![Sample our Politics & International Relations journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5947/TOC-Subject-Sample-2021-Politics-International-Relations.jpg"})
ABSTRACT
This paper investigates the public responses of members of Congress to the deaths of Michael Brown and Eric Garner and the subsequent protests and grand jury decisions. To do so, we examine members’ engagement with the issue on Twitter, which became a platform for public protest with such hashtags as #BlackLivesMatter and #ICantBreathe. We find that a member’s race is a more robust predictor of their engagement on the issue than is the member’s partisanship or the partisan and racial demographics of their district. By showing that descriptive representation may overwhelm more traditional notions of district-based representation in responses to a racially charged issue, we further highlight the role descriptive representation in Congress plays in ensuring that the diversity of voices coming out of Congress reflects the diversity of voices in the public at large.
Acknowledgments
We thank the Wesleyan Quantitative Analysis Center for invaluable assistance, in particular, Mansoor Alam and Husam Abdul-Kafi for their work collecting the data and Manolis Kaparakis and Pavel Oleinikov for his help overseeing data collection. We also thank Lainey Hellman for research assistance and participants in the Identity Politics Research Group for helpful feedback. We also thank Wesleyan University for a Research Apprenticeship Grant.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1. Since 2013 Wesleyan University’s Quantitative Analysis Center has been automatically collecting and storing tweets from all known MCs’ accounts using Twitter’s REST API (https://dev.twitter.com/rest/public) and the Twython package in Python. Members’ accounts are queried multiple times a day to search for new tweets. For the analyses here we retrieved and coded all tweets from the members’ accounts during the timespans under study. This included retweets and replies, as we view the decision to retweet or reply to a tweet about Brown or Garner’s death as substantively engaging with the issue.
2. It was possible to verify nearly all Twitter accounts by one of these methods. A handful of accounts could not be verified. In those cases we read recent tweets to assess the authenticity of the account.
3. We were also following several accounts of members that did not tweet during this timespan.
4. The coders read the text of the tweets and, in the case of ambiguous tweets, followed any links to assess whether they were related to Ferguson or Eric Garner’s death. The coding sheet specified that tweets referencing the following should be coded as related to Ferguson or Garner: Tweets about Ferguson, Michael Brown, Eric Garner, Darren Wilson, Daniel Pantaleo, protests in Ferguson, police militarization, criminal justice reform, racial profiling, tensions between police and African American community/people of color, chokeholds, and the Defense Department’s 1033 program (the program by which some police departments received military equipment). About 65 Spanish-language tweets were coded after using Google Translate to translate them into English.
5. The analyses to follow exclude the MCs who represent Ferguson – Lacy Clay in the House and Claire McCaskill and Roy Blunt in the Senate. In testing our hypotheses, we are primarily interested in the relationship between a member’s race, partisanship, the demographics of her district, and tweets about Ferguson or Eric Garner. We expect that the motivations for tweeting about the events in Ferguson for representatives or senators who represent Ferguson are different from all other MCs. The analyses that follow can, therefore, be thought of as models analyzing the predictors of responding to the events in Ferguson among members who did not represent the affected area. The figure also excludes members who were running for other office, as their behavior may not be well explained by district characteristics. Finally, since the figure graphs the proportion of tweets across the timespan, members who did not tweet from August 9–25 are excluded. This latter group is included in future analyses that rely on a count of total tweets as opposed to a proportion.
6. The number of non-black Democrats in this category increases to 42 if you include members excluded from this analysis due to the lack of Twitter activity.
7. Emmanuel Cleaver – a black Democrat whose district includes Kansas City, Missouri – is an obvious outlier, with over 80% of his tweets about Ferguson despite representing a district that is not majority black. As a robustness test, we excluded Cleaver from the analysis and still found sizeable differences (19% compared to 2%, p < .001).
8. Again, sizeable differences remain if Cleaver is excluded as a robustness check (12% compared to 1%, p < .001).
9. The dependent variable is a non-negative count, which makes the use of a count model appropriate. The assumption that the variance equals the mean is violated in all models that follow, suggesting a negative binomial regression model should be used in place of a Poisson model (Long Citation1997). The results are similar if we use a logistic regression model just predicting whether a member tweeted at least once about Ferguson (see supplemental material).
10. Information on members’ race comes from the CQ Press Congress Collection database (Member Profile Results Citation2015). The information on members’ districts comes from the 2013 American Community Survey, with the exception of the Obama vote share variable, which comes from the Cook Political Report (Wasserman Citation2013) for House districts and the CQ Press Voting and Elections Collection (Presidential General Election Citation2013) for states.
11. The sign and statistical significance for the coefficients in the model are unchanged if members running for another office are included in the analyses. The results are also robust to the inclusion of members representing Ferguson (with a dummy variable for these members).
12. There are a quite a few members who did not tweet about Ferguson at all, which results in many zero values for the dependent variable. If the model is run using a zero-inflated negative binomial regression, the substantive effects are similar (a 3.61 increase in the predicted number of tweets for black Democrats compared to white Democrats in the district described in the text).
13. The tweets during this time period were coded by one of the authors.
14. The one difference is that members who retired from Congress are dropped since the election to select their replacement had already occurred by the time of the grand jury decisions. The results are substantively unchanged if all members are included along with dummy variables for the members representing Ferguson and New York City.
15. Both authors read and content coded all relevant tweets during this timespan. Any disagreements were resolved by the authors. The authors only coded the text of tweets and not any press releases, floor speeches, photos, or newspaper stories that were linked from a tweet. The decision to only code the content of the tweets is not meant to diminish the importance of other forms of member communication. Our interest here is in the extent to which members engaged with the issue on Twitter, however, and analyzing what members said in media appearances, press releases, and on the floor of the House would take us away from that focus. In addition, some links were no longer active by the time we analyzed the tweets or truncated during collection of the tweets, meaning it was not possible for us to access all the linked material.
16. The analyses again exclude the representatives and senators who represent Ferguson in the August 9–25 timespan. If these tweets are included, the proportion of tweets about federal and legislative action is still the only difference that approaches conventional levels of statistical significance (p = .06).
17. Tweets from senators and representatives who represent New York City or Ferguson are excluded. If the tweets from these members are included, the difference in the proportion of black and non-black Democrats’ tweets with overt racial content increases (difference between black and non-black Democrats equals 0.13, p < .01). In addition, a larger proportion of non-black Democrats’ tweets mention federal or legislative action, and the difference is significant at p < .05. This is the opposite of the pattern observed in the August 9–25 timespan, suggesting there is no consistent relationship between member race and the proportion of tweets calling for federal or legislative action.