Friends that Matter: How Social Distance Affects Selection and Evaluation of Content in Social Media

8/13/2019 Friends that Matter: How Social Distance A ffects Selection and Evaluation of Content in Social Media

1/43

Friends that Matter: How Social Distance Affects

Selection and Evaluation of Content in Social Media

Solomon Messing Sean J. Westwood

March 20, 2013

Abstract

As Americans shift from following a limited set news outlets to casually browsing social me-dia, they empower their friends and acquaintances to regulate their information environment.This means that interpersonal relationships and social cues, in the form of endorsements byindividual contacts or groups, affect the content that individuals select and how they processit. Though every relationship is unique and multifacted, each can be characterized along asingle important dimensionsocial distance, comprised of the related phenomena of frequentcontact, tie strength, social similarity and social group membership. We document the ef-fects of social distance on selectivity and content evaluation in two experimental studies. Thefirst study, conducted in an experimentally controlled Facebook application, operationalizessocial distance as frequent contact and measures its effect on news story selection. Our sec-ond experiment shows how ethnic and socio-economic group membership affects consumers

decisions to read content, and demonstrates how consumers differentially process the samecontent depending on their social distance to the recommender, which we show affects sub-sequently measured political preferences on related issues. Our results have implicationsfor our understanding of selective exposure, consumption, and other media effects in socialmedia.

Department of Communication, Stanford University. Equal contributions. Thanks to Shanto Iyengar,Kimberly Gross, Kyle Dropp, Cliff Nass, Eytan Bakshy, Rebecca Weiss, Thomas Leeper, Nathaniel Swigger,Lilach Nir, Danny Hayes, Patricia Joseph, and other panel participants at the 2012 Annual Meeting of theMidwestern Political Science Association in Chicago, Illinois 2012. This research was generously supportedby a Google Research Award and Stanford University.

1


2/43

Social media change the nature of getting the newswhereas traditional news outlets

deliver news and information curated by an editorial board in a centralized location, social

media display news and other content shared by the consumers social network contacts.

Hence, the ways that consumers find, select, and process content are influenced by the re-

lationship between the consumer and each individual in her social network of contacts who

share content. Of course, each relationship is unique and multifaceted, but each can also be

characterized along a single important dimensionsocial distance. Social distance includes

two main components from the social science literature: (1) the frequency of communication

and the related sociological concept of tie strength (Allport 1954;Zajonc 1968;Granovetter

1973;Alba and Kadushin 1976), and (2) social similarity: common ethnic, racial and socio-

economic group membership (Park 1924; Bogardus 1925; Tajfel and Turner 1979; Turner

1982;Homans 1951;Karakayali 2009), unconcious evaluations of physical (and possibly ge-

netic) similiarity (Bailenson et al. 2008), percieved attitude similarity (Park and Schaller

2005), and even coincidental shared history and prior activities (Brock 1965; Burger et al.

2004). If social distance affects how consumer find, select, and process content, the shift

from loyally frequenting traditional media outlets to browsing social media has tremendous

consequences for political communication.

We show that social proximitysocial distance between the consumer and the recommender

affects how consumers decide to allocate attention and consume content. This process priv-

ileges information shared by socially close friends at the expense of socially distant (often

heterogeneous) contacts. This changes the way we process political information. We show

that social proximity shapes these outcomes when operationalized both in terms of communi-

cation frequency (closely related to tie strengthAllport 1954;Zajonc 1968;Granovetter 1973;Alba and Kadushin 1976), and when operationalized as common ethnic, racial and socio-

economic group membership (Park 1924; Bogardus 1925; Tajfel and Turner 1979; Turner

1982;Homans 1951;Karakayali 2009). We also establish a theoretical expectation that un-


3/43

der certain conditions the tendency to select content from socially proximate recommenders

will be magnified over time by feedback loops created by the algorithms that integrate past

selection preferences to determine which content to display (the related phenomena of filter

bubbles was suggested inPariser 2011).

We document the effects of social distance on selectivity and content evaluation in two

experimental studies. The first study, conducted in an experimentally controlled exact copy

of the Facebook interface, measures of the effect of social proximity on news story selection.

This design documents how social networks and peer endorsements affect the choices made

when deciding which news stories to read in social media. We then present the results of a

second web-experiment on a national sample designed to document the causal effect of racial

and economic attributes of recommenders (friends) on our decisions to read content, and

document how social distance between the recommender and the consumer shapes evaluation

of the same content and affects subsequently measured political preferences on related issues.

Traditional Media Models versus Social Media Models

Consumers are increasingly shifting from a mode where habitual viewership and reader-

ship of a traditional source is replaced with happenstance encounters with stories found in

social media, with implications for the kind of news we ultimately consume. From 2000 to

2012, local and national news audiences shrank by more than 10 percentage points, while the

percentage of American adults who regularly get news from Internet sources has grown from

nine percent to 36 percent (Pew 2012).2 By 2011, 47 percent of Americans reported that

they get at least some local news and information on their cellphone or tablet computer. By

2012, the percentage of Americans who report that they learned something about the 2012

2During the same time period, the audience for cable news saw more limited growth, with 34 percent ofAmericans in 2000 and 41 percent of Americans in 2012.

1


4/43

campaign on Facebook was 42 percent, according to the same report ( Pew 2012).3 More

than half of Americans now use social media on a regular basis (Facebook 2011) and Amer-

ican Internet users spent more time on Facebook than any other single Internet destination

in 2011 by nearly two orders of magnitude according to Nielsen, with an average time per

month of 7-8 hours per person (Nielsen 2011).

As usage patterns shift to a mode where the vast majority of viewers visit any given site

only once or twice per month (Pew 2012), themeansby which the viewer finds a given story

becomes critically important. In such an environment, what drives attention to a story is

not a newsroom editor deciding the story that leads, nor a homepage spot with a large font

and a big photo, but prominence of the story on social media. Social media users provide

links to these stories, referringtheir contacts to the actual news content on the originating

website. Figure1presents a heatmap showing representing the top sources of referrals (rows)

to each of the top traditional news websites (columns); the darker each cell, the higher each

referral source ranks in importance for that site (data from Neilsens Netview database, see

Pew 2011).4 As the figure shows, Facebook was among the top sources of referral traffic to

each of the top 21 news websites in the Neilsen data; Twitter was a significant source as well

(Pew 2011). Given the important relationship between the mix of information to which one

is exposed and political attitudes (Klapper 1960; Sears and Freedman 1967), and the way

people update those attitudes in response to persuasive arguments (whether or not selective

consumption is possible,Leeper 2013), these patterns have significant implications for media

fragmentation and political polarization.

Much of the past work on the political implications of new media focuses on the

politically polarizing effects of the political blogoshpere (Adamic and Glance 2005; Baum3We include those reporting that they LEARN SOMETHING about the PRESIDENTIAL CAMPAIGN

or the CANDIDATES regularly (12%), sometimes (20%), or hardly ever (10%).4Some of the top referrers to particular sources were not in the top 25 and so do not appear in the

plotfor example, AOL.com is the largest referral source to AOLNews but doesnt even register for othernews sources so we exclude it.

2


5/43

Figure 1: Referral Traffic to Traditional Online News Sources (Neilsen)

ask.com

fark.com

my.yahoo.com

my.msn.com

aolnews.com

news.search.yahoo.com

news.yahoo.com

swagbucks.com

stumbleupon.com

huffingtonpost.com

traffic.outbrain.com

realclearpolitics.com

reddit.com

twitter.com

yahoo.com

hotair.com

search.aol.com

drudgereport.com

search.yahoo.com

bing.com

facebook.com

news.google.com

google.com

ABCNews

AOLNews

CBSNews

CNN.com

ChicagoTribune

DailyMail

Examiner.com

FoxNews.com

HuffingtonPost

LATimes

MSNBC.com

NYDailyNews

NYTimes

SFGate.com

USAToday

WasingtonPost

News Website

Refe

rralSource

and Groeling 2008; Sobieraj and Berry 2011) and the potential for blogs to drive civic

involvement (Kerbel and Bloom 2005;Perlmutter 2008), while other work has examined the

consequences of increased political selective exposure in the context of visiting traditional

media websites and news aggregators (Iyengar and Hahn 2009;Messing and Westwood 2012).

Yet loyal consumption of these media remains confined to a relatively small proportion of the

3


6/43

public.5 Other work has discussed the potential for social media to recreate the inadvertent

audience (Pasek, More, and Romer 2009) or drive voter behavior through social influence

processes (Bond et al. 2012). But no study of which we are aware has studied how the

features of this new mode of communication affect how we select and process news.

How Social Media Differ and Why It Matters

From the viewers perspective,6 the most important distinction between traditional news

sources and social media websites like Facebook is that the latter serve to aggregate a much

more diverse array of content in a single location. Encountering news in this context repre-

sents a fundamental departure from the way in which news consumers have typically engaged

content: they no longer first decide on a news source. Instead, viewers of social media select

the story itselfand we argue here that this choice must be informed by evaluations of the

person who recommended this content in the first place.

