SHARES, LIKES, AND ENDORSEMENT: EXAMINING THE INFLUENCE OF FACEBOOK FRIENDS ON ONLINE DISTRIBUTION OF HEALTH BASED MISINFORMATION By JEFFREY KYLE RILEY 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 2014
Â© 2014 Jeffrey Kyle Riley
To everyone who helped me get to this point
4 ACKNOWLEDGM ENTS I would like to thank my mother , who is the reason I have managed to achieve the goals I have set in my life. I woul d also like to thank Dr. Julie E. Dodd for her patient supp ort through this whole process, as well as my committee: Dr. Kim Walsh Child ers, Dr. Christopher McCarty, and Dr. Wayne Wanta, all of whom contributed greatly to my research and personal knowledge. Another important thanks is deserved by Holly Cowart, who was often a voice of calm in the trying times that a dissertation causes. So many people contributed to my ability to reach this point that it becomes difficult to write out specifics. Therefore, I would like to thank anyone and everyone who played a part in that process, no matter how small.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 4 LIST OF TABLES ................................ ................................ ................................ ........................... 7 LIST OF FIGURES ................................ ................................ ................................ ......................... 8 LIST OF DEFINITIONS ................................ ................................ ................................ ................. 9 ABSTRACT ................................ ................................ ................................ ................................ ... 11 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .................. 12 2 REVIEW OF LITERATURE ................................ ................................ ................................ . 22 Theoretical Foundations ................................ ................................ ................................ ......... 22 Diffusion of Innovations ................................ ................................ ................................ ......... 25 History of Media Hoaxes ................................ ................................ ................................ ........ 26 Rumors and Misinformation ................................ ................................ ................................ ... 29 Online Social Networks and Social Media ................................ ................................ ............. 30 Rumor Cascades ................................ ................................ ................................ ..................... 36 Health Communication in the Digital Age ................................ ................................ ............. 37 Credibility Assessment ................................ ................................ ................................ ........... 39 Using Social Network Analysis ................................ ................................ .............................. 43 In Summary ................................ ................................ ................................ ............................ 46 Hypotheses ................................ ................................ ................................ .............................. 48 3 METHODOLOGY ................................ ................................ ................................ ................. 49 4 RESULTS ................................ ................................ ................................ ............................... 63 5 CONCLUSIONS & DISCUSSION ................................ ................................ ........................ 80 Diversity of Endorsement ................................ ................................ ................................ ....... 80 Health related Messages ................................ ................................ ................................ ......... 81 Power Endorsers ................................ ................................ ................................ ..................... 82 Editorial Control ................................ ................................ ................................ ..................... 84 Rumor Theory ................................ ................................ ................................ ......................... 88 Social Capital ................................ ................................ ................................ .......................... 89 Conformity ................................ ................................ ................................ .............................. 92 Opinion Leade rs ................................ ................................ ................................ ...................... 93 Limitations ................................ ................................ ................................ .............................. 94 Future Research ................................ ................................ ................................ ...................... 96
6 APPENDIX A : FIRST SESSION MATERIALS ................................ ................................ ............................. 100 B : HAND OUT MATERIALS ................................ ................................ ................................ .... 103 C : SECOND SESSION MATERIALS ................................ ................................ ....................... 105 LIST OF REFERENCES ................................ ................................ ................................ ............. 107 BIOGRAP HICAL SKETCH ................................ ................................ ................................ ....... 115
7 LIST OF TABLES Table page 4 1 Frequency of demographics and characteristics for participants ................................ ....... 72 4 2 Facebook usage ................................ ................................ ................................ .................. 73 4 3 Frequency of questions determining reported Facebook use ................................ ............. 74 4 4 One way ANOVAs comparing collapsed cell, illness, and diversity of network analysis with their connecting Like, Share, and Click variables ................................ ....... 75 4 5 Means of collapsed cells, diversity of modules, and illness seen in the Facebook post with the three dependent variables: Like, Share, Click ................................ ..................... 76 4 6 One way ANOVAs comparing collapsed cell with statement of knowledge of the illness seen in the Facebook post ................................ ................................ ....................... 76 4 7 A linear regression with statement of how often the Facebook post as the dependent variable ................................ ................................ ........... 77 4 8 A bivariate correlation table bet ween the statement of how often the participant uses stimuli ................................ ................................ ................................ ................................ 77 4 9 A bivariate correlation table between the statement of how often the participant uses stimuli ................................ ................................ ................................ ................................ 78 4 10 Descriptives of averaged score for trust level of i .............................. 78 4 11 Fa cebook post as the dependent variable ................................ ................................ ........... 78
8 LIST OF FIGURES Figure page 3 1 Layout of cells indicating which cell received which variable in the exper imental portion of the study ................................ ................................ ................................ ............ 57 3 2 Example of an average Facebook ego centric network when randomly diagrammed ..................... 58 3 3 Example of a typical Facebook network once it has been modulated. The colors and numbers are randomly generated by Gephi. The percentage on the left indicates the size of that module within the network. Example taken from sample of participants. ...... 59 3 4 What an typical Facebook network looks like once modul ation and separation has been applied. Example taken from sample of participants. ................................ ............... 60 3 5 Fake Facebook post created for the second portion of the experiment. This link used the common cold as t he health message and was seen by half the participants. ............... 61 3 6 Second fake Facebook link created for use in the second portion of the experiment. s the health message and was seen by half the participants. ................................ ................................ ................................ ........... 62 4 1 Facebook compared to trusting the content po sted on Facebook. ................................ ..... 79
9 LIST OF DEFINITIONS Cascade A flow of information that is perpetuated by individuals who are aware that the information is popular and being shared actively by others. Ego When a social network anal ysis is based on the perspective of one person, When one asks one user to name some of his or her they know each other, the user being asked is considered the ego. Ego network A social network from the perspective of one node with whom a ll other nodes connect Edge In social network analysis, this is the connection between two nodes, forming the network. For example, on Facebook, two users being connection on the website. When visualized, they are usually seen as lines Facebook status A small piece of information posted by a given Facebook user on his or her own wall, usually not including any kind of hyperlink or image. Usually a small piece of personal information updati own Timelines what is happening with the user. Facebook timeline The list of content a given Facebook user sees when looking at the front page of the website. The Timeline consists of any content posted by individual in chronological order of when they were posted. Facebook wall content can be items posted by the user themselves, or by others posting items directly to their page. A connection made on Facebook with another user. When one user on Facebook can be seen by anyone that mean that the two users are actually friends in the interpersonal sense. Hoax An intentionally created false rumor, often times created specifically to deceive the reader. Often times follows a lighthearted tone. A button that appears below any piece of content on Facebook. The image portion looks like a small blue thumbs up symbol, and next to it is a small the content will see that t form of endorsement for information. It is also used as a verb for clicking said button.
10 Mertenitis An invented illness used by this study in order to have an illness that no person could have any existin g knowledge. It was presented as having the same symptoms as the common cold. Misinformation Information that began as factual information, or was originally based in some kind of fact, but over time was either intentionally or unintentionally skewed. M odule In social network analysis, a module is a calculated grouping of nodes based on their connected edge patterns. Node In social network analysis, this is the single unit of item being analyzed or connected together, forming the network. For example, i n Facebook, usually seen as dots, squares, or other symbols. Individuals who perform certain activities on Facebook, such as posting status updates, photos, and content , at a higher rate than most users. Rumors Pieces of information that cannot be verified at the time they are read. They do not have to be non factual . They simply cannot be completely fact checked at the moment they are read. A button that appears below any piece of content on Facebook. It appears the content will then be copied onto the wall of the person who clicked, form of endorsement fo r information. It is also used as a verb for clicking said button. Social network analysis The analysis of social relationships using nodes and edges which is utilized by a number of different hard sciences and social sciences. Social capital The commo dity produced by social behavior as if it were an economic system, where actions are met with an expected result, often including preferential treatment. The small micro blog style posts made on the social media website Twitter. It is limited to 140 characters or fewer.
11 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 SHARES, LIKES, AND ENDORSEMENT: EXAMININ G THE INFLUENCE OF FACEBOOK FRIENDS ON ONLINE DISTRIBUTION OF HEALTH BASED MISINFORMATION By Jeffrey Kyle Riley August 2014 Chair: Julie E. Dodd Major: Mass Communication This was a two stage experimental analysis that set out to find the impact o f network construction on rumor spread at the individual level on the popular online social networking site Facebook. The research was designed by forming a theoretical basis in rumor theory, opinion leade r theory, and known research concerning online soci al networking websites. The study also approached rumors from a health communication standpoint. The first portion of the experiment networks and asked the participa n ts to answer questions about their onli ne activity. During the second portion, participants were asked if they would choose to endorse a health message if it was posted to their Facebook pages. In the 2X2 experiment, two variables were used: diversity of e ndorsement from different modules, and using a real versus fake illness. The study revealed that regardless if it is true or false, which sets a poten tially pro blematic implication dealing with the
12 CHAPTER 1 INTRODUCTION As of 2014, Facebook play s a tremendous role in the day to day communication of millions of people . On a daily basis, millions of pieces of information are posted, shared, and liked on Facebook ( Facebook statistics, stats & facts , 2011 ). On a typical day in 2011 , Facebook wa s the conduit for about 1 million posted hyperlinks, 1.8 million status updates, and 1.5 millio n messages ( Facebook statistics, stats & facts , 2011). That represents an enormous amount of content that is being shared within a network of connected users. Accordin g to the analytic data from 2014 collected by Alexa (2014), an Amazon owned website that tracks analytic data, Facebook was the second most accessed website on Earth in total daily pageviews, behind only Google. Alexa (2014) data also showed a global avera ge of nearly 30 minutes of daily time on site across all users, one of the largest time on site numbers of any website. The next closest social networking site on the Alexa (2014) global rankings was LinkedIn, at No.7, with an average of only about 7 minut es of time on site. YouT ube, a video player website with social media components, only averaged about 18 minutes of time on site; Twitter, a microblog with social networking components, only had about 9 minutes of time on site. No other website in the top 20 listing created by Alexa (2014) had more time on site. With Facebook passing the 1 about one Facebook account for every seven people on Earth (Ortutay, Octob e r 4 , 2012). Based on that data, it is clear that Facebook has a hegemonic presence in the marketplace of online social networking websites. Given the millions of pieces of information that are posted to Facebook every day , Facebook is one of the biggest c ommunication tools in the world. The top ranked news site, according to Alexa (2014), is CNN.com at number 17, well below Facebook in every statistic
13 available, from time on site to number of links in. CNN.com has about 250,000 active hyperlinks that are l inking in from around the web (Alexa, 2014). Facebook has about 6.7 million active hyperlinks (Alexa, 2014). From breaking news stories in the form of shared stop site for the in formation people seek on a regular basis online (Mitchell, et. al., 2013). T his enormous, diverse array of content and traffic was not what Facebook was originally designed for. The online social networking site that now controls the online network market started in a dorm at Harvard and was initially intended to be used as a dating type website where students could learn more about other students (Zeevi, 2013). Initially only about half of inventor and owner, Mark Zuckerberg, allowed students at other Ivy League schools to join (Phillips, 2007). The site soon expanded, first to Stanford and other universities in the greater Boston area, then allowing anyone with a coll ege specific educational email address access. From 2004 until 2005 the site grew in popularity, but it consistently maintained a relatively steady and manageable population. That changed when the site opened up to allow high school students and corporatio ns in 2005, and anyone with a n email address in 2006, to create an account (Phillips, 2007). The number of accounts rapidly increased yet the basic layout of the site was still about the same as it was when it first started. Facebook is a networking softwa re design developed for students at a handful of Ivy League schools that has evolved into a sharing site that , in 2012, had about 1 billion unique accounts. The site has undergone changes since its dorm administrators a up below all pieces of user created and user li nked content on Facebook. T he stated goal of the button is
14 provide a way for people to share their inte and provi de recommendations to there is a form of communication and endorsement, and the name of the person who clicked n, the content itself is then posted on the own wall rising , up nearly 52% from the previous year (Wong, 2013). However, not all of that enormous volume of content has been factual. Many e xamples of rumor s spread throughout the Facebook network have occurred both on the small scale, interpersonal level, and large scale, websi te wide basis, in the form of malicious personal rumors, unfounded breaking news misinformation, and intentionally created hoaxes (Chen, 2013). Facebook does not have an editorial staff to review site content, making it vastly different from a newspaper, n ews website, or magazine that has a staff who profess ionally edit all content. Facebook has lawyers and reviewers who scan for content that could place Facebook in legal danger, such as nudity, illegal content, or copyrighted content, but those individuals are not previously a reporter and editor for Bloomberg News (Bercovici, 2012). In January 2014, p opular news stories and topics of the day with short descriptions of their content, the result of an expansion into user dominated
15 only shows aggregation of content into one sp ot, not the removing of non factual content. demonstrated itself as a form of actual editorial control the type seen at newspapers, or news websites, for example. Face must be solved by answering the question: How is it that a piece of incorrect information, be it in the form of a linked news story or a doctored image or any other form, is perpetuate d on the Facebook network? In 2014, a team of researchers from Facebook analyzed the phenomenon of quick and far reaching rumor spread on Fa cebook, noting that very little was known about how rumors propagate on Facebook (Friggeri, et al., 2014). Calling the researchers measured the replication and longevity of a number of popular rumors on Facebook during 2013, noting how the spread of rumors changed based on the feedback individual users received. Their study was based in r umor theory, whic h often pays little attention to differentiating between a true rumor and a false rumor. In rumor studies, a rumor is defined as a piece of information that c annot be verified at the time at which it is received, which does not automatical ly confirm it as being false (Buckner, 1967). Frigeri, et al. (2014) looked at the overlap between widespread rumors on Facebook and rumors that were cataloged on the popular hoax database Snopes.com. By doing so, the researchers could track what happened to the rumor after it was seen corrected online. The correction occur r on Snopes.com page indicating that the rumor was either false or not wholly true. On Snopes.com, about 45% of the reviewed rumors are completely false, about 29% are considered ambiguous
