Divided We Stand: Conference Presentation on the Relationship between Social Media and Polarized Partisan Politics

I recently had the wonderful opportunity to present some of my research at the 87th annual conference of the Indiana Academy of the Social Sciences. It was an amazing experience, and I was so thrilled that my fiancée and parents were able to come support me. The following is a shortened version of the paper I presented in a panel session alongside three other researchers who also discussed their own, very interesting, research projects:

Introduction

Evidence from numerous surveys over the years clearly shows that there is a growing chasm between Democrats and Republicans, not only ideologically, but also in terms of animosity.  At the same time, as we’ve traversed headlong into the digital age, social media use has increased exponentially.  While, according to more optimistic scholars, social media offers the promise of a public sphere where people with varying perspectives, ideologies, and ideas can engage in productive dialogue, the literature to date makes clear that the promise of a truly open public sphere is far from the reality of the online world.  Several researchers have found that people’s political behavior on social media is, on the whole, polarized, taking place within view-reaffirming echo-chambers. 

This behavior is encouraged by people’s desires to selectively expose themselves to information concordant with their worldview, especially when their worldviews are deeply held and ideologically extreme.   This tendency is itself born of the fundamental propensity of those who are most alike to have little social distance from one another and share a common social network, beliefs, culture, identity, and sources of information. 

Whether the highly polarized online world has contributed to the development of hyper-partisanship in American politics has been the subject of much speculation; however, little hard evidence connecting online political engagement with heightened partisan loyalty exists.  To contribute to our understanding of the role social media might play in shaping party identity and electoral outcomes, I analyzed post-election tracking survey data collected by the Pew Researcher Center during the 2010 midterm congressional election to test the hypothesis that there is an endogenous relationship between online political engagement and ideologically extreme views (by ideologically extreme views I mean views that are farther to the left or right from the ideological center), and to answer the following research question:  does online political engagement impact the likelihood of partisan identity and partisan voting?  In other words, does engagement in online political activity make a person more likely to identify as a Republican or Democrat?  And, does such activity make a person more likely to vote for the party that most aligns with their political ideology? 

The Pew survey I used contained details, not only about how respondents voted in their congressional districts, but also about their engagement with politics online and on social media, their party identification, their political ideology, and their demographic information. 

To analyze Pew’s survey data, I utilized binomial logistic regression (logit) using a quasibinomial dispersion parameter to account for the effect of Pew’s probability case weights on the binomial distribution of my dependent variables.  I also used ordinal logistic regression (ordered logit) rather than ordinary least squares (OLS) to model the effect of covariates on ideological intensity (measured on an ordinal scale from 0 to 2) since using OLS on a non-interval dependent variable would yield unreliable results.

Findings

The below figure displays a graphical representation of the plotted regression coefficients for five logit models where the dependent variable for each was one of five binary measures of online political activity:  1) whether an individual had ever shared political content online (“share”); 2) whether an individual had ever engaged in online discussions about political issues or candidates (“talk”); 3) whether an individual was a friend of a political group, organization, or candidate on social media (“friend”); 4) whether an individual had ever posted political content on social media (“post”); and 5) whether an individual had ever joined some sort of political group online (“join”).  Each point on the plot represents the estimated model coefficient for each covariate while the lines represent one (the thick line) and two (the thin line) standard deviations of the estimate (or the 95% confidence interval).  If the 95% confidence interval for a coefficient does not cross zero (denoted by the line in the middle), the result is statistically significant.  Positive points, of course, represent coefficients with positive values while negative points represent coefficients with negative values.

Figure 1.

Effect of Ideological Intensity on Online Political Engagement

The results offer support for the hypothesis that ideological intensity is positively associated with online political engagement; however, the statistical significance of the results depends on how online political engagement is measured.  Controlling for age, gender, race (whether the respondent identified as Hispanic had to be dropped due to singularities), education, employment, and income, greater ideological intensity (0 = moderate; 1 = conservative/liberal; 2 = very conservative/very liberal) had a significant positive effect on the likelihood of sharing political content and of having online political discussions.  Meanwhile, ideological intensity’s positive effect on the likelihood of being a friend of an online political group and of posting political content on social media only approached significance at p < 0.1.  Moreover, ideology failed to have even a moderately significant effect on the likelihood of joining an online political group.

