I am an assistant professor of political communication at Louisiana State University's Manship School of Mass Communication.  This site is where I bring together my love for studying politics, my love for analyzing data, and my love for the state of Louisiana.   

Polls, Fundamentals, and Predictions... Or, Would a Runoff Really Be That Close?

Polls, Fundamentals, and Predictions... Or, Would a Runoff Really Be That Close?

Two polls released in the past week have generated a lot of buzz in the Louisiana gubernatorial election.  Both polls – one conducted by Public Policy Polling and sponsored by an anti-Vitter PAC, the other conducted by Clarus Research Group and sponsored by The Advocate and WWL – show Vitter in a tighter than expected race if he lands in a runoff with Edwards. 

Unexpected poll results offer dramatic storylines and make for interesting conversation, but that doesn’t mean they tell us a lot about what would happen on Election Day.  When making predictions it is important to keep in mind not only what we see in these polls but also what we know about how these kinds of elections tend to go. 

The polls should not be dismissed…    

Neither pollster gets a great grade in 538’s ratings based on methodology and predictive power (a B- for PPP and a C+ for Clarus).  PPP uses automated polling and does not call cellphones.  That is part of the reason many media outlets do not cover their polls; they just do not satisfy the accepted standards for high quality scientific polling.    

Still, the fact that the two polls have similar patterns is telling.  One unexpected poll result could be random error, two showing a similar pattern are less easily dismissed as a fluke of method.  

Also, looking at PPP’s final poll projections from last year’s gubernatorial and senate races, they came within 4.1 percentage points.  This means that the margins between the top two candidates in their final polls came within about four points of the actual vote margin.  This is quite good.  The average miss by pollsters in the 538 database of polls from previous years is 5.3 percentage points.  In other words, despite their methodological issues, PPP’s late campaign polls from last year hit the actual results pretty well.           

…but early polls are not good predictors of what happens on Election Day

These latest polls in the Louisiana gubernatorial contest are out two months before the runoff.  Polls this early are far less accurate at forecasting the Election Day margin of victory.   For example, last year’s PPP polls from about two months before Election Day missed the actual margin between leading candidates by 8.2 percentage points.  (To find this, I pooled PPP’s 2014 polls for governor or senate taken two months before Election Day plus or minus three weeks, 17 polls in all.) 

In about a third of those early polls, the margin between candidates was off from the Election Day result by double digits.  In most of those elections the early polls forecast tight contests, but eventually those races became comfortable victories.  In Louisiana, for example, the early PPP senate poll last year forecast a slim Cassidy win in a runoff against Landrieu, 48% to 45%.  In the December election Cassidy’s margin was four time greater, 56% to 44%.  It works the other way too.  In the Virginia senate race the early PPP poll showed a comfortable lead for the Democrat, but the contest turned out to be a nail-biter on Election Day.       

This is not a PPP problem.  In fact, the accuracy of their forecasts from late campaign polls suggests that these early polls are accurately measuring voters’ intentions at that time too.  It’s just that their intentions are a moving target, changing over the final months of the campaign. 

So, if Edwards and Vitter end up in a runoff we should not be surprised if the Election Day result looks quite a bit different from the current polling. 

The fundamentals matter, a lot

Political scientists usually think of elections in terms of the ‘fundamentals,’ which lay the ground on which an election unfolds and usually predict outcomes very well.  Chief among the fundamentals is the partisan leanings of the electorate.  Partisans almost always vote for their party.  That’s not a sexy story, but it’s one borne out by data time and time again. For example, 90% of Republicans and Democrats vote for their party’s nominee in a presidential election.   

Gubernatorial elections are not much different.  To calculate the share of partisans who vote for their party’s candidates in gubernatorial elections, I looked at all such elections since 2006 for which an exit poll was available, about 80 elections in all.  (Louisiana is not part of these data, a consequences of having elections in odd numbered years.)  Over this period, roughly 85% of partisans ended up voting for their party’s candidates for governor.  The trend has been on the rise in recent years: 81% in 2006 and 2008, 85% in 2010, 90% in 2012, and 89% in 2014. 

