Why We Shouldn’t Make Much of Early Polling in the Louisiana Senate Race
We are five months from the first round of voting to fill David Vitter’s U.S. Senate seat in Louisiana. There has been very little polling in the race so far, but last week Southern Media & Opinion Research released an early snapshot of the contest. John Kennedy, who led in the poll, already has it up on his campaign website touting it as “great news”.
And, that’s my cue…
There is no reason to put any stock in what a poll says about this race five months out.
This is not because early polling in U.S. Senate races are inherently problematic. In fact, they are usually fairly predictive on average. Two years ago, 538’s Harry Enten had a fantastic analysis of early poll in U.S. Senate races. He looked at races across the country from 2006 through 2012 and found that averages of polls taken six months before Election Day are quite predictive of the final margin of victory in the election. So, why am I skeptical of an early poll of the U.S. Senate in Louisiana this year?
The polls in Enten’s study mostly involve a simpler question than the one Louisiana voters face. Those polls typically ask voters to choose between two major party candidates, perhaps with a smattering of independent or third party contenders. That kind of election (and the polling that goes on for them) happens a lot, so we can access a lot of data to draw out systematic patterns. From this pattern, we can make judgments about how much polls in a current election might mean.
Louisiana’s U.S. Senate elections – particularly for an open seat – do not look like that. They involve many candidates listed together on a single ‘primary’ ballot regardless of party. If no one gets a majority on the first ballot we get a runoff of the top two, again regardless of party.
A Louisiana election could look like a general election more typical in other states if we make it to a runoff that includes one Democrat and one Republican. At this point, however, we do not know who those candidates would be (or, necessarily, if there would even be a runoff). Early polls here cannot ask about a race just between two major party candidates. In other states, it more typical to know who will be (or who is likely to be) on a general election ballot as the major party nominees by this point in the campaign.
So, we are dealing with a different kind of election and a different kind of poll question in Louisiana in 2016 than what Enten worked with in his analysis.
It is a far rarer scenario as well. Louisiana has only three open seat U.S. Senate contests races under these election rules: 1986, 1996, and 2004. Those elections included 14, 15, and 7 candidates respectively. There are a lot more moving pieces, than the typical two-candidate matchup found in the other 49 states.
Even worse, there has been very little early polling in these contests. I found just two publicly-released polls take from March through June in 2004 and just one in 1996.
None of them comes anywhere close to the actual margin of victory either for the first round of voting or for the runoff in 1996 (there was no runoff in 2004).
In 1996, Mary Landrieu led a March poll with 23 percent. Woody Jenkins placed fifth at 8 percent. In the first round of voting, held in November, Jenkins led at 26 percent and Landrieu placed second at 22 percent. The following month, Landrieu beat Jenkins by about 6,000 votes.
The March poll leader ultimately won in December, but it did not predict the November vote leader. In fact, it did not even have Jenkins in the top two. The poll margins are also quite different from the election margin of victory. Landrieu was up +15 over Jenkins in the March poll, but she trailed him -4 in the November vote and ended up essentially tied in the runoff.
Things were not much better for early polls in 2004. Sure, Vitter led in both (24 percent and 38 percent) and then won the contest. But these early polls did not indicate he would secure 51 percent of the vote and win without a runoff. The rank order of candidates by support also differed from early polls to actual vote. John Kennedy – yes, the same John Kennedy – placed second in both early polls (22 percent and 25 percent) but ended up third on Election Day (15 percent). Instead, Chris John came in at 29 percent after showing just 10 percent and 19 percent in the early polls.
Vitter’s lead over his next closest competitor was +2 in one poll and +13 in the other. His margin on Election Day was +22. His lead expanded in 2004, but Landrieu’s shrunk in 1996.
Let’s take stock:
- The leaders in the early polls did ultimately win, but…
- These early polls did not tell us who would make the runoff (or even if there would be a runoff).
- These early polls did not tell us what share of the vote each candidate would end up with.
- These early polls did not tell us what the margin of victory would be, or even whether the poll margin would be larger or smaller than the actual vote.
It looks like these early polls have a modest connection to what happens in November at best. Maybe it is just harder to poll a multi-candidate extravaganza than a more typical two-candidate general election.
But even this is just a hunch…based on only three polls, from just two elections both of which occurred more than ten years ago.
There just is not enough data to even say how poorly these polls might be at predicting votes. Maybe they’re always this bad, maybe better…maybe worse. If we know this little about how these polls fit with election outcomes, then there is no basis for drawing a line from last week’s poll to what might happen on November 8, 2016.