This is part of a series of short posts about politics that seeks to show how we use data science to learn more about the real world. Follow along here.
In the aftermath of Donald Trump’s election in 2016, there was ample inquiry into whether Americans facing more financial pressure — those that were “economically anxious” — were more likely to switch from voting for Obama in 2012 to the unexpected Republican nominee in 2016. These theories are not correct. Diana Mutz, a political scientist at the University of Pennsylvania, wrote in her 2018 article “Status threat, not economic hardship, explains the 2016 presidential vote” that
there is little to no evidence that those whose incomes declined or whose incomes increased to a lesser extent than others’ incomes were more likely to support Trump.
The issue of economic anxiety relates to a tangential long-standing pattern in US politics, economic voting. This popular thesis claims that voters reward the incumbent party in presidential elections if the economy is doing well and punish them if it’s getting worse. As I wrote in my newsletter last week, there is good reason to believe that this is no longer the case. 1 The relationship between presidential approval ratings and consumer sentiment has fallen apart precipitously in recent years. Even under Obama’s presidency, there was little relationship.
The curious mind wonders if there is anything we can test in the 2018 midterm elections that informs the debate further. In fact, in the form of the 2018 Cooperative Congressional Election Study, a time-series survey from political scientists at Harvard and elsewhere, there is. In this post, I walk through a first-draft of an analysis of the 2018 CCES. (Per usual, use the button at the top of the article expand the code chunks and explore the process — and also per usual I’ve left ample comments for you!)
My first cut at determining what role economics plays in support for Trump — really, defection from Trump — includes running a logistic regression to predict the chance that a 2016 Trump supporter voted for a Democratic House candidate in 2018 based on their demographic and political profile and, importantly, their attitude toward health care. The economic variable of interest is whether someone reports higher or lower income since 2017. The results, predicted for a constant respondent while varying their economic condition, is null. See here:
As we would expect, opinions on health care policy did make a difference in whether or not Trump voters defected. That fits with out theories, as health care has been a very salient issue for a few years now, and the biggest dip in Trump’s approval rating came after his attempt to partially repeal the Affordable Care Act and decrease Medicaid funding, a bill known as the American Health Care Act of 2017. I’ve plotted the effects of each included variable here:
If the self-reported income variables are accurate, this analysis should suggest we throw some more cold water on theories of economic voting in 2019. I do think that claims that the status of the economy “doesn’t matter” should be qualified, though. It seems more likely to me that a very bad economy could hurt president Trump in 2020, by giving his supporters a reason to distance themselves from him, even though a good one hasn’t given Trump’s opposition a reason to support him. The question warrants further investigation, but this is a good start.
It’s also good example of using logistic regression in R to answer important questions about our current state of politics. If you’re into that, you’re in the right spot….
PS: See, this is why you should subscribe to my newsletter!↩
R for Political Data Science Week 11: Is Beto the Media Sweetheart?
R for Political Data Science Week 10: What If Each State Allocated Their Electoral College Votes Proportionally?
R for Political Data Science Week 9: The “Strongest” Democrats and Republicans (That Ran for Office in 2018)