The effect of ideology also varied between respondents who identified as moderate to very conservative and respondents who identified as moderate to very liberal.  Figure 2 displays the plotted coefficients for five logit models where the cases included in the analysis were restricted to SNS users who reported having either moderate (0), conservative (1), or very conservative (2) views.  The results show that greater levels of conservative ideology among those identified along the moderate to very conservative spectrum significantly increased the likelihood of engaging in online political discussions, being a friend of an online political group, and posting political content of social media, ceteris paribus.  However, greater conservative ideology among moderates to conservatives only had a moderately significant positive effect on the likelihood of sharing political content (p < 0.1) and an insignificant impact on joining an online political group.

Figure 2. 

Effect of Ideological Intensity on Online Political Engagement among Moderate to Very Conservative Respondents

Among SNS users along the moderate to very liberal spectrum, the effect of more liberal ideology had a less pronounced overall effect on the likelihood of online political engagement compared to more conservative ideology.  The results depicted in Figure 3 show the estimates from five logit models where the effect of ideological intensity on online political engagement was modeled only for cases where respondents either identified as moderate (0), liberal (1), or very liberal (2).  Ideological intensity’s impact among moderate to very liberal respondents had only a significant positive effect on the likelihood of engaging in online political discussion and posting political content.  Meanwhile, more liberal views failed to have a statistically significant relationship with sharing political content, being a friend of an online political group, and joining an online political group.

Figure 3.

Effect of Ideological Intensity on Online Political Engagement among Moderate to Very Liberal Respondents

Aside from ideology having an impact on online political engagement, as already discussed, there are good theoretical reasons for thinking that online political engagement might reinforce and intensify ideology.  The results from five ordered logit models depicted in Figure 4 offer strong support for this hypothesis.  Each of the five measures of online political engagement had a statistically significant and positive effect on the likelihood of greater levels of ideological intensity, all else being equal.  However, again, like in the case of ideology’s effect on online political engagement among moderate to conservative versus moderate to liberal respondents, the effect of online political engagement varied depending on whether the ideological intensity being measured was conservative or liberal; although, the difference between groups was not nearly as pronounced.

Figure 4.

Effect of Online Political Engagement on the Likelihood of Having Farther from Center Ideological Views

Figure 5 displays the results from five ordered logit models where the dependent variable was an ordinal multinomial identifier of conservative ideological intensity that ranged in order from “moderate,” to “conservative,” to “very conservative.”  Each of the measures of online political engagement had a positive effect on the likelihood of greater reported levels of conservative ideology, ceteris paribus.

Figure 5.

Effect of Online Political Engagement on the Likelihood of Having More Conservative Ideological Views among Moderate to Very Conservative Respondents

Meanwhile, the results depicted in Figure 6 from five ordered logit models where the dependent variable was an ordinal identifier of liberal ideological intensity (that ranged from “moderate”, to “liberal,” to “very liberal”) indicate that online political engagement’s effect on liberal ideology among moderate to liberal respondents was only positively significant when online political engagement was measured by whether a respondent had engaged in online political discussion, posted political content on social media, and joined a political group online.  Conversely, the effect of being a friend of an online political group only had a moderately significant positive effect (p < 0.1) while sharing political content online had no significant impact.

Figure 6.

Effect of Online Political Engagement on the Likelihood of Having More Liberal Ideological Views among Moderate to Very Liberal Respondents

Turning to the relationship between ideology and online political engagement with partisanship, results from 11 logit models depicted in Figure 7 indicate that ideological intensity’s impact on the likelihood of identifying with either the Republican or Democratic parties, as well as the impact of online political engagement was positive and statistically significant.

Figure 7.