16 enough so that they cannot be proven false, but also were certainly not true, and the remaining 26% of rumors are considered outright truths. much lower than the 26% found on Sn opes.com. Friggeri, et al. (2014) found that outright truthful rumors were much more viral than outright false rumors. However, the researchers found that false rumors had a higher proportion of being com article correcting the rumor than did true rumors. That finding presents a dangerous result a false rumor on Facebook can break through the barrier of being exposed as false if it is viral enough. It did not matter if the Sno pes article proving the rumor false followed the rumor along through the cascading process. Once a false rumor was popular enough and was being spread fast enough, the false rumor was going to continue to spr ead and propagate at that speed until the rate of propagation declined d ue to saturation or disinterest. Because Facebook operates without an editorial staff to delete known false rumors, Friggeri et al. (2014) found that false rumors thrived in the Facebook environment, making thousands of new connections per hour. Although m ost of the rumor examples used by Friggeri, et al. (2014) were light hearted in nature, such as a photo claiming that a month was lucky for having five Fridays, Saturdays, and Sundays, the potential exists for a more dangerous situation. In their research, it was noted that medical rumors represented an above average
17 corrected, the question presents itself: What happens if a false rumor about a piece of potentially life threatening health based or medical infor mation cascades on Facebook? The work done by Frigg eri, et al. (2014) is helpful for establishing that simply correcti ng a false rumor is not enough the false rumor will continue to propagate, if the rumor is viral enough. but not the closer picture of the psychological decisions that could potentially manifest when the rumor managed to spread from one single Facebook user to another. The researchers noted in explaining the limitations of the study: the re sharing mechanism on Facebook contributes to the observed data in unknown ways . . . Furthermore it is unclear how individuals chang e their attitudes toward rumors This study , begun several months before the release of the Friggeri, et al. study, Facebook by addressing the issue of rumor spread at the small scale. Instead of analyzing the terioration from beginning to end, this study took a closer look at the The study did so by setting up a two book network data and asked questions concerning Facebook usage, and the second stage showed the participants what looked like a link on a Facebook wall. Participants were then asked if they The purpose of this dissertation was to examine rumor spread on Facebook by testing two different concepts, both with a basis in rumor theory. The first was the message itself. Two different images were used, each looking like a realistic link posted to Facebook. One contained
18 information about the common cold, while the other contained information about an illness that does not exist Mertenitis and was created just for this study. The other variable involved of the participants saw the link as network. The two variables were used to construct a 2X2 experimental design . are two ways that false rumors on Facebook have the potential to have hurtful outcomes: The first is the small form rumor, which often ta kes the form of cyberbullying. Rarely is cyberbullying enormous and wide spread across the network, as it usually stays within a series of smaller interpersonal networks. Yet, cyberbullying is still extremely harmful f or those who find themselves victims. Research on the topic of cyberbullying has focused on social networks and their potential same NCPC (2001) information states that one way to stop cyberbullying is to not forward to any other people any online information about someone in which you do not know if it is true. Spreading malicious rumors about some one online is type of cyberbullying most reported by victims and the second most reported type of cyberbullying stated by offender s (Hinduja & Patchin, 2010; Patchin & Hin duja, 2010). One example is the case of 17 year old high school junior Austin Carner. A rumor spread across social media that Carner was planning on brin g ing a gun to school, resulting in the police searching his room and questioning him (Leitsinger, 2012) . The rumor began and spread on Facebook while Carner was working on a community service
19 project. The rumor was considered a malicious act of bullying (Leitsinger, 2012). The implication of rumors, bullying, and Facebook was taken a step further in August 2010, when a rumor began circulating around the town of Millersville, Pennsylvania, that a 14 year old student named Emma Kuhlin had committed suicide after finding a rumor online (Murse, 2012). that contains rumors and insults about a list of people , made popular by the 2004 film Mean Girls . The story of Kuhlin spread first in person, then online, becoming a viral case. People made tribute YouTube videos and Facebook pages and rallied in support of Kuhlin and against cyberbullying. The only problem was that Kuhlin never existed office (Murse, 2012). Authorities did not know who invented Kuhlin. The archetype individual who acted as a warning about the dangers of online rumors was, herself, a rumor (Murse, 2012). There is another type of Facebook rumor the mass hoax. The mass hoax tends to resemble chain mail both in physical form and e mail chain mail and is normally designed to perpetuate across a wide audience by using emotionally charged stories and images, or the a large company, or a randomly drawn prize (Bea, 2013). The mass hoax is often th e r esult of contests intended to get hoax has led to the creation of a number of online sites dedicated just to showcasing and cataloging them, such as Hoax Slayer.com , ThatsNonsense.c om, and the popular Snopes.com. Some popular examples of mass hoaxes include images of a child prepped for a heart valve repair surgery with the incorrect caption claiming t doctor would perform a fr ee heart transplant (Mikkelson, 2012). Th e image received thousands who
20 would withhold treatment from a patient based on the number of times a photo was endorsed on Fa cebook . In 2012, a software developer named Nolan Daniels posted a fake winning Powerball ticket on Facebook, stating that he would randomly give peop le $1 million if they clicked ed, almo st 200,000 people friends with (Chen, 2012). Daniels (2012) later wrote in a response to the public that he had made the original post s imply to see what would ha ppen an d felt remorseful that people had gotten their hopes up of being saved in tough economic times. The use of sites like Facebook, Twitter, and Reddit in the aftermath of the Boston Marathon bombings in April 2013 showed the unintended consequences of widespread rumors on social media. In the wake of the horrific tragedy at the Boston Marathon, communities on the Internet, such as Reddit, Something Awful, and Twitter, attempted to assist the police investigation by scouring publically available images o f the scene and highlighting suspicious individuals (Miners, 2013). Not only did this action lead to racial profiling but a lso to dozens of incorrect leads. Among those incorrectly identified as the potential bombers were t wo off duty police officers who w ere false ly targeted as suspects, the father of the 8 year old child hurt in the explosion, a Moroccan high school student who was a championship track and field runner, and a missing Brown University student who was later found dead (Chen, 2013). Images o f people circled in the crowd found their way onto Facebook, where the information continued to spread . O n Facebook, the spread was different however. Unlike on Reddit, where the posts were made by disassociated account names, Facebook is mostly real peopl e with real names associated with their accounts. By posting these incorrect images pointing the wrong people out as suspects, Facebook users forced the FBI and others to prematurely release the names of actual suspects
21 due to the widespread targeting of i nnocent people (Montgomery, Harowitz & Fisher, April 2013). Misinformation and rumor on Facebook again caused a harsh negative The research presented in this study builds on existing research about how a content platform in this case Facebook transmits rumors. The research also indicates potential ervision for false, but not illegal , rumors. Aside from the theoretical implications, the practical implications of this study are al so clear for journalism educators who are preparing beginning journalists in best practices for use of social media and for health organizations in terms of the potential for false health information spreading on Facebook.
22 CHAPTER 2 REVIEW OF LITERATURE Although a number of academic areas in communications, social network analysis, psychology, and sociology have examined rumors and social media, there has been no direct research looking at rumors and Facebook from the perspective of how the information is Theoretical Foundations Starting in the 1940s, researcher Solomon Asch (1940) pro posed that social conformity has ternal suppression of dissenting voice. Asch (1955) tested that co n cept with the now point out which of the lines on another poster matched the orig inal line in length. There was one that was clearly too small, one that was clearly too long, and one that was the correct length (Asch, 1955). However, there really was only one true participant in the study all of the others were confederates who inten tionally selected the incorrect line. When the confederates vocalized their selection of the incorrect line, in a majority of cases, the actual participant would, despite objections and internal disagreement, go along with the rest of the group (Asch, 1955 ( 1955) work supports a basic assumption into the core foundation of this study that people will endorse seemingly incorrect information if put in a position of social pressure. Asch (1955) proposes that action taken due to social pressure occu rs through a three option process, in which there is a distortion of perception, judgment, or action. Distortion of perception happens when someone honestly believes that the group made the best decision due to their increased numbers. Distortion of judgme nt happens when someone believes he or she originally made the correct assessment, but because of the group agreement on another answer,
23 they must have been incorrect as a form of insecurity. Distortion of action occurs when someone believes the group is i ncorrect and his or her own assessment is correct but does not speak up because of fear of upsetting the status quo. ly lacks the nuance needed to determine which particular kinds of social pressu re could influence someone to endorse a piece of misinformation. suggest that people tend to conform, and that conformity can lead to incorrect internal assessments and action, but what cognitively occurs at the individual level, and how that is addressed through the contact and friendship of a structured network, is not addressed. In addition to Asch (1955), t wo other classical areas of communication theory are used in this study as foundational materials . The first is the two step f low of communicat ion, largely pioneered by Katz and Lazarsfeld (1955). The sec ond is diffusion of innovations as explored by Rogers (2003). Two step flow hinges on the idea that the impact of messages presented through various mass media to an individual, initially examined as print, television , and film, is not simply from the message itself (Katz & Lazarsfeld, 1955) th oughts and opinions on the mediated message. The opinion leader, often seen by the individual as someone with existing knowledge, understanding, and connection to the topic at hand, will express his or her own on the subject, informing and supplementing the opinion of the individual. It is important to note that an opinion leader is not necessarily someone who is socially powerful, in that it does not have to be someone with an important job, wealth , etc. Being an opinion leader also do es not imply that the individual is automatically an early
24 adopter of new ideas or consumes more media than everyone else the concept simply means that opinion leaders are perceived as having a heightened sense of leadership on a given subject. Katz and Lazarsfeld (1955) show communication theory has th e strong potential to explain Facebook interactions as well. who dominate conversation and skew statistics on a regular basis (Hampton, et. al., 2012). The average Facebook user tends to get more from his or her friends in the form of information and content than he or she give s , larg ely because of those average user receives for power users to become a sort of dif fused opinion leader becomes increasingly important as more people use online social networks specifically to find information. Hermida et al. (2012) found that almost two fifths of online social network users claim to get news from people they follow, or are friends with, on Facebook and other online social networks, and that sociality is quickly becom ing a part of the information gathering process. Katz & two step flow of communication during an age of analog technologies, when newspapers and printed materials were the main source s of information , radio and film were close second s , and television was the growing source of information . The theory of the two step flow was co unter to the thought that someone simply received a piece of information and decided to either act upon it or not. The Internet, and specifically online social networks, has taken the role of the new powerful communication tool. Online social networks such exemplified by online profile, create a new set of issues when examining two step flow.
25 Even in the 1960s, Katz (1963) began to see that the two step flow was no t adequate at only two steps. Diffusion of Innovat ions The second fundamental theoretical component to this study is diffusion of innovations. The theoretical proposal, pioneered by Rogers (2003), analyzes and identifies the patterns of adaptation to new technologies and ideas. Diffusion of innovations h as been used to explain a wide range of situations, from adoption of farming practices to the use of snowmobiles to herd reindeer. Diffusion studies have the capacity to work at the larger macro level, but also the smaller, micro level. At the macro level, the theory has the capacity to explain the growth patterns of new technologies from early ado pters to laggards. At the micro level, Rogers (2003) explain s how smaller, more nuanced terms of interpersonal communication can have a great impact on adaptation of new technologies, patterns, and ideas. Diffusion is the area this study uses as a fundamental basis in the process of explaining the existence and influence of opinion leaders. While Katz and Lazarsfeld (1955) deal with how individuals turn to opinion leaders for guidance or opinion, Rogers (2003) deals with opinion leaders in terms of progression of new thoughts, items, or concepts. Rogers (2003) argues that opinion leaders have particular attributes that allow them to be leaders and drivers of forwar d thought. An opinion leader is often someone who is exposed to an above average amount of media, is easily accessible and approachable in society, has a slightly above avera ge socioeconomic status, and is recognized as competent and trustworthy. Another key point made in diffusion theory is that there may be, at times, unintended consequences of rapid adoption of a new item, be it tangible or theoretical. It is important to note that Rogers (2003) makes it clear that innovation is not always positive, and early adoption is not always for the better. There are examples of rapid adoption of technologies that turned out to
26 have vast negative consequences, such as the drug Thalidomide being given to pregnant women to ease morning sickness and pregnancy related anxiety, when in fact the drug was causing an outbreak of fatal birth defects an outbreak that lasted for almost seven years (Fintel, Samaras, & Carias, 2009). Researchers have suggested that, at times, diffusion theory has been approached in only a posit ive light, wherein the goal appears to be increasing the number of early adopters and decreasing laggards (Kreps, 2010). This study analyzed how the rapid adoption of two s , could be cau sing problems in rumor spread. In the case of figur ing out the impact of the psychological impact upon endorsement tools, rapid adoption may be seen as a negative. Diffusion studies and two step flow studies have a history of being analyzed together due to the overlapping nature of considerin g interpersonal communication as being of the utmost of importance. T his study uses that overlap both the micro and macro possibilities of opinion leaders as a way to explain how rumor based psychological impacts can operationalize on Facebook. History of Media Hoaxes Media in this ca se, traditional mass media have a h istory with the dissemination of hoaxes, rumors, and misinformation. The historical perspective of previously mass communicated pieces of misi nformation allows for the formation of a proper working definition of what this study examined , as well as establishing that deceiving people is not a new phenomenon. Rumors, as defined by this study, are simply pieces of information for which the truthful ness of the information cannot be determined at the time it is received, while misinformation is the skewing of information into an incorrect form as it passes through a network (Buckner, 1967). Hoaxes, however, are intentionally created pieces of misinfor mation designed to be passed along, sometimes treated as rumors by their consumers (Fedler, 1989).
27 H oaxes were often the result of journalists and publishers trying to be humorous or trying to entertain their readers rarely did a purposefully created hoa x cause any serious harm (Fedler, 1989). A hoax, therefore, can be in the form of a rumor when it is received, but a hoax has to be created as such, and if it is not true, it is misinformation. Some of the implications of this study fall within studying th e one on one mechanical spread of intentional misinformation, so understanding hoaxes and their history with mass communicated messages is important. Hoaxes, after all, were created to be spread (Fedler, 1989) . rizing media hoaxes showcases some of the prime examples of early media crafting and spreading hoaxes. Fedler (1989) argues that hoaxes follow three trends. First, they are often designed to entertain, not harm. If one were to look at modern examples of th is on Facebook, the winning Powerball ticket Nolan Daniels p osted in 2012 fits this framework (Chen, 2012) . Daniels admitted he intended no harm when he Ph million h e simply wanted to see what would happen if he did it (Kendall, 2012) . Next, hoaxes are often owned. Journalists in the 1800s and early 1900s who crafted and published hoaxes would admit to their hoax when questioned, stating they wrote the piece as a form of entertainment or humor, never insisting the information was 100% truthful. Some examples include colorful tales of explorers attaching balloons to ships to sail to Mars, explorers in the jungles of Africa finding 9 foot tall giants, or chemicals being discovered that can turn a copper penny into gold. In all of those examples, the authors, when asked, never insisted their stories were real they always admitted their writing was fiction designed to entertain an audience through deception. Again, this fit s in with the Nolan Daniels story. When questioned, Daniels admitted to fabricating and posting the image (Kendall, 2012) . Finally, the widespread
28 implications of the hoax are often out of the control of the original creator, sometimes yielding hoaxes wher e ones were not intended. Fedler (1989) used the example of Orson Welles and , the radio play mimicking news coverage of an alien invasion that sparked widespread panic. Felder (1989) argued that Welles did not intend on people perceivin g his novella as real life, however the audience did anyway what started as fiction was interpreted as fact. Although Fedler (19 89) stated that hoaxes a re rarely intended to harm, he wa s careful to point ou t that harm can indeed occur. A lthough the inten t was not to deceive listeners , some in the audience were severely frightened and in turn barricaded themselves in their homes, overloaded police and emergency contact lines, and had panic attacks. Fedler (1989) also explained the context in which most of the historical examples of media driven hoaxes occurred. In the golden age of the hoax, the mid to late 1800s, there was little opportunity to q uickly fact check information. Contextually, the world was going through a second industrial and scientific revolution, and there was a post Civil War enthusiasm for the future. The combination of those factors made people more willing to accept such wild stories. (Fedler, 1989). In the modern day context, with Internet availa bility, a newspaper story about a Dutch explorer attaching hot air balloons to his schooner and floating to the moon can easily be detected as fiction (Fedler, 1989). But at the time, an intersection of lack of knowledge of how space works and hope for the future allowed some to consider, for a few short moments, that such a thing was possible especially considering that the submarine was once considered just as implausible but was invented shortly before that particular moon hoax (Fedler, 1989). That con text lends itself well to examining the modern state of hoaxes and rumors on Facebook.