The fact that news source is often unclear in the context of social media motivates the

need to study what drives the decision to select from a series of items that appear in social

media (e.g., those present in ones Twitter Feed or Facebook News Feed). We know that when

choosing a media item from a set of headlines, consumers generally seek to maximize utility

by employing cue-based heuristic processing (Kahneman 2003;Tversky and Kahneman 1974)

rather than pursuing a cognitively-taxing optimization strategy, especially when considering

criteria on more than one dimension (e.g., professional relevance and the ability to hold a

conversation about current events, see Messing and Westwood 2012). Heuristic processing

is also likely when we lack unambiguous information about the costs and benefits of each

5E.g., the audience for blogs is around nine percent of Americans (and much smaller if we define theHuffington Post as a partisan news source rather than a blog), according to (Pew 2012)

6Throughout, we refer to people who are passively browsing social media websites as viewers to dis-tinguish the activity of passive consumption from content creation-sharing/posting, endorsing, or writingcontent. Of course most people will engage in multiple activities during any given visit to a social mediawebsite.

4


7/43

outcome (Conlisk 1996;Simon 1972), which is certainly the case when trying to anticipate

whether we will like or benefit from reading/viewing a news item.

The heuristic cues that people utilize in conventional media have been identified as

source (most often operationalized as the media outlets brand, rather than the storys

byline/author, seeAlthaus and Tewksbury 2002;Iyengar and Hahn 2009;Sundar, Knobloch-

Westerwick, and Hastall 2007), story placement, the presence of a photograph, and other

editorial cues (Graber 1988) to help them judge the relevance, credibility and importance

of a news story. In social media there is no editor and the only traditional cues from cur-

rent models of news consumption are story title and source (though source often appears

as unobtrusive and in small font). In fact, in social media the notion of what constitutes a

source becomes ambiguousfor some viewers, it might be the person who shared and/or

recommended the information, the initial author of the content, or the media brand itself,

(e.g.,Metzger 2007).

Hence, social media affords a different and expanded set of cues compared to those

previously available, most notably social endorsement cues. There are two dominant types

of social endorsement cues visible in social media today: aggregated social endorsements that

summarize the number or proportion of people who read, endorse, and/or comment on a

media story; and personal recommendations, which constitute endorsements of content from

fellow members of that readers social network. Messing and Westwood (2012) show how

the presence of aggregate social endorsements change the way that people select content:

readers rely on social cues to a dramatically higher degree than source cues, at least with

respect to the ideological orientation thereof. We explicate how we expect social distance to

affect selection and processing of content in social media, but first we describe the interfacefeatures that serve to emphasize social attributes of recommenders.

Social media emphasize recommenders personal identity, which provides a basis for our

strong expectations that social distance shapes how viewers allocate attention and evaluate

5


8/43

content shared by peers. Social media purposefully encourage people to use their real names

and photographs, which diminishes anonymity and makes assessments of social proximity

possible (anonymity has been used to manipulate social distance, showing that it decreases

reciprocity (in dictator games, Hoffman, McCabe, and Smith 1996), elicits less sympathy

(Loewenstein and Small 2007) and altruistic behavior toward others (Charness and Gneezy

2008), and increases willingness to harm another person (Milgram 1963; Zimbardo 1969))

while photographs increase perceived likability and leads to greater hesitance to violate

social norms (Bailenson et al. 2005), and lead to greater cooperation (Parise et al. 1996).

Photographs in particular convey appearance and hence trigger complex cognitive processing

to evaluate group membership, status, attraction and competence.7 This emphasis on social

identity means that social attributes of recommenders are conveyed to viewers browsing

social media and should be expected to affect selectivity and the way we process content .8

Personal Recommendations: How Social Distance Af-

fects Exposure, Selection, and Evaluation

We now turn to an examination of how personal recommendations affect selectivity and

the way we process news content within social media. Integral to all of these processes is

the concept of social distance or proximity, which has often been used to measure affect

between members of various socioeconomic, ethnic, racial, and/or national groups (Park

7For examples of studies operationalizing these variables with photographs, see for group membership:(Blair and Jenkins 2002; Bond and Cash 1992; Dixon and Maddox 2005; Eberhardt et al. 2004; Gilliamet al. 1996; Hugenberg and Bodenhausen 2004; Klatzky, Martin, and Kane 1982; Maddox and Gray 2002;Messing, Plaut, and Jabon 2010;Ronquillo et al. 2007;Schaller, Park, and Mueller 2003;Tajfel et al. 1971;

Valentino, Hutchings, and White 2002), status: (Berger and Zelditch 1985; Lerner and Moore 1974;Pettyand Wegener 1999;Ridgeway 1987;Strodtbeck, James, and Hawkins 1957), attraction: (Lerner and Moore1974; Maddux and Rogers 1980; Cunningham et al. 1995; Eagly et al. 1991), and competence: (Todorovet al. 2005;Antonakis and Dalgas 2009).

8We recognize that it is also possible that people orient to the recommender as if she were the source ofthe message, especially in light of the heuristic processing involved in source orientation in human-computerinteraction(Sundar and Nass 2000).

6


9/43

1924; Bogardus 1925). It is conceptualized as a product of social interaction (a notion

supported by the mere exposure literature on the preference for familiar attitude objects

Zajonc 1968,2001) and is closely related to the concept of tie-strength, a (probably linear)

combination of the amount of time, the emotional intensity, the intimacy (mutual confiding)

and the reciprocal services which characterize the tie, or the relationship between two people

(Granovetter 1973; see alsoHomans 1951;Karakayali 2009).

Social distance affects how we form our networks of online contacts, the people to whom

we pay most attention online, and hence whose recommendations we are likely to follow.

Prior to the decision to select and consume an article as discussed above, the content to

which a person is exposed is bounded by the structure of a given users social network of

online contacts. Individuals tend to cultivate social ties with others that are similar to them-

selves, resulting in a social structure characterized by groups that are largely homogeneous

along socio-demographic traits and attitudinal orientations (Lazarsfeld and Merton 1954)a

pattern widely referred to as homophily. This patterns results partially from the lack of di-

verse individuals likely to be accessible to an individual (structural homophily), and partially

from the tendency to actively seek out and foster ties with other similar individuals while

avoiding those who differ (choice-homophily) (McPherson, Smith-Lovin, and Cook 2001).

Of course, when individuals meet new individuals, it is often by meeting friends of existing

friends (Parks 2008), who are likewise similar.

It is thought that the content that these individuals share is driven, at least in part,

by performance considerations. Liu (2007) found that people attempt to convey prestige,

differentiation, authenticity, and theatrical personas in their usage of social media. How-

ever, (Parks 2011) found that only a small portion of MySpace users actually engaged intaste performancesmany simply remained passive consumers of content. Despite the fact

that people attempt to convey differentiation when engaging in this behavior, it is worth

speculating that those engaged in performances are likely to take into account the perceived

7


10/43

preferences of those in their network when deciding what to share. Of course, it is also

worth noting that actual attitude-homophily is lower than individuals estimate and lower

than accounts in the theoretical literature have previously suggested (Goel, Mason, and

Watts 2010).

Social distance also affects the people to whom we pay most attention online, and hence

our exposure to news content. People often employ the liking-agreement heuristicthey

trust others they like and assume that they agree with their decisions (Chaiken 1987). This

heuristic predicts that viewers select content based on tie strength, or how close two people

are to each other, and there is evidence that people utilize this heuristic when selecting and

endorsing articles that friends endorse in the context of news aggregation services (Lerman

2007). In addition, the social network literature on choice homophily (McPherson, Smith-

Lovin, and Cook 2001) implies a related heuristic: people like me like what I like. This

predicts selection based on homophily, the similarity between people on an array of personal

attributes, including race, income, political attitudes and/or partisan affiliation, gender, age,

education, and profession. The two are most certainly relatedan extensive literature across

the social sciences demonstrates that people are often drawn to others perceived as similar

(seeBaumeister 1998, for a review).

