Does Unemployment Make People More Likely to Vote?

by John Sides on November 2, 2012 · 2 comments

in Campaigns and elections,Political Economy

This is a guest post by Matthew Incantalupo, a doctoral candidate in politics and social policy at Princeton.

*****

The United States is slowly emerging from a once-in-a-generation period of high unemployment and the economy is the central issue of the waning 2012 campaign. With millions of Americans still looking for work, how might experiencing job loss affect whether people will turn out to vote on Tuesday?

In my research, I find that Americans perceive job loss as a more personal problem if they live in areas with low unemployment.  In these areas, the experience of job loss is politically demobilizing and makes people less likely to vote.

But in areas with high unemployment, it becomes a more salient political issue.  Americans perceive job loss as part of a broad problem that government should address. In these areas, job loss helps to mobilize unemployed Americans.

One problem with studying how unemployment affects participation is that age and socioeconomic status confound the relationship between job loss and turnout. Young, poor, and less educated Americans are at risk of being unemployed; those three factors do a good job predicting who votes as well. I get around this by exploiting the timing of job loss around Election Day as a quasi-experiment. Using the Current Population Study (the “household survey” used by the Bureau of Labor Statistics to compute the monthly unemployment rate), I compare individuals who lose their jobs 1-4 weeks before Election Day to individuals who lose their jobs 1-4 weeks after Election Day.

Here is the effect of job loss on turnout at different levels of state unemployment:

The effects of unemployment on turnout depend on the state unemployment rate.  When the rate is low, unemployed people are less likely to turn out, controlling for other factors.  Thus, the line dips below 0. When the rate is high, they are more likely to turn out.  Local economic conditions, measured at the state level, matter.

The same finding does not emerge when I compare individuals who quit their jobs just before and after Election Day, which suggests that the finding  cannot be explained by the fact that people who do not have to go to work simply have more free time to vote.  The key is losing your job.  Americans who lose their jobs in high-unemployment contexts are more likely to report voting than otherwise similar Americans who are employed on Election Day, but who will lose their jobs shortly.

With unemployment still high and salient, I expect many unemployed Americans will turn out on Tuesday (if they haven’t already). This could be problematic for President Obama if they want to punish the incumbent. On the other hand, the unemployed are also younger, poorer, and more likely to be members of racial minority groups than gainfully employed Americans, and these groups tend to support Obama. Regardless, these findings are hard evidence that Americans’ personal hardships can become politicized and mobilize them to participate.

{ 2 comments }

Andrew Gelman November 2, 2012 at 11:37 am

Interesting post, but I don’t like the graph. The visual makes it look like you are plotting 16 data points, but what you’re really showing is an estimated regression line. I’d suggest ditching the dots and the error bars; just show the estimated line with gray area indicating uncertainty. Also, x-axis could be labeled at 5%, 10% rather than at every percentage point. We don’t have any data at 0% unemployment, right? I’d also like to see some dots showing where the actual data are.

Let me re-emphasize that last point. To me (and, I expect, to many other readers), a claim of “hard evidence” will be much more persuasive if we can see the data plotted. A statistically significant regression coefficient is fine, but ultimately we can understand such patterns via engagement with individual cases (in this case, specific years and specific subgroups of the population, not individual people).

I also recommend labeling your graph as “estimated effect,” not “effect.” No matter how good your unemployment instrument is, your background variable (the national unemployment rate) is not an assigned treatment; it is observational and could very well be correlated with any effects.

Doug November 3, 2012 at 1:46 am

I think “Adjusted Predictions” is a good name for the kind of analysis that is shown in the graph (to avoid the “effects” language), and I hope people make greater use of them. (I assume this was done in Stata, which still needs to clarify some aspects of their -margins- command.) I don’t see a problem with plotting predictions as points; it seems a silly distinction to plot a line and show or not show estimated points. Having the graph only cover the observed unemployment rates is a good idea. Not sure how else for the range to be shown on the same graph (unless a caption below the graph explains this). Not sure what your second paragraph is suggesting should be done to the graph.

Comments on this entry are closed.

Previous post:

Next post: