The past few days have seen several stories on waiting times at the polls in November, spurred by the release of a couple of studies, including one by Charles Stewart at MIT with an N of over 10,000. Key findings from the MIT survey are illustrated here in the NYT piece. With these data we can’t be at all certain that these resulted from the much discussed Republican vote suppression efforts, but the effects that are observable did have a substantive effect on voters, and were systematically related to politically predictive demographic categories such as race and income.
Democrats waited an average of 15 minutes, Republicans 12.4. Waits were longer in more urbanized settings and for those on lower income brackets. Most damning is that Latinos and African-Americans waited an average of 20.2 minutes, while whites 12.7 minutes. The state with the longest wait was Florida at 45 minutes. This Nation piece on Florida, which conducted its own examination into the problem, includes this brilliant quote by the incumbent Florida Secretary of State: “I can confidently say Florida conducted a fair election in 2012.” Perhaps when compared to some past Florida elections of note, but perhaps not when compared to 2012 cross-nationally. Work done “by an Ohio State University professor and The Orlando Sentinel, concluded that more than 200,000 voters in Florida “gave up in frustration” without voting.” The NYT article states that the overall cost to Democrats numbered “hundreds of thousands of votes”.
Positive reform (not to be confused with voter ID) at the Federal level is encountering the usually justified obfuscation from Republicans:
Conservatives have complained that Democrats are politicizing an issue that should be handled by the states, not the federal government. “It’s ridiculous to stand in line a couple of hours to vote,” said Senator Charles E. Grassley of Iowa, the senior Republican on the Senate Judiciary Committee. “But I think it’s also ridiculous to make a political issue out of it when it’s very easily handled.”
And that’s one of the more ridiculous things I’ve read today. Voting is a political issue, and making voting easier (or harder) is all about politics. When African-Americans and Latinos have to wait nearly twice as long as whites, it’s political. We know, and they know, that efforts to suppress votes masquerading as fraud prevention result in Republicans having a larger percentage of a given electorate, while easing the costs to voting (e.g. easier / no registration, more polling places, shorter lines, etc.) increase the Democratic percentage of a given electorate. While it’s a starkly political issue, there is one key difference: we’re right on the normative merits. Anybody who wants to vote should be able to vote. As a society, we should be reducing, not erecting, roadblocks to the act of voting.
I cobbled together a state level dataset to quickly examine if there are determinants of the average wait by state. While the studies discussed in the NYT and Nation articles are based on survey data, what does this look like at the state level? The dependent variable in the little model that follows is the average wait time courtesy of the MIT survey. I included some standard state level measures, including PVI, wealth, aggregate population, poverty rate, as well as percentages African-American and Latino of the overall state population. I also hypothesized that unified Republican control of the state government (defined as both chambers of the legislature and the executive) would lead to longer wait times, as it was the Republicans pulling back on early voting windows and introducing Voter ID laws (which they achieved anything, achieved longer lines). There are 25 such states. Finally, I included the margin of victory for the state winner in the Presidential election.
Several notes prior to viewing the table in all its glory are warranted. First, interpretation of the effects will be a product of your own view of the significance of statistical significance. I think it is often mis-applied. In this model, it could be argued that I have the universe of cases at my disposal, hence I’m not trying to ascertain the probability that the effect I’m observing in this sample is the result of random chance. Furthermore, with an N of only 48 (two states are not included in the model; Washington relies heavily on postal voting, while Oregon is exclusively so, rendering waiting lines at polling places an irrelevant concept), only the strongest substantive relationships will be significant. A lack of significance does not equate no relationship, it simply means that the relationship has not been observed with the precision necessary to be arbitrarily comfortable generalizing from our sample to the target population. Second, the measures are not normed to a common metric, meaning size is not relative. Third, as this is state-level data, it is not fine enough to capture precinct-level variations, and I suspect a lot of the vote suppression tactics were conducted at precinct level.
“Significant” relationships are found with state wealth, percentage African-American, and poverty rate (the bi-variate correlation between poverty rate and per capita GDP is low.) The actual estimate for wealth required moving the decimal a few spaces to show a real number, but a rough norming of the measures indicates that it has the second strongest substantive impact on wait times: the wealthier the state is, the shorter the wait. Counter-intuitively, the higher the poverty rate, the shorter the wait as well, and I’m not sure what to make of that. The overall winner, in terms of both significance levels and normed substantive effect is % African-American. For each percentage point increase in a state’s black population, the average wait increased nearly half a minute. This doesn’t seem like a lot, but this measure ranges from 0.8% in Montana to 37.6% in Mississippi, hence the overall range effect is around 18 minutes. Moving on to the measures that did not report significant estimates: there really isn’t anything going on with overall population size, states with larger Latino populations had longer waits, the greater the margin in the election, the shorter the wait, and states under unified Republican control waited nearly two minutes longer when everything else in the model is taken into account. A straight bi-variate analysis is starker: the average waiting time for states under unified Republican control was 13.42 minutes, while 9.13 minutes for those with at least a modicum of Democratic input (and this relationship is significant with a one-way ANOVA).
Finally, the overall model fit is only .31: these variables only explain 31% of the variance in waiting times. While some variation is just random, there’s with near certainty several additional unobserved determinants of waiting times.
As I only just slapped all this together, I’m not completely sure what the story is beyond the obvious: your wait time will be shorter in a richer state, but longer in a more heterogeneous state. I think, given the nature of the data and the significance of the bi-variate relationship, we can also be confident that states under unified Republican control had significantly longer lines to vote.
And that’s pretty much exactly the way they like it.