How Feedback Loops Magnify the Effects of Social Distance

Even small selective tendencies can have a big effect on the actual content someone con-

sumes due to feedback loops created when selectivity affects outcomes that in turn affect

future selectivity or exposure, and we argue that social media contain structural features

that serve as additional feedback loops. Selective exposure to attitude-consistent informa-

tion (Festinger 1957) often occurs as an effort to defend ones personal position (Chaiken,

Liberman, and Eagly 1989), but can also happen when media consumers are motivated to

seek objective information but evaluate confirmatory information as more objective (Fischer

8


11/43

et al. 2005;Fischer, Schulz-Hardt, and Frey 2008). Of course, past work has hypothesized

that feedback loops related to selectivity reinforce attitudes in all media, which in turn drive

future selectivity, in what scholars call spirals of selectivity (Berelson, Lazarsfeld, and

McPhee 1954;Slater 2007). For example,Berelson, Lazarsfeld, and McPhee(1954) found in

a longitudinal field study that exposure to political campaign content not only reinforced ex-

isting beliefs, but also that increased exposure to the campaign led to more selective exposure

to media content consistent with their beliefs after the campaign. Furthermore, having more

extreme attitudes is associated with additional selective exposure over-time (Stroud 2010),

while at the same time, exposure to partisan media causes affective polarization (Levendusky

2013); and similarly a lack of attitudes and knowledge is thought to lead to less information

processing in regression toward zero (Eveland, Shah, and Kwak 2003). When people can

select from a range of opinions, they search out information consistent with the framing of

their pre-existing attitude or the initial persuasive message on about the topic/issue in ques-

tion encountered, and become more polarized and certain about their attitudes (Druckman,

Fein, and Leeper 2012). Indeed, when opinion become sufficiently strong, individuals will

completely disregard counter arguments and perceive the source as hostile (Vallone, Ross,

and Lepper 1985).

In addition to these selective spirals that characterize the way that an individual tends

to consume media, social media contain other structural features that serve as additional

feedback loops. As discussed above, the friendship networks that structure exposure to

content are likely to be similar on many dimensions. But our networks are likely to grow

even more similar over time due to conformity or social influence (Newcomb 1943; Heider

1944; Lazer et al. 2010). And, we are likely to distance ourselves from those with whoseopinions we dislike (Heider 1944;Festinger 1957). Thus, as our networks grow more similar

and socially close over time, so too should the content recommended by those in our networks

grow more homogeneous.

9


12/43

Algorithmic filtering serves as yet another source of feedback that can serve to reinforce

and magnify the effects of social distance between people under certain conditions. In social

media, we see what our friends share, but we only see about 25% of it and the order in

which we see that content is determined by algorithms that rank content based on imputed

preferences (to increase engagementBernstein et al. 2013). These algorithms model a per-

sons preferences based on her demographics, past choices, and/or the popularity of a item.

Hence these algorithms can serve to magnify the effects of social distance: the more the

viewer clicks on content posted by socially close individuals, the more frequently that con-

tent posted by these individuals will appear in the viewers news feed. Simultaneously, this

increasingly homogeneous content will crowd out content posted by those who are socially

distant from the viewer.

In order to demonstrate exactly how algorithms that privilege content from favored social

contacts can serve to magnify the effect of tie strength and homophily in particular when

selecting content, we simulate the process of visiting a social networking website and selecting

content posted by socially close and distant friends over time, depicted in Figure2.9 The

simulation is designed to establish a theoretical expectation establishing the magnitude and

conditions under which we might expect ranking algorithms to privilege content from friends

who are socially close on some dimension. We model this as a complex stochastic process

(full simulation details and code available in the appendix).

The simulation suggests that while de-facto network homophily should be the largest

factor driving exposure to content in social media, algorithmic feedback loops can be intro-

duced if people prefer and consume content recommended by homophiles friends. However,

9This simulation was created based on a simplified public description of the key components of socialmedia ranking algorithms, commonly referred to as EdgeRank and defined as

e

ue

we, where ue is the

affinity score between the viewer and the content creator, wc is the weight for any given activity (e.g.,posting, liking, commenting, re-sharing), and de is the decay constant based on how long ago the activityoccurred, which we set to 1 here, see also http://edgerank.net/. The simulation represents a theoreticalexpectation, not an empirical finding.

10
http://edgerank.net/http://edgerank.net/http://edgerank.net/


13/43

Figure 2: A Model of Feedback in Social Media

ContentShared

Exposure Consumption

Probabilityof Exposureto Friendj

Algorithm Selectivity

Algorithm Updates

Figure 3: Results of 500 Simulations, over Time (T [1, ...,365])

No Selectivity, No Algorithm No Selectivity, Algorithm (.15) Selectivity (.05), No Algorithm Selectivity (.05), Algorithm (.15)

0.6

0.7

0 100 200 300 0 100 200 300 0 100 200 300 0 100 200 300

Time (days)

ProportionofItemsSelected

fromS

ociallyClose/SimilarFriends

Low Network Similarity/Proximity (.55) Moderate Network Similarity/Proximity (.60) High Network Similarity/Proximity (.65)

in the absence of clearly biased selective consumption, these algorithms do introduce feed-

back, which motivates the study of selective consumption in the following research presented

below. The assumptions this simulation relies on may differ from the real world in important

ways that are worth mentioning. Most importantly, we do not model selective spirals in

which selective consumption polarizes opinion and leads to greater selective consumption.

Also, we do not model variation in the popularity of content (e.g., the number of endorse-

ments an item receivesFacebook likes or Twitter favorites), which should increase the

likelihood that algorithms will insert such content into a users news feed, and the likelihood

that the user will select that content (see Messing and Westwood 2012), which we would ex-

11


14/43

pect to diminish tendencies privileging socially close friends. Furthermore, news feed ranking

algorithms are thought to include a decay factor for peer-interactions such that older inter-

actions are not weighted as heavily (Facebook 2010;Kincaid 2010), which might also serve to

diminish the impact of socially proximity over time. Finally, Facebook and other companies

might very well implement other factors in their models designed to increase the diversity of

content people see, which would certainly diminish the magnitude of any feedback compared

to our simulation.

Social Distance and Information Processing

Even after a social media viewer selects content, social proximity continues to affectconsumers, shaping the way they process information. Recommenders social attributes

activate stereotypes and affective associations with the group (Devine 1989) in the mind of

the viewer, affecting how they process content. Such attitudes can be primed by implicit

cues, subtle references to phenomena, including images as Mendelberg(2001) documented

by showing that pairing a picture of a Black Willie Horton with the issue of crime in the

presidential campaign of 1988 primed racial resentment in candidate evaluations and policy

opinions. Others have shown that racial animus can be triggered by exposure to a wide

variety of cues, including relatively innocuous visual references to Blacks, which are thought

to boost the accessibility of racial schemas in memory, affecting vote choice (Kinder and Sears

1981; Valentino, Hutchings, and White 2002).10 On the other hand, counter-stereotypical

10In each case, the implicit-cue priming hypothesis has been tested by experimentally manipulating thevisual cue and examining its interaction with the respondents measure of racial resentment, which consistsof a battery of questions about attitudes toward progressive race policies. Sniderman and Tetlock (1986)point out a range of problems with this conception: (1) it tends to discount traditional racism as a spent force

when it is most certainly not, (2) it labels people who oppose progressive race policies as symbolic racists,(3) it is not clear about the relationship between anti-black affect and traditional values, (3) the boundaryconditions for when to conclude that racial motives determine a policy preference is not clear. Nonetheless,it predicts a wide range relevant behavior including vote choice for black candidates (Kinder and Sears 1981;Iyengar, Banaji, and Hahn 2009) and pro-black policies (Feldman and Huddy 2005), independent of otherexplanatory variables. We rely on this measure in this study.

12


15/43

exemplars can serve to do the exact opposite, lowering the impact of racial attitudes on

behaviorPlant et al. (2009) found evidence that exposure to counter-stereotypical Black

exemplars reduce implicit racial bias.

Social proximity to the recommender also affects whether readers are ultimately per-

suaded. First, social proximity to the source serves as a social cue that strengthens per-

suasion by increasing personal relevance (Petty and Cacioppo 1979;Brinol and Petty 2009)

when people are not sufficiently motivated or able to think deeply about the message (see the

dual-processing literature, e.g., Chaiken 1980;Petty and Cacioppo 1986; Chaiken 1987).

Second, social proximity increases the amount of thinking about high quality arguments

conveyed in a message, if thinking is not otherwise constrained (Petty and Cacioppo 1979;

Brinol and Petty 2009). This is especially true for common in-group membership (Mackie,

Worth, and Asuncion 1990), and more likely to occur when the message topic is group-related

(Van Knippenberg and Wilke 1992;Mackie, Gastardo-Conaco, and Skelly 1992) and when a

prototypical or representative group member delivers the message (Van Knippenberg, Lossie,

and Wilke 1994). Likewise, the amount of consideration will increase when the argument is

presented by multiple sources (i.e., traditional media source and social media recommender

Harkins and Petty 1987).

Study 1: Selectivity in Social Media

Our first study demonstrates the impact of social distance when selecting social media

news content. Here we operationalize social distance using behavioral measuresparticipants

actual interactions with their friends on Facebook (tie strength). We created a web applica-

tion that serves as an experimentally controlled copy of the Facebook news feed. Participants

saw content from their actual Facebook news feed (made possible by utilizing Facebooks

API), along with randomly assigned current news stories that we inserted in random order.

13


16/43

Each story appeared to have been shared by a friend from each participants actual social

network. We randomly assigned socially close and socially distant friends to recommend

each of these articles, effectively manipulating the social distance of the recommender. We

attained a measure of social distance between the participant and each of the participants

friends at the start of the study based on the number of mutual interactions on Facebook.

Design and Methods

Participants were recruited from various courses at a West Coast research university.

