What Drives Politicians' Online Popularity? An Analysis of the 2010 U.S. Midterm Elections

ABSTRACT

The number of Web site visits, Facebook friends, or Twitter followers that politicians attract varies greatly, but little is known about what drives politicians' online popularity. In this article, we use data from a systematic tracking of congressional candidates' popularity on four Web platforms in the 112 most competitive congressional districts in the 2010 U.S. midterm elections to address that question. Using multivariate regression models, we show that while district-level socioeconomic characteristics have little effect on candidates' online popularity, challengers and candidates in open-seat races tend to attract larger audiences online, as do candidates who are more visible on political blogs. Surprisingly, how intensely candidates are covered in news media, how popular they are in opinion polls, and how much money they spend during the campaign show no significant effect. These findings help us understand the dynamics of Internet politics, and they have wider implications for candidate competition and party politics.

Keywords:

• Online campaigning
• online engagement
• political blogs
• social media
• U.S. congressional elections

Acknowledgments

The authors would like to thank David Karpf, Sandra Gonzales-Bailon, and the participants in our panels at the Midwest Political Science Association's 2012 Annual Meeting for their comments and critique.

Notes

1. Data analyzed in this study will be made available at Dataverse (http://dvn.iq.harvard.edu/dvn/dv/jitp) six months after publication. The data will be provided immediately for replication purposes upon request.

2. The Congressional Districts included are as follows: Alabama 2 and 5; Arkansas 1, 2, 3, and 4; Arizona 1, 5, and 8; California 3, 11, 18, 44, 45, and 47; Colorado 3, 4, and 7; Connecticut 4 and 5; Delaware at-large; Florida 2, 8, 12, 16, 22, 24, and 25; Georgia 8 and 12; Hawaii 1; Iowa 1 and 3; Idaho 1; Illinois 8, 10, 11, and 14; Indiana 2, 8, and 9; Kansas 3 and 4; Kentucky 3 and 6; Louisiana 2 and 3; Maryland 1; Massachusetts 10; Michigan 1, 7, and 9; Minnesota 1 and 6; Mississippi 1; Missouri 4; Nevada 3; North Carolina 2, 8, and 11; North Dakota at-large; Nebraska 2; New Hampshire 1 and 2; New Jersey 3, 7, and 12; New Mexico 1 and 2; New York 1, 13, 19, 20, 23, 24, 25, and 29; Ohio 1, 6, 13, 15, 16, and 18; Oregon 5; Pennsylvania 3, 4, 6, 7, 8, 10, 11, 12, 15, and 17; South Carolina 5; South Dakota at-large; Tennessee 4, 6, and 8; Texas 17 and 23; Virginia 2, 5, 9, and 11; Washington 3 and 8; Wisconsin 3, 7, and 8; West Virginia 1 and 3.

3. http://www.compete.com (accessed September 26, 2011). Estimates of site traffic were usually available in the following month, so we completed the data collection for this variable in December 2010.

4. Two methodological limitations of the Web site traffic data must be acknowledged. First, Compete combines a randomly selected national sample with data from Internet service providers. Because the latter often come from local cable providers, they tend to be concentrated in some geographic areas, which may over-represent some regions at the expense of others. Second, monthly unique visitors is a less precise audience metric than other figures such as page views, visits, or time spent on the site.

5. YouTube video views may be slightly inflated by the fact that the system discounts multiple views of the same videos for users who are registered with the site and log in with their credentials before watching, but does not do so for users who are not registered or who have not logged in during their session.

6. We coded official candidate Web sites for the following 13 functions: donate money, volunteer, find events, organize events, forward the page to friends, mobile phone interaction, share or link the Web site on any social network, “donate your status” on any social network, find your polling place, register to vote, link to Facebook page, link to Twitter profile, link to YouTube channel. The variable measures the number of these functions each Web site offered, so values range from 0 to 13. The empirical distribution ranged from 2 to 11, and the median candidate had 6 functions on his/her Web site.

7. The top liberal and conservative political blogs were identified based on the November 2010 rankings of David Karpf's Blogosphere Authority Index. The liberal blogs are, respectively, The Huffington Post, DailyKos, Talking Points Memo, Firedoglake, and Atrios, and the conservative blogs are Hot Air, Big Government, Townhall, American Thinker, and Ace Of Spades HQ. See http://www.blogosphereauthorityindex.com/default.asp?archive=bai_Nov212010.mdb (accessed on February 3, 2012). For each blog, we ran advanced searches with Google with each candidate's first and last name and the word “Congress” as the text search string and the blog's URL as the domain; the search was limited to the period January 1 to November 1, 2010. Because the average number of mentions for any candidate was greater among liberal than conservative blogs due to the differences in the amount of contents published on these particular blogs, we standardized the values of mentions for liberal and for conservative blogs. The variable is thus the sum of the standardized number of “hits” that Google returned for each candidate's name on all five liberal and all five conservative blogs.

8. For each candidate, we ran two separate Lexis-Nexis searches—one with The Associated Press Online set as source and the other with The Associated Press Local & State Wire. In both searches, the text search string included the candidate's first and last names and the word “Congress,” and the time frame was set from January 1 to November 1, 2010. Because the distribution of the variables was highly skewed, we standardized their values before entering them in the regression models.

9. The fact that these numbers are higher than those for Facebook and Twitter does not necessarily mean that candidates attract substantially more attention on YouTube compared to other online campaign channels. While one can only “like” or “follow” a candidate once on Facebook and Twitter (and can stop doing so), and while our measure of Web site traffic involves unique monthly users, the total number of views on a candidate's YouTube channel is cumulative over time and additive across all videos uploaded. As most campaigns host dozens of videos on their YouTube channels, the total views of videos on their channels are in all likelihood considerably higher than the number of unique video viewers. A measure of the latter is, unfortunately, not available to the public, and is only poorly represented by the number of channel subscribers, as very few people subscribe to YouTube channels. (In November, the mean of subscribers was 61 and the median 14 across all candidates.)

10. Regression models including only district-level variables failed to achieve adjusted R² coefficients above .1 for all dependent variables apart from Web site traffic, where the coefficient was .126, still much smaller than in the complete model.

11. This finding may be at least partially related to the fact that mentions on Associated Press national and local wires may not be a perfect indicator of media coverage of Congressional candidates. To verify the validity and reliability of our indicators, we thus collected additional data on mentions of each candidate on American newspapers, broadcast news, and cable news using the Lexis-Nexis database with the same search strings and time limits used to retrieve AP stories. However, almost no candidate got any coverage on broadcast news, and adding to our model the standardized variables for newspaper and cable news coverage did not significantly change our results. Because these additional variables were also strongly correlated with our AP measures, we decided not to include them in the final models out of concerns with parsimony and also to avoid multicollinearity issues.

12. By “demand,” we mean the candidates that those citizens who do choose to engage with politicians online seem to prefer. The demographics, socioeconomic status, and individual properties of the citizens who do so is a separate issue with well-known biases towards the more affluent, the more well-educated, and the younger.

13. In particular, we employed measures of candidates' policy preferences based on their Congressional voting records, but they were of limited use given that, by definition, they are only available for incumbents, which excluded 133 of our 224 candidates, and 99 of our 112 Republican candidates. We also devised a measure of candidates' mentions on the major cable news talk shows, which are hosted by prominent partisan commentators and tend to attract highly ideological audiences, but this measure failed to improve our models' results and goodness of fit in any appreciable way, most likely because only a handful of candidates garnered the lion's share of visibility on these programs.