................................72 F STIMULI FOR EXPERIMENT 6.............................................................................73 G STIMULI FOR EXPERIMENT 7..............................................................................75 H STIMULI FOR EXPERIMENT 8..............................................................................78 I STIMULI FOR EXPERIMENT 9..............................................................................79 J STIMULI FOR EXPERIMENT 10...........................................................................80 LIST OF REFERENCES...................................................................................................81 BIOGRAPHICAL SKETCH.............................................................................................88

PAGE 8

viii LIST OF FIGURES Figure page 4-1. Patterns of results as anticipated by various accounts (A,B and C) and as observed in Experiment 1 (D)..................................................................................................24 4-2. Possible outcomes in Experiment 2............................................................................29 4-3. The observed pattern of results in Experiment 2........................................................31 4-4. Pattern of results as an ticipated by several accounts (A B and C) and as observed in Experiment 3........................................................................................................36 5-1. Classification of causes as either stable or unstable in Experiment 4........................41 5-2. Average generalization as a function of number of experiences and purchase context (product service) in Experiment 5...............................................................43 5-3. Mean generalization per re plicate in the product and in the service condition in Experiment 6............................................................................................................45 5-4. Mean generalization as a function of replicate and dimension in Experiment 7........48 5-5. Mean generalization as a function of purchase context and behavior in Experiment 8............................................................................................................51 6-1. Mean generalization as a function of replicate and number of experiences in Experiment 9............................................................................................................54 6-2. Mean generalization per condition in Experiment 10................................................56

PAGE 9

ix 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 GENERALIZING FROM PURCHASE OUTCOMES By Wouter Vanhouche August 2005 Chair: Joseph W. Alba Major Department: Marketing Induction is a ubiquitous but rarely investigated process in marketing contexts. Consumers frequently interact with a vendor usually through purchas e, and then must assess the likelihood that future interactions will produce the same outcome. Research in decision science suggests that generalizi ng from small samples is common but ill-advised. Research in cognitive psychology suggests that generaliz ation varies as a function of the perceived typicality of the episode or exemplar. In line with a hypothesis about conceptual coherence in the categorizat ion literature, I argue that induction will be driven primarily by the theories consumers have regarding the reason for an outcome. I refer to this perspective as the “causal lay theory” view of induction. Moreover, I hypothesize that generalization wi ll be greater when the driv ing mechanism is perceived as stable rather than unstable. I find systematic support for this hypothesis in a series of experiments. Specifically, consumers generalize more quickly from a positive outcome than from a negative one, and more quickly in a product context than in a service context. Surprisingly, a strong

PAGE 10

x typicality manipulation failed to impact generalization. With multiple inconsistent outcomes, generalization is determined by th e valenced order of the outcomes and the time lag between those outcomes, which is cons istent with the lay theory perspective. With multiple consistent outcomes, generalization quickly reaches asymptote, even when an unstable mechanism had initially been assumed. Results suggest that some major findings from the induction literature are not transferable to the non-taxono mic stimuli encountered in cons umer contexts. The Law of Small Numbers, too, cannot easily be genera lized to a consumer context. Instead, the results are more in line with a hypothesis in the category-formation literature and with findings in the social realm.

PAGE 11

1 CHAPTER 1 INTRODUCTION The question of how people generalize from a limited sample of observations to a larger population of observations has received considerable attention in decision science and cognitive psychology. Philosophy, too, ha s recognized the funda mental importance of generalization, as well as its ubiquitous nature. Reichenbach (1951) contended that generalization is the essence of knowledge and long before hi m Aristotle (1963) maintained that inductive reason ing, rather than divine revela tion, is the prime source of human beliefs. Psychologists have added that generalization is perh aps the simplest and most pervasive of everyday inductive task s through which people come to know their physical and social world (Krueger a nd Clement 1996; Nisbett et al. 1983). Generalization from instances is surely ubiquitous and important in a marketing context and has obvious implications for repeat purchase and customer loyalty. Surprisingly though, behavioral consumer rese archers have raised hardly any explicit questions on the topic. Is one purchase experience with a vendor enough to induce strong expectations about future quality with that vendor? If not, does a typical purchase experience induce stronger generalization? Does a negative experience lead to stronger generalization than a positive one? Does the nature of the purchase determine the level of generalization, that is, does a product experien ce lead to different generalization than a service experience? Is the objective base rate of the occurrence of a certain outcome the sole driver of generalization? Although insights from a variety of literatures are

PAGE 12

2 suggestive, systematic behavioral research is missing in marketing (Folkes and Patrick 2003 for an interesting exception). The goal of this dissertation is to dire ctly address these unanswered questions by systematically testing the validity of pr e-existing generalization frameworks in a consumer context and ultimately proposi ng a framework that has not been used systematically in the generalization or induction literature. Re sults suggest that pre-existing frameworks do not hold up as well as might have been anticipated. The newly proposed framework fits the overall data pattern better. The focus is on generalization from one or two purchase experiences with a vendor. The dependent measure probes the de gree to which consumers find this small sample representative for a large populat ion of experiences, often expressed in percentages. For instance, in a restaurant setting, the question pertains to the percentage of all meals on the menu that respondents believe will have the same quality as the meal chosen. Presentation and discussion of empirical re search make up the bigger part of this dissertation (Chapters 4-6), but a theoretical overview is outli ned first (Chapter 2). Three pre-existing frameworks are introduced before an alternative approach is outlined, one that is a virtual st ranger to classical generalization or induction research. Chapter 3 explicates the research questions and evalua tes initial evidence in the marketing domain.

PAGE 13

3 CHAPTER 2 LITERATURE REVIEW AND THEORY DEVELOPMENT Generalization and Induction Research Three distinct frameworks emerge from re search closely and directly related to the question of how and when consumers ge neralize: (1) the Law of Small Numbers, (2) the heterogeneity account, a nd (3) the typicality account. A seminal marketing paper relies on the first of these accounts and is the starting point of this literature review. Hoch and Deighton (1989) propose that generalization will occur even when only very small samples are available as input to the estimate. Although not focusing on generalization, they suggest that consumers do not generate many hypotheses when confronted with a certain purchase outcome; for instance, a bad meal in a restaurant. Instead, consumers jump to conclusions fa irly quickly and generalize, even when statistical criteria do not support such conclusions. Hoch and Deighton (1989) based their view on two streams of literature, the first being concerned with the Law of Small Number s, which states that people generalize quickly because they believe that even sm all samples are representative (Tversky and Kahneman 1971). An example in a marketing context is the consumer who labels a restaurant as bad on the basis of one negative experience (e .g., a bad meal). The second, arguably less well-known literature on hypothe sis generation seems to confirm that people often generate no more than one hypothesis (Gettys and Fi sher 1979; Gettys, Mehle, and Fisher 1986). Moreover, hypothesis formation may happen in a rather passive way so that salient aspects of the proble m drive the content of the actual hypothesis

PAGE 14

4 (Hoch and Deighton 1989, p. 4). Thus one bad meal may easily be taken as sufficient evidence of a bad restaurant. The Law of Small Numbers and the work by Hoch and Deighton are clearly in line w ith the claim that people percei ve less variability than is actually present (Kareev, Arnon, and Horwitz-Zeliger 2002). Other researchers have argued that peopl e may not always generalize from small samples. In a reply to Tversky and Kahnema n (1971), Nisbett et al (1983) show that heterogeneity is perceived, in some situations at least, and thus genera lization is low. This view will henceforth be referred to as the “h eterogeneity account.” Empirical evidence is provided in a scenario requiri ng participants to imagine they arrive at a previously undiscovered island. They have to estimate the percentage of tribe members, Barratos, who are obese. The only input available for the estimate is the knowledge that a given sample is obese. The size of the sample is varied: 1, 3 or 20 obese tribe members. Clearly, generalization is low when the sample consists of 1 member (~38%), higher with a sample of 3 (~56%), and highest with a sample of 20 (~75%). In stark contrast, generalization about an object that is found on this imaginary island, floridium, which is said to burn w ith a green flame when heated, is extremely high, even with an n=1 sample. Nisbett et al (1983) argue that obes ity is perceived as containing more heterogeneity in a tribe population than does a popul ation of floridium elements. This difference in perceived heteroge neity allows Nisbett et al. to make their point that, at least in some situations, people perceive heterogeneity even though only a small and arguably homogenous sample is overtly available. In those situations, people are notably less likely to generalize and will instead reason statistically. This “generalization” literature, concerned with the Law of Small Numbers and perceived

PAGE 15

5 heterogeneity, focuses on statistical reasoning and whether or not pe ople engage in it. However, at least two important questions remain largely unanswered. When are people likely to perceive heterogeneity and when are they not likely to do so? Also, and in more general terms, what drives perceived heterogeneity? One seemingly obvious factor that may a ffect perceived heterogeneity is the typicality of the sample. When a given sa mple is atypical for its population, perceived heterogeneity may be high and thus generaliz ation low. Conversely, when the sample is typical for its population, perceived heteroge neity may be low and generalization high (Rothbart and Lewis 1988). This pattern of re sults is anticipated by what is usually referred to as the “induction literature” in c ognitive psychology, which will be referred to here as the “typicality account” (Heit 2000; Osherson et al. 1990; Rips 1975; Sloman 1993). The typicality effect is perhaps the single most -demonstrated effect in the induction literature, a nd typicality as a factor may account for more explained variance than any other factor. A priori, the induction literature may be considered as the single most-important source that can help answer questions raised in this dissertation because of its rich history, its signifi cant volume, and the force of one of its main arguments, that typicality drives generalization. Although the typicality eff ect is compatible with the notion of perceived heterogeneity and the shared focus on gene ralization in the induc tion and generalization literature, the induction literatu re does not build on the gene ralization literature. Reasons for this lack of cross-referencing are unclear but focus in the two streams of research differs slightly. Induction research focuse s on “how people project information from

PAGE 16

6 known cases to the unknown” (Heit 2000, p. 569), rather than on the (non)normativeness of any such generalization, which concer ned Tversky and Kahneman (1971). Also, the induction literature relies almost exclusiv ely on natural or biol ogical categories (e.g., animal categories) for its stimuli, while Tv ersky and Kahneman and Nisbett et al. (1983) rely more on “social” stimuli. And although th ere are a variety of induction models, most assume some kind of typicality calculat ion underlying the induction judgment. The output of the typicality judgment almost au tomatically leads to the induction judgment. That is, very basic processes, sometim es called “bottom-up” processes, drive generalization. In contrast, the framework de veloped in Chapter 3 assumes that higher level processes--"top down” rather th an “bottom up”--drive generalization. Induction research focuses on the typicality of the sample in relation to the population, but explicitly ignores the dimension or characteristic on which the sample is considered (Heit and Rubinstein 1994, for an exception). Like Nisbett et al., the framework developed here focuses less on th e typicality of the sample than on the importance of the dimension or character istic. Finally, generalization in induction research is usually based on category knowledge rather than object knowledge: an entire category such as all sparrows rather than just one or a few sparrows. Despite the differences between the two lines of researc h, a 25-year review of induction research by Heit (2000) concluded that most induction models apply widely. Typical samples lead to stronger generalization than do atypical sa mples (e.g., a “sparrow” sample leads to stronger inferences concerning “a ll birds” than does a “penguin” sample). An equivalent example in a marketing context would be generalization to all meals on a steakhouse menu from a steak rather than a pasta dish. In sum, given the dominant status of the

PAGE 17

7 induction literature and the extensively documen ted typicality effect, typicality may well be the most important factor that drives pe rceived heterogeneity a nd thus generalization. Available insights in the generalization and inducti on literature suggest that consumers are highly likely to generalize, even with very small samples. If they do not, the perception of heterogeneity of the characteristic in its population is the inhibiting force. Despite the lack of cross-referenc ing between the induc tion and generalization literature, typicality of the sample appears to be an extremely important factor in the perception of heterogeneity and thus generalization. Howe ver, a fourth account is introduced to explicitly consider consumer contexts. Causal Lay Theories as an Alternative Framework Although insights from the literature re view seem compellingly documented and comfortably intuitive, easy generalization to a consumer context is not guaranteed. Do consumers always generalize from small samples, as the Law of Small Numbers suggests? The induction literature, as well as research by Nisbett et al. (1983), suggests they may not. If they do not, will typicality be a major determining factor? Or will other factors also drive perceptions of hete rogeneity and thus generalization? In an effort to understand generalizati on in a consumer context, I introduce the concept of causal lay theories as a determin ant of generalization, an alternative option to the frameworks already reviewed. Unlike th e Law of Small Numbers, the causal lay theory framework assumes that consumers will not always generalize. It is also broader than the typicality account in allowing for lo w generalization when the sample is typical and for high generalization when the sample is atypical. It is more specific than the perceived heterogeneity account by assumi ng a definite source for perceived

PAGE 18

8 heterogeneity: the causal lay theory that is invoked to understand the occurrence of a particular outcome. Lay theories are causal knowledge structur es about how the world functions or is organized. Depending on the specific causal la y theory generated, generalization will vary. When the mechanism underlying the theo ry assumes stability or systematicity, generalization tends to be high, even when based on very small samples. When the assumed mechanism reflects instability, genera lization is low. An example of a stable mechanism may be the expertise of the chef in a restaurant scenario, whereas an example of an unstable mechanism may be an “off day” for the chef. If the chef scores low on expertise today, he probably di d so yesterday and is likely to continue to do so tomorrow. However, a chef having a bad day may ha ve had a good day yesterday and may be expected to have anothe r good one tomorrow. Depending on which mechanism is invoked, generalization from a si ngle experience will vary. Note that lay theories--the central con cept of the framework--often do not reach the rigor and consistency expected of a scie ntific theory (Nisbett and Ross 1980; Tversky and Kahneman 1980). While lay theories have be en shown to improve judgment in some situations in the social realm (Wright a nd Murphy 1984), they may hurt judgment in other situations (Chapman and Chapman 1969) However interesting, the appropriateness of the impact of those lay theories is not the focus of this di ssertation. Instead, the emphasis is on whether consumers hold such cau sal lay theories and if so, whether they drive generalization. The induction literature im plicitly suggests that they do not and that typicality calculations drive generalization. A major goal of th is dissertation is to address these questions explicitly, empirically (Chapters 4-6) and theoretically.