They were asked to use our proprietary Facebook application for a study on social media.

After granting permission, participants waited for a few seconds while we collected theirfriend list and previous six months worth of wall posts, newsfeed items, comments, and photo

tags. The application them summed the total interactions between the participant and the

participants friends to yield a social proximity measure. Two highly connected friends and

two less well-connected friends were then randomly drawn, and assigned to recommend

news stories that would appear in the participants newsfeed.

Participants were then shown their current news feed, with the addition of eight news

stories embedded at randomly selected points. External links were disabled to prevent dis-

traction. The eight stories were randomly assigned a highly connected friend recommender (2

stories), a less well-connected friend recommender (2 stories), and to a state in which either

the New York Times or Fox News logo was displayed instead of an actual friend (4 stories).

News stories were sourced from CNN, the New York Times, and the Wall Street Journal on

a variety of topics including politics, international news, sports, and entertainment. Figure

4presents a screenshot showing the screen layout for the experimental application, which

retains the look and feel of the Facebook news feed. All references to the storys source from

within the article were removed from the articles so as not to confound the source manipu-

lation. Participants could click on a story to read the full story in an iframe, a window that

14


17/43

opened within the current web page. A click on a story constitutes our dependent measure,

though it is important to note that participants could read the storys summary which was

embedded in the news feed. After the experiment, we issued a survey to respondents asking

about relevant socio-demographics and inquiring about the people who ostensibly recom-

mended each story in their news feed. Participants were fully debriefed at the conclusion of

the survey.

Figure 4: Racial Homophily in Selection of News Items in Facebook Application

A total of 183 students participated in the study, 34 of whom were removed from the

analysis because they guessed the manipulation.11 Participants comprised a relatively diverse

set of individuals, with 75 identifying as White, 16 as Black, 9 as Hispanic, 27 as Asian, 21

identified as being of another racial/ethnic group, and 1 did not list any such identification.Females were over-represented, with 90 Females to 59 males. The sample skewed politically

11To alert us as to when deception was detected, we asked participants Sometimes it is surprising whatour friends like. How surprised were you at the news stories your friends recommended? Those respondingNo way! My friends would have never recommended those stories were removed from the analysis.

15


18/43

Figure 5: Selection by Tie Strength and News Outlet

Weak Tie

Strong Tie

0.075 0.100 0.125 0.150 0.175

Mean Selection Rate, SE

Source

Selection by Tie Strength

left, with 64 identifying as Democrats, 19 as Republicans, 24 as Independents or Other, and

42 not providing any political affiliation.

Results

Social proximity between the viewer and the person who recommended the article affects

the selection decision. Figure 5 shows that recommendations from contacts with whom

interaction is relatively frequent (strong ties) were more likely to prompt participants to

select an article than recommendations from contacts with whom interaction was rare (weak

ties).12 A paired, two-sided t-test comparing each respondents probability of selecting

content recommended by strong versus weak ties reveals that this difference is significant

(t(148) = 2.287, p= 0.024).13

We also found some tendencies for people to select content recommended by others with

a similar ethnic background. Of course, homophily should be expected to govern the abso-

lute number of articles recommended from friends of various ethnic backgrounds that each

participant encountered, so we compare the selection ratethe number of articles selected

divided by the number of articles that each ethnic group recommended. With some ex-

12All error bars are standard errors. We take the proportion of articles selected per respondent per highand low recommenders to avoid exaggerating our confidence in these estimates by artificially inflating N.

13All tests employed here are two-sided.

16


19/43

ceptions, each group selected an approximately similar ratio of articles to read. Whites

selected at a marginally higher rate articles recommended by Black contacts (= .15) than

Asian contacts ( = .02, t(17.86) = 1.49, p = .16). However, Blacks selected at a lower

rate articles recommended by White contacts (none) compared to Black contacts (= .10,

t(15) = 2.18, p = .05), and similarly, Asians selected at a lower rate articles recommended

by Black contacts (none) than Asian contacts ( = .18, t(19) = 2.53, p = .02). Of course,

we did not manipulate ethnicity in Study 1 and so cannot make a causal claim about its

impact on selectivitywe document the causaleffect of common ethnic group membership

on selectivity in Study 2 below.

Study 2: How Group Membership Affects Selectivity

and Attitudes Toward Content

In this study, we experimentally manipulated recommender attributes independent of

the viewers actual friend network, in order to estimate how such attributes affect selection

and processing of content. This design was flexible enough to allow the use of national

survey sample. Our findings suggest that a recommenders socioeconomic attributes have

implications for how people select content and can shape attitudes on topically related issues

(in this case, on welfare) assessed after reading the article.

Design and Methods

Participants were drawn from the Survey Sampling International (SSI) panel to partic-

ipate in a web experiment and were told that they were testing a news browsing system

we developed. After completing some basic demographic questions, we asked participants

to browse news stories from four topics (U.S. news, health, sports, and entertainment) and

17


20/43

select stories to read later. For each topic, participants could select one of three stories in the

context of a minimalist social media interfaceaffording superior experimental controlas

depicted in Figure6,top. One of these stories contained a recommendation from a person

whose race, gender, and socioeconomic status were manipulated (as per Figure 6, below).

Recommenders were unknown to participants.

Figure 6: Minimalist Social Media Selection Interface

After selecting four stories, participants were asked to read a story that the system

recommended for them. All participants saw the same storya welfare piece written in

a tone typical of U.S. journalistic coverage, aspiring to neutrality. The system displayed a

recommender as depicted in Figure7. After reading the story, participants completed a bat-

tery of questions about their attitudes regarding the importance of welfare, their preferences

regarding welfare spending, and their attitudes about the negative consequences of welfare.

The stimulus images consisted of a head shot, clothing and background. Our socioe-

18


21/43

Figure 7: Minimalist Social Media Reading Interface

conomic status (SES) manipulation consisted of two conditions: the high SES condition

featured models dressed in business attire with a background designed to convey high SES

such as an office, while the low SES condition featured models dressed in t-shirts with a

background designed to convey low SES, such as a brick wall or graffiti mural. Our race and

gender manipulations were operationalized by using head shots of different models who were

either White or Black and male or female. We carefully matched these images on attrac-

tiveness so as not to confound the race manipulation.14 To maintain consistency, the same

outfits, backgrounds, and faces were used to generate the images for each of the 16 cells

(high/low SES, white/black and male/female). Control images consisted of White males

of high SES. We provide the stimulus images in Figure 8. Our treatment images for the

selection study are labelled as Set 1, while our treatment images for the welfare attitude

14We enlisted 225 Mechanical Turk workers to rate the images. The difference in attractiveness ratings (1-7scale) between Whites ( = 3.86, N= 113) and Blacks ( = 3.34, N = 112) was not significant (two-tailedt-test,t= 1.33, P > .10.

19


22/43

study are labeled as Set 2.

Figure 8: Stimuli for experimental manipulations of SES, gender and race

Participants consisted of 1162 respondents from across the country. Participants com-

prised a relatively diverse set of individuals, with 850 identifying as White, 130 as Black, 101as Hispanic, 51 as Asian, while 28 identified as from an other racial/ethnic group, and two

did not list any suchidentification. Females were slightly over-represented, with 622 Females

to 538 males. The sample leaned slightly to the political leftthough at a ratio similar to

voter registration figureswith 428 identifying as Democrats, 340 as Independents or Other,

285 as Republicans, and 85 not providing any political affiliation.

20


23/43

Results

Selectivity

First we turn to the question of whether the social distance operationalized in terms of

racial group membership and SES of a recommender affect the choices people make when

selecting content. Figure 9 (left) shows the impact of race and income homophily in the

selection of content.15 For each respondent, we take the mean selection rate for each recom-

mender, then compare the impact of each recommender race using this metric. We must of

course stratify our analysis by group membership, and we include only Whites and Blacks

in our analysis of racial group membership. Whites were significantly less likely to select

a story when it was recommended by poor Blacks ( = .28), compared to rich Whites

(= .33, t(918.5) = 2.40, p= .016) and poor Whites (= .32, t(1405.0) = 1.68, p= .092).

Conversely, Blacks were significantly less likely to select a news story when rich Whites rec-

ommended it ( = .31) compared to rich Blacks ( = .42, t(126.7) = 2.13, p = .035) but

not significantly so for poor Blacks ( = .41, t(140.7) = 2.09, p = .038). Whites were also

more likely to select stories that did not come recommended (Control in the figure), while

the opposite was true for Blackswe suspect this reflects a stronger sense of social identityamong Blacks, though it could also be due to simple differences in news consumption habits.

Turning to economic homophily, we find a similar pattern, shown in Figure 9 (right).

High income participants16 were more likely to select content that came from a rich White

recommender ( = .33) than a poor White ( = .26, t(236.6) = 2.11, p = .034) or poor

Black ( = .26, t(262.6) = 2.04, p = .041) recommender. Among low income respondents,

there is no clear pattern. We see a similar tendency among high income respondents to select

the article without a recommender.