The Main Effect of Ideological Intensity & Online Political Engagement, along with The Effect of Their Interaction, on the Likelihood of Party Identification

As for the effect of online political engagement, the results largely support the view that such behavior, though it is positively associated with ideological intensity, cannot fully explain partisanship.  There is, however, one exception.  Controlling for the interaction of ideological intensity and sharing political content online, the main effect of sharing political content was negative and statistically significant, suggesting that among SNS users, sharing political content online made a respondent less likely to identify with either the Republican or Democratic parties.

Figure 8.

The Main Effect of Conservative Ideological Intensity & Online Political Engagement, along with the Effect of Their Interaction, on the Likelihood of Republican Party Identification among Moderate to Very Conservative Respondents

The positive effect of ideology on party identification was not a unique feature of those with either leftwing or rightwing leanings.  Results shown in Figure 8 indicate a statistically significant and positive association between conservative ideology and Republican Party identification among moderate to very conservative respondents.  Likewise, the results in Figure 9 show that higher levels of liberal ideology are positively associated with Democratic Party identification among moderate to very liberal respondents. 

Figure 9.

The Main Effect of Liberal Ideological Intensity & Online Political Engagement, along with the Effect of Their Interaction, on the Likelihood of Democratic Party Identification among Moderate to Very Liberal Respondents

However, returning to moderate to very conservative respondents, when controlling for the main effect of conservative ideology and the interaction of online political discussion and ideology, the results show a negative statistically significant association between the main effect of online political discussion and Republican Party identification.  Meanwhile, among moderate to very liberal respondents, the main effect of sharing political content, when holding the main effect of liberal ideology and the interaction of liberal ideology and sharing political content online at 0, was negative and statistically significant.

Regarding the effect of party identification and ideology on partisan voting behavior, the results depicted in Figure 10 show the estimates from two logit models where, for the first, the dependent variable was a binary indicator for whether the respondent cast their vote for the Republican Congressional candidate running in their Congressional district during the 2010 midterm election and where, for the second, the dependent variable was a binary indicator for whether the respondent cast their vote for the Democratic Congressional candidate running in their Congressional district during the 2010 midterm election.  Both models were estimated for all cases reported in the survey.  Party identification was measured via three dummy variables where each was coded as 1 if a respondent identified with the Republican Party, the Democratic Party, or as an Independent.  Ideology was measured on a scale from 0 to 4 where 0 = very conservative and 4 = very liberal.  The results show, both clearly and unsurprisingly, that party identification and ideology positively predict the partisan vote, ceteris paribus.

Figure 10.

Effect of Party Identification and Ideology on Partisan Voting

However, when the effect of ideology is interacted with online political engagement, the results become far more interesting.  Figure 11 displays the results from 11 logit models where the dependent variable was a binary indicator for whether a respondent voted for the Republican Congressional candidate in their district and where cases were restricted to respondents both who identified as either moderate, conservative, or very conservative and who reported having used a SNS.  The results show a positive relationship between more conservative views and the Republican vote among moderate to very conservative respondents.  Meanwhile, the main effect of online political engagement along with the interaction of conservative ideology with online political engagement failed to have a significant impact across all models. 

Figure 11.

Effect of Conservative Ideology, Republican Party Identification, & Online Political Engagement, along with the Effect of the Interaction of Conservative Ideology with Online Political Engagement on Voting for the Republican Congressional Nominee in One's Congressional District among Moderate to Very Conservative Respondents

However, among moderate to very liberal respondents, the results were much different.  Higher levels of liberal ideology failed to have a statistically significant impact on the likelihood of voting Democrat among moderate to very liberal respondents who reported having used an SNS in all but one of the models (model 8) depicted in Figure 12.  In this particular model both the main effects of liberal ideology and of engaging in online political discussion, along with their interaction, were estimated while controlling for party identification as well as demographic control variables.  For this model, the interaction of liberal ideology and online political discussion, along with the main effect of liberal ideology, had a significant effect.  The results show that, among moderate to very liberal respondents, conditional on liberal ideology > 0, engagement in online political discussion decreased the likelihood of voting for the Democratic Congressional candidate running in a respondent’s district.  Meanwhile, for another model (model 4) the results show that being a friend of an online political group had a significant and positive association with voting for the Democratic Congressional candidate running in a respondent’s Congressional district.  Furthermore, in yet another model, the interaction of liberal ideology with joining an online political group had a significant and negative effect on the likelihood of voting Democrat, suggesting that among moderate to very liberal SNS users, conditional on liberal ideology > 0, joining an online political group had a negative effect on the likelihood of voting Democrat.