PAGE 19

9 The remainder of this chapter reviews a variety of literatures that may be theoretically relevant but th at has not been relied on by induction or by generalization researchers. Insights from the literature on causal reasoning and that on hypothesis generation address the question of whether people hold causal lay theories. Evidence from the social psychology literature on stereotyping, the corr espondence bias, the negativity effect, and person versus group pe rception is discussed to provide initial evidence for the second question of whether cau sal lay theories drive generalization. The suggestion is that people can and do hold causal lay theories, even about events that have occurred only once, and that such theories im pact generalization, at least in the social realm. Do People Hold Causal Lay Theories ab out Events that Have Occurred Once or Twice? This question implies two more subquestions. First, is the human cognitive system able to draw causal inferences from very small samples? Second, if so, how many of those inferences (theories) ar e typically made per observation? Is the human cognitive system designed to draw causal inferences from very small samples? If people have a causal la y theory--if they draw a causal inference--about a one-time event, then they must feel th ey know why something happened after having experienced it only once. Such a claim may run counter to co mmon wisdom as well as to insights from some of the most influentia l thinkers in psychology (Heider 1958, Kelley 1967) and philosophy (Mill 1973). For instance at the heart of Kelley's ANOVA model lies the necessity to observe covariation be tween presumed cause and effect before a causal inference can be made. According to th e model, causal inferences will be drawn to the extent that cause and effect coincide in a series of observati ons. One observation is,

PAGE 20

10 by definition, not a series; so according to th is account, it seems unlikely for people to draw causal inferences based on only one observation. When two observations are presented in this dissertation, only outcome information without information about a cause was provided. As such, covaration information is again not overtly available and causal inferences seem unlikely. The ANOVA and similar accounts may make one wonder whether people actually draw causal inferences in the type of situations studied in this dissertation. However, it has been recognized (Hilton and Slugoski 1986) that causal attrib utions are not always made as prescribed by the normative ANOVA or an y other covariational model (Cheng 1997; White 2002). Different approaches have been propos ed (Ahn et al. 1995; Ahn and Kalish 2000; Einhorn and Hogarth 1982, 1986; Johnson, Long and Robinson 2001; Mandel 2003; White 1990). The approach advanced by Ahn and her colleagues may reveal relevant insights. The basic argument is that people seek out causal mechanisms in their knowledge base to develop an explanation for a specific event, rather than relying solely on covariation information to identify a causa l relationship between sometimes arbitrary factors. Somewhat simplified, this means that people confronted with a certain event will search for an explanatory mechanism or theo ry. Clearly, the suggesti on is that it is not impossible for the human cognitive system to draw causal inferences from extremely small samples. A remaining question then is how many reasons people typically generate. How many theories do people typically generate? As previously suggested, people seem ab le to draw causal inferences based on only one observation, at least in some situa tions. A following questi on pertains to the number of hypotheses generate d for a single event or observa tion. Granted that any given

PAGE 21

11 outcome is likely to be caused by multiple factors, it is not unreasonable to expect that people may generate multiple reasons or theori es. If so, uncertainty about exactly which theory holds may increase and the degree of generalization may consequently decrease. The literature on hypothesis genera tion may reveal valuable in sights inasmuch as it deals with the number of hypotheses (theories) people tend to genera te for a particular problem (event or instance in our analysis). (The terms "hypotheses" and "theories" are used interchangeably in this section.) The possible number of hypotheses or th eories one can consider for a given problem can theoretically be large, and peopl e are clearly able to take into account or generate multiple theories for a given problem (Alba, van Osselaer and Vanhouche 2003; Arocha, Patel and Patel 1993; Bockenholt and Weber 1993; Fisher et al. 1983; Gettys and Fisher 1979; Koehler 1994; Kruglanski 1990; Li berman et al. 2001; McClure, Jaspars and Lalljee 1993; Mehle 1982; Trope and Liberman 1996; Weber et al. 1993 ). For instance, it is likely that multiple theori es will be generated by a phys ician faced with a patientÂ’s problem or by a detective dealing with a crime. However, an extensive review on hypothe sis (theory) generation and testing (Sanbomatsu et al. 1998) concludes that hypothesis testing is often a first-come, first-confirmed process. The first viable th eory generated has an enormous comparative advantage (p.202). This research suggests that people often generate only one explanation for a given problem. This is consistent with work by Simon (1 956, 1982): People search for a satisfying but not necessarily optimal so lution (Garst et al. 2002) This tendency is consistent with research that suggests that people tend to pursue confirmatory strategies (Arocha et al. 1993; Hoch 1984; Hodgins and Zuckerman 1993; Klayman 1995; Schaklee

PAGE 22

12 and Fischoff 1982; Tschirgi 1980; Zuckerman et al. 1995): Most of us look for the presence of what we expect, not for what we would not expect. In addition, it has been suggested that once a hypothesis becomes focal (i.e., has been generated) its strength is overestimated (Sanbomatsu, Akimoto and Bi ggs 1993). This renders a causal inference more plausible and makes the “hypothesizer” more confident (Getty s, Mehle and Fisher 1986; Koehler 1994). If the foregoing analysis of the liter ature on causal reason ing literature and hypothesis generation is valid, it makes sens e to expect that people draw causal inferences, even from samples with only one observation. This is exactly what Read (1983, 1984) has empirically demonstrated. In line with the work of Ahn et al. (1995) he has shown that causal inferences based on one observation are more likely when a plausible theory (an analogy) is avai lable (Read 1984; Weber et al. 1993). It seems fair to conclude that people can and do draw causal inferences based on samples with as few as one or two observati ons. We tend to do this when we have a lay theory that explains the datum in a satisfact ory way. As soon as one lay theory has been activated, the search ceases, and uncertainty originating from the plausibility of multiple theories is not experienced. A subsequent que stion is whether a pers on’s causal theories are used as input in generalization judgment s. I address this issue empirically in a marketing context (Chapters 4-6), but first outline evidence from social psychology. Are Causal Lay Theories Used as Input in Generalization Judgments? Evidence supporting the impact of lay theori es comes from research streams that are not necessarily focusing on generalization or induction per se, but instead deal with lay theories in one way or anot her. Four significant bodies of research, all from the social psychology literature, ar e briefly discussed.

PAGE 23

13 Research on person versus group percepti on indicates that pe ople perceive more heterogeneity when a certain behavior is performed by a group rather than by an individual (Hamilton and Sherman 1996). C onsequently, generalization is stronger regarding a person than a group. That is to say, people seem to factor in the origin of a certain behavior and thus they seem to have a certain theory about its causes. Research on stereotyping is rather explic it about the impact of lay theories. It no longer questions whether people’s lay theories-in this case called “stereotypes”--affect judgment, but instead accepts th at they do and asks when they are activated and applied and when they are not (Kunda and Spencer 200 3). For instance, if people hold the causal lay theory that advanced age slows down certain responses, th e question in this literature is no longer whether the causal belief affects futu re judgment about speed-related behavior but when it does. The impact of lay theories is found to be so pervasive that research is even focusing on situations in which judgment is not impacted. For the purpose of this dissertation, the message is that research other than the core induction or generalization work suggests that lay theories impact judgment. Research on the correspondence bias also seems relevant (Gilbert and Malone 1995). Participants in a representative experi ment draw conclusions about the personality characteristics of an actor, even on the basis of one isolated observat ion of behavior that is explicitly said to be caus ed by situational factors rather than personality traits (e.g,. Jones and Harris 1967). The social psychology res earch is interested in the fact that people perceive personality traits to be stable predictors of behavior, whereas situational characteristics often are (set up to be) the actual predictors. Th e point of interest here is that people seem to bring a specific causal la y theory to the scene, one that attributes

PAGE 24

14 behavior to personality traits, not to certain cues in the enviro nment. It is this stable lay theory that determines the level of ge neralization: People ge neralize extensively regarding the future behavior of the actor, even on the basis of a single observation. The implication is that they would have generali zed much less--or at least much differently--if the lay theory concerned situational cues rather than personality traits. That is, not only is a lay theory driving generalization, but impact of the stability of the assumed mechanism is implicitly recognized as well. A fourth significant body of research fi nds that negative information receives more weight than positive information ( Folkes and Kamins 1999; Herr, Kardes and Kim 1991; Peeters and Czapinski 1990; Reeder a nd Brewer 1979; Skowr onski and Carlston 1987). In a social context, Yba rra (2002) translates this observation into the claim that people believe that negative behavior is cause d by stable personality traits while positive behavior is caused by variab le situational cues. The im plication is that stronger generalization occurs based on negative rather than positive behavior which suggests that consumers will generalize more strongly from a negative rather than from a positive purchase experience. Furt her, theories featuri ng a “personality” mechanism lead to strong generalization while those underlying a “s ituational” mechanism lead to weak generalization. Ybarra goes even as far as cl aiming that the percepti on of social behaviors is driven by the goal of in ferring underlying causes in the person. That is, not only do people apply lay theories when available, they actively search for them, at least in the social realm. To summarize, research using social stimuli tends to suppor t the notion that lay theories impact generalization judgment and th at hypothesized stable mechanisms lead to

PAGE 25

15 high generalization while unstable mechanisms lead to comparatively low generalization. In addition, it is argued that people activel y search for causal lay theories. Before considering whether lay theories impact generalization in a marketing context, I compare the four theoretical frameworks, advance hypothe ses to test them, and review the scarce evidence available in the marketing liter ature in light of the four accounts.

PAGE 26

16 CHAPTER 3 RESEARCH QUESTIONS AND INITIAL EVIDENCE IN MARKETING Research Questions and Hypotheses When and how do consumers generalize? This general but ce ntral question is approached by testing the valid ity of the three frameworks that have been extensively investigated in the generalization or induc tion literature and by examining a fourth approach that has remained vi rtually untested as a generaliz ation framework. This chapter explicates how each of the accounts differ and a dvances hypotheses to test their validity. One stunningly simple yet powerful answer to our general question comes from the Law of Small Numbers (Tversky a nd Kahneman 1971) in the generalization literature: Consumers always generalize becau se they find even the smallest sample representative. Several experi ments allow testing of this hypothesis. The first includes two manipulations--typicality and outcome valence--that can be expected to induce lower generalization. The perceived heterogeneity view is noncompatible and posits that consumers will not always generalize. Generalization wi ll be low when perceived heterogeneity is high (Nisbett et al. 1983). Confirmation of th is hypothesis implies disconfirmation of the previous one. Experiment 1 is again applicable in this re gard, and Experiment 2 tests whether generalization is driven only by objec tive heterogeneity (in th e sample and/or the population). The typicality account (Heit 2000) is also incompatible with the Law of Small Numbers but is more specific than the heter ogeneity framework in ex plicitly anticipating

PAGE 27

17 lower levels of generalization when the sample is atypical rather than typical. It further distances itself from other accounts by sugge sting that generalization occurs virtually automatically as a result of low-level typicali ty calculations. If va lid, the implication is that generalization necessarily c ovaries with typicality. This hypothesis is also tested in Experiment 1. The fourth account originates from resear ch other than strict generalization and induction work. Like the typicality and hete rogeneity accounts, the lay theory framework is incompatible with the Law of Small Numb ers in that it assumes low generalization in some situations. It is broader than th e typicality account in allowing for high generalization when typicality is low, and low generalization when typicality is high, and it is more specific than the heterogeneity account in assuming a specific source of perceived heterogeneity: the lay theory invoke d to account for the outcome of a specific purchase situation. Although the exact content of lay theories can vary greatly, the stability of the underlying mechanism is one-dimensional. This stability supposedly drives generalization. The lay theory framewor k, contrary to the othe r three accounts, has remained virtually untested as an explicit generalization account. To make up for this lapse, this research project started testing specific marketing-relevant hypotheses, as outlined below. Most are derived from the extant literature. If the causal lay theory pers pective is to be a valid ge neralization account, it must (1) show that consumers hold causal lay theori es and (2) that the underlying stability of these theories drives generalization. For this approach to differentiate itself from the typicality account, it is further desirable (3 ) that it shows stabili ty and instability in contexts independent of typica lity. Experiment 4 explicitly addresses the first and third

PAGE 28

18 requirements, testing whether negative outco mes elicit stable theories while positive outcomes elicit unstable mechanisms (Ybarra 2002) or whether th e opposite occurs, as claimed by Folkes and Patrick (2003). Experiments 3, 5 and 6 address the sec ond requirement. The question is whether consumers’ lay theories about product versus service expe riences affect generalization, and whether their theories are similar to those held by academics. For instance, Zeithaml et al. (1985) recognized that hete rogeneity is higher in a serv ice than in a product context. If consumers’ lay theories conform, generalizat ion is expected to be higher in a product than in a service context because more stability is assumed in the former. Causal lay theories about single outcomes are not the onl y theories investigated. Experiment 2 tests the presence of lay theo ries about a sequence of two inconsistent outcomes. This allows for a test of the wi dely held belief that “the first impression matters most” (Nisbett and Ross 1980) and thus that more stability is inferred from a “first” rather than a “recent” experience. If a lay theory perspective turns out to be defendable, another question pertains to the speed of belief (theory) updating. For in stance, consumers may hold a positive belief about a certain company but encounter a nega tive experience that is attributed to an unstable cause. In such a situ ation they may generalize only to a limited degree. It is imperative to investigate how many more nega tive experiences are needed before the negative outcomes are perceived to be caused by a stable mechanism. Experiments 3, 4, 9 and 10 test the hypothesis that consumers upda te beliefs slowly (Boulding, Kara and Staelin 1999) but also consider the possibi lity that consumers update quickly. Fast

PAGE 29

19 updating would not be inconsistent with a lay theory perspect ive that assumes an unstable mechanism after one experience but a stab le mechanism after two experiences. Initial Evidence in Marketing Context In evaluating previous research in market ing, I identified three articles that deal explicitly with the generalization questi on. These give supportiv e evidence for the typicality and the perceived heterogene ity accounts. An unmodified Law of Small Numbers does not seem to fit the scant avai lable data and evidence for the lay theory framework is mixed. Using a paradigm similar to that used in standard induction re search, Joiner and Loken (1998) studied the typica lity of the “population” rath er than of the sample. For instance, employing the traditi onal induction paradigm whereby participants are exposed to the following format, they found that (2) is rated as a stronger argument than (1). (1) Sony TVs have attribute X; therefore, Sony bicycles have attribute “X”. (2) Sony TVs have attribute X; therefor e, Sony cameras have attribute “X” In other words, generalization from one sp ecific category (e.g., TV s) to another is stronger when the latter is more typical (e.g., camera rather than bicycle). Although this is not the “typical” typicality effect, the result s clearly show an effect of typicality on generalization. It is important to note, however, that as in most induction experiments, typicality information is made salient by e xplicit comparison of the typical and atypical information. Boulding et al. (1999) do not investigat e the impact of typicality but advance suggestions about the pace with which evalua tive beliefs about a se rvice expe rience are updated as a function of the number of experi ences with the servic e. Specifically, the authors investigated generali zation from one hotel experience (sample) to the overall