15We pool over gender in this analysis as it proved inconsequential in shaping selection or informationprocessing. We also pool our White recommenders.

16We define high-income here as those in the top 33 percentile in the sample.

21


24/43

Figure 9: The Impact of Social Distance on Selection

White N = 850

Black N = 130

Poor, Black

Rich, Black

Poor, White

Rich, White

Control

Poor, Black

Rich, Black

Poor, White

Rich, White

Control

0.30 0.35 0.40 0.45


RaceofRecommender

Ethnic Group Membership

High Income N = 229

Low Income N = 931

Poor, Black

Rich, Black

Poor, White

Rich, White

Control

Poor, Black

Rich, Black

Poor, White

Rich, White

Control

0.25 0.30 0.35 0.40


IncomeofRecommender

Economic Strata

Stratifying on racial resentment reveals evidence that attitudes related to social distance

serve to strengthen its effects. Figure10shows the those with high racial resentment were

less likely to select a poor Black recommender ( = .27) compared to a poor White rec-

ommender ( = .36, t(891.10) = 2.78, p = 0.006). High-resentment respondents were also

more likely to select content from a rich White recommender (= .33, t(586.441) = 2.516,

p = 0.012). However, they were not significantly more likely to select content from a rich

Black recommender (= .32, t(893.79) = 1.51, p= 0.131).

Processing

We now turn to the question of how respondents process article content differently based

on the race and socio-economic status of the recommender. After reading the article at

the end of the experiment, we asked respondents a standard battery of questions related

to welfares social consequences and whether they support increasing or lowering current

levels of funding for welfare. We compute a social consequences index by taking the mean

of the four responses (after reversing relevant responses) and rescaling to 0-1 such that 0

22


25/43

Figure 10: The Impact of Racial Resentment on Selection

High Resentment N = 545

Low Resentment N = 615

Poor, Black

Poor, White

Rich, Black

Rich, White

Control

Poor, Black

Poor, White

Rich, Black

Rich, White

Control

0.250 0.275 0.300 0.325 0.350 0.375


RaceofRecommender

Racial Resentment

represents most negative consequences while 1 represents most positive. We also rescale our

welfare spending preference question similarly to 0-1, such that 0 represents a preference

for a significant decrease, while 1 represents a preference for a significant increase. As

shown in Figure11, there is evidence that recommender race affects judgments about the

social consequences of welfare and support for funding social welfare programs for Whites

and Blacks alike.

Among Whites, seeing a Black recommender with high socioeconomic statusa counter-

stereotypical exemplarwas more likely to elicit positive support for welfare (= .53) than

a poor White ( = .48, t(357.329) = 2.309, p = 0.022). This result is especially important

as it crosses the midpoint of the index. This exemplary Black recommender was also more

likely to be selected an article recommended from a poor Black recommender ( = .49)

to an extent that approaches statistical significance (t(365.728) = 1.511, p = 0.132). A

similar pattern holds for preferences regarding welfare spending. Among Whites, seeing

a Black recommender with high socioeconomic status resulted in a preference for greater

23


26/43

spending ( = .50) than seeing a poor White recommender ( = .43, t(359.391) = 2.579,

p = 0.010). This preference for greater spending compared to a poor Black recommender

was not significant, however (= .48, t(391.18) = 0.548, p= 0.584)

Figure 11: Participants Race and Welfare Attitudes

White N = 850

Black N = 130

Poor, Black

Rich, Black

Poor, White

Rich, White

Poor, Black

Rich, Black

Poor, White

Rich, White

0.45 0.50 0.55 0.60 0.65 0.70

Consquence index ( harmful, + helpful), SE

RaceofRecom

mender

Welfare's Social Consquences

White N = 850

Black N = 130

Poor, Black

Rich, Black

Poor, White

Rich, White

Poor, Black

Rich, Black

Poor, White

Rich, White

0.4 0.5 0.6 0.7

Spending Preference ( decrease, + increase), SE

RaceofRecom

mender

Welfare Spending Preferences

Racial and socioeconomic attributes matter in selection irrespective of content, even when

recommenders are strangers. Here, weve isolated the effect of these attributes independent

of story topic or content, as well as tie strength (by virtue of the fact that the recommenders

are strangers). Weve also established that patterns of selection among racially distinct rec-

ommenders are due to actual preferences for those races, not merely that racially similar

individuals happen to be generally more connected to each other. Perhaps more impor-

tantly, and somewhat paradoxically, this suggests that recommender attributes alone can

affect public opinion on racially charged issues. Specifically, when exposed to a counter-

stereotypical exemplar, Whites were more likely to give more positive evaluations of racially

charged social policies like welfare.

24


27/43

Discussion

We showed that the attributes of those who recommend content in social media affect how

readers select and process content. In particular, the social distance between the consumer

and the recommender, including tie strength, racial, and socio-economic similarity are key

factors when a consumer decides how to allocate attention and consume the vast array of

content embedded in social media. Our designs allow us to make a direct causal assertion

that social proximity, operationalized as communication frequency or tie strength (Study

1) and common group membership (Study 2) drive the selection of content, independent

of common interests or other sources of similarity/homophily. This study adds to growing

evidence that socially proximate contacts (strong ties) drive selection and play a key role in

the information diffusion process (Bakshy et al. 2012).

Furthermore, we demonstrated that patterns in content selection have direct implications

for attitudes about American politics. In particular, the social attributes of the storys

recommender interact with an individuals attitudes about policy, and affect stated policy

preferenceson social welfare policies in this case. We also showed the conditions under

which this tendency to select content from socially close recommenders can be magnified over

time as a byproduct of algorithms that integrate past selection preferences when determining

what to display to users.

If social distance affects how consumer find, select, and process content, the shift from

loyally frequenting traditional media outlets to browsing social media has tremendous conse-

quences for political communication. Agenda-setting theory posits that the media determine

the issues and events that enter public consciousness, and those that do not make media

coverage generally fall to the wayside (e.g., McCombs and Shaw 1972;McLeod, Becker, and

Byrnes 1974;Erbring, Goldenberg, and Miller 1980;Hill 1985); priming theory goes further,

suggesting that people make political decisions based on topical and/or social considerations

that are salient in the mass media (Krosnick and Kinder 1990; Gilliam and Iyengar 2000;

25


28/43

Mendelberg 2008;Valentino, Hutchings, and White 2002); on the other hand, indexing the-

ory posits that the press sometimes functions as a mere conduit for messages from officials,

especially in times of war (Bennett 1990;Cook 1994; Zaller and Chiu 1996;Entman 2003;

Slantchev 2006).

But when information flows between groups of people rather than from media companies

directly to consumers, these effects depend on the extent to which interpersonal informa-

tion flows mirror the medias agenda. If the medias agenda is distorted by interpersonal

information flows (as in the days before the broadcast era of political communication, see

two-step flow modelsKatz and Lazarsfeld 1955;Katz 1957) and selective consumption (an

increasingly common finding, see Klapper 1960; Sears and Freedman 1967; Iyengar et al.

2008;Iyengar and Hahn 2009; Stroud 2010; Messing and Westwood 2012), mass media ef-

fects on social media consumers will be the exception, not the rule (Bennett and Iyengar

2008). Indeed, scholars are already finding weaker aggregate and individual-level media ef-

fects, such as agenda-setting (Shehata and Stromback 2013). Our findings provide evidence

that agendas in the context of social media should be considered highly interpersonal, re-

sembling something more akin to the two-step flow model of political communication, rather

than mass media effects or priming (see alsoMessing and Westwood 2012;Mutz and Young

2011).

Because people are more likely to select content from socially close contacts, the likelihood

of exposure to attitude-inconsistent information in social media will be lower insofar as our

close friends have similar viewpoints. On the other hand, social media encourage users

to maintain a vast array of online relationships comprising of both socially proximate and

distant ties (Hampton et al. 2009), including work contacts with whom the potential forcross-cutting discourse that introduces counter-attitudinal information is substantially higher

(Mutz and Mondak 2006). These socially distant, weak ties are responsible for propagating

novel content that viewers would not otherwise encounter (Bakshy et al. 2012).

26


29/43

Nonetheless, our results open new questions related to technologys role in polarization

and fragmentation beyond exclusively partisan lines (Sunstein 2007; Slater 2007; Stroud

2010), but also along an array of existing social cleavages (as alluded to in Bennett and

Iyengar 2008). Our results also suggest that we should expect key media effectsagenda-

setting, priming, framing and indexingto occur differently in different socio-economic,

ethnic, and political strata. Furthermore, as the attributes of those who share the story

affect how we process that story and update our policy preferences, we might see additional

factionalization within socioeconomic/ethnic strataeven if all strata are consuming the same

content. Regardless of whether social media can recreate the inadvertent audience (Pasek,

More, and Romer 2009) and overcome partisan bias (Messing and Westwood 2012), our

findings suggest the potential for the social media ecosystem to fragment along existing

social cleavages.

