Tag Archives: trail heat polls

Is Obama “Unbeatable”? Whistling in the Graveyard of Trial Heat Polling

Is Obama unbeatable in 2012?  With the economy showing no signs of recovery, joblessness hovering above 9%, the poverty rate on the rise, median income dropping, Obama facing the lowest approval ratings of his presidency and with his favorability rating now below 50%, and with the recent release of Ron Suskind’s book that purports to show a president not in charge of his own White House, you might think the obvious answer is a resounding “NO!”

According this article by Tim Noah at The New Republic, however, you might be wrong. Noah rests his case on recent trial heat polls that show Obama easily coming out ahead in one-on-one match ups against his main Republican rivals. For example, in this Public Policy Polling survey that was in the field shortly after the President’s job speech, Obama bests each of the five top Republican candidates.

For Noah, the fact that Obama is still beating his opponents in trial heat polls despite the dismal economic climate and his own administrative struggles indicates his strong position going into 2012.  Indeed, Obama’s biggest worry, Noah warns, may be that “these [trial heat polling] numbers might make him overconfident. How lucky can you get?”

How lucky indeed?  Alas, for Noah and other Obama supporters, given the current electoral fundamentals, pinning one’s hopes on trial heat polls taken this far before the election is the epitome of whistling in the graveyard.  In fact, history shows that these early polls are not reliable predictors of actual election results. As evidence, Sarah Pfander and Owen Witek went back and examined previous trial heat polls between the incumbent president and the opposition candidate at roughly this same point in the calendar, and compared the surveys to the election outcomes. Because prior to 1980 only Gallup consistently polled more than a year before the election, I’ll focus here on trial heats for the five most recent elections involving an incumbent president, dating back to 1980.  The findings suggest that Obama supporters should not rest their hopes on the results of trial heats this far out.

Let’s start with the 1980 election.  In trial heat polls in April and August of the previous year, Ronald Reagan ran about even with Jimmy Carter, but with large numbers of voters saying they were undecided.  By January, 1980, however, Carter was leading Reagan by 62%-32%! As late as June, 1980, Carter was still polling ahead of Reagan by 6-7% in trial heats.  Of course, Carter lost that election by almost 10% in the popular vote.  Flash forward to 1984. In August, 1983, and again in January, 1984, Reagan and Mondale were tied in trial heat polls.  Reagan, of course, went on to crush Mondale by almost 20% in the popular vote. It doesn’t get better from here. In 1992, of course, Bill Clinton was a virtual unknown, but in the earliest trial heat poll from January of that year, the incumbent George H.W. Bush was beating him by 15%.  In fact, Bush lost to Clinton in a tight race.  In 1996, the trial heat results are complicated because pollsters assumed Ross Perot would again be the third-party candidate, and he was often included in early surveys between Clinton and Dole. Nonetheless, the earliest survey we could find, from late 1995, has Clinton ahead of Dole by 11%, with Perot running third.  However, a Gallup Poll from January 1996 without Perot has Clinton’s lead over Dole down to 4%.  Clinton, of course, won handily by almost 9%.  Finally, in 2004, two polls from 2003 have Bush leading Kerry by 15% and 3%.  The latter survey came closer to the actual results; Bush won by a bit more than 2%

The following table put together by Witek summarizes the results of the those trial heat polls taken closest to the current point in the 2012 election calendar, and compares them to the actual poll results.  This is a rough guide, of course, but as you can see, except perhaps for 1996, they are completely unreliable indicators of the likely popular vote results and in three of the cases they don’t even predict the winner. (I don’t count the 1984 trial heat as a correct prediction since it falls within the poll’s margin of error.)

Election Year Trial Heat Prediction 10-14 Months Before Election Actual Winning Margin Net Difference Projected vs. Actual

1980

Even Reagan +9.75

9.75

1984

Reagan +1 Reagan +18.2

17.2

1992

Bush +15 Clinton +5.5

20.5

1996

Clinton +11 Clinton +8.5

2.5

2004

Bush +15 Bush +2.5

12.5

If trial heat polls this far out don’t tell us much with a great deal of confidence, when can we begin to rely on them to accurately predict the election outcome?  Brendan Nyhan, in this article cites more systematic research by political scientists Christopher Wlezien and Robert Erikson indicating that, not surprisingly, trial heats’ predictive power increases as we get closer to the election.  Indeed, their accuracy grows in almost linear fashion, as indicated by the following chart from the Wlezien/Erikson paper.  It shows trial heat polls’  predictive reliability increasing the closer we get to election day.  (Think of the left-hand axis as measuring how much of the final outcome can be predicted from trial heat polls, with “1” indicating that polls are in effect perfectly predicting the outcome.)

This is because as voters begin to pay attention to the race, the fundamentals that influence how they are likely to vote also begin to drive the trial heat results.  In short, when we get to Labor Day – or about 60 days before the election- we should see a closer convergence between trial heat polls and what our forecast models predict based on these fundamentals.  At this point about two months before the actual election, of course, many political scientists issue their forecast models based on those fundamentals.

Meanwhile, there will be undoubtedly be many more trial heat polls during the coming months and they will receive a good deal of media coverage, particularly as pundits’ cherry-pick the results that seem to support their preferred candidates.  In truth however, the entertainment value of these polls is greater than their predictive value, and the media coverage of them should be judged accordingly.