29 Rumors and Misinformation for rumor orientation. Buckner (196 5) proposed that the context in whi ch someone receives a rumor, the setting of the person and the rumor, influences how likely it will be that someone will be critical of the rumor. According to Buckner (1965), an individual approaches a rumor in either a an The uncritical orientation, which suggests that the individual will be more likely to pass along the information without checking it, occurs when one wants the rumor to be true, when the individual has no prior knowledge of the subject, when there is a crisis situation, or when the rumor is on a topic about which few people have knowledge. Buckner (1965) also showed that the incorrectness of a rumor tends to be exaggerated when it travels through a network, moving from person, to groups of people and back again, as opposed to a straight line as a child might play the Telephone Game, where one child shares a piece of information with another child one at a time, and the information 965) work s uggests that networked people leading to but a large webbed network of individuals who have connections with other individuals (boyd & Ellison, 2006). I nformation on Facebook is passe d along t he chain in scattered pattern not a straight line (Friggeri, et al., 2014). Jaeger, Anthony and general circulation wi thout certainty as to its truth (pg. 3). They propose that th e study of rumor must come in the form of careful control of the source of the message, the recipient, the message itself, and the context therein. The contexts that can lead to a higher rate of rumor transmission message from a peer as opposed to an authority figure (Jaeger, Anthony & Rosnow, 1980).
30 -usually based on -co mes with existing social capital, trust, and credibility heuristics (Vitak, Ellison & Steinfield, 2011). For example, just as physical attractiveness leads to a heightened sense of credibility in interpersonal communications, the same result has been found when dealing with pro social posting of information on Facebook (Walther, et. al., 2008). Physical attractiveness is simply one of many potential heuristics. Surveys tend to show that Facebook use is tied with higher social trust, however , there is no spe cific way of using Facebook that is the harbinger of that trust about 150,000 other users (Biddle, 2012) . That nu mber is based on the average number of F riends button to distribute information cross his or her network. On the high end, in a network dominated by an above average number potential to be seen by more than 7 million users not accounting for the concept of massive virality discussed in the introduction. Online Social Networks and Social Media boyd and ) research on online social netw orks helps set the stage for knowing the way people use Facebook. For example, most people tend to use online social networks not to find new friends and branch out, but to maintain their existing relationships from network profile that mimics their existence through using their real name, posting their real photo, and associating their profile with the things that they enjoy of Facebook tended to have upwards of 900 to 1, riends, but reported having
31 close, regular contact with only about 15 t o 20. This is supported by social network analysis research, in which the spread of network size has been noted, but researchers tend to find that people can still only remember specific details about the same amount of people as those in pre Facebook time s (Kadushin, 2012). The specific reasons why people choose to use Facebook can be difficult to fully categori ze, although most uses tend to involve the process of becoming a part of a group, maintaining fr i endships, entertaining themselves, and determining a sense of group norms and social identity with their peer groups (Pempek, Yermolayeva & Calvert, 2009; Quan Haase & Young, 2010; Cheung, Chiu & Lee, 2011; Nadkarni & Hofmann, 2012 ). Ryan and Xenos (2009) suggest ed that there are some potential negative r amifications to those activities, including elements of diagnosable narcissis Research examines the exten as boyd & Elliso n (2007) state. Zhao, Grasmuck and Martin (2008) suggest that this self presentation often comes in the form of visual representation through particular pages, creating an online identity through the things an individual consumes and purchases as opposed to narra ted ideas , such as stories and updates from everyday life . This gives credence to the assum ption followed by this study, one that will be explored later in the literature review, that social networks a lso expression than a utility or social obligation (Tufekci, 2008). The amount of interaction with predictive of ho w much someone would use a social networking site ( Tosun, 2007; Tufekci, 2008). Specific motivations for using Facebook as a way to represent real world persona online, aside from the generic maintenance of relationships, includes
32 games and entertain ment, posting photographs, and organizing social activities however, maintaining relationships remains the top priority of the majority of Facebook users (Tosun, 2007). Even something as seemingly simple as the words one chooses to use in a Facebook can act as powerful predictors of gender, levels of extroversi on, socioeconomic status , and age ( Schwartz, et. al., 2013). The next step is establishing that the concept of soc ial capital exists on Facebook a concept that will be important in test ing the validity of endorsement of information on Facebook. Vitak, Ellison & Steinfield (2011) suggest that there is a give and take component to the bridging of social capital on Facebook, namely that a given individual posting inf ormation in the form of bridge capital, but the bridge is not completed until source of the material. for users to ma intain vast F often contain people from close loved ones to casual acquaintances, social capital grows on a single uniform platform (Vitak, Ellison & Steinfield, 2011). For those without much social capital in the physical world, Fac ebook, and other online social networks, act as an entry point into increased social capital acting as an entry point into social situations (Burke, Kraut & Marlow, 2011). Network density also acts as yet another demonstration of the strength of weak ties that social capital growth is possible at all levels within the uniform platform. For the purposes of this study, this is important, as social capital can potentially act as a form of impact upon endorsement. Compared to other online social networks, the re is a high level of trust within the network itself (Acquisti & Gross, 2006). This may stem from the fact that most people who use Facebook, and any online social network, are using it to maintain relationships with their existing
33 friend network, not exp and and make new friends, so the social capital that exists in the physical world is transferred over to the digital (boyd & Ellison, 2006) . Research by Acquisti and Gross (2006) suggests that users of Facebook highly trust the network itself, their own fr iends on Facebook, and information coming from other Facebook users including friends of friends. Those findings support the concept of transferrable social capital and endorsement presented by both boyd (2007) and Buckner (1965). It is important t o note , however, that Acquisti and (2006) work was conducted before Facebook opened access to the website in 2007 to non college students, companies, and then the general public, as much of their conclusion is based on the idea that the heightened trust of Facebook comes from t he network being one designed and used by college students. , in 2014, both a pre social and post social activity, one deeply routed in a process of social capital and trust building as a form of friendship and intimacy (Lambert, 2013). Different people may give different responses as to connections often foll ow an in person contact . If the post contact process is followed, people are using Facebook as a form of further vetting the new connection as one that may be useful for growth of social capital, one with a potential romantic interest, or simply one with similar enough interests. The decision to meet the new connection again in person may be determined by what the viewer sees on the new connection Lambert (2013) argues that day to day behavior on Facebook is largely determined by the need to protect the image that could fall into the pre and post social procedure. Facebook users tend to protect their friendships happens. Lambert (2013) suggests that much like the real world dynamic of intimacy, the
34 i ntimacy formed th rough friendships on Facebook dies indeed come with a component of insecurity. Lambert (2013) work supports boyd (2007) in the idea that people tend to use online social networking sites to maintain existing friendships, however, the ar gument is made that the useful (Fischer, 1982) . Just like someone may denote different levels of in person friendships with special labels, such as acquaintance, co form of distinction when dealing with Facebook something that the site combines into a binary ctually four latent rien on Facebook when referencing a Facebook (pg. 163) are comfortable interacting both online and in person, and have a large overlap in interests and value systems. Lambert (2013) calls this the strongest level of friendship. The second tier of friendship is not as meaningful, in that there is less overlap in interests in values, there is less in person interaction, and less of a history leading up to the point of co nnecting on Facebook, yet a connection and sense of positivity about said connection still exists. Second tier Facebook often rely on Facebook as a primary mode of communic ation and are colloquially what people tend to think about when they think of the term F riend The third tier of does not actually warrant the use of the word friend in the colloquial sense but is still a connection on the site. Lambert (2013) describes th is level of (pg. 164) yet the two individuals rarely have any history or knowledge of each other. The second tier and third tier are the one s often described by Granovetter
35 considered a meaningful weak tie (Lambert, 2013). Connections on the fourth tier of are often the result of forgotten connections they are connections that started as second tier or third tier, but over time were demoted and neglected. The topic of rumors on online social networks has been explored before, but rarely using Facebook. Twitter has been the focus of resea rch on social media and rumor analysis, mostly due to its speed, use during current events, and its strong reliance on opinion leaders (Mendoza, Poblete & Castillo, 2010; Oh, Kwon & Rao, 2010; Qazvinian, Rosegren & Radey, 2010). Situngker (2011) suggests t hat Twitter is indeed a social network, whereas other researchers , such as Qazvinian, Rosegren & Radey, (2010) have suggested that Twitter is structurally more like a blog , or micro blog, than a true social networking site. However, Facebook and Twitter ar e strikingly different, both in how they look and how they operate, and this creates differences not only in how their networks function, but what kind of different satisfaction users get from the product (Raacke & Bonds Raacke, 2008; Johnson & Yang, 2009) . Work by Situngker (2011) suggests that on Twitter, the rumor can travel quickly across a number of networks due to the , rumor to his or her followers, a given percentage of the followers will foll ow along and Retweet the message again. This causes the stream of rumor to spread out at a steady pace until the rumor or someone with an extraordinary number of followers. That causes a burst of attention in the rumor, and ofte n causes the rumor to be seen by more of the general public (Stiungker, 2011). However, this burst can also tend to be the harbinger of the end of the rumor, as the massive a m out of atten tion it receives after being Ret weeted by a Twitter celebrity causes a higher chance of someone, or a number of people, correcting the rumor. Little research
36 has examined the small scale psychological heuristics that may be at play during the spreading of a rumor. Social science has tracked rumors across large scale network s, but the goal of this study is to examine the process of endorsement on the micro transaction, to see what may be impacting Th two working functions on Facebook that allow people to endorse and diffuse information, play a role in shaping the online social network experience. Research suggests that the items (i.e., photo s, articles) people choose act as active predictors of many different demographic characteristics, demonstrating a counter direction to the concept boyd (2007) created (Kosinski, the web, with non sociality and endorsement as something not confined to Facebook rather, it enco urages the user to endorse across the Web and allows Facebook to become the hub (Gerlitz & Helmond, 2013). Rumor Cascades Facebook has funded research to study rumors and information spread on their site. Dow, Adamic & Friggeri (2014) found that informat ion flow on Facebook tends to work in a pattern Cascade user in a viral manner. Cascades are largely fueled by endorsement tools on the site such as the through others who have previously
37 of the piece of information, at that point, is reaching the viral state, wher e the spread is called a information can cross from group to 2014) note that in order for a piece of information to enter into the cascade phase, it must appeal to some s ort of wide audience. The vast majority of content on Facebook will never enter into & Friggeri, 2014). Friggeri, et al. (2014) expanded on the idea of cascade s by examining how rumors Friggeri, et al. (2014) found that rumors have a propensity for entering into the cascade process due to their wide appeal and sense of evoked curiosity. Much li ke the general information cascades, cascades button. However, when looking specifically at rumors, Friggeri, et al. (2014) found that rumors can be impacted by how if corrected, tend to cease being a part of a cascade pattern. However, if a rumor reaches a level being corr ected will . Friggeri, et al. (2014) were careful to note they did not know the specific level of viral speed needed for a rumor to bypass the correction process. Health Communication in the Digital Age Fox (2011) found that i ndividua ls in the United States we re using the Internet, and within that, social media and online social networks, to find answers to health questions . According to data collected by the Pew Internet & American Life Project, 80% of Internet users have look ed
38 to websites for information on health topics, specifically diseases and treatment options, which Pew suggests is about 25% of all adults in the U.S. (Fox, 2011). From that, about 34% of Internet users, or 25% of adults in the U.S., have looked at news g roups, review sites, and blogs, for commentary or experience on a health related issue (Fox, 2011). The survey also found that about 18% of Internet users have looked for other people online who might have health concerns or health complications that are s imilar to their own. There is also data to suggest that individuals are specifically seeking online social networks for their information in increasing number (Fox, 2011) . Of the Pew sample, about 23% of online social network users say they follow their f riends personal health experiences or updates online, 17% say they have used online social networking sites to remember or memorialize other people who have suffered a health condition, and 15% say they have gotten health information from an online social network site (Fox, 2011). That latter number represents, according to Pew, abo ut 7% of all adults in the United States. Further research suggests that people turning online for health related information, particularly on social media type sites such as bl ogs, is not a small temporary trend (Fox & Duggan, 2013). Those t urning online tend to be people starting with search engines and finding general health related information sites, like WebMD. Less than 1% actually start their search for health information potential as a place for secondary information. For example, what happens if someone misinterprets a piece of information on WebMD, or perhaps picks up a piece of information from less repu table health website, and then posts that information to his or her own Facebook . Just as previous literature found also been posted by users on online social networking websites, with misleading information
39 included (Moreno et. al., 2009). Korda and Itani (2013) suggest that s ocial media defined as the content produced o n and through social networks may very well offer a fast and affordable format for health promotion and behavioral cha nge, but so far the full potential has not been utilized. Credibility Assessment The concept of credibility, and what it means for how people consume and endorse media, is pertinent to understand in this study, as well as any study dealing with misinfo rmation and the potential to endorse. In the classical equation of trust, credibility is the aspect most grounded in perception rather than reality ( McGinnies & Ward, 1980) between consumer and producer, and can be impacted by McGinnies and seen as both honest and with integrity. The section of this literature review concerning rumors and misinformation largely d ealt with the potential problems with susceptibility of individuals. This section, therefore, will deal with the idea of producer side credibility. Determining c redibility is a problem that is rooted in the intense rise of people using the Internet to find the information they are actively seeking out (Johnson & Kaye, 2000). The more people reported using the Internet to seek out info rmation, the more they tended not only to state that the Internet hosted information was credible, but that it was the most c redible format for information , higher than other traditional formats such as news p apers (Johnson & Kaye, 2000). What Johnson & Kaye (2000) propose, that time spent in the Internet environment is associated with evaluating information on the Internet as c redible, forms an interesting dilemma when paired with the statistics found by Fox (2011) that people of all demographics and characteristics are turning to the Internet for health information.
