A look at Spring Polling vs. Fall Voting in Texas

Now that we've had our fun thinking and talking (and talking some more) about the poll released Wednesday by Quinnipiac University showing Democratic Senate hopeful Beto O'Rourke within the margin of error of Senator Ted Cruz in their Texas Senate race, it's worth putting those polling results into context. Below, we've compiled statewide polling results from April through July for elections going back to 2010 for President, Senator, and Governor. While it's certainly fair to say that Cruz's lead over O'Rourke is not as stout as one might expect given historical polling (in the polling data below, the lead for the GOP candidate at this point in the election cycle is 9 points on average), there's little evidence from the eventual election results that Cruz's lead isn't likely to grow as the campaign season begins in earnest, let alone when voters actually begin to cast their ballots.

The table above illustrates at least two things. First, polling collected during the Spring and early Summer before an election isn't terribly predictive of the actual election outcome. Second, more polling is better than less.

Historically, polling data collected this early has underestimated the Republican electoral advantage by 4 points on average, with a high as large as 16 points! (And by the way, this is a national phenomenon , not limited to Texas polling alone.*) This happens for a number of reasons, a few of which we'll touch on here. 

First, the sample of most spring polls differs significantly from the composition of the electorate for the general election. Most surveys collected at this point in time sample registered voters (as did the Quinnipiac Poll, understandably), instead of the smaller pool of likely voters sampled to look as close as possible to the group of individuals who show up on Election Day. That difference is an important one, as Republican voters tend to be older and whiter than the overall electorate, two characteristics that tend to correlate with a higher rate of participation. In short, Republican voters, because of their underlying characteristics, tend to vote more reliably, especially in midterm elections.

While polls of likely voters tend to produce more accurate trial ballot estimates, it's hard to say who a likely voter will be seven months before an election. This is well before the actual campaign season traditionally begins (after Labor Day), and before voters have been inundated by multiple local, state, and national campaigns. In short, voters are unlikely to be paying much attention to the candidates or actively considering their vote choices.

Another related reason that early polling tends to be less accurate than polling conducted closer to Election Day is that often some or even none of the candidates are well-known to the electorate. In  Quinnipiac University’s first poll in Texas, roughly half of the Texas electorate was unable to express an opinion about O'Rourke (a result similar to recent University of Texas / Texas Tribune Polling). What this tells us is that a trial ballot at this point in time is unlikely to measure the result of a considered comparison between two candidates. Instead, it is more likely measuring some combination of attitudes towards the more well known candidate – in this case, Senator Cruz – or simply measuring partisanship. While this information is valuable in its own right, it's not likely to be predictive of what people will actually do come Election Day. 

The inherent ambiguity in interpreting Spring trial ballots’ meaning for the General Election leads to our second point: more polls are better than fewer. Given all the variation on display in the table above, it's good to see more polling outfits taking an interest in Texas. Polls sample a share of the population of interest to generate an estimate of an attitude of interest, in this case, vote intentions. While this exercise may be more complicated than it appears, one truth exists: more polls are better than less. More estimates, from quality pollsters like those at Quinnipiac University, only add to the knowledge we have about public opinion here in Texas. One can always quibble about approach – believe it or not, around the edges, pollsters have their own styles – and sampling by definition will have its vicissitudes. The only way to gain clarity is with more data, ideally transparently presented. 

Even with the best information available to us, it's still really important to interpret that information in a fair and even-handed way that considers the data itself, what we know about the past, and our best read of the present. O'Rourke looks closer to Cruz in the limited, early public polling than previous statewide candidates in recent elections, and there are signs of Democratic enthusiasm. But the historical patterns in polling tell us that, whatever baseline emerges, we should expect to see larger Cruz leads in polling as we make our way to Election Day. The obvious caveat is that historical data is good for predicting outcomes where the present can be reasonably said to resemble the past.

Here, as in so many other instances, the impact of Donald Trump’s presidency on the electoral environment seem at the same time assured yet difficult to predict in its particulars. The 2018 election already appears less than ordinary, but there’s no sure way to assess whether or how much this disruption will shape Texas elections.

*Luskin, Shaw, and Hetherington. 2018. "Under-Forecasting the Republican Vote? A Tale of Campaign Spending and Partisan Homecoming across Two Eras." Presented at the 2018 Midwest Political Science Association Conference, Chicago, IL.