The 2020 US Presidential Campaign: Policy Issues
Twitter data is now increasingly used to measure public opinion on a range of issues.   Given that use of the platform is not universally or equally distributed across populations, and it is difficult to accurately locate all users geographically, political analysts need to treat any inferences drawn from these data about national electorates, with caution. We know for instance that among those expressing views on Twitter in the U.S. there is a clear bias toward the more politically active and interested.  Two recently published reports by the U.S. Pew Research Center also demonstrated an ideological bias within the American Twitter-verse, with the conversation typically dominated by those from the left and supporters of the Democratic party.  The key question we pose in this short series of DiCED blog posts is to what extent these biases in the Twitter conversation were challenged during the recent U.S. Presidential election campaign? In this first blog post, we investigate which issues were most discussed on Twitter during the campaign. In the second blog post, we examine the idea that Twitter is used by specific users by focusing on the presence of partisanship on this platform. The onset of the global pandemic and consequent lockdown across America meant that this election saw a sudden and significant shift of Americans working and social life into the online sphere. Did this migration also extend to the political environment? Was there a more widespread use of social media platforms such as Twitter for political expression by the U.S. public, given that more regular informal channels and outlets for discussion were blocked? We tackle this question by comparing the virtual conversation about the election on Twitter with the ‘real world’ version, drawn from survey responses, to see how representative the former was of the latter in the closing weeks of the campaign.
To investigate this question, we compared the salience of issues that were discussed on Twitter to those that were considered important by nationally representative sample of the U.S. population, during the final four weeks of the campaign. Specifically, we conducted a pre-election survey with online pollsters YouGov between September 16 and October 20 (N = 5379). The figures were weighted and are representative of all US adults (aged 18 and above). As part of the survey, we offered respondents a list of 17 issues and asked them which they considered to be important in the 2020 US presidential election. These were: the economy, health care, immigration, abortion, terrorism, the federal budget deficit, Medicare, the rights of the LGBTQ+ population, the environment, taxes, law and order, the rights of racial and ethnic minorities, gun policy, social security, trade policy, drug use/opioid addiction, and COVID-19. As a follow-up we asked which of these issues would be the most important in deciding their vote. For our Twitter analysis we harvested a sample of Tweets daily across the period October 6 to November 10. The tweets were captured using a list of keywords that were identified by the research team as relevant to the 17 issues and saved in an ‘Election Issues’ Data Dictionary. We extended the list of YouGov issues to include a range of secondary topics that we considered might also feature (less prominently) in the public conversation, such as election integrity and postal voting. We monitored Twitter trends and popular Twitter posts on a weekly basis to update and extend the keywords in our Data Dictionary to ensure we captured tweets mentioning these issues. The total number of issues measured in our Twitter data was 25 and thus exceeded our YouGov list. These issues include the economy, housing, energy costs, conspiracy, education, defence, healthcare, immigration, abortion, terrorism, international issues, international issues (specific), international issues (general), international security, LGBT rights, rights of racial and ethnic minorities, women’s rights, the environment, law and order, social security, trade, drugs/opioid crisis, COVID-19, election integrity, and postal voting.
The tweets were identified as emanating from the account of a Twitter user based in the USA, by using the Twitter geographic filter, which relies on IP addresses. More specifically, we used the geocode parameter in the Twitter API to set up a radius centred in the USA to identify those tweets. A user’s location is determined internally by the Twitter API through its Geotagging API. In those cases when no geocode was available in real-time, we relied on the location specified on users’ profiles. To make the Twitter data as comparable as possible to the YouGov data we first truncated the Twitter data to cover the period October 6 to October 20 and then pooled the results across the two-week period, rather than reporting them daily. We then expanded the Twitter data to the cover the period from October 6 to November 10 and reported issue mentions daily, to paint a picture of how the issue conversation on Twitter changed during the campaign period. Our measure of salience for issues discussed on Twitter was based on the daily rate in which the issue (and related key terms) featured in posts, rather than absolute number of total mentions in each 24-hour period. We refer to this as the tweet rate. The specific details of how we calculated the relative daily rate of issue mentions is set out in the methodological appendix.
We first consider the findings from our YouGov survey. Figure 1 reports the ten most cited issues that respondents identified as important in deciding how they would vote. While no single issue dominated, the economy, health care and the coronavirus pandemic proved to be most important influences on voters’ choices.
When asked to narrow the list down for respondents to the single issue they deemed would be the most important to their vote (N = 5379), Figure 2 shows that the most influential problem in determining how they would decide on Election day was Covid-19 (22%). Close behind, however, was the economy and then in third place healthcare.
We then compared these responses to the findings from our Twitter data. Figure 3 reports the top ten issues that featured in U.S. Twitter posts, aggregated across the same 14-day period, based on the tweet rate. The findings show there is a strong similarity in the range and ranking of issues that preoccupied Twitter users compared to the general public. Figure 3 presents the average tweet rate (tweets per minute) for the top 10 terms for the subcategories of each issue, between October 6 and October 20, 2020. This figure shows that, notably, out of these ten issues, Covid-19, and the economy were tweeted most frequently.
This underscores the findings of Figure 2, which indicated these factors were also most important in determining individual’s vote choice. That said, the Twitter data also reveal that Covid-19 received featured much more intensively in the Twitter conversation than any other issue by a factor of almost three to one. Thus, while Twitter was clearly a hub for discussion about the pandemic in the lead- up to the election, and the issue obviously mattered a great deal to voters, relying on Twitter data instead of our YouGov ranking would lead us to view it as significantly more important in determining how voters made up their minds.