40 All other vari ables controlled for, Flanagin and Metzger ( 2000) found that individuals tended to rank information found online to be as credible as other platforms even if there was no clarification if the information was actually credible . In their study, t he same author, for example, was seen as equally credibl e in print as he or she was online. Heavy users of the Internet were not more likely to view the Internet as a non credible platform on which to find information, despite their exposure to more of the rampant hoaxes and widespread misinformation contained on the Internet. Actually, the opposite tended to be true heavy users tended to rank information from the Internet as equally credible to information on other platforms. The explanation for this is potentially that exposure to hoaxes on the web has made heavy users suspicious, and , therefore, they are more likely to cross check Internet based information with other Internet based information or with a physical format , such as a newspaper (Flanagin & Metzger, 2000). McGinnies & Ward (1980) suggest that, all things being equal with source message, Facebook users expertise and trustworthiness on assessment of credibility on delivered messages. The messages took the form of pamphlets a bout maritime borders in different countries, with the pamphlets being distributed in four different countries, with half being presented by persuasive individuals without any expertise in maritime law and half being presented by ordinary people acting as maritime law experts. The responses from the participants indicate those who received the message from the more persuasive speaking presenter were more likely to change their opinion on the subject, as well as see the presenter as more credible (McGinnies & Ward, 1980). The study challenged the idea that people tend to assess credibility from the trustworthiness of the
41 source, and that trustworthiness is determined by credentials and expertise. O n the contrary, the study found that credibility was based on . Eastin (2006) analyzed web based credibility of health messages, using a 2x3 experimental model to determine the impact of source expertise and content knowledge. Interestingly, Eastin (2006) found that neither expertise of source, with such variables as medical expert, professional expert , researcher, or unknown, nor the existing knowledge of the content statistically impacted the participants of the study. Instead, participants tended to rank the majority of material as credible with the final conclusion that users of online health content are not as selective when it comes to applying a base level of credibility to online content. Eysenbach & Koheler (2002) suggest that base level credibility exists on sites with prof essional similar type details for the author or site itself. Further, few participants in Eysenbach and bout the website they had just read. Together, s suggests that credibility on the Internet is a tricky phenomenon, with both suggesting that traditional heuristics for content credibility are seemingly not important to the active audience of Internet users. One explanation for this may be that Internet users, particularly younger Internet users, do not frequently seek out health information (Eysenbach, 2002). Health information is sought out when it is needed and the moment it is needed, but seeking out health information i s not a part of the regular routine or regimen the same way that email, news websites or social media may be. Because of this , users tend not to develop a sense of base level credibili ty, which inhibits the ability to cross check and internally compare information (Eysenback, 2002).
42 Kunst et. al., (2002), however, suggests that the design of the website itself may not be a very good indicator for the credibility of the information con tained within in looking at some of the top available websites for five different health conditions, while cross comparing for currency, source, and hierarchical structure of information, the results suggest that the actual accuracy of the information va ried wildly among those three variables. However, combining the findings of Kunst et. al. (2002) with Eysenbach & Koheler (2002), the implication is that people largely use general layout as the main assessment tool for credibility but that accuracy and ge neral layout are not connected. Their research indicates that people are potentially thinking a well designed site is much more accurate, despite not knowing at all where the information came from and with no evidence backing up the suggestion that the inf ormation is accurate. Younger people tend to have the same problems of widespread credibility on the Internet, however some level of media literacy concerning social media does help with a younger age bracket (Gray, et. al., 2004). Younger people tend to b e better at identifying something like a personal blog page, which anyone can make and host, versus an official website of a hospital or clinic, for example (Gray, et. al., 2004). The research suggesting that credibility assessment is wholly increased on line is not without its detractors though. Stavrositu & Sundar (2008) suggest that there is also a connection with credibility given to newspapers. In a study asking participants their level of assessed credibility toward newspapers, and then toward strict ly online information, Stavrositu & Sundar (2008) found that credibility toward the print product predicted the level of online credibility. The implication is that individuals may see online information more as a supplement to other more credible informat ion rather than a replacement for it, although the trend may very well be shifting as people begin using the Internet as their first source for information (Stavrositu &
43 and Internet news, assessing the cross platform c redibility of both. Stavrositu and work also used basic news and information websites as the platform for Internet messaging. With a growing percentage of people u sing Facebook, Twitter, Tumb lr, and other social media platforms to get their information, and with more people using it as their initial contact with information, the question remains as to what impact heavy social media use has on credibility especially considering that social me dia often has an element of implied endorsement. Using Social Network Analysis The purpose of this study was to investigate the impact of network structure, as well as information knowledge, on rumor spread on Facebook. In order to do so, the network st ructure itself must be examined as a dynamic component of regular decision making. An area of study called social network analysis therefore must be used. Social network analysis is, in its most simple form, the study of social connections and the impact t hey potentially have (Knoke, 2008). For example, one of the best applications o f social network analysi s was the Framingham Heart Study, which used the social network of a small town in Massachusetts to track health patterns. One social network analysis fo und that obesity epidemics had distinct patterns in social circles (Christakis & Fowler, 2007). Social network analysis is versatile and multi disciplinary, with applications both in the hard sciences and social sciences. Because of that multi disciplinary approach, theoretical foundations have been established over time. At the base level, the use of social network analysis in this study comes from the assumption that weak connections between individuals that form a less dense network of connectivity amo or its importance in the grand scheme of a
44 network the individuals on the outer perimeter of the network who bridge together multiple groups are the ones who have a great impact. Granovetter (1973) also argues that the strong impact of weakly looped ties is also present in ego centric networks. Granovetter (1973) looked at arguments about the impact, when viewing ego centric networks, of friends of friends upon the ego, concluding that although frien ds of friends form a very sparse and indirect field of influence for the ego, the variation of ideas compensates for the strength of infl uence. Granovetter (1973) stated: sector; such ties that they are the channels through which ideas, influences or information socially distant from xamined from the perspective of the ego, the same kind of lo o se looping occurs . Because of the nature of Facebook, an ego centric network will not have friends of friends appear with the study, as information post ed by friends of friends would not appear w Facebook user. However, as research such as Ellison (2008) has show cased, there are varying levels of connectedness and so cial capital when dealing with Facebook The concept of social capital is of concern when discussing social network analysis as well. Previous research in this literature review has discussed how social capital is present in online social networking sites, both through the mirroring of in person networks and the way the websites are built. Kad ushin (2012) discusses the importance of social capital in fully understanding social networks. Kaduskin (2012) argues that based on established social network analysis research, social capital is not limited to geographic distance. If that is indeed the c ase,
45 the social ne twork analysis portion of study that would be present in interpersonal relations. Further, Glanville and Bienenstock (2009) stated: apital may also imply the presence of community v alues. Kaduskin (2012) discussed social capital as a form of bond within a community, potentially as one of the defining foundations of what makes a community more than just a seemingly random group of people in the first place. Name generators are often used as a way to collect data on social capital across in person social networks. In a study using a name generator, the participant is asked to name people they are related to, are friends with, or would rely on if discussion was needed on an important matter (Kaduskin, 2012). The researcher then takes those names or initials, along with detailed characterist ics such as age, gender, education, etc., and creates an item called This dissertation research will treat Facebook, and the perception of Facebook therein, as an ego network. The participant will provide names in the f orm of allowing their Fa cebook connections to be collected from the site. If Kaduskin (2012) is correct, then the geographic differences should not apply as a limitation of social capital within the context of analyzing Facebook, and the presence of social capital should show itself as a form of trust. Gossip is another area where the impact of a piece of information traveling throughout a network has been examined using social network analysis and acts as a crossover from the area discussing trust and social capital. As gossip is diffused throughout a network, groups of loosely tethered clusters tend to have a harmful end result, with the breaking down of edges, whereas tight clusters tend to show the opposite end result small clusters of tight friend ship groups actually become tighter when faced with disseminated gossip (Malarz et. al., 2006; Lind et. al.,
46 2007; Shaw, Tsvetkova & Daneshvar, 2011). Most of that work was conducted using interpersonal communication, however, not an online social networki ng website. X iang, Nelville and Rogati (2010) propose that online social networks, especially Facebook, use rigid binary formulas that are simply not nuanced enough for proper social network analysis. On riend iend . Although Lambert (2014) states that website itself. That lack of nuance causes interpersonal best friends or significant others to be classified statistically the same as very weak connections, which leads to what Xiang, Nelville and Early usage of social network analysis to study online social networks and communities took the form of examining community building webs on earlier si tes, such as LiveJournal and MySpace. For example, Backstrom et. al., (2006) proposed that the decision to join an online network and become active within that network was determined both by how many friends were members of the network and how dense the ne twork was when quantifying connectedness. Gilbert & Karahalios (2009) suggest that tie strength can be further clarified and supported by asking supplemental questions, such as if the Facebook friend would be willing to write the individual a letter of rec ommendation, lend him or her $100, or if they would be upset if the In Summary T he established literature indicates the potential existence of psychological elements in the decision making process that come from the (1965) work into rumor theory drives that direction of the study, especially the claims of how networks, as opposed to straight lines, tend to yield less correct information when faced with a rumor. The literatu re on social network analysis shows that weak connections can have a great
47 statistical tools that will be helpful in breaking down the shap network namely modularity and a history of research into how rumors tend to spread. The study also was based on the assumption that people tend to make their online identity on an online social networking site based on their real g themselves into being (boyd & Ellison , 2006 , pg. 14 ) and that they tend to use online social networks to maintain existing friendships, not make new ones. Research also suggests that other social heuristics have been established already on Facebook, su ch as trusting someone with a physically attractive profile picture. Therefore, the study progresses with the assumption that people interact with Facebook as a method of self presentation, that they are putting themselves online, and when they view the pr ofiles of others, they are seeing the presented selves of that person. And , due to the already established existence of psychological influences in the decision making process , there could very well be other psychological influences in play. Also importan t take an uncritical stance toward a rumor if they do not have any base knowledge of the topic of the rumor. That intersects with the literature suggesting that more and more people a re turning to the Internet, including social networks, for health information. That leads to the need to study and examine the endorsement process on Facebook, how endorsement is perceived, and how placement within the network of those doing t he endorsemen t impacts decision making. Facebook is being analyzed as a platform for potential rumor spread due to the natur e of the site itself, which originally was designed to be used by undergraduates at Harvard. Although there have been software changes to Faceboo k since its launch in 2004, including adding in the
48 changed. Facebook is now being used as a global communications tool, something it was not in tended for. Th e use of Facebook as a too l for distributing news could be leading to rumor spread. Hypotheses Based on the established literature, this study examined a research question and two hypotheses. The research question is : R1 ) How does endorsement of misin dorse the information ? Breaking down the wider goal of the research question, two hypotheses were formed. The first deals with the concept of diversity of network clusters within Fa cebook itself. The stance leads to the first hypothesis, which will prog ress as: H1 ) Individuals will be more likely to endorse a piece of health misinformation on different network subgroups based on modularity . The second hypothesis deals with the health aspect of the established literature. According to the literature suggesting that people are increasingly using the Internet and various networks to find health information, combined with the work by Buckner (1965) suggesting that lack of knowl edge about the contents of misinformation tends to lead to the recipient acting non critically toward it, the second hypotheses is: H2 ) Individuals will be more like ly to endorse a piece of health misinformation on Facebook if the misinformation is concern ing a topic they kno w little about.
49 CHAPTER 3 METHODOLOGY This study was designed to test the impact that specific, individual influences would have a piece of information on Facebook. The study used two variables to analyze. The first variable already endorsed the information. The second variable was the content of the message that the participant received. In this 2X2 experiment format , participants were randomly assigned to one of four cells. The layou t of that 2X2 format can be see n in Figure 3 1. To test the first variable, half of the most populated social group on Facebook. The other half of the participants saw a Facebook post endorsed by three social group. To test the second variable participants were shown either a Facebook post about the common cold or a Facebook post about a non existent illness created for this study called i n the link in the post to get more information. The study was conducted as a two step experiment. Participants came to the lab twice with one week between the first and second step s . The first step was a session to collect data on These data were used in the second session so participants saw his or her During the week between the first session and the second session , the researcher designed custom questionnaires for each participant based on their own Facebook data. When the participants returned for the second session, they were shown a Facebook post. The post was an image designed to look like a Facebook post, but was actually
50 presented in Qualitrics. ( See Appendix C ) . Depending on w hich cell the participant was assigned , he or she the experiment, that list was developed by the researcher using Facebook data obtained during the first session. Participants in the second session were instructed to make choices about the post based on the idea that their six friends actually endorsed it. In the second session, participants w ere asked to indicate how likely they would be to saw. Each of the three options was presented on a 1 to 10 scale with 1 representing would not at participants were asked to assign a number, 1 to 10, to indicate their level of trust in each of the During the first s ession, participants , all of whom were undergraduate college students, were asked a series of demographic questions including age, major, and gender. Then they were asked about their use of Facebook, including how many days in a week they access Facebook, and how many times in a day they access Facebook for more than 20 minutes at a time. by a series of questions asking how much, on a 10 point scale, they trust the c ontent they see on their level of health knowledge . Finally, participants were asked how much they used the they returned for the second ses sion, participants were asked to use a 10 point scale to show how
51 10 point scale. Finally, th ey were asked to indicate how were shown endorsing the information. A full list of questions and the forms used in both sessions of the experiment can be found in Appendix A, Appendix B, and Appendix C. A total of 246 p otential participants were initially sent an email asking for their participation in the study, with explanation of how the procedure would work and asking for them to contact the principle investigator. Emails were sent to students in a college of commun ications who were taking at least one of four undergraduate courses: Principles of Advertising, TV & American Soc iety, Video Games & Culture, or Social Media & Society. The targeted courses were selected for their open access, which would provide a diverse population of majors and other background characteristics. No journalism classes were specifically targeted, as there was a c oncern journalism students might have approached the study materials in an overly skeptical manner du e to their classroom training . The researcher had contacted the four course instructors to secure their participation, with part of the incentive for student participation being that the students who participated would receive a small amount of extra credit. A total of 162 students m ade initial contact with the principle investigator , offering to volunteer to participate in the study. Those 162 potential participants were then sent a calendar using Doodle.com with options of three days in March 2014 during which they could come in to participate in the first session . Each of those three days had three time options 9 a.m, 2 p.m., and 7 p.m. Of those 162 students , 113 initially signed up for a time to come in and complete the first portion of the study. Of those 113, 69 arrived at thei r scheduled time. Of those, 52 returned to complete the second portion of the experiment. The experiment was repeated in early April 2014, with those from the original potential pool of 162 who did not sign up for a time, or who signed up and did not arriv e for their scheduled time, receiving an email to sign up for another
52 time. From that second attempt, 25 scheduled times to come in, 23 arrived to complete the first portion, and 18 returned for the second portion. The second group of participants came in at identical time sets as the first group . Each session was limi ted to 20 participants , as there were 20 computers in the laboratory room. Three sessions were scheduled on three testing days Monday, Wednesday and Friday. The number of participants range d from three to 18. The first session lasted about 25 minutes and was guided by the researcher using a PowerPoint presentation. At the end of the first session, the participant was asked to return in one week at the same time and day to the lab room for th e second session. During the second session, the participant was given an individualized URL to their custom questionnaire and asked to complete the questions without going to other places on the Internet or asking fellow participants any questions. The se cond session lasted about 5 extra credit could be awarded. In creating t he individualized questionnaire for each participant, the researcher did not randomly session, participants were shown a list of Friends and their corresponding modules asked to provide the researcher with a list of 10 names from one module or two modules, depending on which cell he or she were in . The selection was done at the end of the first session. P articipants were handed a form (See Appendix B) and asked to follow a set of verbal instructions from the researcher. Participants were shown the s creen in Figure 3 3, except based on their own network. The numbers on the left of Figure 3 3 were randomly assigned. The colo rs in the middle of Figure 3 3 were randomly assigned. The percentages on the right side of Figure 3 3 represent how much size, in percentages, that module takes up within its network. Using Figure 3 3 as an
53 biggest by percentage, and half of the participants were asked to li 2 with whom they interact with in a regular basis, or they rely on for inform the questionnaire in the second session. During the first session, as participants were listing the names for later use in the study, include dentistry, medicine, or pharm later analyses. In soc ial network analysis, modularity analysis looks at patterns in connections among nodes and performs groupings of the nodes based on com mon connections (Newman, 2006). Essentia know each other, and then putting them into smaller groups based on those common connections. The grouping action is calculated based on a particular mathematical alg orithm performed by the software. It is important to note that the calculated mod ularity is not based on the internal network Facebook uses to connect two users together, such as employment, geographical area, education, etc. Although there is often overla p between those two schemes based on the nature of networked connections, it not precise enough to use for an academic analysis, as one person may share both a geographical
54 area and emplo yment networks with the ego, while one person can only be in one calculated modularity group. T wo different software programs were used to aid in the analysis. The first was a Facebook application called Netvizz, which was developed by Bernhard Rieder, an associate professor of media studies at the University of Amsterdam. Netvizz was designed for academic purposes, and it did not save any of the data it collected in any other place in order to protect the privacy of participants. The Netvizz software coll ected the names of all people with whom the sense. After creating a d ata file in Netvizz, the participants opened the data file in Gephi, another free social network analysis software. Gephi uploaded their Facebook network into a layout that could be easily seen by the researcher and analyzed. The basic appearance of a Face book network in Gephi can be seen in Figure 3 instructions by the researcher, during which they separated their n etwork based on modularity. When modularity is applied, and the nodes are pushed away from each other in a process called 4 shows what a Facebook network looks like after modularity has been calculated a nd the nodes pushed apart. Note that the network seen in Figure 3 4 is the gray network from Figure 3 2, but with the modularity listing from Figure 3 3 applied. As seen in Figure 3 4, the light blue and drab green modules represent the two biggest modules by population. The Facebook posts the participants saw during the second session were created specifically for this study using Photoshop. The two posts can be seen in Figure 3 5 and Figure
55 3 6. For the user profile area, a blank silhouette was chosen to avoid creating any confounding issues. The area in which the names would ordinarily be was left blank with a small box Below that was a generic photo of a pil e of medication. The image of pills was created specifically for the purposes of the study and was the same for both templates. It also was fitted headline reading either hich this information , supposedly, was coming . A fictional website was used, as using an existing m edical information website, such as WebMD.com, could potentially introduce unwanted heuristics such as appeal to authority. Instead, the fictional he description under the , read either : Check out what you can do to spot the symptoms, as well as these simple tricks you can use to s looking like a bad season for Mertenitis on college campuses this semester. Check out what you can do to spot the symptoms, as well as these simple tricks you to its lack of similarity in name to other illnesses . By using an illness that does not exist meant it would be impossible for participants to ha ve any existing knowledge of that illness coming into the study. It is important to note that during the week in which the second session of the first data collection of the study was set to begin, Facebook underwent a significant layout change. That
56 the template used by t he study was technically outdated, however , the design change was so recent, and all the users in the study established as Facebook users before the layout change, it was assumed that the participants in the study were familiar with the older layout. The study was granted approval by the Institutional Review Board at the university where the study was conducted. Following the initial IRB submission, the IRB requested additional information about the nature of data collection namely that participants were b eing asked to any Facebook user who had previously set his or her Facebook to private would not appear in the agreement, the IRB approved the study. Also, precedent had been established in the field of social network analysis that people do not have the right not to be named in a study as a standard protection of privacy, providing those named are not contacted. The specific instructions on privacy set by the IRB were followed throughout the study, and participants were asked to sign Informed Consent paperwork before beginning any work on the study.