Partisan voting may be a little weaker in gubernatorial elections than in presidential elections, but it is still quite strong. 

Polls move in the direction of the fundamentals

Early in a campaign, voters rely less on partisanship.  By Election Day they tend to fall in line.  Consider the race for governor of Arkansas last year.  In the early PPP poll, the Republican candidate led by six points.  In the late poll, he led by ten points.  Then on Election Day he won by 13 points, more than double the margin in the early poll. 

How were voters changing?  In the early poll, 77% of partisans were backing their party’s candidate.  In the late poll, 83% were backing their party’s candidate.  In the Election Day exit poll, 90% of partisans said they voted for their party’s candidate. 

How might the polls in Louisiana move?

If typical patterns hold, then the margin between Edwards and Vitter (if they end up in a runoff) will likely shift from what we see in the recent polls.  What’s more, that shift will likely involve more partisans “coming home” to their party’s candidate. 

The PPP Edwards-Vitter runoff question has just 59% of Republicans backing the Republican candidate.  If that were to hold on Election Day, it would be almost unprecedented.  In only five of the 80 gubernatorial races since 2006 for which I have exit poll data did a candidate do that poorly (or worse) among voters of the same party: 2006 Tennessee, 2006 Wyoming, 2008 West Virginia, 2008 Vermont, and 2010 Colorado.  That’s just one out of every 16 elections.  It is extremely rare that a candidate fares that poorly among voters from his or her own party. 

The WWL/Advocate poll has Vitter’s Republican support at 72% in a runoff with Edwards.  This comes closer to typical patterns, but is still well below the norm.  Candidates in gubernatorial election do that poorly or worse among their own party in only about one in four elections.  (To be fair, Edwards is polling lower than expected among Democrats in this poll, especially among blacks).              

So, can a Democrat beat Vitter in a runoff? 

If typical patterns hold, a Republican will have the inside track in a runoff against a Democrat in Louisiana - regardless of who that Republican is.  The state’s electorate has grown more Republican over recent decades (more on this in a future post). 

But, partisan voting patterns are not the only “fundamental” in gubernatorial elections.   The second big one is the overall mood of the electorate.  The Louisiana electorate appears ready for a change.  The Reilly Center for Media and Public Affairs at LSU’s Manship School of Mass Communication released a poll this summer showing that a majority of Louisiana voters think the state is heading in the wrong direction (Full disclosure: I worked on that poll).  That could matter too, especially if one of the candidates in a runoff is closely linked in the minds of voters with the current administration. 

The question is:  Can the mood of the electorate combined with the idiosyncrasies of the particular candidates in a runoff (which tend to play only a minor role relative to partisanship) be enough to move voters out of their partisan patterns?  The only answer today is: We. Just. Don’t. Know. 

The recent polls indicate we could have a competitive runoff, but we should take them with a grain of salt knowing that the typical patterns suggest otherwise. 

Good predictions draw upon all the relevant data – both polls and fundamentals – and weight each accordingly.  Good predictions are also probabilistic, not deterministic (hence my frequent use of “likely” and “unlikely”).  So, to say that Vitter definitely wins over Edwards if the two meet in a runoff would be an overstatement that ignores the information these two polls offer.  However, to say that Vitter would be in trouble in a runoff is an even bigger overstatement that ignores the fundamentals.  Those statements might make for interesting conversation, but neither is fully substantiated by the data.  

Slow Starts, Negative Strikes, and the Role of Super PACs: Everything You Wanted to Know about This Year's Ad War

Slow Starts, Negative Strikes, and the Role of Super PACs: Everything You Wanted to Know about This Year's Ad War

How much do the polls in the governor's election tell us about who will win?

How much do the polls in the governor's election tell us about who will win?