Note: I released this blog post early for the readers who support the site on my Patreon page. Subscribe now for a host of benefits, including early access and a premium newsletter.
For years, this day has been marked by national media attention and prognostication of party fortunes. Strategists flock to the Lone Star State and close observers look for marks of coming party waves. Candidates and party bosses look for clues of strategy in similar races around the nation.
Today is primary day in Texas. And it is no ordinary primary.
That’s because the national spotlight has already been shone on the state’s elections. “Young grassroots candidate O’Rourke positions himself for a competitive primary,” this piece would have read last week. It could have just as well read “Democrats’ explosive growth in TX early voting signals trouble for vulnerable Republicans.”
By now you are sure to have seen the numbers, but here’s the recap (data from the largest 15 counties):
- Democrats cast 17,000 more early votes than they did in the 2016 primary
- Republican turnout was down a huge 37% from last cycle
- Democrats cast more votes that Republicans for the first time since 2006
These data paint a fairly blue picture for Democrats in November. However, they are not sure proof of a Democratic wave. What if Democrats are energized now, for example, but Republicans will surge again in the fall? Perhaps the competitive nature of down-ballot Democratic races has spurred their turnout. Whatever the case, there is inherent noise in using primary votes to predict general election intention.
… or so we thought…
Aided by county-level turnout data provided by Chris Grimes, a Graduate Student at the University of Arizona, I found a rather stable relationship between the share of the vote that Democrats receive in November and the percentage of primary ballots cast in Democratic contests. Below I present these results (which have a 0.86 correlation) that signal a fair outlook for Democratic Senate candidate Beto O’Rourke.
The chart above graphs a county’s Democratic percentage of early primary ballots on the horizontal axis with the share of Democratic ballots cast in general elections on the vertical axis. The relationship between these measures — which I assume reflect aggregate levels of partisanship between candidates, plus-or-minus some differences based on candidate quality and contest dynamics — predict that Democrats should expect to see slightly better performance in November than they did in the primaries, boding well for the party that received roughly 53% of primary ballots in the top 15 counties this cycle.
But what about statewide contests? Can we predict races, not at the county level, but with statewide information? Of course we can, but that does not mean the results are sound. That is in fact exactly the case; below I graph results of an analysis predicting the Democratic share of general election ballots with those from primary contests.
The image above shows a weak relationship (a 0.3 correlation and negative R-squared) between the statewide Democratic share of early primary and general election votes. So weak that there is a +/- 9% margin of error on our November predictions of total Democratic votes. Not usually do we run across political data that is — frankly — so useless. Statewide primary turnout doesn’t tell us anything about the general election.
So, does that mean we cannot learn anything from Texas primary elections? Of course not. If we assume that the county-level data reveal a statistical relationship that can transfer to statewide contests — perhaps the latter is only void of meaning for lack of data — then a prediction of November that can assess the competitiveness of top-of-the-ticket candidates (say, U.S. Senate candidates Beto O’Rourke (Dem.) and Ted Cruz (Rep.)?) may be meaningful. And it also may be valid: this approach would have estimated Hillary Clinton to lose to Donald Trump by 10 points in the 2016 presidential election. She lost by 9.
This method predicts that young Democratic congressman Beto O’Rourke would beat Ted Cruz by eight points in November, with a margin of error of a whopping 26 percentage points (meaning he could lose by up to 18 points). If we’re handicapping the 2018 Texas Senate race with this long-range quantitative approach, the signal is mostly hidden within the noise.
It is a tall task to forecast who will win elections 8 months away. Of course, some data can predict these things. Special election results and generic congressional ballot polls are famously helpful. But if you’re reading into Texas early voting and primary data to glean insight into the events of November 6, 2018, you may be led more astray than helped. At best, these data push our expectations slightly towards the Democrat. At worst, they’re entirely useless.