Why You Should Ignore "Analysis" of the Early Vote
Early voting for the October 12 election in Louisiana began on Saturday and will continue throughout the week. Early voting may be convenient for voters, but it is bad for election analysis. It prompts otherwise seemingly intelligent pundits to make all kinds of silly claims they have no business making.
This week, you will likely hear these pundits say things like “More voters cast their ballots early this year, so enthusiasm must be up for this election.” Or, “All those registered Democrats [or Republicans] casting early votes is good [bad] news for Governor Edwards.”
The only problem is that anyone who tells you what the early vote numbers mean does not actually know what they are talking about. If they did, they would not be trying to read the tea leaves of the early vote.
Not all data are equally useful for understanding what is going on in an election (or for forecasting what will happen, if you’re into that sort of thing). Early voting - especially in Louisiana - does not predict elections well.
1. Early votes don’t tell us much about enthusiasm and turnout
Political watchers often take a boost [or drop] in the number of early votes as an indication of higher [or lower] voter interest leading to higher [or lower] turnout.
There is some truth to these claims. Generally, elections with more early votes tend to have more total votes. Yet, besides just being an insipid take (“more votes means more votes”), this heuristic often breaks down when making inferences about specific elections.
This table, which ranks all statewide elections in Louisiana since 2007 by early votes, total votes, and turnout rate shows just how often the heuristic fails.
Sometimes the heuristic works. To account for the fact that different offices attract more voters, let’s compare within similar offices, starting with U.S. Senate races. The number of early voters fell from the 2014 U.S. Senate runoff to the 2016 U.S. Senate runoff. Sure enough, the overall turnout rate fell precipitously. But, let’s back up and compare the 2010 general election to the 2014 runoff (in 2010, Louisiana used a more traditional system for federal elections in which the general election functioned more like the runoffs we use today). The number of early votes rose from 2010 to 2014, but the overall turnout rate dropped.
We see a similar pattern with governor’s primaries. The number of early voters rose from 2007 to 2011 to 2015, but the total turnout rate was lower in the latter two elections.
Or, look at what happened between the 2015 primary and runoff. The number of early votes shoots up by about 35,000 from October to November, but overall turnout barely budged.
Until last year, the November 2015 election held the record in Louisiana for most early votes in a non-presidential contest at 270,144. Compare that to the next highest, the November 2014 election, which had about 25,000 fewer early votes. Which election had higher turnout? Not the one with more early votes….and it is not even close. Turnout in November 2014 was 51.5 percent. Turnout in November 2015 was 40.2 percent.
Or, look at presidential contests. The number of early votes has steadily climbed from 292,213 in 2008 to 356,603 in 2012 to 531,555 in 2016. Turnout? Holds steady at 67 or 68 percent despite all those early votes.
I am intentionally cherry-picking elections that cast doubt on the ‘more early votes mean higher turnout’ heuristic, but it is not hard to find them because the relationship between early votes and turnout is weak and imprecise.
Changes in early voting from election to election reflect at least three things: 1) Actual changes in voter enthusiasm, 2) Differences in levels of office, and 3) A secular increase in early voting that has nothing to do with variation in engagement across elections. The problem is that pundits have a hard time separating the first from the third.
Early voting in its current form has not been around for very long in Louisiana. Over the past decade, there has been a greater propensity for voters to become familiar with and use early voting. To illustrate this, here is the share of total votes that were cast early in all statewide elections since 2007.
The trend is clear. More and more voters are choosing to cast their ballots early rather than on Election Day. Indeed, some of the highest shares of early votes were in 2017’s primary and runoff for State Treasurer when overall turnout barely topped ten percent.
This trend is even clearer when broken out by office at the top of the ticket.
A naïve glance at early voting numbers can cause people to misinterpret this pattern of habitual voters moving their votes from Election Day to early voting as a boost in enthusiasm about a specific election.
