A surprising fact about the 2016 election is that Trump received fewer votes from whites with the highest levels of racial resentment than Romney did in 2012

Who Put Trump in the White House? Explaining the Contribution of Voting Blocs to Trump’s Victory. Justin Grimmer, William Marble. ecember 12, 2019. https://williammarble.co/docs/vb.pdf

Abstract: A surprising fact about the 2016 election is that Trump received fewer votes from whites with the highest levels of racial resentment than Romney did in 2012. This fact is surprising given studies that emphasize “activation” of racial conservatism in 2016—the increased relationship between vote choice and racial attitudes among voters. But this relationship provides almost no information about how many votes candidates receive from individuals with particular attitudes. To understand how many votes a voting bloc contributes to a candidate’s total, we must also consider a bloc’s size and its turnout rate. Taking these into account, we find that Trump’s most significant gains came from whites with moderate attitudes about race and immigration. Trump’s vote totals improved the most among swing voters: low-socioeconomic status whites who are political moderates. Our analysis demonstrates that focusing only on vote choice is insufficient to explain sources of candidate support in the electorate.


4 Conclusion
In this paper we demonstrate that the common practice of regressing vote choice on individual characteristics is largely uninformative about where a candidate support lies in the electorate. This is because vote choice is only one component of the contribution of voting blocs to a
candidate’s vote total. We must also know how prevalent a group is in the electorate and the turnout rate of the group to know how much a group contributes to a candidate’s vote total. Taking these three components into account, we first show that even though racial and ethnic attitudes were activated in 2016, they did not contribute a distinctive number of votes to Trump. We show that Trump’s net vote total among whites with the highest levels of racial resentment was smaller than Romney’s. Further, we find that Trump’s relative support grew more among white moderates on immigration than among white conservatives on immigration, and that Trump received an almost identical share of votes from former Obama voters as Romney. Rather than these explanations, we show that Trump received an increase in relative support among low-SES whites who are independents and political moderates. We find Trump gained support among whites who are disabled and retired, but we see only limited evidence that Trump gained support among whites who reside in depressed economic contexts. Our analyses cannot answer causal questions about the optimal campaign strategy for a candidate, nor does it provide conclusive evidence about who campaigns should target to win elections. Yet, our results do show that Trump improved over Romney among voters who are regularly identified as “swing” voters (Hill, 2017). These voters casted more votes for Trump in 2016 than they did Romney in 2012, and this group of voters supported Clinton at much lower rates than they supported Obama. Despite concerns about ideological polarization, increased partisan acrimony, and low engagement among independents, these findings imply that white swing voters comprise an important voting bloc for presidential campaigns and are likely to remain so in future elections. Of course, as the electorate becomes less white, other racial groups are likely to comprise this crucial group of swing voters (Barreto and Segura, 2014; Fraga, 2018). Our results also show that analyses that focus on activation of attitudes overstate the importance of racial and immigration conservatives to Trump’s victory. For example, Sides, Tesler and Vavreck (2019) argue white racial and immigration conservatives who switched from Obama to Trump were pivotal to Trump’s victory in close states. Their evidence, how
ever, is based on regressions of vote switching on racial and ethnic attitudes among white voters in close states and not an explicit calculation of votes. To do the vote calculation, we replicate our analysis on immigration attitudes subsetting to respondents who voted for Obama in the prior election.15 We find that among former Obama voters, Trump improved his relative support most among moderates on the immigration scale, despite clear evidence from the vote choice term that immigration was activated in 2016 among former Obama voters. However, there are very few people who reported voting for Obama in the previous election with very conservative immigration attitudes. Thus, our main findings about immigration—the Trump benefited from gains among moderate—are replicated even among former Obama voters. Of course, in close elections small groups of voters can swing the outcome. However, there are many such groups of potentially pivotal voters. Our results show that explanations that focus on activation alone miss the largest changes in the electorate. Methodologically, our paper demonstrates that if the goal is to explain election results, studying only the correlation between attitudes and vote choice among those who turn out to vote is insufficient to know where a candidate receives votes. And worse, focusing only on vote choice can produce misleading or outright incorrect estimates of where a candidate receives support. The implications of this are far reaching for how social scientists explain the results of elections and how they use experiments to make recommendations for campaign strategy. If the goal of the activation literature is understanding why a candidate won an election, then much of the current practice of how elections are analyzed needs to be expanded to also include measures of turnout and composition. Further, the activation literature’s focus on vote choice and attitudes does not eliminate the need to consider turnout rates or changes in composition. In fact, regressions of vote choice on attitudes could still be deeply biased by differential turnout across attitude levels or changes in composition of attitudes in the electorate. For example, by focusing only on vote choice of those who turnout, there is a clear selection issue: only those individuals who vote can report a vote choice. As a result, even in studies focused on activation, differential turnout could create an impression of attitude
activation when actually the differences across elections are due solely to differential changes in turnout (Nyhan, Skovron and Titiunik, 2017; Knox, Lowe and Mummolo, 2019). Our results also have implications for experimental analyses of campaign strategies. For example, experimentalists regularly run interventions focused on vote choice and use the results to assess the efficacy of particular campaign strategies. But our analysis shows that it is also essential to consider the share of the electorate who could receive the treatment, how the treatment affects turnout, and the vote choice among those treated. Without including this information, experimental analyses could provide misleading estimates on how a strategy could affect a candidate’s vote total. Our simple statistics and quantities of interest provide the relevant quantities for understanding where a candidate’s support increases in the electorate.