57 1 One Module Common Cold 2 One module Mertenit is 3 Two modules Common Cold 4 Two module Mertenitis Figure 3 1. Layout of cells indicating which cell received which variable in the experimental portion of the study
58 Figure 3 2. Example of an average Facebook ego centric network when randoml y Example taken from sample of participants.
59 Figure 3 3. Example of a typical Facebook network once it has been modulated. The c olors and numbers are randomly generated by Gephi. The percentage on the left indicates the size of that module within the network. Example taken from sample of participants.
60 Figure 3 4. What a typical Facebook network looks like once modulation and separation has been applied. Example taken from sample of participants.
61 Figure 3 5. Fake Facebook post created for the second portion of the experiment. This link used the common cold as the health message and was seen by half the participants.
62 Figure 3 6. Second fake Facebook link created for use in the second portion of the seen by half the participants.
63 CHAPTER 4 RESULTS This research was based on conducting a two stage, 2X2 experiment with college students who were Facebook users. The purpose of the experiment was to determine if a the message it self and by the pattern of endorsement of the message The first step in breaking down the data was to determine the demographics and characteristics of the participants. As the participants all came from four different undergraduate classes at a large public university, the age range was representative of the traditional undergraduate age range of 18 to 22. Of the 70 participants, 97% were in that age range, with only 4 p articipants who were age 23 or older. The demographics breakdown can be seen in Table 4 1. The participants indeed were youn ger than the average in the United States , with the plurality of 20 and about 50% of the participants falling between ages 19 and 2 1. The sample was predominately women (76%) with 52 of the 70 total participants being women. Past analyses into Facebook research does note that studies of Facebook tend to skew toward women (Ross, et. al. 2009). However, data collected by the Pew Researc h Internet Project (2014) indicates that women were more present than men on social networking websites, they are ere about 58% women to 42% men on average (Smith, 2014). Those same demographic statistics for Facebook in 2014 point to about 34% of all Facebook accounts belong to someone between the ages of 18 and 25. Also, the department
64 from which the participants we re sampled is about 75% women, which indicates the demographic makeup was accurate (McFarlin, 2013). As two of the four class participants were recruited from the telecommunication department , 26 students were telecommunication majors, 10 were public relat ions, 7 were advertising, 4 were media & society, and only 2 were journalism. There was concern at the initial planning of the study that journalism students might be more prone to be skeptical toward information on Facebook. Many journalism courses emphas is the importance of being overly critical of online information. The overall question asking participants to state their perceived level of medical knowledge was on a scale of 1 to 10, with 10 representing no medical knowledge and 1 equating high levels of medical knowledge. The inversion for the question was emphasis verbally by the researcher as to ensure participants were paying close attention. In the scale, 5 indicated the participant had an average amount of medical knowledge. In the written directi ons, participants were asked to pay careful attention to the scale. The average score for medical knowledge was 5.35 with a median of 5 and a mode of 3. These numbers indicate that in general, participants believed their existing level of medical knowledge fell in the middle of the scale, with the mode actually indicating that participants believed they were closer to having above average medical knowledge. The next step was establishing how much the participants perceive they used the Facebook endorsement stimuli they were the result of the initial questioning of endorsement to set a level of how often ed count plurality of
65 count plurality for the , participants were more likely to the content they saw . Those results can be seen in Table 4 2. The next set of questions dealt with how much participants state they trusted the people h, and how much they state they trust the content that tho se can be seen in Figure 4 1. P articipants tended to state they t hey see on Facebook in the form of posted links. The average for trusting their Facebook a 7 on a 10 point scale, with 10 representing an extremely high level of trust. Using those numbers in analyzing the data from the study , the numbers sup port research from the literature review that indicated people do indeed seek bonds in the physical world with people with whom they share some connection on Facebook (Labert, 2014). The average scores for an 5, indicating that reported an overall above average level of trust on Facebook. The final set of data from the initial session with participants was a series of questions concerning their stated usage of Facebook in an average day and week and some de tails about that usage. The plurality, about 34% (n=70), stated they used Facebook more than 20 minutes at a time during all seven days of the week. Only three participants stated they did not use Facebook more than 20 minutes at a time on an average day. A majority of participants stated they check Facebook once every few hours in an average day. The results of the question asking for the amount checked in a day tended to fall in to a normal distribution curve. Nineteen of the 68 participants stated they ha Having more than 1,000
66 y tended to be average to more than average users of Facebook. Based on available statistics from 2014). That brings up a question that will be addressed later, which is the potential of the growth side data suggests that Facebook users average about 400 minutes of on site time per mont h, which works out to about 13 or 14 minutes per day (McGee, 2014). Thus, the participants stating they spend more than 20 minutes at a time on Facebook every day suggests that they are in the demographic of 18 to 25 year olds who are an be seen in Table 4 3. Before analyzing the variables of endorsement post stimuli, first the cells themselves were collapsed to account for both the health message and module diversity variables. The cells were labeled with numbers as demonstrated earli er in the Methodology chapter. That can be seen in the Figure 3 1. Then those collapsed cells were broken into two different newly created variables in SPSS, with one representing just th e diversity of modules variable and the other representing just the d ifferent health message module, so the two variables could be analyzed independently. A series of one way ANOVAs were then conducted comparing the means from the collapsed cells, diversity of modules, and health message with their response for the question asking if participants in the study were also asked about clicking the link to find out more information. The means were then analyzed in an ANOVA. The results can be seen in Table 4 5. None
67 .843, p = .475). Looking at the analysis of variance in grouped means, the stimuli had no recognizably significant impact on the participants reported likelihood of endorsing the information . With the collapsed cells analyzed, the next step was to look at the two variables independently . The colla psed cells included the potential influence of both variables. The first stating: Individuals will be more likely to endorse a piece of health misinformation on Facebo ok from different network modules. An ANOVA of the means from the diversity of modules variable, without the he alth message variable included, was not statistically significant, indicating that the means were not (MS = 4.765, F = 678, p = .413). The numbers indicate th did not have a statistically significant impact. The next variable was the message itself, with the hypothesis stating: Individuals will be more like ly to endorse a piece of health from misinformation on Facebook if the misinformation is concerning a topic they know little about. The ANOVA was repeated with just the health message variable, excluding any influence of the diversity of modules v ariables. Again, there were no statistical significance resul ts The sets of ANOVAs specific to the two variables indicate that the null forms of H1 and H2 shoul d be accepted. The numbers indicate the content of the message itself did not have a statistically significant impact.
68 With the two hypotheses tested at the ANOVA level, the next step was to examine the means of the collapsed cells and ungrouped variables to see if any significant trends existed. Those findings can be seen in Table 4 6. Participants consistently stated they were more likely to to regardless of individual variable, with the exception of the group who were in the single module and common cold cell, who reported an equal likelihood to of 3.63. Those results followed the pre stimuli responses, which indicate that regular basis. In all of the conditions of the experiment , participants stated they were more likely to click on the link to find out more information than either of the two endorsement tools. The high number for likelihood to click potenti ally points to the participants wishing to receive more information before choosing to endorse the information themselves. Overall , none of the means scale, indica The results from Table 4 6 indicate that, when only looking at the diversity of modules were themselves on the piece of information. However, it was only a .24 difference, and as previously indicated, was not enough to indicate statistical significance. The diffe rence in sharing between the two was only .35, again not enough for significance. For the illness variable, the means were actually counter to the hypotheses those who received the created illness actually stated they the common cold.
69 Ther e was a lower set of means for C ell 3 compared to the other cells. Cell 3 was for about t below the next other cells at 2.24. Although not an endorsement tool, participants were also le ss likely to choose to click on the link in Cell 3, with a mean of 4.53. As a part of the experiment, participants were asked to state how much they knew about the illness referenced in a Facebook post they had seen , with half of the participants seeing t he Facebook post for the common cold and half seeing it for Mertenitis, the non existent illness. When the means were compared using a one way ANOVA, the means indicate that the collapsed cell itself had a statistically significant impact o n how much parti cipants stated they had knowledge of the given illness (MS = 136.4, F = 50, p = .000) . Those who received a Facebook post about the common cold tended to say they knew a lot about the common cold, whereas those who received a Facebook post about Mertenitis tended to say they knew little or nothing about Mertenitis. The difference between stated knowledge was one of the strongest statistically significant results in the study. This indicates that the participants did indeed cognitively differentiate between the common cold and Mertenitis, as the results show less than 3% of participants who received the Mertinitis post stating they actually knew about the nonexistent illness. Important to the findings of the study was the ok. At the beginning of the first session, participants indicated how often they button, and at the end of the study, they were asked how likely they would be to click the linear regr ession was performed using
70 from both portions of the experiment, and the results indicated a strong connection and statistical significance at the p<.05 level ( = .489, t = 4.555, p = .000). It is important to note that beta is negative in the linear regression due to the language in during an average session on Facebook, The coding system designated Because of that, the negative beta indicates the connection is between people who gave low numbers on the first question a nd high numbers on the second with the understanding that both indicate a higher e second session, causing seemingly negative regression scores. Finally, a bivariate correlation was completed to support the linear regression. The correlation test indicated the same thing a statistically significant connection betwee button often in general and likelihood to .489, p = .000). n later, the same barrage of tests was and how likely they were to A linear regression was completed using the same two variables, and again, they w ere found to be statistically significant ( = .343, t = 2.968, p = .004). Finally, the bivariate correlation confirmed the significance (Pearson correlation = .343, p = .000).
71 whether the participant saw the message about the common cold or Mertenitis. The Facebook n a lot , regardless of message . The same set of tests were conducted using how acted a s a predictor. session of the experiment were medical professionals. That was calculated in SPSS as a simple binary system of 0 indicating not a medical profession al and 1 indicating a medical professional. Then, each participant was given a score for how medically professional Ultimately, that number was compared to a number of variables, including the likelihood to uli, however nothing statistically significant was found.