PAGE 30

20 population of experiences with the hotel. They conclude that "any time a firm wants to change consumers' beliefs of its perceptual positioning through the delivery of goods or services, managers must recognize that these beliefs may be slow to adjust” (p. 481). It may be informative to note that some of the data reported by Nisbett et al. (1983) seem in line with this claim. Even with a sample of 20 observations generalization remains short of asymptote in estimates of tribal ob esity. Boulding et al. argue that multiple observations are needed to update belief. This seems counter to a lay theory perspective, which could conceivably allow for fast updating as a function of an increasing number of observations. For instance, while one nega tive experience in a restaurant may be perceived as being caused by an unstable m echanism, a second negative experience may induce the perception of a stable cause a nd thus high levels of generalization. The third and most recent study (Folkes and Patrick 2003) is less pessimistic about the impact of lay theories on ge neralization. Although th ese authors did not manipulate typicality either, they observed differential generaliza tion as a function of valence of the outcome. Generalization to th e population of colleague service providers was more pronounced when the sample consiste d of a friendly, rather than unfriendly, service employee. That is, a positivity effect is observed in a serv ice context. The mere observation that generalization is rather lo w, based on negative outcomes, may point to the limited generalizability of the Law of Small Numbers in a marketing context. The combination of low generalization in the negative outcome condition with high generalization in the positive outcome condition is taken by Folkes and Patrick as evidence for the existence of lay theories. Such lay theories would imply that a good outcome is perceived as caused by a stable mechanism (e.g., the policy of the firm),

PAGE 31

21 whereas a negative outcome is perceived as caused by an unstable mechanism (e.g., the personality of one person who decided not to follow the firmÂ’s policy). Direct evidence for these theories is not provided, however. To summarize, the Law of Small Number s is not always consistent with the available evidence. For example, generali zation is lower in the negative outcome condition in Folkes and Patric k (2003). Both the typicality (Joiner and Loken 1998) and lay theory accounts (Folkes and Patrick 2003), and thus also the perceived heterogeneity perspective, seem to receive at least so me support. However, Boulding et al. (1999) anticipate slow updating of beliefs, which is not necessarily anticip ated by a lay theory perspective.

PAGE 32

22 CHAPTER 4 TEST OF EXPLANATORY POWE R OF COMPETING ACCOUNTS The empirical section is organized ar ound three sets of experiments, each presented in a separate chapter. Chapter 4 explores the explanatory power of the four theoretical accounts. Chapter 5 targets the lay theory account and examines the characteristics and impact of specific lay th eories on generalization, especially when the sample is n=1. Chapter 6 focuses on those situ ations when two observations are available as input, the first of which is perceive d as caused by an unstable mechanism. Experiment 1 The main goal of Experiment 1 is to investigate the effect of typicality on generalization in a marketing context. At the same time the valence of the purchase outcome is manipulated to investigate whethe r the negativity effect widely observed in the psychology literatur e can be replicated. The typicality effect has been demons trated so extensiv ely in the induction literature that it may seem redundant to aim fo r replication. However, as noted before, a specific set of stimuli and a specific paradigm have been used in the induction literature, and both may have favored the impact of typica lity in induction research in an artificial way. The induction literature does not make differential predictions regarding generalization about positive versus nega tive purchase outcomes. However, this distinction is obviously relevant to marketers, especially si nce most available evidence

PAGE 33

23 suggests that generalization is more extrem e when based on negative rather than positive experiences. Interestingly, some consumer re searchers have reported a weak negativity effect (Ahluwalia 2002) or even a positiv ity effect (Folkes and Patrick 2003). Method Seventy business majors at the Universi ty of Florida participated in the experiment for partial class cr edit. Participants imagined they went to a new restaurant with a group of friends. For at least 30 seconds they read a menu with seven meat dishes (e.g., New York steak of black angus beef, prime rib of beef) and one pasta dish (Appendix A for the full stimuli). Half of the participants were told they had selected a beef item (typical), while the ot hers learned they had select ed the pasta (atypical item). Contrary to the standard induction experiment, typicality in this expe riment needs to be inferred. Typicality of the selected item is crossed with the outcome of the meal. After having selected their meal, half of the partic ipants learned the quality of their meal was low while the other half lear ned it was high. Participants were next asked (1) what percentage of all meals in this restaurant did they think would be of the same quality as the meal they had selected and (2) what was the rationale behind their generalization estimate. This experiment, as well as all the others, was conducte d entirely on computer. Predictions Various predictions can be made. First, th e induction literature anticipates a main effect of typicality in that generalization is more extreme in the typical than atypical condition (Figure 4-1A). Second, a larg e body of evidence anticipates higher generalization with a negative rather than positive outcome (Figure 4-1B). Third, the Law

PAGE 34

24 of Small Numbers proposes that generalizati on will be high across al l conditions (Figure 4-1C). 0 10 20 30 40 50 60 70 80 90 typicalatypical positive negative A 0 10 20 30 40 50 60 70 80 90 typicalatypical positive negative B 0 10 20 30 40 50 60 70 80 90 typicalatypical positive negative C 0 10 20 30 40 50 60 70 80 typicalatypical positive negative D Figure 4-1. Patterns of results as anticipated by various accounts (A, B and C) and as observed in Experiment 1 (D). A) Pred icted pattern by typicality account, B) Anticipated pattern of negativity effect is observed, C) Pa ttern anticipated by Law of Small Numbers, and D) Observed pattern in Experiment 1. Results Manipulation check. In an independent study, 32 undergraduate students judged the degree to which they found their item (steak or pasta) typical, given the menu. Despite the relatively low cell sizes (n= 16), the difference between the typical and atypical conditions was highly significant [t (30)=5.4; p<.001], there by demonstrating the strength of the typi cality manipulation.

PAGE 35

25 Numerical data of main experiment. Figure 4-1D clearly s hows that neither of the three predicted re sults is actually obtained. A 2x2 (typicality x outcome valence) between-subject ANOVA revealed no main effect of typicality (F<1); instead there is an outcome effect of lower generalization in the negative condition [F(1,66)=4.3; p<.05]; no evidence for an interaction is obtained (F<1). Cognitive responses. Two independent judges analy zed the cognitive responses that were recorded after the generalization estimate had be en made. In all experiments except Experiment 2 the coding categories we re “stable,” “unstable,” and “ambiguous.” “Stable” indicates a stable mechanism, such as the expertise of the chef, while “unstable” refers to an unstable mechanism such as th e chef having a bad day. Responses that did not fit either category were coded “ambi guous.” The relation between the level of generalization and the assumed underlying me chanism is established by calculating the correlation between the two, with the “mech anism” variable being a dichotomous variable (unstable=0, stable=1). The goal of this measure is to provide directional evidence for the effect of lay theories, even when a specific manipulation does not reveal the anticipated impact of lay theories. In E xperiment 1, the judges ag reed in 91% of the cases and came to a joint conclusion in th e other 9%. The resul ting correlation between generalization and stabil ity was r=.56 (p<.001). Discussion At least three observations are surprising in that they do not correspond to predictions of prominent literatu res: (1) contrary to prediction s in the induction literature, typicality failed to impact generalization; (2 ) a positivity instead of a negativity effect is observed since generalization is higher in the positive condition than in the negative

PAGE 36

26 condition; (3) contrary to predictions of the Law of Small Numbers, one condition showed significantly lower ge neralization than another. Do the results imply that typicality does not impact generalization “naturally”? A claim based on this one marketing experime nt would be far-fetched, but the results suggest that generalization in the market place is not driven by mere typicality calculations as suggested by th e induction literature. Moreov er, the results may indicate that, if lay theories drive generalization, typicality may not always be the primary input for such theories. Perhaps typicality does not “jump out” as much as previously believed. The second surprising observation refers to the non-occurrence of a negativity effect. Instead, generalization is higher in the positive than in the negative condition. This result, too, is interesting for several reasons. First, it validat es the “null effect” of the typicality manipulation in that the latter cannot be attributed to negligent or inattentive participants who slid the scale without thinki ng about or even readi ng the scenario shown to them. Second, the mere observation that generalization varies between any two conditions rules out an unmodified Law of Small Numbers as a framework to account for the results. Third, whereas large bodies of literature anticipate a negativity effect, Experiment 1 shows a positivity effect. One may wonder why a positivity effect is observed in this experiment (and in Folkes and Patrick’s 2003 study) while a nega tivity effect is found in so many others. Also, what drives the positivity effect? In line with the positive correlation between generalization and stability of the underlying mechanism, one could argue that lay theories drive the results and thus the positiv ity effect. If so, consumers believe that positive outcomes are caused by stable mechanisms, such as the expertise of the chef,

PAGE 37

27 while negative outcomes are cau sed by unstable mechanisms, su ch as the chef having a bad day. This would suggest that people ha ve different lay theories about positive and negative outcomes in the commercial world--whe re a positivity effect is observed--than in the social world, where a negativity e ffect has been reported (Ybarra 2002). It is also possible, however, that gene ralization is not driven by typicality calculations or by lay theories. Instea d, people may know and use the objective percentage of positive or negative experiences in restaurants. That is, base rates may drive the positivity effect and thus gene ralization. Although we do not know the exact base rates of positive and negative experiences in restaurants, it seems reasonable to expect more positive than negative experiences, rather than the opposite. In sum, exactly which factors are driving the results in Experiment 1 is impossible to determine at this point. It is clea r, however, that typicality does not drive generalization. Furthermore, the overall data pa ttern is inconsistent with the Law of Small Numbers, but not with the perc eived heterogeneity framework, the lay theory account, or a base rate explanation. The plausibility of a base rate explanation is addressed more explicitly in Experiment 2. Experiment 2 Experiment 2 allows further investig ation of the impact of base rates on generalization by presenting pa rticipants with two experi ences, one positive and one negative. The order in which they appear and the time interval between them are manipulated. Neither factor s hould impact generalization when objective base rates about the occurrence of good and bad outcomes drive generalization. However, manipulation of both factors may induce differential lay theori es about why the specific set of outcomes

PAGE 38

28 occurred. If lay theories driv e generalization, differences ma y be expected as a function of those factors. A third goa l is to investigate whether consumers reason statistically when confronted with overt heterogeneity: the positive and the negative purchase experience. Nisbett. et al. (1983) argued that statistical reason ing is more likely to occur when people have reason to believe that there is heterogeneity on the considered dimension. This experiment explores whether ov ert heterogeneity is a sufficient source to fuel perceived heterogeneity, induce statistical r easoning and thus inhibit generalization. Method In anticipation of a manipulation in Experi ment 3, the restaurant context used in Experiment 1 was changed to an explicit product or service context. Participants from the University of Florida (n=182) and Erasmus Univ ersity (n=110) read a scenario in which a consumer experience--either a product or serv ice experience--was depi cted with either a positive or a negative outcome. Before any assessment was recorded, participants were told they experienced the same brand again--e ither one year or one week later, this time with the opposite outcome as result (Appendix B for the full stimuli). The dependent variable pertained to expecta tions about a third experience being positive, negative, or unpredictable. A 201-point scale was presente d, ranging from very negative (-100) over fifty-fifty (0) to very positive (100). As in Experiment 1, participants were allowed to express the rationale behind their re sponse in an open-ended question. A 2x2x2 between-subject design was employe d, crossing the order of outcomes (bad-good vs. good-bad) with delay between out comes (one week/one year) and purchase context (product/service).

PAGE 39

29 Predictions If people reason statistically--that is, they treat their two outcome s in the same way they would treat two coin flip outcomes--the average estimate in all conditions should not deviate from zero (Figure 4-2A). However, it is also possible that co nsumers plug in base rates when generalizing. As Experiment 1 suggests, people may believe that positive purchase experiences outnumber negative experiences. If such base rates are the basis for generalization, a pattern similar to the one shown in Figure 4-2B can be expected. -100 -80 -60 -40 -20 0 20 40 60 80 100 1 week1 year good-bad bad-good A -100 -80 -60 -40 -20 0 20 40 60 80 100 1 week1 year good-bad bad-good B -100 -80 -60 -40 -20 0 20 40 60 80 100 1 week1 year good-bad bad-good C -100 -80 -60 -40 -20 0 20 40 60 80 100 1 week1 year good-bad bad-good D Figure 4-2. These are some possible outcomes in experiment 2. A) Anticipated pattern of people reason statistically, B) Expected pattern of base rates dominate, C) Pattern of results indicating a primacy effect and D) Pattern of results indicating a negativity effect. A third possibility is that consumers have a theory as to why their specific set of outcomes occurred. A variety of theories can be adopted, two of which are considered here. First, consumers may emphasize either the more recent or the first experience

PAGE 40

30 (Figure 4-2C). This would result in either a recency or a primacy effect. Both phenomena have been reported extensively by a variety of researchers. For instance, in the social realm, Nisbett and Ross (1980) stated that “primacy effects are overwhelmingly more probably” (p.172). Research on c ognitive ability (Jones et al. 1968) as well as a review paper on order effects suggest that, indeed, th e first impression is often considered the most important one, at least in a short and eas y task such as that used in Experiment 2 (Hogarth and Einhorn 1992). However, res earch in other doma ins suggests recent information may be more dominant (Cu ccia and McGill 2000 regarding accounting, Davis 1984 regarding jury decision-making). Alternatively, but not necessarily mutu ally exclusively, consumers may hold theories in which either the negative or pos itive experience dominates while the other is perceived as an outlier or the beginning or end of a trend. In line with the result of Experiment 1, it may be that the positive outc ome will dominate. In that case, a recency effect would be observed in the bad-good and a primacy effect in the good-bad condition. Note that such a result woul d be very much in line with a base rate explanation. Alternatively, as suggested by a large body of other evidence, it is possible that negative information will dominate the estimation responses (Figure 4-2C). Results Brief inspection of the means in Figure 43 indicates that none of the predicted patterns is obtained. Instead, the means s how a combined emphasis on recent and positive information. In addition, generalization is st ronger when there is a one-year, as opposed to a one-week, delay between the two outco mes. This pattern is confirmed by a 2x2x2 [delay (one week/one year) x order of outcome (good-bad/bad-good) x purchase context (product-service)] ANOVA on the absolute values with all three factors considered as