There are several limitations to this research. The sample for study 1 was comprised of

college students and had disproportionate allocations of liberals and Whites. Moreover, the

results in study 1 suggest a general disinterest in news stories, or perhaps a lack of interest

in this study in particular, as only about half of the studys participants selected a single

news story to read. Study 2 required participants to select one of a number of news stories,

but this design may have encouraged satisficing. Of course, any such disinterest or deficit of

attention on behalf of participants during our studies compared to actual social media use

should be expected to obscure the true effect sizes, meaning this design may understate the

true effect size.

Nonetheless, our findings suggest that social distance is indeed a powerful force driving

news consumption and serves to privilege information shared by socially close friends at theexpense of heterogeneous contacts. Furthermore, our findings add to the evidence that in the

context of social media, factionalization will not only be driven by anticipated agreement,

but the mix of content we encounter (Sears and Freedman 1967;Leeper 2013), both of which

27


30/43

are driven by social distance. Our findings also suggest that the racial and socioeconomic

diversity of our social networks impact our policy preferences, by virtue of the fact that

the storys recommender affects both the importance we place on a story and our political

attitudes related to the storys content.

References

Adamic, Lada A., and Natalie Glance. 2005. The political blogosphere and the 2004 U.S.election: divided they blog. In Proceedings of the 3rd international workshop on Linkdiscovery. New York, NY, USA: ACM p. 3643.

Alba, Richard D., and Charles Kadushin. 1976. The Intersection of Social Circles A NewMeasure of Social Proximity in Networks. Sociological Methods & Research5(August):

77102.Allport, Gordon W. 1954. The Nature of Prejudice. Oxford, England: Addison-Wesley.

Althaus, Scott L., and David Tewksbury. 2002. Agenda setting and the new news: Patternsof issue importance among readers of the paper and online versions of the New YorkTimes.Communication Research29(2): 180207.

Antonakis, John, and Olaf Dalgas. 2009. Predicting Elections: Childs Play! Science323(5918): 1183.

Bailenson, J. N., S. Iyengar, N. Yee, and N. A. Collins. 2008. Facial Similarity between

Voters and Candidates Causes Influence. Public Opinion Quarterly.Bailenson, Jeremy N., Kim Swinth, Crystal Hoyt, Susan Persky, Alex Dimov, and Jim Blas-

covich. 2005. The Independent and Interactive Effects of Embodied-Agent Appearanceand Behavior on Self-Report, Cognitive, and Behavioral Markers of Copresence in Im-mersive Virtual Environments. Presence: Teleoperators & Virtual Environments14(4):379393.

Bakshy, Eytan, Itamar Rosenn, Cameron Marlow, and Lada Adamic. 2012. The Role ofSocial Networks in Information Diffusion. arXiv1201.4145(January).

Baum, Matthew A., and Tim Groeling. 2008. New Media and the Polarization of American

Political Discourse. Political Communication25(4): 345365.

Baumeister, Roy F. 1998. The Self. In The Handbook of Social Psychology, ed. Dan T.Gilbert, Susan T. Fiske, and Gordon Lindzey. New York: McGraw-Hill pp. 680740.

Bennett, W. Lance. 1990. Toward a Theory of Press-State Relations in the United States.Journal of Communication40(2): 103127.

28


31/43

Bennett, W. Lance, and Shanto Iyengar. 2008. A new era of minimal effects? The changingfoundations of political communication. Journal of Communication58(4): 707731.

Berelson, Bernard R., Paul F. Lazarsfeld, and William N. McPhee. 1954. Voting: A Studyof Opinion Formation in a Presidential Campaign. Chicago: University Of Chicago Press.

Berger, J., and M. Zelditch. 1985. Status, rewards, and influence. Jossey-Bass Inc Pub.

Bernstein, Michael, Eytan Bakshy, Moira Burke, and Brian Karrer. 2013. Quantifying theInvisible Audience in Social Networks. In Proceedings of CHI. Paris: .

Blair, Irene, and Christopher Jenkins. 2002. The role of Afrocentric features in person per-ception: Judging by features and categories.Journal of Personality & Social Psychology83(1): 524.

Bogardus, Emory S. 1925. Social Distance and Its Origins.Journal of Applied Sociology9: 216226.

Bond, Robert M., Christopher J. Fariss, Jason J. Jones, Adam D. I. Kramer, CameronMarlow, Jaime E. Settle, and James H. Fowler. 2012. A 61-million-person experiment insocial influence and political mobilization.Nature 489(September): 295298.

Bond, Selena, and Thomas F. Cash. 1992. Black beauty: Skin color and body images amongAfrican-American college women.Journal of Applied Social Psychology22: 874888.

Brinol, Pablo, and Richard E. Petty. 2009. Source factors in persuasion: A self-validationapproach.European Review of Social Psychology20(1): 4996.

Brock, T. 1965. Communicator-recipient similarity and decision change.Journal of Per-

sonality and Social Psychology 1: 650654.

Brown, Rupert, and Miles Hewstone. 2005. An Integrative Theory of Intergroup Contact.In Advances in Experimental Social Psychology, ed. Mark P. Zanna. Vol. Volume 37Academic Press pp. 255343.

Burger, Jerry M, Nicole Messian, Shebani Patel, Alicia del Prado, and Carmen Anderson.2004. What a Coincidence! The Effects of Incidental Similarity on Compliance. Per-sonality and Social Psychology Bulletin 30: 3543.

Chaiken, S. 1987. The heuristic model of persuasion. Social Influence5: 339.

Chaiken, S., A. Liberman, and A. H. Eagly. 1989. Heuristic and systematic informationprocessing within and beyond the persuasion context. Unintended thoughtp. 212252.

Chaiken, Shelly. 1980. Heuristic versus systematic information processing and the use ofsource versus message cues in persuasion. Journal of Personality and Social Psychology39(5): 752766.

29


32/43

Charness, Gary, and Uri Gneezy. 2008. Whats in a name? Anonymity and social dis-tance in dictator and ultimatum games. Journal of Economic Behavior & Organization68(October): 2935.

Conlisk, J. 1996. Why bounded rationality? Journal of economic literature34(2): 669700.

Cook, Timothy E. 1994. Domesticating a Crisis. In Taken by Storm: The Media, PublicOpinion, and U.S. Foreign Policy in the Gulf War, ed. W. Lance Bennet, and David L.Paletz. Chicago: University of Chicago Press pp. 105130.

Cunningham, Michael, Alan Roberts, Anita Barbee, Perri Druen, and Cheng-Huan Wu.1995. Their Ideas of Beauty Are, on the Whole, the Same as Ours: Consistency andVariability in the Cross-Cultural Perception of Female Physical Attractiveness. Journalof Personality & Social Psychology68(2): 261279.

Devine, P. G. 1989. Stereotypes and prejudice: Their automatic and controlled compo-nents.Journal of Personality and Social Psychology56(1): 518.

Dixon, Travis L., and Keith. B. Maddox. 2005. Skin tone, crime news, and social realityjudgments: Priming the stereotype of the dark and dangerous Black criminal. Journal ofApplied Social Psychology35: 15551570.

Dovidio, John F, Anja Eller, and Miles Hewstone. 2011. Improving Intergroup RelationsThrough Direct, Extended and Other Forms of Indirect Contact. Group Processes &Intergroup Relations 14(March): 147160.

Druckman, James N., Jordan Fein, and Thomas J. Leeper. 2012. A Source of Bias in PublicOpinion Stability. American Political Science Review106(02): 430454.

Eagly, A. H, R. O Ashmore, M. G Makhijani, and L. C Longo. 1991. What is beautiful isgood, but : A meta-analytic review of research on the physical attractiveness stereotype.Psychological Bulletin110: 107128.

Eberhardt, Jennifer, Phillip Goff, Valerie Purdie, and Paul Davies. 2004. Seeing Black:Race, crime, and visual processing.Journal of Personality and Social Psychology6: 876893.

Entman, Robert M. 2003. Projections of Power: Framing News, Public Opinion, and U.S.Foreign Policy. 1 ed. University Of Chicago Press.

Erbring, L., E. N Goldenberg, and A. H Miller. 1980. Front-page news and real-worldcues: A new look at agenda-setting by the media. American Journal of Political Sciencep. 1649.

Eveland, William P., Dhavan V. Shah, and Nojin Kwak. 2003. Assessing Causality in theCognitive Mediation Model A Panel Study of Motivations, Information Processing, andLearning During Campaign 2000.Communication Research30(August): 359386.

30


33/43

Facebook. 2010. f8 Live on Facebook. https://apps.facebook.com/feightlive/.

Facebook. 2011. Statistics. http://www.facebook.com/press/info.php?statistics.

Feldman, Stanley, and Leonie Huddy. 2005. Racial Resentment and White Oppositionto Race-Conscious Programs: Principles or Prejudice? American Journal of Political

Science 49(January): 168183.