72 Table 4 1. Frequency of demographics and characteristics for participants Gender Female 52 Male 16 Age 18 5 19 13 20 23 21 16 22 7 23 1 24 1 25 or older 2 Major Telecommunications 22 Public Relations 10 Advertising 7 Media & Society 4 Journalism 2 Health knowledge (On 10 point scale) Mean 5.35 Median 5 Mode 3 St.d 2.10
73 Table 4 2. Frequencies for reported u Facebook usage Very often Almost every link I see 5 0 Often Multiple times while on one Facebook session 43 6 Occasionally Once per day 13 8 Not very often Once per week, if that 7 45 Never I do not use the button 0 9 Total: 68 68
74 Table 4 3. Frequency of questions determining reported Facebook use "How many days in an average week do you use Facebook more than 20 minutes at a time? Frequency None 3 1 2 days 10 3 4 days 20 5 6 days 11 7 days 24 Frequency Once in the day 1 Three times in the day 12 Once ev ery few hours 44 Once per hour or more 10 the average day. 1 Frequency 1 to 100 0 101 to 200 1 201 to 300 3 301 to 400 8 401 to 500 8 501 to 600 8 601 to 700 5 701 to 800 6 801 to 900 6 901 to 1,000 4 More than 1,000 19
75 Table 4 4 . One way ANOVAs comparing collapsed cell, illness , and diversity of network analysis with their connecting Like, Share, and Click variables SS df MS F p Collapsed cells Like Between 12.747 3 4.249 .644 .590 Within 422.194 64 6.597 Total 434.941 67 Share Between 21.600 3 7.200 1.147 .337 Within 401.871 64 6.279 Total 423.471 67 Click Betwe en 17.815 3 5.938 .843 .475 Within 450.655 64 7.041 Total 468.471 67 Diversity of network modules Like Between .941 1 .941 .143 .706 Within 434.000 66 6.576 Total 434.941 67 Share Between 2.118 1 2.118 .332 .567 Within 421.353 66 6.384 Total 423.471 67 Click Between 4.765 1 4.765 .678 .413 Within 463.706 66 7.026 Total 468.471 67 Illness seen in the description Like Between 5.346 1 5.346 .821 .368 Within 429.595 66 6.509 Total 434.941 67 Share Between 2.058 1 2.058 .322 .572 Within 421.413 66 6.385 Total 423.471 67 Click Between 7.906 1 7.906 1.133 .291 Within 460.565 66 6.978 Total 468. 471 67
76 Table 4 6 . One way ANOVAs comparing collapsed cell with statement of knowledge of the illness seen in the Facebook post SS df MS F p Collapsed cells Between 409.401 3 136.467 50.077 .000 Within 174.408 64 2.725 Total 583.809 67 Table 4 5 . Means of collapsed cells, diversity of modules, and illness seen in the Facebook post with the three dependent variables: Like, Share, Click Like Share Click Collapsed cell Single Module/Cold 3.63 3.63 5.63 Single Mo dule/Mertenitis 3.56 2.94 5.72 Two Modules/Cold 2.76 2.24 4.53 Two Modules/Mertenitis 3.94 3.59 5.76 Diversity Single Module 3.59 3.26 5.68 Two Modules 3.35 2.91 5.15 Illness Common Cold 3.18 2.91 5.06 Mertenitis 3.74 3.26 5.74
77 Table 4 7 . A linear regression with statement of how often the participant uses the le Variable B SE(B) t p 1.607 .353 .489 4.555 .000 Table 4 8 . A bivariate correlation table between the statement of how often the the Facebook post post stimul i How often the participant stated they used the Share button How likely the participant was to the Facebook post How often Pearson Correlation 1 .489 P .000 N 68 68 Share the post Pearson Correlation .489 1 P .000 N 68 68
78 Table 4 9 . A bivariate correlation table between the statement of how often the the Facebook post post stimuli How often the participant stated they used th e Share button How likely the participant was to Facebook post How often Pearson Correlation 1 .343 P .004 N 68 68 Pearson Correlation .343 1 P .004 N 68 68 Table 4 10 . Descriptives of aver N Range Min Max Mean St.D Average of trust 68 7.67 2.33 10 7.348 1.575 Table 4 11 . A linear regression with statement of how often the participant uses the ook as independent factor and the likelihood they Variable B SE(B) t p 1.143 .385 .343 2.968 .004
79 Figure 4 1. Bar chart comparing how often participan Facebook compared to trusting the content posted on Facebook. 0 5 10 15 20 25 1 2 3 4 5 6 7 8 9 10 Frequency of Selection Scale of Trust 1 = Low trust; 10 = High trust Trust Content on Facebook Trust People on Facebook
80 CHAPTER 5 CONCLUSIONS & DISCUSSION This study was designed to examine the spread of health based misinformation on Facebook, and it did so by ta king two different variables into account. Although many other studies have examined the spread of misinformation on Facebook within a larger frame, such as tracking a piece of misinformation from its origin to see how many times it gets posted, little had been done examining the variables decision to choose to endorse a piece of misinformation. This question is important because widespread cascades of information are perpetuated by groups of indi vidual actors. Therefore the study progressed with the initial research question: R1) dorse the information ? Diversity of Endorsement The study approache d that research question from two directions. The first hypotheses of the study dealt with the concept of modulation of the network itself. Based on literature concerning the tenants of rumor theory combined with fundamental elements of conformity, the fir st hypotheses was: H1) Individuals will be more likely to endorse a piece of health misinformation on different network modules. The data collected by this study suggests th at the diversification of modules did not have . That was determined by first running ANOVAs, then post hoc tests from those ANOVAs, then regression analyses, and finally corr elation tests. No statistical test found any significance at the p<.05
81 threshold. Those three statistical tests were first performed on each individual cell of the study, which included the other tested variable. They were also performed individually on ju st the variable itself, where the means of those with one module and those with different modules were compared. The lack of significant statistical analysis was found for both of the endorsement tools It did not make a difference to th e participant, when choosing to endorse the health message, whether a piece of health information was riend s one network . Health related Messages The second hypotheses dealt with the health me ssage itself, with half the participants receiving a message concerning the common cold and half receiving a message concerning what was presented as an actual illness but was, in fact, an invented illness. Based on the basic components of rumor theory, th e participant should have been more likely to endorse the message for the invented illness, as they would have been more likely to take an uncritical stance due to a lack of existing knowledge with which to compare. Therefore, the second hypothesis was: H 2) Individuals will be more like to endorse a piece of health misinformation on Facebook if the misinformation is concerning a topic they know little about. The testing for H2 yielded no statistical significance using ANOVA with post hoc, regression, and correlation. Individuals were not more likely to endorse a piece of health information based on a topic they know little or nothing about. The groups analyzed were the same as well. Each individual cell was analyzed for means, as well as the specific divi sions only pertinent to the health message endorsement. Individuals were not more likely to endorse a health message whether they knew about the topic or not
82 based on whether they receive one module of their Facebook network. The low rates of endorsement, on average, mean that health messages may not be the most risky form of rumor on Facebook , however at the time, based on the data, that is unclear . Also, very few participants claimed the y knew anything about the invented illness when asked, which potentially shows that on average, most of the participants were unwilling to lie about their stated knowledge of something that did not exist. Testing the unwillingness to lie could also be pres ent if the methodology was repeated using a sensitive subject. The sensitive subject could be something health related, like STDs or sexual health, or it could be something political, like issues involving racism. Another potential element in play here is the common cold. Power Endorsers An import a nt finding in this study is the impact o on the spread of misinformation on Facebook. The study showed that the more a participant stated that to select both in the study , regardless if the post was for the common cold or the fake illness. This was confirmed using ANOVA at p<.05 (p=.000, F=4.733), linear regression (p=.000, t=4.555), and bi variate correlation (p=.000, Pearson correlation .489). The statistical results suggest a strong significance, especially the Pearson correlation, which indicates a rather high level of correlation. This study found significant results suggesting that tho se who state they often use the lth based
83 information while taking an uncritical stance towards the factualness of the information. The results suggest those who are more l , the Fa ng status updates, posting photos, commenting on statuses and photos, and posting links. Hampton et al. (2012) more pictures than average. Importantly, the Hampton et al. (2012) study found that it is more common for an average Face piece of content themselves. The results of this study suggest that there may be a group of endo rse at increased rates may perhaps be endorsing at that rate without concern for the truth of feedback of continued updates one gets from endorsing items than b eing seen as factually correct. The work by Fribberi et. al. (2014) points to that being the case. In their work on they had been corrected as being seen false. There be the cause of what Fribberi et. al. (2014) found , that false rumors persist on Facebook despite
84 evidence being posted to the contrary. en if the message is false. constitute a special group of Hampton et al. (2012) found, questions arise as to the potential impact on the everyday user when the action on the site bu ilt to allow one user to tell other user s that a piece of information is valid, trustworthy, or of high quality, is diluted to the point it no longer carries those associations. It also could lead to a number of real world implications apart from the theoretical implications. Ed itorial Control In 2014, Facebook, which had about 1 billion accounts and was the second most accessed website on Earth , behind Google, did not have any editorial control over the factualness of information presented on the site. Non factual content was no t deleted or edited in any way, even if it led to Service, the administrators of the site will only delete content from the site if it contains something illegal in the U.S ., nudity or pornography, or is a thre at toward another Facebook user (Facebook Terms of Service, 2014). Facebook should consider enacting some form of editorial However, Facebook en acting a system where paid employees filter through all of the posted contend on the site seems like an unlikely solution due to the time and money needed. Just based on the number of users alone, aside from the sheer amount of content posted on a regular basis, it would be infeasible and near ly impossible for Facebook to control messages on their site at the administrative level. public relations, legal, and ethi cal difficulties. Facebook most likely would struggle with issues of freedom of speech if Facebook paid moderators were the ones in charge of filtering content,
85 F acebook would accept the legal responsibilities of those posting s ; and there would be a huge a mount of ethical considerations to consider when determining which content should not be allowed to be posted. One solution to dealing with the potential of false information on Facebook could be crowd sourcing. News websites have taken this approach, allo wing commenters on web stories to flag other comments that do not properly contrib ute or contain intentionally misleading content (Rollins, 2013). editors then review the comment and make the decision whether the comment is to be removed and in some cases the person who made the inappropriate comment is even banned from the site (Rollins, 2013). (Thoma s, et al., 2011). However, Twitter has not expanded this into specifically asking users to harassing content. There is the potential for Facebook to build in a flagging system, wherein users who spot intentionally misleading content can flag it as such. The flagging system could then be overseen by administrators of the site, putting much less workload on any single administrator and removing some of the legal implicatio ns of Facebook removing content . Such an approach would potentially give a sense of empowerment to Facebook users, that they were sourced system would have m has been abused by marketing firms, those with differing political opinions, and as a tool of cyber bullying by forming groups to flag content and users enough times as to silence them (Thomas, et al., 2011). There is no perfect solution to editing a mas sive online social networking website like Facebook, but allowing for a system of
86 crowd sourcing could be a step in the right direction. Just as the content on Facebook has changed over time from the comments of a handful of Harvard students to literally m illions of pieces of information on politics, health, and personal updates, the method Facebook uses for determining the content it does not want on the site should change as well. The use of the web for health information is on the rise ( Fox, 2012). The increase of reliance on the Internet for health information is not just the Internet itself, but specifically the use of social networking websites and social media. Health information in this capacity, on Facebook, is not limited to just finding out abou t new information it is about receiving crowdsourcing opinion. If one considers the number of people seeking health information on online social networks in the context of this study, some troubling implications form. What if a dangerous health related false rumor was to begin cascading on Facebook? The data collected by this study suggests that a health related rumor the rumor potentially for propagating a dangerous health rumor that even others attempting to correct the rumor cannot fix. Dangerous health rumors could take many forms, from a panic inducing warning based hoax to a positively packaged false rumor about homemade, yet useless, remedies to cure a dangerous illness . being which would also help prevent rumors from making the jump from the digital world to the physic al.
87 Organizations in the United States with a vested interest in health literacy and communication should consider creating an easily accessible website to vet rumors seen on Facebook and other large online social networks. The Centers for Disease Control and Prevention, the Mayo Clinic, and the National Institute of Health all have an online presence of some kind with available information. All have a website, yet none have a large presence on Facebook. These organizations should consider a more interacti ve and proactive step that could Cascade, Organizations such as FactCheck.org and PolitiFact.com have been operation for years by a llowing readers to submit claims, and they Perhaps the CDC should co nsider having a site that acts as a submission based editorial process, where users submit links they have found on Facebook, and employees of the CDC can comment on their truth or falsity. Those comments could be made in a way to promote easy linking to F a stopping mechanism, the CDC perhaps carries more authority on health related matters. Such a system, which would encourage user feedback and participation, could a lso help contribute to a culture on Facebook where people are more likely to question what they read. Facebook, as a company, has recognized the potential effect of widespread rumors on their website. The company conducted research that studied the cascade of information across the site, including analyzing what spread, the anatomy of a cascade of information itself, and an attempt to predict the direction of cascades. Such research shows that the company is interested in the impact that those phenomena
88 are having on their site. Online social networking sites are used by individuals as extensions of the m selves online (boyd & Ellison, 2006). Online social networking websites are outlets for vast int erpersonal connections and networking, where everything from friendship to self esteem can be fostered (Ellison, Stienfeld & Lampe, 2010). If online social networks become plagued with uncontrolled rumors with dangerously false content, these networks run the risk of backlash from their own users, who may either seek out another online social network or reduce the amount of time they spend on the site. Facebook users may also become much less willing to participate in the activities needed to keep online so cial network sites alive if there is a concern about inter Rumor Theory rumor theory helped e stablish the understanding that individuals tend to take either an uncritical stance or critical stance when faced with a rumor (Buckner, 1965). A critical stance means that the receiver of the rumor is more likely to look deeper into the rumor to determin e if it is true or false, whereas an uncritical stance means the receiver is more likely to simply pass on a rumor. re are a number of different overlapping variables that can increase the likelihood someone takes an uncritical stance, this study used two specific variables. The variable used to create the differences in the health message was the idea that someone is m ore likely to take an uncritical stance if they have nothing on which to base the truthfulness of the rumor. The results of the study did not find this effect. Participants were not more likely to pass along the rumor about the fake illness. Perhaps the re ason for that can be explained by another facet of rumor theory, which states that when faced with a rumor, an individual is much more likely to take an uncritical stance if the rumor concerns something
89 dangerous or if there is a context of heightened tens ion. In this study, the fake illness, Mertenitis, used the symptoms of the common cold, which are not generally related to illnesses that cause feelings of heightened tension. Also, during the time in which this study was being conducted, there was no heal th concern making headlines in the news. If, perhaps, the data was being collected during a particularly bad influenza season, or during a time when the influenza season was being covered in the news, end up having a greater impact, as the two heuristics could be impacting the participant at the same time. based on the idea in rumor theory that rumors tend to sp read faster in a network scenario versus a straight line. An individual is therefore more likely to take an uncritical stance toward a rumor if the rumor is present in a network where unconnected people are also spreading the rumor. Although Buckner (1965) formulates this finding on interpersonal relations, this study overlaid at about the same percentage. Facebook is very clearly a network, so it is interesting that the endorsement diversity variable had no impact. More work should be done examining if there is a threshold for network variability in the endorsement process. Tha t could be done by further diversifying the modules the endorsement comes from, so that the control for number of modules is less important than the maximum diversity. Social Capital At a deeper level, this study used known components of social capital to help establish the idea that users of Facebook participate in a process wherein certain social benefits can occur based on actions performed on the site. The hypotheses of this study attempted to predict that Facebook users would choose to endorse a piece of health based information based on that
90 information being endorsed by others. As such, this study approached endorsement as a form of social capital. Lambert (2014), for example, argued that social capit al not only exists on Facebook but is the core fou same kinds of social capital that in person friendships have due to Facebook profiles acting as (Vitak, Ellison & Steinfield, 2011). The s a problem for social capital as it is understood on Facebook. Vitak, Ellison & Steinfi eld (2011) y selecting the been seen and deemed important. Deeper meaning can be found the base level it indicates that the content was s If the buttons are overused, there is the potential for the buttons to lose their understood capacity to ac t as endorsement tools, and therefore a quick and easy way to contribute for social capital. Ease of use and lack of consequence play into the equation of figuring out action as social capital as one popular example goes, in deciding if one will loan a nei ghbor a cup of sugar so they can complete a cake, one can think about the fact they have plenty of sugar, and loaning one cup will not hurt their ability to cook. Some of the more popular forms of social capital in society are ones that are not taxing towa rd the one performing the act, or at least not taxing enough that act as social capital tools, Facebook would be left without a quick endorsement tool. Th e only form of endorsement left on the site would be to comment, explicitly stating that the user writing the comment has seen and acknowledges the content, which is the way the site existed before the
91 ite was smaller. It would seem that if the endorsement tool, the overall number of endorsements would go down, as commenting represents a higher time commitm ent and higher buttons were introduced in 2009, and in that time they have grown into buttons seen across the web, even not on Facebook, to endorse content. The influence of deciding to endorse information may a lso be impacted by the size of the network , as those with a smaller network know they may not be influencing as many people as those with a larger network. ion can be The overall stated level of trust is also re asked, the majority of participants in this study responded they had relatively high levels of trust where par in general, and again at the ending of the study when the participants were asked to rank how st. If the participants would have been . That brings up one of the interesting contradictions of the had endorsed. Perhaps, then, there is a deeper concept of what trust actually means on Facebook.