PAGE 41

31 between-subject factors. Absolute values are used rather than the raw data because the raw data may fail to pick up a time lag effect even if it is present. The main effects of time lag [F(1,284)=9.3; p< .01] and order of outcome [F(1,284)=18.2; p<.001] confirm, respectively, that generalization is more extr eme after one year rather than one week, and more extreme in the bad-good condition th an the good-bad condition. No evidence is provided for a main effect of purchase context [F(1,284)=1.8; p<.18 )] or for any interaction (highest F=1.9; p<.17). -20 -10 0 10 20 30 40 50 1 week1 year good-bad bad-good Figure 4-3. The observed pattern of results as a function of time delay and order in which the outcomes are presented. Because this is the only experiment with inconsistent outcomes, the coding scheme for the cognitive responses is di fferent in this experiment. J udges classified the responses either as evidence for statistical reasoning or for a lay theory (other than statistical reasoning). “Statistical reasoning” respons es refer to a 50/50 chance of the next experience being positive/negative. A lay theo ry response was defined as a reference to one experience dominating the other, such as ‘the company is improving’, or ‘the negative experience was a fluke’. The judges agreed in 94% of the cases. References to a lay theory tend to dominate as generalization became more extreme. For instance, in the bad-good condition with a one-year time interval, 85% of the responses were classified as referring to a trend or an outlier, while 14% were classified as statistical reasoning (1%

PAGE 42

32 was not classified in either category). Refere nces to statistical reasoning were relatively more dominant when generalization was le ss extreme. For instance, in the good-bad condition with a one-week interval, 57% of the responses were classifi ed as a lay theory, and 36% as statistical reasoning (7% were not classified in either category). Discussion Experiment 1 failed to suppor t the typicality acco unt, but could not rule out a base rate account as an explanation for its findings. Indeed, consumers may have had knowledge about the overall percentage of positive and negative purchase outcomes and used that knowledge exclusively to make th eir generalization estimate. In a similar way, participants in Experiment 2 could have plugge d base rate knowledge into their estimate. For instance, if 60% of purchases can be objectively classified as having a positive outcome, at least a positive number should have appeared in each of the four conditions without significant differences across the conditions. Howeve r, the observed pattern of results is very different. In addition, the pattern of results deviat es significantly from what could be expected if consumers reason statistically. A ggregate generalization levels were strongly different from zero in all four conditions, de spite the strong overt heterogeneity. Clearly, the combination of one positive and one negative purchase experience is not perceived in the same way as a head-and-tail result from two coin flips. In other words, the results in Experiment 2 cannot be fully explained by the objective heterogeneity of the sample or by the objective heterogeneity of the larger population of positive and negative purchase experiences; that is, the base rates. Still, considerable systematic variance is left: A combined recency and positivity effect is even more pronounced as the time interval increases. What is causing this

PAGE 43

33 variance? One possible interpretation is that consumers try to give meaning to--come up with a lay theory about--the sequence of events encountered and that this meaning is reflected in their generalization estimate. For instance, and in line with the results of the cognitive responses, participants in the bad-good condition may believe that this sequence of outcomes occurred because th e company is improving. Understandably, there is more room for improvement in a time span of one year as opposed to one week. The result is therefore more extreme in the one-year condition. The opposite seems true for the good-bad condition, but to a lesser degree. These results, then, may provide the first pi ece of evidence, albe it indirect, in favor of a lay theory perspective, while at the same time making a simple base rate interpretation less likely. The Law of Small Numbers is supported in that people do not tend to reason statistically, even in the face of strong heteroge neity cues, but it is contradicted in that consumers do not find a small sample representative. The specific content of the lay theories is not a first concern, but it is certainly not inappropriate to ask why these theories and not others? One reason positive rather than negative information may dominate theories is that consumers may assume the company they interacted with is in business. Given this assumption, it is more logical to expect that a company is improving or performing well (bad-good condition), rather than deteriorating or performing badly (good-bad condition). The other question pertains to why lay theo ries are dominated by recent rather than primacy information. Research on cognitive--r ather than commercial--skills has shown a primacy effect (Jones et al. 1968). The same pattern dominates in the social realm (Nisbett and Ross 1980) and was anticipated in an impressive review on recency and

PAGE 44

34 primacy effects (Einhorn and Hogarth 1992). Einhorn and Hogarth expect a recency effect when more complex and cognitively demanding tasks are involved or when two estimates are made (one after each outcome) in stead of one. The task in Experiment 2 is classified as short and easy, and participants make only one estimate, as soon as they received information about the valence of both outcomes. The only way to align our findings with Einhorn and HogarthÂ’s framework is to assume that participants made two estimates--one implicitly after the first experience and the other explicitly after the second experience. Future research may want to explore this possibility, for instance, by explicitly having par ticipants make two estimates--one after each outcome. If the result does not change, Einhorn and HogarthÂ’s framew ork is supported. However, it is not inconceivable that the procedural change might induce a primacy--instead of a recency-effect, given that consumers may make up th eir mind quickly after a first experience. Taken together, the results of Experiment 2 suggest (1) that consumers do not reason statistically in the face of overt heterogeneity, (2) that simple base rates do not (always) drive generaliza tion, and (3) that, instead, lay theories may impact generalization. A recurring observation is th at generalization is more extreme on the positive than on the negative side. Experiment 3 tests the boundaries of this positivity effect. Experiment 3 Experiment 3 explores the boundaries of the positivity effect. It also tests the impact of lay theories about purchase experien ces in a product versus a service context. Experiments 1 and 2 showed that generalization is more extreme on the positive than on the negative side. An emerging question is how persistent this positivity effect is.

PAGE 45

35 Are consumers incurably positive? In other words, is even a set of multiple negative experiences perceived as an exception and is the updating process c onsequently slow, as suggested by Boulding et al (1999)? Or does a second nega tive experience lead rather quickly to the same level of extreme gene ralization as does one positive experience? Such a finding would not be inconsistent with a lay theory perspective. It would suggest that consumers perceive one negative e xperience as an excep tion, but two negative experiences as an indication of stable negative performance. The second goal of Experiment 3 is to explore whether consumers hold different theories about various sets of purchase expe riences and, more importantly, whether those lay theories drive generalization. For instan ce, many scholars may agree with Zeithaml et al. (1985) that heteroge neity is higher in a service context than in a product context. If so, consumers should generalize to a larger degree in a product than in a service context. The specific theories underlying ge neralization may recognize that products are the output of an automated production process whereas servic es are generated by a much more variable mechanism. A final question is whether the positivity effect obser ved in a restaurant scenario in Experiment 1 can be re plicated in a different context. Method In this experiment, 93 students at the Univ ersity of Florida were asked to imagine they had had one or two experiences with “a pr oduct” or “a service,” with the description left vague. If two experiences occurred, the out comes were consistent. In both the product and the service conditions, the scenario show ed four examples of a product/service in parentheses (Appendix C for the full stimuli).

PAGE 46

36 This resulted in a 2x2x2 de sign with purchase context (product vs. service), number of experiences (one vs. two) and outcome valence (positive vs. negative) as three between-subject factors. As in Experiment 2, the task was to estimate the likelihood that the next experience would be positive or negative, and cogni tive responses were recorded after the generalization estimate had been submitted. Predictions A persistent positivity effect and thus slow updating, as suggested by Boulding et al. (1999), would be evidenced by a pattern si milar to the one depicted in Figure 4-4A. -100 -80 -60 -40 -20 0 20 40 60 80 100 positivenegative 1exp 2exp A -100 -80 -60 -40 -20 0 20 40 60 80 100 ProdServ Positive Negative B -100 -80 -60 -40 -20 0 20 40 60 80 100 ProdServ positive negative C -100 -80 -60 -40 -20 0 20 40 60 80 100 positivenegative 1exp 2exp D Figure 4-4. Pattern of results as anticipate d by several accounts (A, B and C) and as observed in Experiment 3. A) Expected pa ttern if the positivity effect is to persist when n=2, B) Pattern as anticipa ted by work of Zeithaml et al. (1990), C) Expected pattern if a positivity eff ect occurs in a service context and a negativity effect in a product context. D) The actually observed means as a function of number of experi ences and outcome valence.

PAGE 47

37 Figure 4-4B shows the possible impact of a product-service effect: The direction of generalization is identical in both conditions, but the magnitude is higher in the product condition (Zeithaml et al. 1990). If, how ever, a positivity effect occurs in the service condition, and a negativ e effect in the product condition (Folkes and Patrick 2003), the pattern should look like the one depicted in Figure 4-4C. Results Descriptively, Figure 4-4D shows a replicat ion of the positivity effect when one experience is involved, but no evidence for a positivity effect when two consistent episodes are experienced. A 2x2x2 [valence (pos itive-negative) x num ber of exposures (one-two) x purchase context (product-ser vice)] ANOVA confirmed this pattern. The number of exposures by valence of outcom e interaction is significant [F(1,85)=7.8; p<.01], thereby confirming the annihilation of the positivity effect after two exposures [t(44)=1.2; p<.24 for the difference between positive and negative conditions after two episodes on absolute values]. Although the main effects of number of exposures [F(1,85)=4.7; p<.05] and valence [F(1,85)=540.7; p<.001] are significant, the main effect of purchase context is not (F <1). Overall, generalization is more extreme after two experiences than after one, and more extrem e on the positive than negative side. Whether the experience involves a produc t or service, makes no difference. In analyzing the cognitive responses, the two judges agreed in 96% of the cases. Th e overall correlation between generalization and stability is r=.39 (p<.053).

PAGE 48

38 Discussion Experiment 3 was designed to test boundari es of the positivity effect and thus the speed with which consumers update their beliefs. At the same time, a first test probed the impact of productand service-rela ted lay theories on generalization. Despite the changed context and the use of a slightly different dependent measure, the positivity effect observed in Experiment 1 is compellingly replicated in Experiment 3, thus supporting the robust ness of the phenomenon. In addition, the results show that consumers confront ed with two experiences generalize equally extreme on th e negative as on the positive side. That is, the positivism of consumers has clear limits. One implicati on is that consumers may switch easily from an unstable to a stable mechanism when ge neralizing. In other words, the increased generalization in the negative outcome cond ition may be suggestive of a fast updating process, which is inconsistent with Boulding et al.Â’s (1999) predictions but not with a lay theory perspective. The other interesting observat ion is the lack of differentiation between the product and the service condition. Experiment 3 sugge sts that consumers do not hold the same theories as academic scholars. Another possibi lity is that consumers do hold the same lay theories, but that these do not impact generaliz ation. One could argue that this is evidence against the lay theory perspective. Still anot her interpretation, however is that these lay theories are held by consumers, but that th e manipulation in Experiment 3 was too weak to elicit them. The full pattern of results in Experime nt 3, then, seems at least partially inconsistent with each of the three framewor ks that reasonably could make predictions.

PAGE 49

39 The Law of Small Numbers has difficulties with the low level of generalization after one negative experience, while both the heteroge neity and the lay theory framework would have anticipated differences as a function of purchase context. When taken together, the full pattern of results observed in Experiments 1-3 cannot easily be accounted for by any of the reviewed frameworks. An unmodified Law of Small Numbers and the typicality account ca nnot be retained, wh ile the heterogeneity and lay theory perspectives receive mixed support. Because the heterogeneity account does not add much to the lay theory perspectiv e, the next set of experiments targets the validity of the lay theory pers pective and further explores the degree to which consumers hold theories about positive and negative experiences. They also further examine purchase experiences in a product versus a service context.

PAGE 50

40 CHAPTER 5 IN SEARCH OF LAY THEORIES AND THEIR IMPACT ON GENERALIZATION The major goals of experiments reported in this chapter are to obtain direct evidence for the existence of lay theories in a consumer context and for their impact on generalization. Experiment 4 Given that the evidence in favor of lay th eories has been indirect and even then inconclusive at best, the main goal of Experi ment 4 is to provide direct evidence. Do consumers believe that positive outcomes are caused by stable mechanisms, such as the expertise of the chef, and ne gative outcomes by unstable mechanisms, such as the chef having a bad day? Such beliefs are necessa ry to interpret the positivity effect in Experiments 1 and 3 in terms of a lay theory perspective. A varian t of the restaurant scenario employed in Experiment 1 is used in Experiment 4. Method This experiment included 55 students, 27 of whom were randomly assigned to the positive outcome condition. All participants were presented with a scenario in which a restaurant critic goes to a particular restaurant to write a review. The critic’s meal is said to be “very good” for half of the participants and “not very good” for the other half. As part of the review, the critic wants to explain why the restaurant produced a good/bad meal. The task of each participant is to provide reasons that could be used in the critic’s review (Appendix D for full scenario). In a second phase, two independent judges who

PAGE 51

41 were unfamiliar with the hypothe sis of the experiment coded each reason as either a “stable” or “unstable” mechanism. Results There were 120 separately codable units (causes) in the negative condition and 109 in the positive condition, all coded as “sta ble,” “unstable” or “non-codable.” The two judges agreed in 74% of the cases. Inconsis tencies were resolved through deliberation by the two judges. 0 10 20 30 40 50 60 70 80 90 100 positivenegative stable unstable Figure 5-1. Classification of causes as either stable or unstable. Of the codable responses, 94% was classi fied as “stable” (6% as “unstable”) in the positive outcome, compared to 51% in the negative outcome condition, where 49% were coded “unstable” (Figure 5-1). The difference in the number of stable/unstable causes between the positive a nd negative outcome condition is statistically significant [ 2=23.7 ; p<.001]. Non-codable responses comp rised 52% of all responses in the positive outcome condition and 61% in the negative outcome condition. Discussion Experiment 4 provides the first direct evidence for lay theories. Consumers are more likely to believe that good outcomes (e.g., a good meal) are caused by a stable