Festinger, Leon. 1957. A Theory of Cognitive Dissonance. Evanston, IL: Row Peterson.

Fischer, Peter, Eva Jonas, Dieter Frey, and Stefan Schulz-Hardt. 2005. Selective Exposureto Information: the Impact of Information Limits.European Journal of Social Psychology35(4): 469492.

Fischer, Peter, Stefan Schulz-Hardt, and Dieter Frey. 2008. Selective Exposure and In-formation Quantity: How Different Information Quantities Moderate Decision MakersPreference for Consistent and Inconsistent Information.Journal of personality and social

psychology94(February): 231244.Gilliam, F. D., and S. Iyengar. 2000. Prime suspects: The influence of local television news

on the viewing public. American Journal of Political Science44(3): 560573.

Gilliam, Franklin D., Shanto Iyengar, Adam Simon, and Oliver Wright. 1996. Crime inBlack and White: The violent, scary world of local news. Harvard International Journalof Press/Politics1(3): 623.

Goel, Sharad, Winter Mason, and Duncan J. Watts. 2010. Real and Perceived AttitudeAgreement in Social Networks.Journal of Personality and Social Psychology99(4): 611621.

Graber, D. A. 1988. Processing the News: How People Tame the Information Tide. WhitePlains, NY: Longman.

Granovetter, Mark. 1973. The Strength of Weak Ties.The American Journal of Sociology78(6): 13601380.

Hampton, K. N., L. F. Sessions, E. J. Her, and L. Rainie. 2009. Social Isolation and NewTechnology. Pew Internet & American Life Project.

Harkins, Stephen G., and Richard E. Petty. 1987. Information utility and the multiplesource effect. Journal of Personality and Social Psychology52(2): 260268.

Heider, F. 1944. Social Perception and Phenomenal Causality. Psychological Review51(6):358374.

Henderson, Marlone D. 2009. Psychological Distance and Group Judgments: The Effectof Physical Distance on Beliefs about Common Goals.Personality and Social PsychologyBulletin35(October): 13301341.

31


34/43

Hill, D. B. 1985. Viewer characteristics and agenda setting by television news. PublicOpinion Quarterly49(3): 340.

Hoffman, Elizabeth, Kevin McCabe, and Vernon L. Smith. 1996. Social Distance andOther-Regarding Behavior in Dictator Games.The American Economic Review86(June):

653660.Homans, George Caspar. 1951. The Human Group. Transaction Publishers.

Hugenberg, Kurt, and Galen V Bodenhausen. 2004. Ambiguity in Social Categorization:The Role of Prejudice and Facial Affect in Race Categorization. Psychological Science15(5): 342345.

Iyengar, S., M. Banaji, and K. Hahn. 2009. Explicit and implicit attitudes: black-white andObama-McCain comparisons. Toronto, Canada: .

Iyengar, Shanto, and Kyu S. Hahn. 2009. Red media, blue media: Evidence of ideological

selectivity in media use. Journal of Communication59(1): 1939.

Iyengar, Shanto, K. S Hahn, J. A Krosnick, and J. Walker. 2008. Selective exposure to cam-paign communication: The role of anticipated agreement and issue public membership.The Journal of Politics70(01): 186200.

Kahneman, D. 2003. A perspective on judgment and choice: Mapping bounded rationality.American psychologist58(9): 697.

Karakayali, Nedim. 2009. Social Distance and Affective Orientations.Sociological Forum24(3): 538562.

Katz, Elihu. 1957. The two-step flow of communication: An up-to-date report on anhypothesis.Public Opinion Quarterly21(1): 6178.

Katz, Elihu, and Paul F Lazarsfeld. 1955. Personal Influence: Social Forces. New York: TheFree Press.

Kerbel, Matthew R, and Joel David Bloom. 2005. Blog for America and Civic Involvement.Harvard International Journal of Press/Politics10(4): 327.

Kincaid, Jason. 2010. EdgeRank: The Secret Sauce That Makes Facebooks News FeedTick. http://techcrunch.com/2010/04/22/facebook-edgerank/.

Kinder, Donald R, and David O Sears. 1981. Prejudice and Politics: Symbolic RacismVersus Racial Threats to the Good Life. Journal of Personality and Social Psychology40(3): 414431.

Klapper, Joseph T. 1960. The Effects of Mass Communication..

32


35/43

Klatzky, R., G. Martin, and R. Kane. 1982. Influence of social-category activation onprocessing of visual information.Social Cognition 1: 95109.

Krosnick, Jon A., and Donald R. Kinder. 1990. Altering the Foundations of Support for thePresident Through Priming. The American Political Science Review84(June): 497512.

Lazarsfeld, P. F., and R. K. Merton. 1954. Friendship as a Social Process: A Substantiveand Methodological Analysis. InFreedom and Control in Modern Society, ed. M. Berger.New York, NY: Van Nostrand pp. 1866.

Lazer, David, Brian Rubineau, Carol Chetkovich, Nancy Katz, and Michael Neblo. 2010.The Coevolution of Networks and Political Attitudes. Political Communication27(3):248274.

Leeper, Thomas J. 2013. The Informational Basis for Mass Polarization. Public OpinionQuarterly.

Lerman, Kristina. 2007. Social Information Processing in Social News Aggregation.arXiv0703087(March).

Lerner, R. M, and T. Moore. 1974. Sex and status effects on perception of physical attrac-tiveness.Psychological Reports34: 10471050.

Levendusky, Matthew. 2013. Partisan Media Exposure and Attitudes Toward the Opposi-tion. Political Communication.

Liu, Hugo. 2007. Social Network Profiles as Taste Performances. Journal of Computer-Mediated Communication13(1): 252275.

Loewenstein, George, and Deborah A. Small. 2007. The Scarecrow and the Tin Man:The Vicissitudes of Human Sympathy and Caring. Review of General Psychology11(2):112126.

Mackie, Diane M., Leila T. Worth, and Arlene G. Asuncion. 1990. Processing of PersuasiveIn-Group Messages. Journal of Personality and Social Psychology58(5): 812822.

Mackie, Diane M., M. Cecilia Gastardo-Conaco, and John J. Skelly. 1992. Knowledge of theAdvocated Position and the Processing of In-Group and Out-Group Persuasive Messages.Personality and Social Psychology Bulletin18(2): 145151.

Maddox, Keith B., and Stephanie A. Gray. 2002. Cognitive representations of Black Ameri-cans: Re-exploring the role of skin tone.Personality and Social Psychology Bulletin28(2):250259.

Maddux, J. E., and R. W. Rogers. 1980. Effects of source expertness, physical attractive-ness, and supporting arguments on persuasion: A case of brains over beauty. Journal ofPersonality and Social Psychology39(2): 235244.

33


36/43

McCombs, M. E, and D. L Shaw. 1972. The agenda-setting function of mass media.Publicopinion quarterly36(2): 176.

McLeod, J. M, L. B Becker, and J. E Byrnes. 1974. Another look at the agenda-settingfunction of the press. Communication Research1(2): 131.

McPherson, M., L. Smith-Lovin, and J. M Cook. 2001. Birds of a feather: homophily insocial networks.Annual Review of Sociology27: 415444.

Mendelberg, Tali. 2001. The race card: Campaign strategy, implicit messages, and the normof equality. Princeton, NJ: Princeton University Press.

Mendelberg, Tali. 2008. Racial Priming: Issues in Research Design and Interpretation.Perspectives on Politics6(01): 135140.

Messing, S., E. Plaut, and M. Jabon. 2010. Bias in the Flesh: Attack Ads and the Effectsof Visual Cues in the 2008 Presidential Campaign. In Proceedings of the Annual Meeting

of the International Communication Association, Singapore.

Messing, Solomon, and Sean J. Westwood. 2012. Selective Exposure in the Age of SocialMedia: Endorsements Trump Partisan Source Affiliation when Selecting Online NewsMedia.Communication Researchforthcoming.

Metzger, Miriam J. 2007. Making Sense of Credibility on the Web: Models for EvaluatingOnline Information and Recommendations for Future Research. Journal of the AmericanSociety for Information Science and Technology58(13): 20782091.

Milgram, Stanley. 1963. Behavioral Study of Obedience. Journal of Abnormal SocialPsychology67: 371378.

Mutz, D. C, and J. J Mondak. 2006. The Workplace as a Context for Cross-Cutting PoliticalDiscourse.Journal of Politics .

Mutz, Diana C, and Lori Young. 2011. Communication and Public Opinion Plus CaChange? Public Opinion Quarterly75(5): 10181044.

Newcomb, Theodore Mead. 1943. Personality & Social change; Attitude Formation in aStudent Community. New York: Dryden Press.

Nielsen. 2011. State of the Media: Social Media ReportQ3. http://www.nielsen.com/content/corporate/us/en/insights/reports-downloads/2011/social-media-report-q3.html.

Parise, Salvatore, Sara Kiesler, Lee Sproull, and Keith Waters. 1996. My partner is a real dog:cooperation with social agents. InProceedings of the 1996 ACM conference on Computersupported cooperative work. CSCW 96 New York, NY, USA: ACM p. 399408.

