Long-Term Economic Decline and the 2016 Trump Vote

• In Event: Cycles in Electoral Behavior

Abstract

In 2016, three typically blue states (Michigan, Wisconsin, and Pennsylvania) switched to the Republicans by tiny margins and gave Donald Trump the victory. In all three states, most people voted their partisanship or continued to abstain from the polls as usual. However, two smaller groups defected from their customary behavior--Democrats converting to Trump and habitual non-voters mobilizing to vote for him. Conversion plus mobilization made the difference.

In the wake of the election, many commentators noted that the white, rural, and less-educated parts of the U.S. swung to the GOP by substantial margins (for example, Scala and Johnson 2017). Much was made of deindustrialization, wage stagnation, the impact of globalization, and other long-term economic factors in the 2016 vote and before, both by journalists (for example, Goodhart 2017) and by scholars (Gest 2016; Autor et al. 2017; Cerrato et al. 2018). What is striking about such arguments, however, is that they contradict the standard finding that only very recent economic conditions affect vote choices, typically those of the prior year or two—a phenomenon known as “myopic retrospection” (Kramer 1971; Fiorina 1981). Researchers have repeatedly found that economic conditions three to eight years ago (the period of two presidential terms) have no effect on vote choices. (Recent reviews include Healy and Lenz 2014 and Achen and Bartels 2016, chap. 6.) However, almost no one has looked statistically at longer-term secular decline in economic fortunes as a predictor of the Trump vote. The central aim of this paper will be to assess the impact of long-term retrospection on the U.S. election of 2016. We will consider its effects on both conversion and mobilization.

The great majority of studies of the Trump vote have used survey data. Surveys have many advantages, but their sample sizes are too small to permit detailed analysis of within-state variation. To assess whether middle-income Pennsylvanians in Philadelphia suburbs behaved the same way as middle-income citizens in Pittsburgh suburbs, for example, only aggregate election returns matched to Census data can provide the answer. In addition, the aggregate returns permit the researcher to study the impact of long-term economic decline in a region. Nearly all surveys, even in the unusual case when a researcher has access to the name of the particular area where a respondent lives, have too few respondents from chronically depressed areas for statistical analysis.

A handful of election surveys use an expensive panel design; almost none includes detailed human-coded validation of voter turnout. Surveys lacking those features are bedeviled by turnout over-report and by faulty memories of vote choices four years ago and earlier. Hence, it is very hard to use them to determine whether Trump’s margin came more from Democrats who switched to him or from customary non-voters who were mobilized to his cause. Aggregate electoral data are much more suitable for those purposes, since they include validated turnout rates for comparable geographic units over time. They also include reliable prior presidential votes back to 2000 and earlier for each area, as well as party registration data in the case of some states such as Pennsylvania. We will exploit all these advantages.

In summary, this paper will assess the sources of the Trump vote in the three “blue wall” states that made him president. Using aggregate electoral data matched to Census records, we will determine (1) whether long-term economic hardship and income stagnation affected the vote more than myopic retrospections, and (2) whether the electoral impact of long-term economic factors was due more to their effect on conversion or to their influence on mobilization.

Authors

• Christopher H. Achen, Princeton University

• Jeremy D. Darrington, Princeton University