PAGE 52

42 mechanism and bad outcomes (e.g., a bad meal) by an unstable mechanism. This result is consistent with the positivity effect observed in Experiments 1 and 3. However, as interesting and important as this observation is, it is not helpful unless consumers use such theories when generalizing. Experiments 5 and 6 further investigate the impact of lay theories on generalization by probing th e product-service dis tinction employed in Experiments 2 and 3. Experiment 5 Experiment 5 takes both Experiments 3 a nd 4 one step further. Having provided evidence for the existence of lay theories (Expe riment 4), the next step is to show the impact of those theories on generalizati on. The context of product versus service (Experiment 3) is selected again, with th e manipulation slightly strengthened. The generic product and service de scription in Experiment 3 is made more specific in Experiment 5. The rationale is that the more specific the description, the more likely that specific theories will be triggered, and ther efore the more likely that differences in generalization will occur between th e product and the service condition. Method The goal in the stimuli selection was to in clude products that typically show little variance in the production process and to choos e service stimuli in wh ich more variability can be perceived. Toothpaste, a printer cartr idge, and a battery were used as product stimuli; financial advice, a hotel stay, and delivery service were used as service stimuli (Appendix E for full stimuli). In this experiment, all the purchas e experiences were negative. The 127 participants imagined they had had either one or two negative expe riences. The outcome

PAGE 53

43 information was still rather general, just as in Experiment 3. A 2x2 [number of exposures (one vs. two) x purchase context (product vs. service)] was employed. Results -80 -70 -60 -50 -40 -30 -20 -10 0 1exp2exp products service Figure 5-2. Average generalization as a functi on of number of expe riences (one/two) and purchase context (product/service). The results are shown in Figure 52. Once again, despite the stronger manipulation, no significant difference is obs erved between the product and the service condition (F<1). Instead, the main effect of “number of exposures” is highly significant [F(1,123)=50.5; p<.001]. The interaction is not si gnificant (F<1) and neither is the single main effect of replicate in the product [F (2,62)=1.6; p<.21] or service condition (F<1). Agreement in coding the cognitive responses was reached in 91% of the cases and the correlation was r=.39 (p<.05). Discussion Given the intuitively plausible difference in heterogeneity between a product and a service context, the lack of any significant difference is striking. Again, one possible explanation is that consumers do not hold the same theories about products and services that most academics do. Or, consumers may hol d those theories but not apply them when generalizing. Experiment 6 further probes a th ird possibility: Consumers hold lay theories

PAGE 54

44 about products and services similar to those that academics hold, but a stronger manipulation is needed to elicit them. Experiment 6 The rationale behind Experiment 5 was ta ken one step further in Experiment 6. The purchase scenario was made even more specific. This time participants were not only provided with specific product or service inform ation, they were also informed about the specific dimension on which the product or service did not perform appropriately. Method A pool of 95 students evalua ted either four service or four product replicates. Replicates in each condition were selected as intuitive representatives of exemplars in their respective categories and to fall short in a specific characteristic. Replicates in the product condition are “a razor blade that provides a somewhat rough shave,” “a bad tasting chocolate bar,” “a wristwatch that is running behind,” and “a pen that does not distribute the ink evenly.” “A n undercooked meal,” “rude service in a coffee shop,” “your lawn is not carefully mowed,” and “being pla ced on hold for a long time at a call center” are the replicates in the serv ice condition (Appendix F for the full scenario). The question to the participant was, “If you purchased 100 ‘replicate,’ what perc entage would perform equally poorly?” Results Figure 5-3 shows the means in the produc t and service condition for each of the replicates. Descriptiv ely, average generalization seem s higher in the product condition than in the service condition. Statistical analyses confirm the main effect of purchase

PAGE 55

45 context when the replicate is treated as a repeated measure, [F(1,93)=11.5; p<.001] and almost when treated as a between-s ubject factor [F(1,87)=3.1; p<.09]. 0 10 20 30 40 50 60 70 80 90 prodserv Figure 5-3. Mean generalization per repli cate in the product a nd service condition. The main effect of replicate in the product [F(3,141)=20.3; p<.001], but also in the service condition [F(3,138)= 19.5; p<.001] was highly signif icant when the replicate was treated as a repeated measure in two one-way ANOVAs, which indicates high variability even within the product and service contexts. Ag reement among judges in the coding task was 88%. The corr elation was r=.57 (p<.001). Discussion The results of Experiment 6 extend thos e observed in Experiments 3, 4, and 5 in various ways. Relative to Experiments 3 a nd 5, overall generalization in the negative condition of Experiment 6 increased in the pr oduct, but not in the se rvice, condition. That is, the specific description of the purchase si tuation in Experiment 6 led to the pattern inferred from Zeithaml et al. (1990): higher ge neralization in a produc t context than in a service context. A product-ser vice distinction in Experime nt 6 that is absent in Experiments 3 and 5 suggests that, even if consumers hold generic lay theories about purchase experiences in a pr oduct versus service contex t (e.g., the “assembly line”

PAGE 56

46 theory), more particular lay theories about sp ecific purchase situati ons, rather than broad product-service differences, may be elicited more naturally. By showing evidence for the product-service distinction, Experiment 6 extends the results of Experiment 4--consumers hol d lay theories--in multiple ways. First, it provides evidence for the existen ce of another set of lay theori es, a set reflecting the work of Zeithaml et al. (1985). More importantly, it reinforces the lay theory perspective by showing that those lay theori es do affect generalization. However much Experiment 6 provides important support for the lay theory perspective and answers crucial questions, it ce rtainly sparks new ones. It is informative about the specific level of pr oductor service-related lay theories that consumers may apply when generalizing. Although the obser ved main effect of product-service is interesting and in line with the proposition inferred from Zeithaml et al. (1985), one should not disregard the considerable syst ematic variance within both the product and service conditions. This variance suggests that the specific dimension and/or the specific product or service category may be as crucial as the broad product-service distinction. Indeed, if just one of the replicates had b een selected for each of the two conditions, it would have been possible to observe a main ef fect of purchase contex t that is opposite to the observed pattern: higher generalization in a service than in a product context. Such an outcome is not anticipated by the classical heterogeneity distinction between products and services (Zeithaml et al 1990) and may add an importa nt insight to the existing theoretical literature, as we ll as to the knowledge base of the manager. Whether the specific dimension or the specific category or a combination of both determines the

PAGE 57

47 specific lay theory is impossible to infer from the design employed in Experiment 6. Experiment 7 aims at disentangli ng the impact of both factors. Experiment 7 Although Experiment 6 shows differences in generalization between a product and a service context, in line with Zeithaml et al. (1990), the considerable variance within the product and service conditions suggests that this distinc tion does not suffice to predict an appropriate level of gene ralization. Instead, the specifi c product/service category may be crucial just like the dimension on which the product/s ervice performs inadequately. Experiment 7 disentangles the impact of th e two factors by keeping the category constant and varying the dimension. Because more overall variance is expected in a service than in a product context, only service replicates are included in Experiment 7. Dimensions that had been considered relevant by previous re searchers (Zeithaml et al. 1990; Coulter and Coulter 2003) were selected and cro ssed with three service categories. Method Four dimensions of service quality were selected from Zeithaml et al. (1990): competence, reliability, courtesy, and credibil ity. Each dimension was crossed with each of three replicates: a restaurant scenario, a car repair scenario, and a painter scenario. Operationalization for the four dimensions in the restaurant scenario was as follows (Appendix H for the full stimuli): “The waiter mixed up the orders and therefore no one gets exactly what he ordered” (competence); “The waiter is slow to take your order” (reliability); “The waiter seems to be abrupt and unfriendly” (courte sy); “The waiter has overcharged you” (credibility). The dependent measure was worded as follows: “For each

PAGE 58

48 set of 100 customers, what percentage do you feel would have the same experience with this waiter/mechanic/painter?” The st udy participants were 259 students. Results Descriptively, the results in Figure 5-4 show that generalization is a function of a combination of the replicate (category) and the dimension. This observation is confirmed by the statistical analyses in a 3x4 (replicat e x dimension) ANOVA, with both factors treated as between-subject factors. The inte raction between dimens ion and replicate is significant [F(6,247)=7.6; p<.001] and so are the main effects of dimension [F(3,247)=16.7; p<.001] and replicate [F(2,247)=14.6; p<.001]. 0 10 20 30 40 50 60 70 80 restaurantcarpainter comp reliab court cred Figure 5-4. Average generalization as a function of replicate and dimension. Agreement among judges in coding the cognitive responses was 89%. The overall correlation between the numeric level of ge neralization and the dichotomous level of stability was r=.43 (p<.001). To illustrate the extremely low generalization in the credibility-restaurant scenario is backed up by an “unstable” rating for 14 out of 15 codable responses.

PAGE 59

49 Discussion No specific directional predictions were adva nced at the outset of this experiment. Instead, the main goal was to explore whether generalization varies as a function of the specific dimension and/or category. The in teraction between dimension and category confirms that both the category and the dimens ion matter. Together with the results of Experiment 6, the data in Experiment 7 indicat e that generalization can vary as a function of the broader category (product vs. service: E xperiment 6) but also as a function of the specific category and dimension within a se rvice context. The suggestion is that consumers come up with a different theor y, depending on the very specific purchase situation in which they find themselves. For instance, overcharging seems to be perceived more of an exception when performed by a waite r than when performed by a mechanic or painter. Overcharging by a waiter also seems to be perceived as more of an exception than unfriendliness of a waiter. That is, not only is the product-se rvice distinction too broad to determine an accurate generalization le vel, so is the specific product or service category. The performance dimension is al so needed to predict generalization. An even more extreme example of how specific lay theories can be is suggested by a comparison across experiments. Experi ment 7 includes a re staurant/unfriendly (courtesy) scenario while Experiment 6 desc ribes a coffee shop/rude scenario. Although both scenarios seem highly similar and ev en interchangeable--especially against a background of broad product-servic e differences--generalization is considerably higher in the restaurant/unfriendly scenario (mean = 59) than in the coffee shop/rude scenario (mean = 38). It is entirely possi ble that this difference is not systematic and is attributable

PAGE 60

50 to error. However, it would be interesting to see whether su ch subtle differences spark differential theories. Experiment 8 pursues this possibility. Experiment 8 Context (restaurant/coffee shop) and beha vior (unfriendly/rude) are orthogonally manipulated. The suggestion is that consum ers may perceive negative outcomes to be caused by more stable mechanisms in a co ffee shop than in a restaurant and that unfriendly behavior is seen as more stable than rude behavior It would be interesting if consumers are sensitive to these subtle di fferences, but not to such seemingly obvious marketing variables as “product” and “servi ce” at the broad generic level (Experiments 3 and 5). Method and results Sixty-five undergraduate stude nts participated in this 2x2 experiment that crosses context (restaurant/coffee shop) with behavior (rude/unfri endly) as between-subject factors. The students were told to imagine an encounter with a rude /unfriendly waiter in a restaurant/coffee shop and were asked what percentage of custom ers they feel would have the same experience in this place (A ppendix I for the full stimuli). Confirming the across-experiment differences twice (Figure 5-6), the two main effects are significant. Generalization is higher in a coffee shop than in a restaurant [F (1,61)=4.7; p<.04], and higher with the unfriendly than the rude behavior [F(1,61)=4.1; p<.05]. There is no evidence for an interaction (F<1). Coding of the cognitive respons es resulted in 89% agreement and a correlation of r=.66 (p<.001).

PAGE 61

51 0 10 20 30 40 50 60 70 restaurantcoffee rude unfriendly Figure 5-6. Mean generalization as a func tion of purchase context and behavior. Discussion The differences observed across experime nts between rudeness and unfriendliness on the one hand and a restaurant and coffee s hop context on the other hand are replicated in a well-controlled environment. This suggest s that consumers percei ve true differences between seemingly interchangeable contexts or behaviors. This finding, together with the positive correlation between generalization and st ability, is important from a marketer's perspective and is interesti ng because it suggests that rudeness/unfriendliness are perceived as being caused by a more unstable me chanism in a restaurant than in a coffee shop. At the same time, rudeness is perc eived as being caused by a more unstable mechanism than unfriendliness. However, one cannot rule out a simple base rate explanation: Rude behavior may be less likel y in a restaurant than in a coffee shop and rude behavior may occur less frequently than unfriendly behavior. If so, future research may want to investigate the interplay betw een causal lay theories and base rates in determining generalization. Before turning to Chapter 6, I summari ze the evidence collected in Experiments 4-8. The second set of experiments supports the lay theory account where the first set did not. Experiment 4 shows that consumers hold la y theories consistent with the positivity

PAGE 62

52 effect observed in the first set. Experiment 6 indicates that cons umers not only hold lay theories but also apply them when generali zing. Experiments 7 and 8 suggest that those lay theories tend to be more specific th an could reasonably be anticipated. The suggestion--when n=1--is that generalization is dependent on the very specific lay theory that is applied, and not easily predictable. To what degree the same is true as the sample size increases is not clear. The question is es pecially pertinent in those situations when n=1 and where an unstable mechanism is assu med. The third set of experiments explores this issue more systematically.

PAGE 63

53 CHAPTER 6 MULTIPLE “UNSTABLE” OBSERVATIONS Experiment 9 Given that consumers seem to be unwilli ng to generalize in some situations, it becomes important to understand the boundaries of this reluctance as the number of observations increases. It is possible that Boulding et al.' s (1999) prediction holds and that an unstable mechanism dominates even after multiple negative outcomes have been experienced. Alternatively, consumers may quickly assume a stable mechanism when n=2 after initially having perceived an unstable mechanism. In other words, the assumption of an unstable mechanism may be short-lived. If true, the implication may well be that the Law of Small Number s holds when n=2, if not when n=1. Method Ninety-five students participated in the study. The two scenarios producing the lowest mean generalization in Experiment 7 were included in Experiment 9: the wristwatch running behind and the undercooke d meal. In addition, two scenarios from Nisbett et al. (1983) were included for compar ison: the element floridium burning with a green flame on an imaginary island and obese members of a tribe (Barratos) on this island. Extreme generalization has been observe d with floridium, even after one trial, while moderate to low generalization has b een observed for obesity (Appendix I for full stimuli).