92 Most studies dealing sp ecifically with Facebook that use trust as a concept utilize one of those existing definitions of trust. This stud y attempted to do the same , using standard definitions of trust when constructing the hypotheses based on the existing literature. Yet the par ticipants did not act upon their stated trust. It could very well be that trust means something completely different to people within the confines of Faceb ook and online social networks. Conformity At a deeper level, the basic concept of testing if people will pass along a piece of conformity. When Asch (1951) conducted his famous line tests, the results showed that an individual was likely to go against his or her own observations in order to fit in. An element of conformity could have influenced those who decided to either endorse or not endorse the Facebook post they were shown in the study. For those who decided to endorse the information, the element of conformity is obvious by seeing others endorse the information, even if it seemed like something that was not real, they chose to be a part of the group. Perhaps, then, not endorsing the information was also influenced by some kind of conformity in that mo st users did not also endorse it. There was a disconnect between some classical elements of conformity and the decisions of the majority of the participants in the study. Most participants did not decide to endorse the message they were presented with, ind icating that they were not willing to conform their own the inc orrect line length were in the same room as the participant and could use stares, for example, to enhance the pressures to conform. That physical presence does not exist in that pure form on Facebook.
93 This research suggests a number of possible future res earch studies. Future research could investigate the impact of conformity on Facebook users and how the conformity heuristic. Future research should attempt to f ind the threshold needed to trigger the conformity heuristic? Alternatively, researchers may seek to discover if there is a bottom limit for conformity. Future research should work on finding out of it takes 10, or 100, or 1,000 o have an influence, or if the psychological heuristic of conformity is somehow something entirely different on Facebook. Essentially, the f uture research should examine whether b efore a given user also chooses to endorse something. Conformity is, at its absolute core, a social phenomenon. There has to be something, be it a behavior or an aesthetic or any other repeatable component, to which an individual can conform . An indiv idua l in a vacuum has nothing with which to conform. This study went forward with the assumption that conformity would be present on Facebook. This study also went forward with the understanding that the reason why figuring this out was important was because o f the potential of a wide spread health related false rumor. Future research can continue Opinion Leaders Opinion leaders have been one of the most consistent concepts in communications studies a concept vital to understanding the two step flow of communication. This study proposed the idea that the concept of opinion leader may be entering an age where opinion leaders are more diffused than ever before because online social networks allow for everyone to potentially be an opinion leader to others. Katz and Lazarsfeld (1955) said opinion leaders often
94 connections. Because every Facebook user has the potential to use th to endorse information, then every Facebook user has access to the tools needed to allow them to have Limitations This study had a number of understood limitations inherent in the methodology and approach that could be addressed in further studies. The first was sample size. The goal was to have 80 participants, and the final number of participants was 68, most likely due to the two stages of the experimental methodology, as well as the time required to complete the first portion of the experimental process. The retention rate between those who arrived for the first session and the second session of the experiment was rather high. Instead, the issue was a large number of potential participants who initially emailed an d stated they were interested in participating and were t hen given a scheduled time, never arrived for their scheduled time. However, the results of the study suggested that the stimuli did not have a significant impact on the decision of the participants to endorse the information, so it is unlikely that more participants would have impacted this specific study. Another limitation to be addressed is in the social network analysis portion of the study. The methodology of this study used the Gephi software to form calculated modules of from both the most populated module and t he second post populated module. This created some potential issues within the design. First, there was not a control for knowing the relationship that each individual participant had with any of their selected modules. For instance, a module could
95 have be en distant family members or college classmates with whom the participant regularly interacts. was connected with other nodes. Many participants commented that the modules ten ded to line up with groups of individuals in their lives, such as friends from high school, family members, friends from social groups, etc. However, the study did not record the level of personal relationship the participant had with that given module. So me participants may have been asked to select individuals from their most populated module when, in general, the participant does not been if a participant old er than the mean was asked to select from a module comprised of mostly high school friends with whom they have not interacted recently . Although they may still partici of higher connectedness. Other participants may have had their modules filled with with whom they have a closer bond. In future studies, care should be taken to ensure that the emotional bond a participant has to a given module is understood. That could also be the topic of a future analysis to examine which are the most common themes of modules formed from This study used a medical messag e as the basis for testing the variable of critical stance, which was based in concepts of rumor theory. According to theories presented by Buckner (1965) concerning rumor theory, individuals are more likely to take an uncritical stance towards a rumor if the individual has little to no existing knowledge of the topic of the rumor. Buckner (1965) explains that the uncritical stance is often due to having no basis of comparison for the piece of information. In this study, a fictional illness was presented to half of the participants,
96 while the other half received a message concerning the common cold. The illnesses chosen could have proven as a limitation of the study, particularly concerning the relatively low potential worry factor of the common cold. The st udy could be repeated using the variable dealing with the health related message, but use another, more dangerous illness such as meningitis. button posts a linked message related messages res earch study could explore how often individuals try to confirm the accuracy of information before sharing it. This experiment was set up so that individuals did not have access to the Internet while completing the second session of the experiment. Some par ticipants may have checked the false illness, attempting to learn about it before endorsing or liking. Participants were asked not to look up information about their questionnaire , but they could not be completely monitored the entire time they were in the lab room and may have used a cell phone. Future Research The most significant findings in this study were based on the presence of what has been t e r who use the site -at a far above average rate, and that the average user tends to receive more feedback than they give. uld be studied . The factual rumors to see if there is a line past which they will not endorse information.
97 whether or not to use endorsement tools. Interviews could be done with those who behave like as endorsement tools, if they see themselves as having influence on Facebook, and what process they use to determine if a piece of information is factual or not. Future work should also attempt to examine how trust works in online social networking litative interviews could be a good way to address this disconnect. Participants could be asked through namely, if they would up where participants could b if a college student sample is used, as most of the participants in this study had well ove r 1,000 ne twork to make the study viable. T hat could be done by using the same NetVizz system used in this study, then copy and pasting the list of Friends from the data editor into a random sampling equal four point scale proposed by Lamb ert (2014), it could clear up one of the limitations of this
98 study, as well as form a viable platform for either quantities methodologies or qualitative interviews. This study should be repeated using a health related message that has more medical urgenc y. Although it seems like an obvious finding, the participants also generally did not attempt to say they had knowledge of the illness that was not real. As noted previously in this section, the lack of significant reaction to the invented illness, as well as the common cold, could have been because there is nothing either new or dangerous about the common cold or about the Meretinitis, based on the description of the symptoms in the Facebook message. Research should be conducted similar to this study but u sing more dangerous health related messages. Especially related messages could be tes ted, ranging from something urgent, such as a warning of a coming meningitis epidemic to dangerously benign, misinformed or naive, such as suggesting people do not need to vaccinate their children. h technological advancement that two specific variables and their potential impact on rumor spread on Facebook . The study found that the diversity of modules did not have a significant impact. The content of the message itself also was not significant, yet there were connections between health message and stated level of knowledge. However, other significant findings related to endorsement were noted and discussed further, with the hope that it will advance knowledge on the subject of rumors and Facebook. intended to be treated like a rumor, much less a hoax (Fedler, 1987 ). Yet, that is what happened.
99 Some l isteners who missed the introduction of the program as a radio drama believed the United at once, all interconnected, with powerful endorsement tools present, there is even more potential for context to be lost. From unintentional rumors to intentional hoaxes, it becomes crucially important to analyze the best ways to target and stop false information on Facebook.
100 APPENDIX A FIRST SES SION MATERIALS 1. Please select your gender: a. Man b. Woman 2. Please select your age: a. 18 b. 19 c. 20 d. 21 e. 22 f. 23 g. 24 h. 25+ 3. Please select your major: a. Advertising b. Journalism c. Public Relations d. Media & Society e. Telecommunications f. Other 4. On a scale of 1 to 10, with 1 representin g not at all, and 10 representing very much, how much would you say you trust the people with whom you interact with on a regular basis on Facebook. 5. On a scale of 1 to 10, with 1 representing not at all, and 10 representing very much, how much would you say you trust the the content in links posted and shared by the people with whom you interact with on a regular basis on Facebook. 6. On a scale of 1 to 10, with 1 representing very knowledgeable , and 10 representing not at all knowledgeable , how much medi cal knowledge would you say that you have, particularly in assessing if symptoms are worth speaking to a doctor about? 7. How often do you use the "Share" button on Facebook? a. Very often almost every link you see b. Often multiple times while on Facebook c. O ccasionally once per day d. Not very often maybe once per week, if that e. Never
101 8. How often do you use the "Share" button on Facebook? a. Very often almost every link you see b. Often multiple times while on Facebook c. Occasi onally once per day d. Not very often maybe once per week, if that e. Never button 9. How many days per week do you access Facebook for more than 20 minutes at a time? a. None b. 1 2 days c. 3 4 days d. 5 6 days e. 7 day 10. In an average day that you use Facebook, how often during the day do you access Facebook? a. Once in a day b. Three times in a day c. Once every few hours d. Once per hour or more e. I do not access Facebook in an average day 11. Roughly how many "friends" do you have on Facebook? By "friends," thi s means the number of people who have officially connected with you through the system. (If you need to log on Facebook to check this, that is fine. You'll be asked to log on when you complete this survey anyway) a. 1 to 100 b. 101 to 200 c. 201 to 300 d. 301 to 400 e. 40 1 to 500 f. 501 to 600 g. 601 to 700 h. 701 to 800 i. 801 to 900 j. 901 to 1,000 12. On what kinds of different devices do you normally access Facebook? Please check all that apply a. Personal computer / Laptop b. Smartphone c. Tablet d. Electronic Reader e. Other
102 13. Where are you in a norma l day when you access Facebook? Please check all that apply. a. Home b. Work c. School d. Other 14. How much would you say your close Facebook friends those who you interact with on a regular basis through wall posts and Facebook Chat are like you? This can be things suc h as major, common interests, and ways of thinking? a. Not at all like me b. Not very much like me c. Neutral d. Somewhat like me e. Almost identical to me
103 APPENDIX B HAND OUT MATERIALS Form 1 & 2 Name: __________________________________ ______________________________ Write the number of your biggest module Write the names of 10 people with whom you are Facebook instructions from the researcher on how to select Place a checkmark in this box if the person to the left is a medical professional. Form 2 Name: ________________________________________________________________ Write the number of your biggest module Write the names of 10 people with whom you are Facebook to instructions from the researcher on how to select Place a checkmark in this box if the person to the left is a medical professional.
104 Form 3 & 4 Name: _______________________________________________________________ _ Write the number of your biggest module and second biggest module Write the names of 10 people with whom you are Facebook instructions from the researcher on how to select Place a checkmark in this box if th e person to the left is a medical professional. Form 4 Name: ________________________________________________________________ Write the number of your biggest module and second biggest module Write the names of 10 people with whom you are Facebook instructions from the researcher on how to select Place a checkmark in this box if the person to the left is a medical professional.
105 APPENDIX C SECOND SESSION MATERIALS 1. For this study, ass ume that the "Shared" post below appeared on your Facebook news feed. You notice that the before it says "...shared a link" it has the following of your Facebook friends: [NAME 1] [NAME 2] [NAME 3] [NAME 4] [NAME 5] [NAME 6] That means that those six fri ends also chose to "Share" the link on their own Facebook profiles, and that is why it is appearing on your own Facebook news feed.
106 Carefully read the language in the headline and description of the "Shared" link. When you are ready to proceed and an swer questions based on this "Shared" link, please click next. 2. How likely would it be that you would click "Like" on the Facebook post from the previous page? On a scale of 1 to 10, with 1 representing "I would absolutely not click 'Like' on the post" and 10 representing "I would absolutely click 'Like' on the post," indicate how likely it would be for you to click "Like." 3. How likely would it be that you would click "Share" on the Facebook post from the previous page? On a scale of 1 to 10, with 1 represe nting "I would absolutely not click 'Share' on the post" and 10 representing "I would absolutely click 'Share' on the post," indicate how likely it would be for you to click "Share." 4. How likely would it be that you would click on the hyperlink itself in t he Facebook post from the previous page? On a scale of 1 to 10, with 1 representing "I would absolutely not click on the post" and 10 representing "I would absolutely click on the post," indicate how likely it would be for you to click "Share." 5. On a scale of 1 to 10, please indicate your knowledge level about the common cold, meaning that you know the symptoms, how to treat it, and how it tends to spread from person to person. The scale is 1 representing "I know absolutely nothing about the common cold" an d 10 representing "I know everything about the common cold." 6. On the left hand side you will see a list of some of your Facebook friends used earlier in this survey. Please indicate, using the 1 to 10 chart, how much you trust that Facebook friend. Please use 1 to represent "I do not trust this person at all" and 10 to represent "I trust this person with everything."
