How much do the polls in the governor's election tell us about who will win?
The gubernatorial election in Louisiana is one month away. We have already seen about a dozen publicly released horse race polls for the election over the past year (that is, polls asking which candidate the respondent will vote for). If historical patterns hold, we should see several more in the remaining weeks of this election cycle. Polls tend to grab media attention because that tap into the intriguing storylines of elections: Who looks to be a winner come Election Day? Who’s gaining ground? Who looks destined for the pack of the ‘also ran’?
The problem is that for most of the campaign the polls don’t really answer these questions. As a predictor of how well candidates fare on Election Day, the polls tend to miss by a wide mark until the closing weeks.
Looking across all gubernatorial elections in Louisiana since 1995, the graph below shows the average absolute difference between the support a candidate gets in a poll and the percent of the vote he gets on Election Day in polls conducted at various points in the campaign.
Over the course of the year before Election Day there is a steep decline in the gap between what polls say and what ends up happening. Polls taken more than one month before the election have been very inaccurate, by about seven to nine points on average. In the first half of the month before the election, the polls improve slightly as predictors of votes – missing by about five and half points. They become even better predictors in the final two weeks of the campaign, when thy miss by only about three points. That amounts to more than a 50 percent reduction in error over the course of the campaign.
To be fair, many of these polls are conducted before the qualifying deadline. It is not always clear who will and who won’t run for governor, so many polls list potential candidates who never end up running. The presence of non-candidates in the question can distort support for actual candidates. But even when polls with non-candidates are set aside, the ability of polls to forecast the election results follows the same general pattern as in the chart.
But even if polls don’t forecast what voters will do in late October, maybe they tell us what voters are thinking in late September, or August, or June, right?
Until the final weeks of a governor’s campaign, most voters simply are not paying much attention. Lack of attention is not going to keep many of them from answering a poll question. A July poll by the Reilly Center for Media and Public Affairs at LSU’s Manship School of Mass Communication showed that only one in five voters were paying attention to the gubernatorial campaign. As you might guess, most of them (91%) were able to offer an opinion of the candidates. But most of the people who were not paying attention to the campaign (79%) also gave an opinion on the candidates.
If they don’t know much about the campaign, where do these opinions come from? When asked about something they have not given any real thought to, poll respondents tend to answer from the top of their heads, relying on what is familiar. This is why early polls are a better measure of name recognition than actual support.
For example, the drop in support for Roemer and the rise in support for Foster in the 1995 polls, did not reflect people switching away from real support for Roemer to Foster. Most of these voters were never likely to end up as Roemer voters (the people who voted for Foster look quite distinct geographically, culturally, and ideologically from Roemer voters). Instead, they had not paid much attention to the election until the final month and simply named the Republican whose name they had heard. Once they tuned in, however, they found a Republican that better matched their political leanings.
Of course, this doesn’t mean that early leads in the polls will necessarily crumble. Despite missing the mark on their vote shares, about half of the polls taken between one month and one year before recent primary elections for Louisiana governor still correctly predicted the outcome in terms of which candidates make the runoff or win outright (18 of 39 polls). Yet, it is probably not name recognition that carries them to victory. Instead, candidate who have name recognition also tend to have other things that matter for elections: Field organization, money for campaigning, endorsements, and experience.
The polls move closer to the actual outcome because there is real movement over the course of the campaign, but it is not so much about changing our minds as it is about tuning in and making up our minds.