Figure 12.

Effect of Liberal Ideology, Democratic Party Identification, & Online Political Engagement, along with the Effect of the Interaction of Liberal Ideology with Online Political Engagement on Voting for the Democratic Congressional Nominee in One's Congressional District among Moderate to Very Liberal Respondents

Discussion & Conclusion

The proclivity to avoid ideas that conflict with one’s own is not a new phenomenon, but with the proliferation of social media, the ability to contact and connect with likeminded people has increased substantially.  Such being the case, social media, for many people, has failed to be the public sphere it was much touted to be and it has instead become an echo-chamber, facilitating the fortification and intensification of individuals’ world views.  Whether this highly polarized online world has contributed to the development of hyper-partisanship in the electorate, however, has remained more a matter of speculation than hard evidence.

Analyses of people’s engagement with political issues online and on social media largely support the view that social media functions for many users as a means for seeking out view-affirming information, and as the electorate in the U.S. grows more and more ideologically polarized, many have theorized that the rise of social media is partially to blame.  Evidence from my own analysis corroborates the view that ideological intensity and online political engagement may have a positive and mutually reinforcing relationship.  However, my results provide mixed findings for the effect of online political engagement on partisanship and the partisan vote.  Where ideological intensity largely succeeds in explaining variance in party identification and the partisan vote, online political engagement largely fails.  Thus, while online political engagement may reinforce and enhance ideologically extreme views, such activity is not a reliable predictor of party identification and the partisan vote.  However, there were some exceptions in my findings worth noting. 

Some types of online political engagement did have a significant association (sometimes positive; sometimes negative) with party identification and partisan voting (under certain circumstances).  For example, the interaction of being a friend of an online political group with ideological intensity had a positive impact on the likelihood of party identification in general, while, more narrowly, the main effect of engaging in online political discussion among moderate to very conservative respondents had a negative effect on Republican Party identification.  Moreover, the main effect of sharing online political content had a negative effect on the likelihood of Democratic Party identification among moderate to very liberal respondents.  Furthermore, regarding the partisan vote, while online political engagement, along with its interaction with conservative ideology, failed to have a significant effect on the Republican vote among moderate to very conservative respondents, among moderate to very liberal respondents, conditional on liberal ideology > 0, engaging in online political discussion and joining an online political group were negatively associated with the Democratic vote.  Meanwhile, the main effect of being a friend of an online political group or candidate had a positive association with the likelihood of voting Democrat among moderate to very liberal respondents. 

More analysis needs to be done to further examine these findings; however, one notable pattern does seem to arise from these results:  sharing political content, engaging in online political discussion, and joining an online political group (what could be called more “active” types of online behavior) had a negative effect on party membership and the partisan vote, while simply “friending” a political party, candidate, or some sort of political group (something that might be considered a more “passive” type of online behavior) had a positive effect on party identification conditional on a respondent having non-centrist views.  Though these findings were by no means the norm in my analysis, they do challenge the idea that online political engagement (particularly active engagement) should have, if any effect on party identity and partisan voting, a promotive impact on partisan commitment.  In fact, these results suggest that in some instances online political engagement may not only harden partisanship, but also soften it—a truly surprising finding indeed given the positive endogenous relationship I found between online political engagement and greater levels of conservative and liberal ideology. 