2. Early votes tell us nothing about election outcomes
One of the things that early vote watchers love to do is to recite the number of votes cast by different groups, especially by registered Democrats and registered Republicans. Inferring anything about election outcomes from these breakdowns is a sketchy practice anywhere, but it is especially problematic in Louisiana where a large number of people vote against their party registration. I will post more about this later this week, but it boils down to the fact that there are many older, white registered Democrats who rarely vote for Democrats.
As a result, any analysis of the number registered Democrats versus the number of registered Republicans who early vote really does not tell you much about their votes.
For example, registered Republicans cast 36% of early votes in the 2015 gubernatorial primary and registered Democrats cast 51% of early votes. Sounds good for the Democratic candidate, right? Not really. The Republican candidates received 57% of the early ballots and the Democratic candidate received 42%. Even if every single early ballot cast by voters who were not registered Democrats went to the Republican candidates (an unlikely scenario), these results would still require nearly a fourth of registered Democrats who voted early to break ranks and support a Republican candidate.
You can see the same when comparing registration of early voters to how elections ultimately played out. This graph shows how weak the correlation is. The dashed line would be a perfect one to one relationship, but the relatively flat solid line shows there is little correlation between early vote and election outcomes. There is a slight upward trajectory but the data are very noisy. For a given breakdown of early voters’ party registration, there are some elections where the Republican candidates’ margin in vote share is quite a bit more than expected and some where it is quite a bit less.
The table below shows the share of early votes cast by individuals registered in each major party since 2008 (all the elections in the graph above) as well as the share of total votes cast for Democratic and Republican candidates.
Even comparing elections of similar types – to see if changes in party share of early vote predicts changes in total vote share across elections – does not tell us anything useful. Consider the 2008 and 2016 presidential contests in Louisiana, which have the highest and lowest shares of early votes cast by registered Democrats, respectively. Surely, this massive shift – with registered Democratic early vote share falling from 58 percent to 44 percent – had a radical impact on how the vote ultimately turned out, right?
Not at all. Although the registered Republican deficit in the early vote fell from 29 points to six points between these two elections, the margin of the total vote did not budge.
So, if we see a similar shift toward registered Republicans in this year’s early voting versus the 2015 gubernatorial primary, should we say Republicans will do better this year than last time? There will be some pundits who say that, but these data do not justify that claim. It may turn out that Republicans do better in the early vote and in the total vote, but these data do not permit inference of the latter from the former. Anyone who says otherwise is just guessing.
Sometimes we get the opposite pattern – the balance between registered Republicans and Democrats in the early vote remains the same from one election to the next, but the margin of the total vote shifts. For example, registered Democrats’ advantage in early voting did not change from October 2011 to October 2015, but the share of the total vote going to Republican candidates fell.
So, if we see no change in the party registration between the 2015 primary and the 2019 primary, should we assume that is good news for Edwards? No. Again, anyone who claims it does is just guessing. Their guess may be based on something, but it is not based on a sound understanding of the statistical relationship (or lack thereof) between early voting and election outcomes.
One day, we may be able to get more use out of the early vote for understanding what is happening in an election. But a few things will have to happen before then. First, we’ll need more elections, so Louisianans can grow more accustomed to early voting and eventually reach a relatively stable equilibrium in the share of votes cast early (and to give us a larger sample of elections to test relationships). At that point, we will be able to assess deviations from the equilibrium as spikes and drops in enthusiasm. For now, though, no one knows if more early votes in 2019 means more enthusiasm (probably not) or just more people using early voting in place of Election Day voting.
Second, voters party registration will need to be more reflective of their actual partisanship and voting behavior. This will likely unfold over time as older white Democrats who stopped voting for Democrats but never changed their registration pass out of the electorate.
In the meantime, don’t let someone fool you into thinking they are smart just because they are talking about what the early vote means for the election. The truth is they do not know. None of us do (yet).
[Portions of this post are revised and updated from a post from 2018.]