PAGE 64

54 Results Descriptively, the low levels of genera lization observed af ter one episode are replicated for the marketing stimuli (F igure 6-1). Similarly, the high level of generalization for the floridium element in Nisb ett et al. (1983) is re plicated, as is the moderate level of generalization for the obese Barratos. Generalizati on after two episodes increases dramatically for the two marketing stimuli, but much less or not at all for the other replicates. 0 10 20 30 40 50 60 70 80 90 watchrestaurantobeseelement 1expos 2expos Figure 6-1. Mean generalization as a functi on of replicate and number of exposes. These observations are confirmed by a replicate by number of exposures interaction [F(3,87)=6.1; p<.01]. The single ma in effect for number of exposures is significant for the marketing stimuli [F( 1,46)=43; p<.001], but not for the two other replicates [F<1]. Overall, the main effect of number of observa tions is significant [(F(1,87)=28.7; p<.001)], as is the main eff ect of replicate [F(3,87)=9.2; p<.001]. Coding of the cognitive responses led to agreement in 93% of the re sponses and an overall r=.63 correlation (p<.01). Discussion All four cells with a sample of n=1 rep licate generalization previously observed either in Experiment 6 or in Nisbett et al. (1983). Generalization is low in the wristwatch,

PAGE 65

55 restaurant, and obese scenario, while extr emely high in the floridium condition. In addition, generalization based on two observati ons in the obesity and floridium cases seems in line with the results that Nisbett et al. reporte d. The observed pattern in the obese condition seems to correspond to th e slow updating process anticipated by Boulding et al (1999). However, the contrast with the marketing stimuli is stark. Whereas generalization was extremely low after only one observation, the addition of a second observation induced a huge increase. In this case, the updating process does not seem to be slow but extremely fast. The suggestion is that even though consumers may initially surmise an unstable mechanism, a stable mechanism is assumed as soon as the same episode is experienced twice. The lesson fo r the marketer may not be that consumers learn and update slowly, as ar gued by Boulding et al., but that they infer consistent low quality easily, even when high quality is init ially expected. However, it is also possible that the high generalization when n=2 is the result of participants’ compliance with what they think is the hypothesis pursued by the experimenter. Experiment 10 pursues this possibility. Experiment 10 Thus far all experiments that incl uded marketing stimuli showed high generalization in the two-experience condi tions. Experiment 10 explores whether the high generalization when n=2 should be attribut ed to “demand” or “blind generalization.” To test this contention, one c ondition is included in which th e target service behavior for the second negative experience is explicitly sa id to be performed by a different rather than the same waiter. The actual outcome is id entical to that when the second experience is caused by the same waiter. If the previous results are attributable to “blind

PAGE 66

56 generalization” when n=2, high generali zation may reasonably be expected in Experiment 10 when n=2, even when the wait er in the second episode is different. Alternatively, if lay theori es--instead of “demand”--dri ve generalization, one might expect low generalization when the waite r is different in the second episode. Method and Results Sixty participants were asked to imagin e they went to a coffee shop and had one or two negative service expe riences with a rude waiter. The second experience either involved the same waiter or a different waiter. When the waiter was the same, participants were asked, “For each set of 100 customers, what percentage do you feel would have the same experience with this waiter ?” When the waiter of the second experience was a different one, the question wa s, “For each set of 100 customers, what percentage do you feel would have the same experience in this coffee shop ?” (Appendix J for the full stimuli) This resulted in a 2x2 design that crosses the number of experiences with the target of generali zation (this waiter/this coffee shop). Because the dependent measure is different in the two “target of ge neralization” conditions, the data are analyzed separately. For purposes of presentation, th e means are presented in one figure. 0 10 20 30 40 50 60 70 80 90 indivshop 1 expos 2 expos Figure 6-2. Mean generalization in each of the four conditions in Experiment 10.

PAGE 67

57 The mean levels of generalization per condition indicate that generalization is considerably higher after two experiences than after one in the “individual” condition but much less in the “coffee shop” condition (Figure 6-2). T-tests confirm that the former is highly significant [t(28)=-4.7; p<.001] wh ile the second does not even approach significance [t(28)=-1.1=p<.3]. The correlation between leve l of generalization and stability is r=.47 (p<.01). The judges agreed in 97% of the responses. Discussion When rude behavior was displayed in th e second experience by the same waiter, generalization increased dramatically, just as in previous experiments. However, when the target behavior was associated with a di fferent waiter, generali zation did not increase relative to the one-expe rience condition. This result suggest s that particip ants did not generalize “blindly” in previous experime nts when n=2 and that the Law of Small Numbers does not always hold, even when n=2. Instead, the suggestion is that generalization can be low when n=2 if the a ppropriate theory is cued, e.g., a different waiter is responsible fo r the negative outcome.

PAGE 68

58 CHAPTER 7 GENERAL DISCUSSION AND CONCLUSION "When do consumers (not) generalize?" and "How should one understand generalization in the marketplace theoretica lly?" are the central questions in this dissertation. The experiments s how that generalization is hi gh in a product context rather than a service context and when the outcome is positive rather than negative, but not when the sample is typical rather than atypi cal. Even within a produc t or service context, considerable variance indicates that generalization can be high in a service context and low in a product context. Although two consistent outcomes tend to be perceived as representative of the larger population, tw o inconsistent outcomes tend not to be. The results are interpreted in line with a causal lay theory perspective. Generalization is high when the assumed mechanism is stable, low when the mechanism is unstable. Direct evidence for the existe nce of such lay theo ries is provided by Experiment 4, which shows that consumers tend to believe that stable mechanisms cause a positive outcome while unstable mechanisms cause a negative outcome. Evidence for the impact of lay theories on generaliza tion is provided by Experiment 6, where generalization is higher in a product context than in a service context. Across experiments, correlational evidence shows that generalization tends to be higher when a stable rather than unstable mechanism is assumed. The level of correspondence be tween these results and previous research differs depending on the domain selected for compar ison. Correspondence is lowest with the

PAGE 69

59 majority of research in the generalizati on and induction domain, but surprisingly high with research in the social psychology lit erature and with category formation research. Relevant literatures are discu ssed and suggestions for future research are incorporated in the following discussion that focuses fi rst on the claim that lay theories drive generalization and then on the actual lay theories. The Lay Theory Account A major inconsistency occurs when the focus of comparison is the “calculation” account suggested by the inducti on literature (Osherson et al. 1990; Sloman 1993; Heit 2000). In many induction studies, typicality exer ts a pervasive and reliable impact, yet it failed to influence generalization in this res earch. Does this discrepancy suggest that two qualitatively different mechanis ms drive generalization in a biological (induction) context compared to a marketing or social context? Do lay theori es drive generalization in marketing and social contexts, while calcul ations predominate in a biological context? Or is the lay theory account a special case of the calculation account or perhaps the other way around? Another look across the borders of strict generalization, induction, and even social psychology research might be instructive. A seminal paper on category formation by Murphy and Medin (1985) suggests a way to resolve the apparent discrepancy. Murphy and Medin rejected the long-held belief th at similarity calculations determine which objects are grouped together to form a categ ory (Medin and Schaffer 1978; Posner and Keele 1968; Rosch and Mervis 1975). They argued that the concept of similarity is too unconstrained (Goodman 1972). Instead, they propos ed that “concepts are coherent to the extent they fit people's background knowledge and naive theories abou t the world”

PAGE 70

60 (Murphy and Medin 1985, p.289; also Rehder an d Hastie 2001; Rips and Collins 1993). Simply put, even if similarity calculations drive category formation, lay theories drive and constrain similarity judgments and thus category formation. Transposed to the domain of generalization, Mu rphy and Medin's claim suggests that even typicality findings in the induction literature are the re sult of a specific set of theories about biological categories rather than the output of simple and “objective” calculations. The suggestion is that generalizati on in the induction lite rature is therefor e not qualitatively different from generalization in a consumer context or in a so cial context. The paradox is resolved by considering the typi cality effect as induced by s ubjective lay theories rather than by objective calculations. Fu ture research may want to investigate this empirically. Still, even if valid, Murphy and Medin’ s (1985) suggestion does not explain why generalization seems hypersensi tive to typicality informati on in the induction literature, but not at all in our context. Are there funda mental differences between lay theories about the biological world and lay theories in the co nsumer and social worl d? This is certainly an interesting hypothesis to be investigated in future resear ch. Alternatively, there may be differences between the paradigms that are responsible for the discrepancy. Perhaps the induction paradigm favors typical ity more than does the paradigm used in this research project. Induction scenarios tend to include both a typical and atypical sample. As such, one can argue that typicality information is ma de salient and a typicality effect is more likely. Should one go as far as implying that the induction paradigm induces artificial effects that do not exist in the real world? Probably not. Instead, the induction paradigm may simply highlight certain aspects of a situ ation, and historically this has often been typicality information. As such, the paradigm can be considered a valuable tool for

PAGE 71

61 detecting specific sets of theo ries. In fact, an interesting challenge for future research would be to employ the induction paradigm as a tool to test the impact of theories not based on typicality. For instance, managers may want to know if consumers hold different lay theories according to whether a company owns or franchises its stores (Agrawal and Lal 1995). Vertically integr ated systems (owning) may induce more generalization across stores than does the franchising format because more independence, and potentially instability or heterogeneity, is allowed in the latter Similar questions can be addressed regarding “make” or “buy” decisions (Anderson and Weitz 1986). Do consumers infer more stability when someth ing is made by the company rather than bought from another company? Also, consum ers may hold different theories about “direct” salespeople versus “representativ es” (Anderson 1985). Perh aps theories about representatives imply lower levels of gene ralization than do theories about “direct” salespeople. Systematically including both optio ns in a stimuli set, may give a sense of the degree to which consumers hold different theories in each of these situations. My results are not necessarily incons istent with the hypothesis generation literature and with Hoch and Deighton’s (1989) suggestion that consumers generate very few hypotheses--often only one--about why a certain outcome occurred. However, my research extends theirs by positing that the stability of the underlying mechanism is important in determining the level of gene ralization, even if only one hypothesis is generated. The results are at least partially incons istent with the Law of Small Numbers. Although generalization tends to be high when the outcome is positive, and in a product context even when the outcome is negative, it is often low in a service context when the

PAGE 72

62 outcome is negative. Even an overtly he terogeneous sample is not perceived as representative. Interestingly, many of the findings that are inconsistent with the Law of Small Numbers are in line with Nisb ett et al.'s (1983) view that people, in at least some contexts, perceive heterogeneity and thus re frain from generalizing. Our results extend Nisbett et al.'s research by identifying f actors that do and do not seem sufficient to explain the variance in generalization. Ne ither the objective heterogeneity of the population (base rates) nor the objective heter ogeneity of the sample proved sufficient to understand variance in generalization. That even typicality calculati ons failed to impact perceived heterogeneity is perhaps even mo re perplexing. Instead, causal lay theories seem to determine the level of perceive d heterogeneity and thus generalization. My results seem most consistent with findings in social ps ychology that people hold and apply causal lay theories in a variety of judgments. Research on stereotypes has long agreed that people hold beliefs--mostly stable--that impact judgment dramatically (Kunda and Spencer 2003). Research on the correspondance bias (G ilbert and Malone 1995), on person versus group perception (Ham ilton and Sherman 1996) and on valenced behaviors (Ybarra 2002) implies that people per ceive more stability in some situations than in others. The results are also consistent with res earch on causal attri bution, which states that people seek out causal mechanisms in developing an explanation for a specific event and that they do not necessarily need covariation information (Ahn et al. 1995). Thus far, the discussion has concentr ated on consistency between relevant literatures and my claim that la y theories drive generalization. At a more specific level, it

PAGE 73

63 may be instructive to compare the actual lay theories that have been observed with what could reasonably be expected on th e basis of previ ous research. Specific Lay Theories The number of lay theories that can be applied is virtually countless. We have started to document a few of them--some a bout individual purchase experiences, others about a sequence of two experiences. First, consumers tend to believe that a positive outcome is generated by a stable mechanism, while a negative outcome is believed to be caused by an unstable mechanism. Consequently, positive outcomes lead to higher generalization than do negative outcomes. This finding is consiste nt with what Folkes and Patrick (2003) observed in a service context, but is opposite to what an entire body of research in the psychology literature anticipated. For instan ce, Ybarra (2002) argued that negative human behavior tends to be seen as the cons equence of a stable mechanism--a personality trait--whereas positive beha vior is perceived as the consequence of an unstable mechanism, e.g., situational factors. Together, these re sults suggest a perplexing dissociation between the social and the commer cial world regarding the valence-stability relationship: Whereas people seem to believe that their own peer is inherently bad, a commercial entity seems to be perceive d as inherently pos itive. Although this dissociation sounds like a scar y thought in a world of ev er-present and sometimes aggressive marketers, it is comforting to know that such positive beliefs about a commercial entity are not longlasting, as suggested by Bouldi ng et al. (1999), but change rapidly as multiple negative outcomes are experienced.