These results may further suggest that moderate to very liberal respondents’ voting behavior is more so predicted by online political engagement than by ideology; whereas, conversely, moderate to conservative respondents’ is more so predicted by ideology than by online political engagement.  What does this finding mean?  It is hard to say.  The results show that among all survey respondents, going from very conservative to conservative, and from conservative to moderate, and so on, significantly increases the mean of the logged-odds that a respondent voted Democrat in the 2010 midterm elections by 0.69, all else being equal.  With only an analysis from one survey from one midterm election as a guide, to suggest that the effect of ideology somehow becomes less predictive once one crosses the boundary from moderate to liberal is a stretch.  More work needs to be done to parse out exactly what these results ultimately say about whether there is meaningful variance in the impact of leftwing versus rightwing ideology, along with variance in the impact of online political engagement, on predicting partisan behavior among moderate to very liberal and among moderate to very conservative voters.  It may very well be the case that liberal ideology’s impact was merely robbed of its significance by Democratic Party identification—a significant and positive predictor the Democratic vote.  The unique impact of online political engagement among moderate to very liberal respondents, however, is an interesting finding worth further research.

The use of more recent surveys from Congressional, as well as Presidential, elections are an obvious next step in examining the relationship between online political behavior and partisan and electoral outcomes.  More recent surveys have continued to show a growing rift between Republicans and Democrats and a continued (and also growing) presence of strong partisan divisions online.  Meanwhile, use of social media has increased substantially over the past few years.  It may be that social media’s role is much more direct now than it was during 2010 when the survey I used for my analysis was conducted. 

Alternative methods for examining the endogenous relationship between ideology and online political engagement should also be considered.  Though my own analysis is suggestive of a two-way relationship between ideology and online political engagement, the use of simultaneous equations models might better capture and account for endogeneity.  Moreover, methods of path analysis, such as structural equation models, would allow for a more rigorous test of the hypothesis that online political engagement has an indirect effect on partisan identification and voting behavior and that this effect is mediated by ideological intensity. 

In sum, the question of whether social media has played a role in the contemporary development of hyper-partisan politics will remain an important area of study for students of American politics for years to come.  With geographical space no longer a limiting factor, people across the U.S. with similar world views, religious beliefs, and political orientations can more easily communicate with one another and mutually reinforce one another’s ideas while simultaneously isolating themselves from discordant points of view.  This phenomenon has likely played no small role in hastening the ideological sorting of the American electorate; yet, while the findings from my own analysis bring us closer to a fuller understanding of the relationship between social media and the intensity of individuals’ ideological views, they raise more questions than they offer answers regarding the impact of online political engagement on partisan identity and voting behavior, meaning there is still more work to be done before our understanding of the role of social media in partisan politics is complete.

Works Cited in Original Paper

Baldassarri, Delia and Andrew Gelman.  2008.  “Partisans without Constrain:  Political    Polarization and Trends in American Public Opinion.”  American Journal of Sociology 114(2):  408-446.

Barbera, Pablo.  2015.  “Birds of the Same Feather Tweet Together:  Bayesian Ideal Point   Estimation Using Titter Data.”  Political Analysis 23(1):  76-91.

Boutet, Antoine, Hyoungshick Kim, and Eiko Yoneki.  2012.  “What’s in Your Tweets?  I Know Who You Supported in the UK 2010 General Election.”   Sixth International AAAI Conference on Weblogs and Social Media.

Colleoni, Elanor, Alessandro Rozza, and Adam Arvidsson.  2014.  “Echo Chamber or Public      Sphere?  Predicting Political Orientation and Measuring Political Homophily in Twitter Using Big Data.”  Journal of Communication 64(2):  317-332.

Dahlgren, Peter.  2005.  “The Internet, Public Spheres, and Political Communication:  Dispersion and Deliberation.”  Political Communication 22(2):  147-162.

De Zúñiga, Homero Gil, Eulàlia Puig-I-Abril, and Hernando Rojas.  2009.  “Weblogs,        Traditional Sources Online and Political Participation:  An Assessment of How the Internet Is Changing the Political Environment.”  New Media & Society 11(4):  553-574.

Doherty, Carroll.  2014.  “7 things to know about polarization in America.”  Pew Research      Center.  June 12.  http://www.pewresearch.org/fact-tank/2014/06/12/7-things-to-know-about-polarization-in-america/ (February 8, 2016).