107 LIST OF REFERENCES Acquisti, A. & Gross, R. (2006). Imagined Communities: Awareness, Information Sharing, and Privacy on th e Facebook. In Golle, P. and Danezis, G. (Eds.), Proceedings of 6th Workshop on Privacy Enhancing Technologies . (pp. 36 -58).Cambridge, U.K. Robinson College. June 28 30. Asch, S. E. (1940). Studies in the principles of judgments and attitudes: II. Determ ination of judgments by group and by ego 465. Asch, S. E. (1955). Opinions and social pressure. Readings about the social animal , 17 26. Backstrom, L., Huttenlocher, D., Kleinberg, J., & Lan, X. (2006) . Group formation in large social networks: Membership, growth, and evolution. Proceedings of 12th International Conference on Knowledge Discovery in Data Mining (pp. 44 54). Bea, F. (2013, May 19). Do not click! These are the Facebook hoaxes you need to avoid. Digital Trend . Retrieved from http://www.digitaltrends.com/social media/facebook scams hoaxes you need to know/ boyd, d. & Ellison, N. B. (2007). Social network sites: Definition, history, and scholarship. Journal of Computer Mediated Communication , 13 (1), 210 230. Brenner, J. (2013). Pew Internet: Social networking. Pew Internet & American Life Project , Retrieved from http://pewinternet.org/Commentary/2012/March/Pew Internet Social Networking full detail.aspx Buckner, H. T. (1965). A theory of rumor transmission. Public Opinion Quarterly , 29 (1), 54 70. Bucy, E. P. (2003). Media credibility reconsidered: Synergy effects between on air and online news. Journalism & Mass Communication Quarterly , 80 (2), 247 264. Burke, M., Kra ut, R., & Marlow, C. (2011, May). Social capital on Facebook: Differentiating uses and users. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (pp. 571 580). ACM. Chen, A. (2012, November 30). Fake winning powerball ticket goe s viral on facebook proving once again that facebook users will share anything. Gawker . Retrieved from http://gawker.com/5964731/fake winning powerball ticket goes viral on facebook proving once again that facebook users will share anything Chen, A. (2013 , April 17). Your guide to the boston marathon bombing amateur internet crowd sleuthing. Gawker . Retrieved from http://gawker.com/5994892/your guide to the boston marathon bombing amateur internet crowd sleuthing
108 Cheung, C. M., Chiu, P. Y., & Lee, M. K. ( 2011). Online social networks: Why do students use F acebook?. Computers in Human Behavior, 27(4), 1337 1343. Christakis, N. A., & Fowler, J. H. (2007). The spread of obesity in a large social network over 32 years. New England journal of medicine, 357(4), 370 379. Chung, C., Nam, Y., & Stefanone, M. A. (2012). Exploring Online News Credibility: The Relative Influence of Traditional and Technological Factors. Journal Of Computer Mediated Communication , 17 (2), 171 186. Daniels, N. (2012, December 18). Why I started a social media hoax -and how I'm using it to help. Huffington Post . Retrieved from http://www.huffingtonpost.com/nolan daniels/powerball facebook hoax_b_2323206.html Eastin, M. S. (2001). Credibility assessments of online health information: Th e effects of source expertise and knowledge of content. Journal of Computer Mediated Communication , 6 (4), 0 0. Journal of Computer Mediated Communication , 12 (4), 1143 116 8. Ellison, N. B., Steinfield, C., & Lampe, C. (2011). Connection strategies: Social capital implications of Facebook enabled communication practices. New Media & Society , 13 (6), 873 892. Eysenbach, G. (2007). Credibility of health information and digita l media: new perspectives and implications for youth. Eysenbach, G., & KÃ¶hler, C. (2002). How do consumers search for and appraise health information on the world wide web? Qualitative study using focus groups, usability tests, and in depth interviews. BM J: British Medical Journal , 324 (7337), 573. Fox, S., & Jones, S. (2009). The social life of health information. Washington, DC: Pew Internet & American Life Project , 2009 12. Facebook statistics, stats & facts for 2011 . (2011, January 18). Retrieved from http://www.digitalbuzzblog.com/facebook statistics stats facts 2011/ Flanagin, A. J., & Metzger, M. J. (2000). Perceptions of Internet information credibility. Journalism & Mass Communication Quarterly , 77 (3), 515 540. Fox, S., & Duggan, M. (2013). Heal th online 2013. Health .
109 Fu, B. (2007). Trust Management in Online Social Networks (Doctoral dissertation, University of Dublin, Trinity College). Gallant, L. M., Boone, G. M., & Heap, A. (2007). Five heuristics for designing and evaluating Web based comm unities. First Monday , 12 (3), 1. Gerlitz, C., & Helmond, A. (2013). The Like economy: Social buttons and the data intensive web. New Media & Society. Gilbert, E., & Karahalios, K. (2009, April). Predicting tie strength with social media. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (pp. 211 220). ACM. Granovetter, M. S. (1973). The strength of weak ties. American journal of sociology, 1360 1380. Granovetter, M. (1983). The strength of weak ties: A network theory revisite d. Sociological theory, 1(1), 201 233. Gray, N. J., Klein, J. D., Noyce, P. R., Sesselberg, T. S., & Cantrill, J. A. (2005). Health information seeking behaviour in adolescence: the place of the internet. Social science & medicine , 60 (7), 1467 1478. Jaeg er, M. E., Anthony, S., & Rosnow, R. L. (1980). Who Hears What from Whom and with What Effect A Study of Rumor. Personality and Social Psychology Bulletin , 6 (3), 473 478. Johnson, T. J., & Kaye, B. K. (1998). Cruising is believing? Comparing Internet and traditional sources on media credibility measures. Journalism & Mass Communication Quarterly , 75 (2), 325 340. Johnson, T. J., & Kaye, B. K. (2000). Using is believing: The influence of reliance on the credibility of online political information among poli tically interested Internet users. Journalism & Mass Communication Quarterly , 77 (4), 865 879. Johnson, T. J., & Kaye, B. K. (2004). Wag the blog: How reliance on traditional media and the Internet influence credibility perceptions of weblogs among blog us ers. Journalism & Mass Communication Quarterly , 81 (3), 622 642. Johnson, P. R., & Yang, S. (2009, August). Uses and gratifications of Twitter: An examination of user motives and satisfaction of Twitter use. In Communication Technology Division of the annua l convention of the Association for Education in Journalism and Mass Communication in Boston, MA . Hampton, K., Goulet, L. S., Marlow, C., & Rainie, L. (2012). Why most Facebook users get more than they give. Pew Internet & American Life Project.
110 Hermida, A., Fletcher, F., Korell, D., & Logan, D. (2012). Share, Like, Recommend. Journalism Studies, 13(5 6). Hinduja, S. & Patchin, J. W. (2010). Cyberbullying Research Summary: Trends in Adolescent Social Networking. Hovland, C. I., & Weiss, W. (1951). The influence of source credibility on communication effectiveness. Public opinion quarterly , 15 (4), 635 650. Hsu, W. H., Lancaster, J., Paradesi, M. S. R., & Weninger, T. (2007). Structural link analysis from user profiles and friends networks: A feature con struction approach. Proceedings of ICWSM 2007 (pp. 75 80). Kadushan, C. (2012). Understanding social networks: Theories, concepts, and findings . New York, NY: Oxford University Press. Katz, E., & Lazarsfeld, P. F. (2006). Personal influence: The part p layed by people in the flow of mass communications . Transaction Publishers. Kosinski, M., Stillwell, D., & Graepel, T. (2013). Private traits and attributes are predictable from digital records of human behavior. Proceedings of the National Academy of Sci ences . Korda, H., & Itani, Z. (2013). Harnessing social media for health promotion and behavior change. Health promotion practice , 14 (1), 15 23. Kunst, H., Groot, D., Latthe, P. M., Latthe, M., & Khan, K. S. (2002). Accuracy of information on apparently credible websites: survey of five common health topics. BMJ: British Medical Journal , 324 (7337), 581. Lambert, A. (2013). Intimacy and Friendship on Facebook . (1st ed.). London: Palgrave MacMillan. Lee, R. S. (1978). Credibility of Newspaper and TV News. Journalism Quarterly , 55 (2), 282 87. Leitsinger, M. (2012, March 08). When rumor, the Internet and school violence fears collide. NBC News . Retrieved from http://usnews.nbcnews.com/_news/2012/03/08/10604539 when rumor the internet and school violence fear s collide Lewis, K., Kaufman, J., Gonzalez, M., Wimmer, A., & Christakis, N. (2008). Tastes, ties, and time: A new social network dataset us ing Facebook. com. Social networks , 30 (4), 330 342. Lind., P. G., da Silva, L. R., Andrade Jr., J. S., Herrmann, H.
111 Acta Physica Polonica B 37 (11) McFarlin, D. (2014, May 22). Ar e Girls Still in the Balcony? Retrieved July 1, 2014, from https://medium.com/@DMcFarlinUF/are girls still in the balcony a52d98e26b91 McGinnies, E., & Ward, C. D. (1980). Better liked than right Trustworthiness and expertise as factors in credibility. Personality and So cial Psychology Bulletin , 6 (3), 467 472. Use of Facebook and Twitter. Communication Research Reports, 29(2), 108 118. Mendoza, M., Poblete, B., & Castillo, C. ( 2010). Twitter Under Crisis: Can we trust what we RT?. In Proceedings of the first workshop on social media analytics (pp. 71 79). ACM. Mikkelson, B. (2012, March 05). 100 Facebook shares appeal. Snopes . Retrieved from http://www.snopes.com/inboxer/medica l/100shares.asp Miners, Z. (2013, April 22). Reddit apologizes for 'online witch hunt' for Boston bombing suspect. PC World . Retrieved from http://www.pcworld.com/article/2036176/reddit apologizes for online witch hunt for boston bombing suspect.html Mon tgomery, D., Horwitz, S., & Fisher, M. (2013, April 20). Police, citizens and technology factor into boston bombing probe. Washington Post . Retrieved from http://www.washingtonpost.com/world/national security/inside the investigation of the boston marathon bombing/2013/04/20/19d8c322 a8ff 11e2 b029 8fb7e977ef71_print.html Moreno, M.A., et al. (2009) Display of health risk behaviors on MySpace by adolescents: prevalence and associations. Archives of pediatrics & adolescent medicine 163.1: 27. Murse, T. (2 012, August 21). Officials: No truth to rumor that 'burn book' led to suicide of penn manor student read more. Lancaster Intelligencer Journal . Retrieved from http://lancasteronline.com/article/local/718391_Officials -No truth to rumor that -burn book -led to suicide of Penn Manor student.html Nadkarni, A., & Hofmann, S. G. (2012). Why do people use Facebook?. Personality and individual differences , 52 (3), 243 249. National Crime Prevention Council, (2001). Cyberbullying tips for teens . Retrieved from websi te: http://www.ncpc.org/topics/cyberbullying/cyberbullying tip sheets/NCPC Tip Sheet Tips for Teens.pdf
112 Oh, O., Kwon, K. H., & Rao, H. R. (2010). An exploration of social media in extreme events: Rumor theory and twitter during the Haiti earthquake 2010 . Qazvinian, V., Rosengren, E., Radev, D. R., & Mei, Q. (2011, July). Rumor has it: Identifying misinformation in microblogs. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (pp. 1589 1599). Association for Computation al Linguistics. Ortutay, B. (2012, October 04). Facebook tops 1 billion users. USA Today . Retrieved from http://www.usatoday.com/story/tech/2012/10/04/facebook tops 1 billion users/1612613/ Patchin, J. W. & Hinduja, S. (2010). Changes in adolescent onlin e social networking behaviors from 2006 to 2009. Computers in Human Behavior, 26, 1818 1821. Pempek, T. A., Yermolayeva, Y. A., & Calvert, S. L. (2009). College students' social networking experiences on Facebook. Journal of Applied Developmental Psycholo gy, 30(3), 227 238. Quan Haase, A., & Young, A. L. (2010). Uses and gratifications of social media: A comparison of Facebook and instant messaging. Bulletin of Science, Technology & Society , 30 (5), 350 361. Raacke, J., & Bonds Raacke, J. (2008). MySpace and Facebook: Applying the uses and gratifications theory to exploring friend networking sites. CyberPsychology & Behavior , 11 (2), 169 174. Rieder, B. (2013). Studying Facebook via data extraction: the Netvizz application. In WebSci '13 Proceedings of the 5th Annual ACM Web Science Conference (pp. 346 355). New York: ACM. Rhee, J., Kim, E., & Kim, H. (2007). Exploring Online Opinion Leadership: A Validity Test of the Concept in the Digital Age. Conference Papers -International Communication Association , 1. Roblyer, M. D., McDaniel, M., Webb, M., Herman, J., & Witty, J. V. (2010). Findings on Facebook in higher education: A comparison of college faculty and student uses and perceptions of social networking sites. The Internet and Higher Education, 13(3), 134 140. Rogers, E. (2003). Diffusion of innovations . (5th ed.). New York, NY: Free Press. Rollins, A. Moderation Policies and Strategies of Four News Sites for User Generated Content. University of Florida College of Journalism & Communicati on .
113 Ryan, T., & Xenos, S. (2011). Who uses Facebook? An investigation into the relationship between the Big Five, shyness, narcissism, loneliness, and Facebook usage. Computers in Human Behavior, 27(5), 1658 1664. Salah, A., Manovich, L., Salah, A., & Ch ow, J. (2013). Combining Cultural Analytics and Networks Analysis: Studying a Social Network Site with User Generated Content. Journal Of Broadcasting & Electronic Media , 57 (3), 409 426. Schwartz HA, Eichstaedt JC, Kern ML, Dziurzynski L, Ramones SM, et al. (2013) Personality, Gender, and Age in the Language of Social Media: The Open Vocabulary Approach. PLoS ONE 8(9): e73791. Stavrositu, C., & Sundar, S. S. (2008). If Internet credibility is so iffy, why the heavy use? The relationship between medium us e and credibility. CyberPsychology & Behavior , 11 (1), 65 68. Stillwell, D.J., Kosinski, M., & Tunney, R.J. (2011) You are what you like. Research Showcase finalist, Nottingham University. Gossip on Social Internet. Computers in Human Behavior , 28 (4), 1510 1517. Tufekci, Z. (2008). Grooming, gossip, Facebook and MySpace: What can we learn about these sites from those who won't assimilate?. Information, Communication & Society , 11 (4), 544 564. Valente, T. W. (2010). Social networks and health: models, methods, and applications . Oxford: Oxford University Press. Vitak, J., Ellison, N. B., & Steinfield, C. (2011, January). The ties that bond: Re examining the relationship between Facebook use and bonding social capital. In System Sciences (HICSS), 2011 44th Hawaii International Conference on (pp. 1 10). IEEE. Walther, J. B., Van Der Heide, B., Kim, S. Y., Westerman, D., & Tong, S. T. (2008). The role of known by the company we keep?. Human Communication Research , 34 (1), 28 49. W esterwick, A. (2013). Effects of Sponsorship, Web Site Design, and Google Ranking on the Credibility of Online Information. Journal Of Computer Mediated Communication , 18 (2), 80 97.
114 Xiang, R., Neville, J., & Rogati, M. (2010, April). Modeling relationshi p strength in online social networks. In Proceedings of the 19th international conference on World wide web (pp. 981 990). ACM. Xu, Q. (2013). Social Recommendation, Source Credibility, and Recency: Effects of News Cues in a Social Bookmarking Website. Jo urnalism & Mass Communication Quarterly , 90 (4), 757 775. Zhao, S., Grasmuck, S., & Martin, J. (2008). Identity construction on Facebook: Digital empowerment in anchored relationships. Computers in human behavior, 24(5), 1816 1836.
115 BIOGRAPHICAL SKETCH Je ffrey Kyle Riley received his Master of Science in journalism from Ohio University, where he completed his thesis on how professional journalism educators view the importance of teaching modern technological based skills and theories. He has been an instru ctor at the University of Florida since 2011, teaching Mass Communication Writing, Multimedia Writing, Multimedia Reporting, and Social Media in Society. He received his Bachelor of Science from the University of Central Florida, where he double majored in j ournalism and political science and served as the first online editor for the student newspaper, the Central Florida Future . He has a professional background in community newspapers and feature reporting, including work for the Seminole Chronicle and Win ter Park Observer . He has also worked in health based public relations for the Colleges of Osteopathic Research & Education.