34


37/43

Pariser, Eli. 2011. The Filter Bubble: What the Internet Is Hiding from You. London:Penguin Press.

Park, Justin H, and Mark Schaller. 2005. Does Attitude Similarity Serve as a HeuristicCue for Kinship? Evidence of An Implicit Cognitive Association. Evolution and Human

Behavior26: 158170.Park, Robert Ezra. 1924. The Concept of Social Distance As Applied to the Study of Racial

Attitudes and Racial Relations. Journal of Applied Sociology 8: 339344.

Parks, Malcolm R. 2008. Encyclopedia of Human Relationships. In Handbook of Relation-ship Initiation, ed. T. Reis Harry, and Susan Sprecher. CRC Press.

Parks, Malcolm R. 2011. Boundary Conditions for the Application of Three Theoriesof Computer-Mediated Communication to MySpace. Journal of Communication61(4):557574.

Pasek, Josh, Eian More, and Dan Romer. 2009. Realizing the social Internet? Offlinesocial networking meets online civic engagement. Journal of Information Technology andPolitics.

Perlmutter, David D. 2008. Blogwars. First edition ed. Oxford University Press, USA.

Petty, R. E., and D. T. Wegener. 1999. The Elaboration Likelihood Model: Current statusand controversies. In Dual-process theories in social psychology, ed. Shelly Chaiken, andYaacov Trope. New York: Guilford Press p. 4172.

Petty, R. E., and J. T. Cacioppo. 1986. The elaboration likelihood model of persuasion.Advances in Experimental Social Psychology 19: 123205.

Petty, Richard E., and John T. Cacioppo. 1979. Issue Involvement Can Increase or DecreasePersuasion by Enhancing Message-Relevant Cognitive Responses. Journal of Personalityand Social Psychology37(10): 19151926.

Pew. 2011. Navigating News Online. http://alturl.com/a9ak2.

Pew. 2012. Internet Gains Most as Campaign News Source but Ca-ble TV Still Leads:Social Media Doubles, but Remains Limited.http://www.journalism.org/commentary backgrounder/social media doubles remains limited.

Plant, E. Ashby, Patricia G Devine, William T. L Cox, Corey Columb, Saul L Miller, JoannaGoplen, and B. Michelle Peruche. 2009. The Obama effect: Decreasing implicit prejudiceand stereotyping. Journal of Experimental Social PsychologyIn Press, Corrected Proof:.

Ridgeway, C. L. 1987. Nonverbal behavior, dominance, and the basis of status in taskgroups. American Sociological Review52(5): 683694.

35


38/43

Ronquillo, Jaclyn, Thomas F. Denson, Brian Lickel, Zhong-Lin Lu, Anirvan Nandy, andKeith B. Maddox. 2007. The effects of skin tone on race-related amygdala activity: AnfMRI investigation. Social Cognitive and Affective Neuroscience2(1): 3944.

Schaller, M., J. H. Park, and A. Mueller. 2003. Fear of the dark: Interactive effects of

beliefs about danger and ambient darkness on ethnic stereotypes. Personality and SocialPsychology Bulletin5: 637649.

Sears, D. O, and J. L Freedman. 1967. Selective exposure to information: A critical review.Public Opinion Quarterly31(2): 194.

Shehata, Adam, and Jesper Stromback. 2013. Not (Yet) a New Era of Minimal Effects:A Study of Agenda Setting at the Aggregate and Individual Levels. The InternationalJournal of Press/Politics.

Simon, H. A. 1972. Theories of Bounded Rationality.Decision and Organization 1: 161176.

Slantchev, Branislav L. 2006. Politicians, the Media, and Domestic Audience Costs. In-ternational Studies Quarterly50(2): 445477.

Slater, Michael D. 2007. Reinforcing Spirals: The Mutual Influence of Media Selectivityand Media Effects and Their Impact on Individual Behavior and Social Identity. Com-munication Theory17(3): 281303.

Sniderman, P. M., and P. E. Tetlock. 1986. Symbolic racism: Problems of motive attributionin political analysis.Journal of Social Issues42(2): 129150.

Sobieraj, Sarah, and Jeffrey Berry. 2011. From Incivility to Outrage: Political Discourse inBlogs, Talk Radio, and Cable News. Political Communication28(1): 1941.

Strodtbeck, F. L., R. M. James, and C. Hawkins. 1957. Social status in jury deliberations.American Sociological Review22(6): 713719.

Stroud, Natalie Jomini. 2010. Polarization and Partisan Selective Exposure. Journal ofCommunication60(3): 556576.

Sundar, S. Shyam, and Clifford Nass. 2000. Source Orientation in Human-Computer Inter-action Programmer, Networker, or Independent Social Actor. Communication Research27(6): 683703.

Sundar, S. Shyam, Silvia Knobloch-Westerwick, and Matthias R. Hastall. 2007. Newscues: Information scent and cognitive heuristics. Journal of the American Society forInformation Science and Technology58(February): 366378.

Sunstein, Cass R. 2007. Republic.com 2.0. Princeton University Press.

36


39/43

Tajfel, Henri, and John C. Turner. 1979. An Integrative Theory of Intergroup Conflict.In The Social Psychology of Intergroup Relations, ed. William G. Austin, and StephenWorchel. Vol. 33 Monterey, CA: Brooks-Cole pp. 3347.

Tajfel, Henri, M. G. Billig, R. P. Bundy, and Claude Flament. 1971. Social categorization

and intergroup behaviour.European Journal of Social Psychology1(2): 149178.Todorov, Alexander, Anesu N Mandisodza, Amir Goren, and Crystal C Hall. 2005. Infer-

ences of Competence from Faces Predict Election Outcomes. Science 308(5728): 16231626.

Turner, J. C. 1982. Towards a cognitive redefinition of the social group. In Social Identityand Intergroup Relations, ed. Henri Tajfel. Vol. 1 Cambridge: Cambridge University Presspp. 1540.

Tversky, A., and D. Kahneman. 1974. Judgment under Uncertainty: Heuristics and Biases.Science185(4157): 11241131.

Valentino, Nicholas A., Vincent L. Hutchings, and Ismail K. White. 2002. Cues that matter:How political ads prime racial attitudes during campaigns. American Political ScienceReview96(1): 7590.

Vallone, R. P, L. Ross, and M. R Lepper. 1985. The hostile media phenomenon: Biasedperception and perceptions of media bias in coverage of the Beirut massacre. Journal ofPersonality and Social Psychology49(3): 577585.

Van Knippenberg, Daan, and Henk Wilke. 1992. Prototypicality of Arguments and Con-formity to Ingroup Norms.European Journal of Social Psychology22(2): 141155.

Van Knippenberg, Daan, Nathalie Lossie, and Henk Wilke. 1994. In-Group Prototypicalityand Persuasion: Determinants of Heuristic and Systematic Message Processing. BritishJournal of Social Psychology33(3): 289300.

Yamakawa, Yoshinori, Ryota Kanai, Michikazu Matsumura, and Eiichi Naito. 2009. SocialDistance Evaluation in Human Parietal Cortex. PLoS ONE4(2): 4360.

Zajonc, Robert B. 1968. Attitudinal Effects of Mere Exposure. Journal of Personality andSocial Psychology9: 127.

Zajonc, Robert B. 2001. Mere Exposure: A Gateway to the Subliminal.Current Directionsin Psychological Science 10: 224228.

Zaller, John, and Dennis Chiu. 1996. Governments little helper: U.S. press coverage offoreign policy crises, 19451991. Political Communication13(4): 385405.

Zimbardo, P. 1969.The human choice: Individuation, reason, and order vs. Deindividuation,impulse and chaos. Vol. 17 University of Nebraska Press.

37


40/43

Appendix: Simulation Code

For each viewer, we generate the number of friends he or she has (nFriend negbin(r=2, p= .01) + 50) and the proportion of those friends who are socially close (or perhaps notsocially distant, propCloseFrnds logist(norm(mean = closeProp)) ). Then for eachfriend, we generate a dummy variable recording whether the friend is socially close to theviewer (isFriendClose bern(propClose)), a likelihood of posting or recommending con-tent (probPost unif(0, 1)2), and initialize the websites algorithmic propensity to displaycontent to the user from this particular friend (displayFriendProb =.5). Then, we simulate365 over-time instances of viewing the news feed (or say, visits to the website). We simulateposts from all of the viewers friends (posts pois(probPost)), and select 20 posts to appearas items in the viewers news feed (newsFeed), using thedisplayFriendProbvalue from eachposts author as sample probability weights. Then, we model which posts the viewer reads bypicking a number of posts to sample (nSelected binom(n= 20, p= .5)), and then sam-pling nSelected items from the newsFeed, using the selection effect (closeEffect = .

Friends that Matter: How Social Distance Affects Selection and Evaluation of Content in Social Media

Documents