In a second step, the aggregate picture formed by the Twitter data in Figure 3 was broken down into daily accounts of how the conversation about issues on Twitter changed across in the full period of data collection i.e., October 6 – November 10. Figure 4 shows the issues that gained most attention from Twitter users daily during this period. The x-axis shows the tweet rate (tweets per minute) for the top 10 terms for the subcategories of each issue, for each day, between October 6 and November 3, 2020.
Based on the tweet rate of the issue-terms, we can see that Covid-19 dominated Twitter discussion on a day-to-day basis although it did fluctuate. Other issues which were discussed throughout, but to a lesser extent, are the economy and law and order. Whereas some issues never receive much attention, others peak on certain dates: attention for these issues fluctuates based on events and developments. For example, we find a surge of interest in healthcare, COVID-19, the environment, immigration, and the rights of racial and ethnic minorities on the 22nd of October. This reflects the topics discussed at the presidential debate which took place that day. According to the Commission on Presidential Debates (CPD), these topics included “Fighting COVID-19”, “American Families”, “Race in America”, “Climate Change”, “National Security”, and “Leadership. We also identified a range of issues which gained attention on Twitter during these final weeks, as the campaign progressed, including conspiracy, election integrity, and postal voting. Due to the nature of Twitter as a social media platform and the possibility for content to go viral and hashtags to trend, we expected for some of these issues would stick, and others to disappear. We found that whilst Twitter did allow for a broadening of the agenda – a discussion for a wider range of issues – attention for these issues was largely temporary. For example, holidays, festivals, and celebrations put issues such as LGBTQ+ rights or indigenous rights on the agenda, but in the weeks that followed, these issues were much less referred to. Furthermore, the issues of conspiracy and election integrity, which gained traction later in the campaign, did not overtake the top 10 issues and in fact received little attention in these weeks leading up to Election Day.
In summary, our analysis has shown there is a clear correspondence in the range and importance of issues regarded as most important in the election by our YouGov respondents and Twitter users. Notably both datasets confirm that Covid-19 and the economy were likely to be the two most decisive issues for voters in the U.S. Presidential election. Given this proximity in findings it would appear that the shift in political talk to the digital sphere prompted by the pandemic may indeed have meant the conversation on social media became more representative of what mattered in the broader public sphere.
That said, we would still draw differing conclusions regarding the intensity of feeling or decisiveness of issues on vote choice based on each dataset. While the two datasets can be seen to point broadly to a similar rank-ordering of issues as drivers of voter decision-making, the Twitter sample would tell us that the election was only about one thing – Covid-19. The YouGov data, by contrast, indicates that the picture was much more mixed, with other issues such as the economy, health care and law and order proving decisive. The question inevitably arises of which ‘version’ of events is the more accurate? We began this article with the presumption that our YouGov data would provide the most representative picture of the issues that mattered most for voters going to the polls on November 2nd. Certainly, ipso facto the YouGov data are expected to provide more valid and accurate insights into the electorate’s views on the importance of an issue a given point in time, given that the underlying sample is more representative of the U.S. population than our Tweet sample. However, the findings from our Twitter data and evidence of disproportionate attention to Covid-19 do give us some pause for thought. The nature of survey methodology means that indicators of issue salience are researcher-driven and constructed, and space-limited. As such, questions about the importance of issues importance in determining vote choice necessarily constrain respondents to choose from a range of pre-ordained items, which one they think is likely to have the strongest impact on their vote decision. The Twitter data, while it may be drawn from a less representative base and be ‘noisier’, arguably offers a more ground- up user generated guide to the issues which voters were finding most compelling during the election.
To an extent, therefore, answering the question of which data set offers the clearest insight into the factors driving vote choice speaks to the wider debates over the extent to which ‘public opinion’ is best measured with ‘designed’ or ‘organic’ data (Groves, 2003). This is not a dilemma or puzzle that we can address in the confines of this post. Closer more rigorous time-series analysis of the dynamics underlying vote choice in this election is necessary to address this question. However, given that Trump lost, despite the economy having grown 2.5% each year since he took office, and his approval ratings being the highest of his term (49%) in January 2020 just prior to the declaration of the pandemic, there would appear at least prima facie evidence for the claim that it was Covid-19 that quashed his chances of victory. If so, then perhaps this Presidential election saw a new-found wisdom in the Twitter crowd.
 Cody, E. M., Reagan, A. J., Dodds, P. S., & Danforth, C. M. (2016). Public opinion polling with Twitter. arXiv preprint arXiv:1608.02024.
 Karami, A., Bennett, L. S., & He, X. (2018). Mining public opinion about economic issues: Twitter and the us presidential election. International Journal of Strategic Decision Sciences (IJSDS), 9(1), 18-28.
 Lopez, J. C. A. D., Collignon-Delmar, S., Benoit, K., & Matsuo, A. (2017). Predicting the Brexit vote by tracking and classifying public opinion using Twitter data. Statistics, Politics and Policy, 8(1), 85-104.
 Barberá, P. (2016). Less is more? How demographic sample weights can improve public opinion estimates based on Twitter data. Work Pap NYU.
 Ross, K., & Bürger, T. (2014). Face to face (book) social media, political campaigning and the unbearable lightness of being there. Political Science, 66(1), 46-62. https://doi.org/10.1177/0032318714534106
 Pew (2019a). Sizing up Twitter Users. https://www.pewresearch.org/internet/2019/04/24/sizing-up-twitter-users/
 Pew (2019b). National Politics on Twitter: Small Share of U.S. Adults Produce Majority of Tweets. Retrieved from: https://www.pewresearch.org/politics/2019/10/23/national-politics-on-twitter-small-share-of-u-s-adults-produce-majority-of-tweets/
 These topics were released on October 16 by Kristen Welker and the CPD. This information can be retrieved from: https://www.debates.org/2020/10/16/moderator-announces-topics-for-oct-22-presidential-debate/