Douglas, Andrew.  2013.  “Moderates Nearly Extinct in the U.S. House.”  FairVote.  August 7.   http://www.fairvote.org/moderates-nearly-extinct-in-the-u-s-house (March 26, 2016).

Erikson, Robert S. and Gerald C. Wright.  2013.  “Voters, Candidates, and Issues in   Congressional Elections.”  In Congress Reconsidered.  10^th ed.  Eds. Lawrence C. Dodd and Bruce I. Oppenheimer.  Thousand Oaks:  CQ Press.

Feller, Albert, Matthias Kuhnert, Timm O. Sprenger, Isabell M. Welpe.  2011.  “Divided They Tweet:  The Network Structure of Political Microbloggers and Discussion Topics.”  Fifth International AAAI Conference on Weblogs and Social Media.

Garzia, Diego.  2012.  “The Rise of Party/Leader Identification in Western Europe.”  Political   Research Quarterly 20(10):  1-12.

Gruzd, Anatoliy and Jeffrey Roy.  2014.  “Investigating Political Polarization on Twitter:  A    Canadian Perspective.”  Policy & Internet 6(1):  28-45.

Hart, William Dolores Albarracín, Alice H. Eagly, Inge Brechan, Matthew J. Lindberg, and Lisa Merrill.  2009.  “Feeling Validated Versus Being Correct:  A Meta-Analysis of Selective Exposure to Information.”  Psychological Bulletin 135(4):  555-588.

Jacobson, Gary C.  2013.  “Partisanship, Money, and Competition:  Elections and the Transformation of Congress since the 1970s.” In Congress Reconsidered.  10^th ed. Eds. Lawrence C. Dodd and Bruce I. Oppenheimer.  Thousand Oaks:  CQ Press.

Kayser, Mark Andreas and Christopher Wlezien.  2011.  “Performance Pressure:  Patterns of Partisanship and the Economic Vote.”  European Journal of Political Research 50(3):  365-394.

Kennedy, Dane.  2016.  “Congressional Partisanship in Historical Perspective.”  AHA Today: A Blog of the American Historical Association.  February 3.  http://blog.historians.org/2016/02/congressional-partisanship-in-historical-perspective/  (February 8, 2016).

Kaufman, Ted.  2015.  “Hyper partisanship and ignorance will do us in.”  delawareonline.com, March 9.  http://www.delawareonline.com/story/opinion/columnists/ted-kaufman/2015/03/29/hyperpartisanship-ignorance-will-us/70630176/ (accessed February 8, 2016).

Kushin, Matthew J. and Kelin Kitchener.  2009.  “Getting Political On Social Network Sites:  Exploring Online Political Discourse on Facebook.”  First Monday 14(2):  n.p.            http://journals.uic.edu/ojs/index.php/fm/article/view/2645 (February 8, 2016).

McPherson, Miller, Lynn Smith-Lovin, and James M. Cook.  2001.  “Birds of a Feather:    Homophily in Social Networks.”  Annual Review of Sociology 27:  415-444.

Pew Research Center.  2015.  “A Deep Dive into Party Affiliation.”  http://www.people-     press.org/2015/04/07/a-deep-dive-into-party-affiliation/ (March 26, 2016).

Pew Research Center.  “Politics Fact Sheet.”  http://www.pewinternet.org/fact-sheets/politics-      fact-sheet/ (February 8, 2016).

Pew Research Center’s Internet & American Life Project.  2010.  “Post-Election Tracking Survey 2010.”  November 30.  http://www.pewinternet.org/datasets/november-2010-post-election/.

Schmitt, Hermann.  2002.  “Partisanship in Western Europe and the US:  Causes and Consequences.”  Annual Meeting of the American Political Science Association, Boston, MA.  August 28 to September 1.  Presentation.

Stroud, Natalie Jomini.  2010.  “Polarization and Partisan Selective Exposure.”  Journal of           Communication 60(3):  556-576.

--.  2007.  “Media Use and Political Predispositions:  Revisiting the Concept of Selective  Exposure.”  Political Behavior 30(3):  341-366.