We recently started exploring patterns in voter turnout in New Jersey, with an eye towards putting together data and analyses that will hopefully be of use to advocates who are interested in improving voter turnout and access. In the last post in this series, we took a look at the relationship between voter turnout and income in New Jersey municipalities. The idea was to see if there were any trends in turnout at the county level that emerged after accounting for the trend of lower turnout in municipalities with lower incomes.
Here we will look at the 2016 congressional turnout, but in a slightly different way than we did before. First, instead of using the number of ballots cast as a fraction of the number of registered voters as the measure of turnout, we'll use the number of votes cast in the congressional race as a fraction of ballots cast. This generally represents the congressional abstention rate, or the rate at which voters cast a presidential ballot but left the congressional question blank. This measure also eliminates difficulty or deterrence in casting a ballot as a factor on turnout, since by definition there are at least as many ballots cast as votes cast in the congressional race.
Second, instead of trying to uncover county level trends after controlling for trends in municipal level income, we will be looking for district level trends. This presents a slight challenge since 15 of New Jersey's 550 or so municipalities are split into two districts. New Jersey uses a bipartisan commission for redistricting, which generally leads to fairer maps than those drawn by partisan state legislatures, and 15 municipal splits is not too bad. We'll just exclude those municipalities from the analysis, since we haven't been able to obtain all the relevant voting data for the split municipalities. Note though that since these 15 municipalities are not excluded randomly, there is the possibility that excluding them could bias our results.
Below we've plotted the log of the median household income in NJ municipalities vs the congressional turnout relative to the total number of ballots cast. The magnitude of the relationship between presidential turnout and income is already pretty jarring, and this shows that on top of that, municipalities with lower incomes had higher congressional abstention rates.
We can't pin this trend on voting access, so it must be explained by things like a voter not liking or not being familiar with any of the candidates, or perhaps perceptions of the competitiveness of the race. We'll investigate the effect of the competitiveness of a race on congressional turnout can by fitting a multilevel regression model to the municipal turnout data. A regression model represents an estimate of a relationship between two variables, and a multilevel regression model estimates relationships in data with a hierarchical structure. In this case we have hierarchical data because there are turnout results on the municipal level, but knowing which district a municipality is in is an additional piece of relevant information which may be related to turnout. Here we'll look at the relationship between the vote margin of the winning candidate and district turnout.
The figure below shows the district regression intercepts, which represent the average turnout rate in a district's municipalities after accounting for the effect of income at the municipal level, plotted against the vote margin of the winning candidate. That means the intercepts don't neccesarily reflect the actual turnout in the district, but rather how high the turnout was relative to our expectations based off of how affluent the municipalities in the district are. The intercepts are marked on the plot by the district's number, which is also color coded to indicate the party of the winning candidate. The trend is pretty strong, with lower turnout in less competitive districts. The closest race was in the fifth district, where Democrat Josh Gottheimer defeated incumbent Scott Garrett. The least competitive races were in the eight and tenth districts, which were held by incumbent Democrats Albio Sires and Donald Payne Jr. respectively. Most districts in the state were not very close races however.
Of course, voters don't know the outcome of the election before it happens, and so any direct relationship with turnout must be due to the perceived competitiveness of the election. Although if that is true then it seems possible that perceptions of competitiveness can influence the actual competitiveness of the election, which puts those who live in districts which are not competitive according to the conventional wisdom in a catch-22. An opposition party may not be willing to commit resources to districts which don't show signs of being competitive, but on the other hand the fact that an opposition party is willing to commit resources to a district could signal to the electorate that the seat is winnable, boosting turnout.
Competitiveness can be signaled by all sorts of things including candidate messaging, local media, grassroots activity, and fundraising reports, but the bottom line is that somebody in these 'noncompetitive' districts needs to decide to ignore the conventional wisdom and get the ball rolling on some grassroots opposition if anything is going to change. If you are unhappy with your representatives, try to find like minded folks in your community and encourage others to speak out about issues that matter to them. If someone tells you not to bother trying to compete in an election, they are probably not saying that out of genuine concern that your efforts will be wasted. It's more likely that they don't want to find out what would happen to the incumbent in a high-turnout election!
In a few days we'll post a follow up looking specifically at New Jersey's 11th district, where differences in congressional turnout in the different counties in that district could have been worth upwards of 20,000 votes. Stay tuned and we'll try to figure out why.
The source code for collecting the voting data from the New Jersey elections website and running the regression can be found on GitHub. The regression model is very similar to the one used the last post, but the hierarchy is municipalities inside districts instead of municipalities inside counties. A diagram of the model in the style of the puppy book is below.