PAGE 74

64 Second, the results suggest that consumers tend to believe that a product is generated by more stable mechanisms than is a service experience. Consequently, generalization tends to be higher in a product than in a service cont ext, as suggested by the work of Zeithaml et al (1990). However, this seems true only when the outcome is negative. When the outcome is positive, stable mechanisms are assumed across the board--in line with the positivity effect--and generalization is high. Another deviation pertains to the specif icity of productand service-related lay theories. Contrary to marketing scholars, consumers do not seem to hold differential theories regarding heterogeneity in a product versus service c ontext at the generic level. Indeed, the product-service difference did not occur until very specific scenario descriptions were introduced. The implication is that, ev en though the product-service distinction may be an interes ting guideline, generalization can vary greatly even within a product or service context, possibly to such a degree that generalizati on can be higher in a service context than in a product context. E qually important in dete rmining the specific level of generalization is the specific pr oduct/service category and even the specific dimension on which the product/servi ce is performing inadequately. Still, the product-service distinction is an interesting parallel to findings in the social psychology literature about pe rson versus group perception (Hamilton and Sherman 1996). Group behavior and a service e xperience seem to induce perceived heterogeneity (instability) much more than in dividual behavior and a product experience. My research suggests that this is true only when the outcome is negative. It would be interesting for future research to investigat e whether valence moderates the effect in the social realm as well. If it does, one may wonder about the direc tion of the moderation

PAGE 75

65 since a negativity rather than a positivity eff ect has been reported in a social context (Ybarra 2002). While theories about individual experiences seem to vary as a function of the valence and the nature of the purchase (p roduct-service), theories about multiple inconsistent experiences tend to vary as a function of the delay between the outcomes and the order in which they app ear. In addition, in line with theories about individual experiences, generalization tends to be more st able as function of a positive rather than negative experience. The order effect is inte resting because it indicates that the most recent--rather than the first--outcome is more likely to be considered as a stable indicator of true quality. This is inconsistent with the majority of findings in the social psychology literature where a primacy-instead of a recency--effect has more often been observed (Nisbett and Ross 1980). Whether the discrepa ncy between my findings in a marketing context and the bulk of evidence in the social psychology literature is attributable to procedural or more fundamental differences may be the target of future research. One possibility is that my paradigm induced recency by probing only one response right after information about the recent outcome had b een processed, rather than two responses-each of them administered after each out come was experienced. More fundamental differences between the two domains cannot be ruled out, and neither can a third possibility. In the social realm, a primacy effect ha s been observed in person perception but a recency effect in group perception (Hamilt on and Sherman 1996). Given the earlier analogy between person percep tion and generalization in a product context on the one hand and group perception and generalization in a service context on th e other hand, it is

PAGE 76

66 possible that a primacy effect would be observed in a product context and a recency effect in a service context. One reason this pattern did not occur in Experiment 2 may be the lack of specificity in the scenario. Conclusion Analogous to a hypothesis on concept c oherence (Murphy and Medin 1985) and to work in the social realm (Hamilton and Sherman 1996; Ybarra 2002), but not anticipated by induction (He it 2000) or generalization research (Tversky and Kahneman 1971), this dissertation claims that generali zation is primarily driven by peopleÂ’s lay theories about how the world functions--rat her than by typicality calculations. To understand how and when consumers generalize, this account suggests, one should know which theory the consumer invokes in the sp ecific purchase situation rather than (1) probe the typicality of the purchase situation or (2) assume that each first experience with a vendor is representative of all future e xperiences with the vendor. More specifically, one wants to know the stability of the underl ying mechanism. A virt ually endless set of theories can be applied, but it is the one-dimensional stability of the underlying mechanism that determines the level of generalization.

PAGE 77

67 APPENDIX A STIMULI FOR EXPERIMENT 1 Imagine it's Friday evening and you're going out for dinner with a group of friends. You decide to go the new restaurant downtow n. YouÂ’ve arrived in the restaurant and you're about to explore the menu. Look at it carefully. --New screen --New York steak of Black Angus Beef with Crushed Black Peppercorn Tenderloin Filet of Beef Baby Back Ribs slow cooked with soy, beer and garlic, glazed with the house made barbecue sauce Beef Kebab In Teriyake Marinade Flat Iron Steak of Black Angus B eef Topped with Fresh Herb Butter Pasta with your choice of Tomato, Pesto or Clam Sauce Prime Rib of Beef Medaillons of Buffalo Tenderloin wrapped in Apple-Wood Smoked Bacon, served on a Grilled Portobello Mushroom All entrees come with a salad, bread and one side dish --New screen --You decided to have the Pasta / Fl at Iron Steak of Black Angus Beef It turns out you liked the meal a lot. / It turns out you didnÂ’t think it was a very good meal compared to other restau rants where youÂ’ve eaten at.

PAGE 78

68 --New screen --What percentage of all meals in this re staurant do you expect to be of the same quality as the meal you had? Please move the sliding scale below to give your answer. --New screen --Why did you guess this percentage of ot her meals that would have the same quality, rather than a lo wer or higher number? Write down your answer in the box below. Don't press Continue before you have finished writing.

PAGE 79

69 APPENDIX B STIMULI FOR EXPERIMENT 2 Imagine that you purchased a product/serv ice (e.g. coffee, t oothpaste, battery, cartridge for printer, disposab le lenses, financial advice, ha irdresser, hotel stay, delivery service, babysitting) from a particular firm for the first time. Your assessment of the product/service was positive/negative. A week /year later you decide to try the brand again. This time, however, your e xperience is negative/positive. --Next screen --Now, imagine that you have the opportunity to purchase a third time from the same firm. Using the scale below, please express your belief regarding the probability that the third experience will be good versus bad. --Next screen --Why did you give the answer you gave? Th at is, please descri be the rationale behind your response. Write down your answer in the box below. Don't press Continue before you have finished writing.

PAGE 80

70 APPENDIX C STIMULI FOR EXPERIMENT 3 Imagine that you purchased a product/serv ice (e.g. coffee, t oothpaste, battery, cartridge for printer, disposab le lenses, financial advice, ha irdresser, hotel stay, delivery service, babysitting) from a particular firm for the first time. Your assessment of the product/service was positive/negative. [Sometim e later you decide to try the brand again. Your experience is positive/negative again.] --Next screen --Now, imagine that you have the opportunity to purchase once more from the same firm. Using the scale below, please express your belief regarding the pr obability that your next experience will be good versus bad. --Next screen --Why did you give exactly the rating you gave instead of a lower or higher number? That is, please describe the ra tionale behind your response. Write down your answer in the box below. Don't press Continue before you have finished writing.

PAGE 81

71 APPENDIX D STIMULI FOR EXPERIMENT 4 Restaurants try to provide good meals to th eir customers. Imagine that you just went to a particular restaurant and you had a meal that was (not) very good. Please list as many reasons as possible as to why this could have happened. Write down your answer in the box below. Don't press Continue before you have finished writing. Please separate different reasons with a star (*).

PAGE 82

72 APPENDIX E STIMULI FOR EXPERIMENT 5 Imagine that you purchased a tube of toothpa ste / battery / cartridge for a printer // financial advice / hotel stay / delivery service from a particular firm for the first time. Your assessment of toothpaste / battery / cart ridge for a printer // financial advice / hotel stay / delivery service was negative. [Sometim e later you decide to try the brand again. Your experience is negative again]. --Next screen --Now, imagine that you have the opportunity to purchase once more from the same firm. Using the scale below, please express your belief regarding the pr obability that your next experience will be good versus bad. --Next screen --Why did you give exactly the rating you gave instead of a lower or higher number? That is, please describe the ra tionale behind your response. Write down your answer in the box below. Don't press Continue before you have finished writing.

PAGE 83

73 APPENDIX F STIMULI FOR EXPERIMENT 6 Imagine that you purchased a brand of razor blade for the first time. You try it and find that it provides a rough shave. If you purchased this brand 100 times, what percentage of the blades do you think would pe rform at the same level as the first one you purchased? Imagine you bought a brand of chocolate candy bar for the first time. You try it and find that it doesn't taste very good. If you purchased 100 of these candy bars, what percentage do you think would taste th e same as the first one you tried? Imagine that you purchased a new wristwat ch. After a week you find out that the clock is running behind. If you tried 100 watche s of this model, what percentage do you think would give an inaccurate time? Imagine that you purchased a brand of ball point pen that you have never purchased previously. You discover that the pen does not distribute the ink evenly, leaving the page with blobs and smears. If you purchased 100 pe ns of this model, what percentage do you think would perform equally poorly? Imagine that you try a restaurant for the first time. You order your meal and find that it is undercooked. If you visi ted this restaurant 100 times, what percentage of all the meals you have do you think would be improperly cooked? Imagine you go to a coffee shop and find the se rvice to be rude. If you visited this coffee shop 100 times, what percentage of the service encounters do you think would be rude?

PAGE 84

74 Imagine you just bought a new home and hi red a company to take care of the garden. The company will send someone to your house once a week to mow the lawn. After the first week you find that the lawn is not mowed carefully. If the company mowed your lawn 100 times, what percentage of times do you think the lawn would be mowed poorly? Imagine that you just bought a cell phone with a new service provider. You have some questions about your plan and decide to call the help desk. It turns out that you are placed on hold for 15 minutes before someone gets to your call. If you were to call the help desk 100 times, for what percentage of your calls do you think the waiting time will be (at least) as long as during your first call?

PAGE 85

75 APPENDIX G STIMULI FOR EXPERIMENT 7 Imagine that you go to a restaurant with a few friends. Each person orders a meal. When the meals are delivered, you find that th e waiter has mixed up the orders and that no one gets exactly what they ordered in th e way they wanted it prepared. For each 100 set of customers, what percentage do you feel would have the same experience with this waiter? Imagine that you go to a restaurant with a few friends. The waiter is slow to take your order and to deliver the check at th e end. For each 100 set of customers, what percentage do you feel would have the same experience with this waiter? Imagine that you go to a restaurant with a few friends. The waiter takes your order but seems abrupt and not very friendly. For each 100 set of customers, what percentage do you feel would have the same experience with this waiter? Imagine that you go to a restaurant with a few friends. At the end of the meal you receive your check and find that the waite r has overcharged you. For each 100 set of customers, what percentage do you feel would ha ve the same experience with this waiter? Imagine that you go to a mechanic for an oil change and some other standard maintenance. Afterward, you pick up your car and when you get home you find that your car is leaking oil. For each 100 set of custom ers, what percentage do you feel would have the same experience with this mechanic? Imagine that you go to a mechanic for an oil change and some other standard maintenance. The mechanic promises that the car will be ready in 2 hours but it actually

PAGE 86

76 takes much longer. For each 100 set of cust omers, what percentage do you feel would have the same experience with this mechanic? Imagine that you go to a mechanic for an oil change and some other standard maintenance. The mechanic agrees to fix it but you find him to be somewhat unfriendly. For each 100 set of customers, what per centage do you feel would have the same experience with this mechanic? Imagine that you go to a mechanic for an oil change and some other standard maintenance. The mechanic quotes you a pric e. When the job is finished, you pick up the car and find that the mechanic charged you more than he initially quoted. For each 100 set of customers, what percentage do you f eel would have the same experience with this mechanic? Imagine that you own a home and would lik e to have some rooms painted. You hire a painter who does the job. Afterward, you notice that the painte r had been sloppy in places and that the end result is not as nice as you expected. For each 100 set of customers, what percentage do you feel woul d have the same experience with this painter? Imagine that you own a home and would lik e to have some rooms painted. You hire a painter who does the job. The painter sa ys that he can get to your job in one week but it actually takes significantly longer. For each 100 set of customers, what percentage do you feel would have the same experience with this painter? Imagine that you own a home and would lik e to have some rooms painted. You talk to a painter who is willing to do the job. You find him to be somewhat unfriendly.

PAGE 87

77 For each 100 set of customers, what per centage do you feel would have the same experience with this painter? Imagine that you own a home and would lik e to have some rooms painted. You hire a painter who does the job. When the job is finished, you get a bi ll that is higher than the price that the pa inter originally quoted. For e ach 100 set of customers, what percentage do you feel would have the same experience with this painter?

PAGE 88

78 APPENDIX H STIMULI FOR EXPERIMENT 8 Imagine that you go to a restaurant / coffee shop. The restaurant / coffee shop employs a lot of waiters and you are being serv ed by one of them. You find him/her to be somewhat rude / unfriendly. For each set of 100 customers, what percentage do you feel would have the same experience in this place?

PAGE 89

79 APPENDIX I STIMULI FOR EXPERIMENT 9 Imagine that you purchased a new wristwat ch. After a week you find out that the clock is running behind. [You decide to te st a second watch of the same model. The performance of the second watc h is identical to the first.] If you tried 100 watches of this model, what percentage do you think would give an inaccurate time? Imagine that you try a restaurant for the first time. You order your meal and find that it is undercooked. [You decide to try the restaurant again. You order a meal and get the same result.] If you visited this restaura nt 100 times, what percentage of all your meals do you think would be improperly cooked? Imagine that you are an explorer who ha s landed on a previously unknown island in the Southeastern Pacific. You encounter several new animals, people and objects. Suppose you encounter a native who is a member of a tribe called the Barratos. He is obese. [You encounter a second member of th e tribe and he is also obese.] If you encountered 100 male Barratos, what pe rcentage do you think would be obese? Imagine that you are an explorer who ha s landed on a previously unknown island in the Southeastern Pacific. You encounter several new animals, people and objects. Suppose you encounter a sample of a new elem ent you call floridium. Upon being heated to a very high temperature, it burns with a green flame. [You encounter a second sample and it also burns with a green flame.] If you encountered 100 samples of floridium, what percentage do you think woul d burn with a green flame?

PAGE 90

80 APPENDIX J STIMULI FOR EXPERIMENT 10 Imagine that you go to a restaurant with a few friends. The restaurant employs a lot of waiters and you are being served by one of them. S/he takes your order but seems abrupt and not very friendly. [Sometime late r you visit the restaurant again. It turns out that you are served by the same/a different wa iter]. For each set of 100 customers, what percentage do you feel would have the same experience with this waiter/in this restaurant?

PAGE 91

81 LIST OF REFERENCES Agrawal, Deepak and Rajiv Lal (1995) “Cont ractual Arrangements in Franchising: An Empirical Investigation,” Journal of Marketing Research 32 (May), 213-221. Ahluwalia, Rohini (2002) “How prevalent is the negativity effect in consumer environments?” Journal of Consumer Research 29 (September), 270-279. Ahn, Woo-Kyoung and Charles W. Kalish (2000) "The Role of Mechanism Beliefs in Causal Reasoning," In F.C. Ke il and R.A. Wilson (Eds.) Explanation and Cognition Cambridge, MA: MIT Press, (199-227). _____, Charles W. Kalish, Douglas L. Medin, and Susan A. Gelman (1995) "The Role of Covariation versus Mechanism Info rmation in Causal Attribution," Cognition 54, 299-352. Alba, Joseph W., Stijn M.J van Osse laer and Wouter Vanhouche (2003) Multicausal thinking and Free Will ," Manuscript under preparation. Anderson, Erin (1985) “The Salesperson as Outside Agent or Employee: A Transaction Cost Analysis,” Marketing Science 4 (3), 234-254. _____ and Bart A. Weitz (1986) “Make-or-B uy Decisions: Vertical Integration and Marketing Productivity,” Sloan Management Review Spring, 3-19. Aristotle (1963), Organon graece (vol. 2) Dubuque, IA: W.C. Brown Reprint Library. (Original translation published 1846). Arocha, Jose F., Vimla L. Patel and Yogesh C. Patel (1993) “Hypothesis generation and the coordination of theory and ev idence in novice diagnostic reasoning,” Medical Decision Making 13, 198-211. Bockenholt, Ulf and Elke U. Weber (1993) “T oward a theory of hypot hesis generation in diagnostic decision making,” Investigative Radiology 76-80. Boulding, William, A. Kalra, and Richar d Staelin (1999) “The Quality Double Whammy,” Marketing Science 18(4), 463-484. Chapman Loren J. and Jean P. Chapman (1969) “Illusory Correlation as an Obstacle to the Use of Valid Psychodiagnostic Signs,” Journal of Abnormal Psychology 71 (3), 271-280.

PAGE 92

82 Cheng, Patricia (1997) "From Covariation to Causation: A Causal Power Theory," Psychological Review 104, 367-405. Coulter, Keith S. and Robin A. Coulter ( 2003) “The Effects of Industry Knowledge on the Development of Trust in Service Relationships,” International Journal of Research in Marketing 20, 31-43. Cuccia, Andrew D. and Gary A. McGill (2000 ) “The Role of Decision Strategies in Understanding Professionals’ Suscep tibility to Judgment Biases,” Journal of Accounting Research 38 (2), 419-435. Davis, J. H. (1984) “Order in the courtroom ,” In D.J. Miller, D.G. Blackman, & A.J. Chapman (Eds.) Perspectives in Psychology and Law New York: Wiley. Einhorn, Hillel J. and Robin M. Hogarth (1982) "Prediction, Diagnosis, and Causal Thinking in Forecasting," Journal of Forecasting 1, 23-36. _____ and Robin M. Hogarth (1986) "Judging Probable Cause," Psychological Bulletin 99 (1), 3-19. Fisher, Stanly D, Charles F. Gettys, Carol Manning, Tom Mehle, & Suzanne Baca (1983) “Consistency checking in hypothesis generation,” Organizational Behavior and Human Performance 31, 233-254. Folkes, Valerie S. and Michael A. Kamins (1999) "Effects of Information about Firms' Ethical and Unethical Actions on Consumers' Attitudes," Journal of Consumer Psychology 8 (3), 243-259. _____, and Vanessa M. Patrick (2003) "The Posi tivity Effect in Perc eptions of Services: Seen One, Seen them all?" Journal of Consumer Research 30 (June), 125-137. Garst, Jennifer, Norbert L. Kerr, Susan E. Harris and Lori A. Sheppard (2002) “Satisficing in Hypothesis Generation,” American Journal of Psychology 115 (4), 475-500. Gettys, Charles F and Stanely D. Fisher (1979) “Hypothesis Plausi bilty and Hypothesis Generation,” Organizational Behavior and Human Performance ,” 24, 93-110. _____, Thomas Mehle and Stanly Fisher (1986 ) “Plausibility asse ssments in hypothesis generation,” Organizational Behavior and Hu man Decision Processes 37, 14-33. Gilbert, Daniel T. and Patrick S. Ma lone (1995) "The Correspondence Bias," Psychological Bulletin ," 117 (1), 21-38. Goodman, N (1972) "Seven Strictures on Similarity," In N. Goodman, Problems and Projects (pp. 437-447). Indianapolis: Bobbs-Merrill.

PAGE 93

83 Hamilton, D.L. and Steven J. Sherman (1996) "Perceiving Persons and Groups," Psychological Review ," 103 (2), 336-355. Heider, Fritz (1958) The Ps ychology of Interpersonal Re lations. New York: Wiley. Heit, Evan (2000) "Properties of Inductive Reasoning," Psychonomic Bulletin and Review 7(4), 569-592. _____and Joshua Rubinstein (1994) "Similar ity and Property Effects in Inductive Reasoning," Journal of Experimental Psyc hology: Learning, Memory and Cognition 20 (2), 411-422. Herr, Paul M., Frank Kardes, and John Ki m (1991) "Effects of Word-of-Mouth and Product-Attribute Information on Persua sion: An Accessibility-Diagnosticity Perspective," Journal of Consumer Research 17 (March), 454-462. Hilton, Denis J. and Ben R. Slugoski (1986) "K nowledge-Based Causal Attribution: The Abnormal Conditions Focus Model," Psychological Review 93 (1), 75-88. Hoch, Stephen J. (1984) “Hypothe sis testing and consumer be havior: If it works, Don’t mess with it,” Advances in Consumer Research Vol. 11 T.C. Kinnear, ed. Ann Arbor, MI: Association for Consumer Research, 478-483. _____ and John Deighton (1989) "Managing What Consumers Learn From Experience," Journal of Marketing 53 (April), 1-20. Hodgins, Holley S. and Miron Zuckerman (1993) “Beyond Selecting Information: Biases in Spontaneous Questions and Resultant Conclusions,” Journal of Experimental Social Psychology 29, 387-407. Hogarth, Robin M. and Hillel J. Einhorn (1992) “Order Effects in Belief Updating: The Belief-Adjustement Model,” Cognitive Psychology 24, 1-55. Johnson, Joel T., Debra L. Long and Mi chael D. Robinson (2001) "Is a Cause Conceptualized as a Generative Forc e? Evidence from a Recognition Memory Paradigm," Journal of Experime ntal Social Psychology 37, 398-412. Joiner, Christopher and Barbare Loken (1998) "The Inclusion Effect and Category-Based Induction: Theory and Applic ation to Bra nd Categories," Journal of Consumer Psychology 7 (2), 101-129. Jones, Edward E. and V.A. Harris (1967) "The Attribution of Attitudes," Journal of Experimental Social Psychology 3, 1-24. Jones, Edward, E., Leslie Rock, Kelly G. Sh aver, George R. Goethals, and Lawrence M. Ward. (1968) “Pattern of Performance a nd Ability Attribution: An Unexpected Primacy Effect,” Journal of Personality and Social Psychology 10 (4), 317-340.

PAGE 94

84 Kareev, Yaakov, Sharon Arnon, and Reut Horw itz-Zeliger (2002) “O n the Misperception of Variability,” Journal of Experiment al Psychology: General 131 (2), 287-297. Kelley, Harold, H. (1967) "Attribution Theory in Social Psychology," In D. Levine (Ed.), Nebraska Symposium on Motivation (pp. 192-242). Lincoln: University of Nebraska Press. Klayman, Joshua (1995) ”Varieties of conf irmation bias,” In D.L. Medin, J.R. Busemeyer, & R. Hastie (Eds.), The psychology of learning and motivation: Decision making from a cognitive perspective New York: Academic Press. Koehler, Derek J. (1994) “Hypothesis ge neration and confidence in judgment,” Journal of Experimental Psychology: Learning, Memory and Cognition 20, 461-469. Krueger Joachim and Russel W. Clement (1996) “Inferring category characteristics from sample characteristics: Inductive reasoning and social project,” Journal of Experimental Psychology: General 125 (1), 52-68. Kruglanski Arie W. (1990) "Motivations fo r Judging and Knowing: Implications for Causal Attribution," in Handbook of Motivation and Cognition: Foundations of Social Behavior Vol. 2, ed. E. Torry Higgins a nd R. M. Sorrentino, NY: Guilford Press, 333-368. Kunda, Ziva and Steven J. Spencer (2003) "When Do Stereotypes Come to Mind and When Do They Color Judgment? A Go al-Based Theoretical Framework for Stereotype Activation and Application," Psychological Bulletin 129 (4), 522-544. Liberman, Nira, Daniel C. Molden, Lorrain e C. Idson and E. Tory Higgins (2001) "Promotion and Prevention Focus on Altern ative Hypotheses: Implications for Attributional Functions," Journal of Personality and Social Psychology 80 (1), 518. McClure John, Jos Jaspars and Mansur Lal ljee (1993) "Discounting Attributions and Multiple Determinants," The Journal of General Psychology 12 (2), 99-122. Medin, Douglas L. and M.M. Shaffer (1978) “C ontext theory of Classification Learning,” Psychological Review 85, 207-238. Mehle, Thomas (1982) “Hypothesis generation in an automobile malfunction inference task,” Acta Psychologica 52, 87-106. Mill, John S (1973) System of Logic (8th ed.) In J.M. Robson (Ed.) Collected Works of John Stuart Mill (Vols. 7 & 8). Toronto Canada : University of Toronto Press (Original work published 1872) Murphy, Gregory L. and Douglas L. Medin (1985) “The Role of Theories in Conceptual Coherence,” Psychological Review 92 (3), 289-316.

PAGE 95

85 Nisbett, Richard A., David H. Krantz, Chri stopher Jepson, and Ziva Kunda (1983), “The Use of Statistical Heuristics in Everyday Inductive Reasoning,” Psychological Review 90 (4), 339-363. Nisbett, Richard E. and Lee Ross (1980) Human Inference: Strategies and shortcomings for social judgment Englewood Cliffs, NJ: Prentice-Hall. Osherson, Daniel N., Edward E. Smith, Or mond Wilkie, Alejandro Lopez, and Eldar Shafir (1990) "Category-Based Induction," Psychological Review 97 (2), 185-200. Peeters Guido and Janusz Czapinski (1990) "Positive-Negative Asymmetry in Evaluations: The Distinction Between Af fect and Informational Negativity Effects," European Review of Social Psychology 1, 33-60. Posner, M.I. and S.W. K eele (1968) “On the Genesi s of Abstract Ideas,” Journal of Experimental Psychology 77, 353-363. Read, Stephen J. (1983) “Once is Enough: Causal Reasoning from a Single Instance,” Journal of Personality and Social Psychology 45 (2), 323-334. Read, Stephen J. (1984) “Ana logical Reasoning in Social Judgment: The Importance of Causal Theories,” Journal of Personality and Social Psychology 46 (1), 14-25. Reeder, Glenn D. and Marilynn B. Brewer (1979) “A Schematic Model of Dispositional Attribution in Interp ersonal Perception,” Psychological Review 86 (1), 61-79. Rehder, Bob and Reid Hastie (2001) "Causal Knowledge and Categorie s: The Effects of Causal Beliefs on Categorizati on, Induction, and Similarity," Journal of Experimental Psychology: General 130 (3), 323-360. Reichenbach, H. (1951) The Rise of Scientific Philosophy Berkeley: University of California Press. Rips, Lance (1975) "Inductive Judg ments about Natural Categories," Journal of Verbal Learning and Verbal Behavior 14, 665-681. _____and Allan Collins (1993) “Cat egories and Resemblance,” Journal of Experimental Psychology: General ,” 122 (4), 468-486. Rosch, E. and C.B. Mervis (1975) “Family Rese mblance: Studies in th e Internal Structure of Categories,” Cognitive Psychology 7, 573-605. Rothbart, Myron and Scott Lewis (1988) "Inf erring Category Attributes from Exemplar Attributes: Geometric Shapes and Social Categories," Journal of Personality and Social Psychology 55 (6), 861-872.

PAGE 96

86 Sanbomatsu, David, M. Akimoto, and Earlene Biggs (1993) “Overestimating Causality: Attributional Effects of C onfirmatory Processing,” Journal of Personality and Social Psychology 65, 892-903. _____, Steven S. Posavac, Frank Kardes and Suzan P. Mantel (1998) “Selective Hypothesis Testing,” Psychonomic Bulletin and Review 5 (2), 197-220. Schaklee, Harriet and Baruch Fischoff (1982) “S trategies in Information Search in Causal Analysis,” Memory and Cognition 10 (6), 520-530. Simon, Herbert A. (1956), “Rational choice and the structure of the environment,” Psychological Review ," 63, 129-138. _____. (1982), Models of bounded rationality Cambridge, MA: MIT Press. Skowronski John J. and Donal E. Carlston ( 1987) "Social Judgment and Social Memory: The Role of Cue Diagnosticity in Negativity, Positivity and Extremety Biases," Journal of Personality and Social Psychology 52 (4), 689-699. Sloman, Steven A. (1993) "Feature-Based Induction," Cognitive Psychology 25, 231280. Tschirgi, Judith E. (1980) “Sensible Reasoning: A Hypothesis about Hypotheses,” Child Development 51, 1-10. Trope, Yaacov and A. Liberman (1996) "Soc ial Hypothesis Testing: Cognitive and Motivational Mechanisms," in Social Psychology: Handbook of Basic Principles NY: Guilford Press, 239-270. Tversky, Amos and Daniel Kahneman (1971), “Belief in the Law of Small Numbers,” Psychological Bulletin 76 (2), 105-110. _____and Daniel Kahneman (1980), "Causal Sche mas in Judgments under Uncertainty," In M. Fishbein (Ed.) Progress in Social Psychology (pp. 49-72). Hillsdale, NJ: Erlbaum Weber, Elke, Ulf Bockenholt, Denis J. H ilton, Brian Wallace (1993) “Determinants of diagnostic hypothesis generation: Eff ects on information, base rate and experience,” Journal of Experimental Psyc hology: Learning, Memory and Cognition 19, 1151-1164. White, Peter A. (1990) "Ideas About Causation in Philosophy and Psychology," Psychological Bulletin 108 (1), 3-18. _____ (2002) "Causal Attribution from Cova riation Information: The Evidential Evaluation Model," European Journal of Social Psychology 32, 667-684.

PAGE 97

87 Wright, Jack C. and Gregory L. Murhpy (1984) "The Utility of Theories in Intuitive Statistics: The Robustness of Theory-Based Judgments," Journal of Experimental Psychology: General 113 (2), 301-322. Ybarra, Oscar (2002) "Nave Causal Understanding of Va lenced Behaviors and Its Implications for Social Information Processing," Psychological Bulletin 128 (3), 421-441. Zeithaml, Valarie, A., A. Parasuraman, and Leonard L. Berry (1985). “The Problems and Strategies in Services Marketing,” Journal of Marketing 49 (2), 33-46. _____, A. Parasuraman, and Leonard L. Berry (1990) “ Delivering Quality Service. Balancing Perceptions and Expectations ” The Free Press, NY Zuckerman, Miron, C. Raymond Knee, Holley S. Hodgins and Kunitate Miyake (1995) “Hypothesis Confirmation: The Joint E ffect of Positive Test Strategy and Acquiescence Response Set,” Journal of Personality and Social Psychology 68, (1), 52-60.

PAGE 98

88 BIOGRAPHICAL SKETCH Wouter Vanhouche graduated w ith a Masters of Science de gree in psychology from the University of Leuven in Belgium in 1995. He held several positions as a research assistant in his hometown university before he decided to take his academic career one step further and pursue a doctorate in the United States in 2001. Four more years of rigorous training at the Marketi ng Department of the University of Florida culminated in the defense of his disserta tion in the summer of 2005.