Don't Count Your Chickens . . .

The most recent (September 19) polling aggregation from Pollster.com shows President Obama with about a three-point lead over Governor Romney.  This, coupled with another bad couple of weeks of media coverage for the Romney campaign, has buoyed the confidence of Obama supporters and has begun to generate a sense of a hapless Romney campaign and an inevitable Obama victory.   Before Obama supporters get too carried away with their euforia, or Romney supporters with their despair, I'd like to remind them both of a guy named Al Gore, who held a fairly commanding lead over George Bush for most the better part of September in the 2000 presidential race:

Note that Gore's lead through about the third week in September was much more substantial than President Obama's current lead in the polls.  Of course, Gore went on to narrowly win the popular vote while narrowly losing the Electoral College vote.  This is not too say I expect a sudden reversal of fortunes in the next couple of weeks.  Just that the current margin really is quite narrow and conditions are still such that a reversal is a real possibility.  Both sides should be anxious; this thing is far from over.

Poll of Forecasters

Jim Campbell has assembled a list of several forecasts that will appear in the October issue of PS.

Here it is.

Forecast

Now that I have final summer data from the Survey and Consumers I can generate a forecast from my National Conditions and Incumbency model.  The gist of my model is that the incumbent presidential party is held accountable for prevailing national conditions but that the level of accountability depends upon whether the incumbent president is running for reelection.  Granted, the idea that the president's party is punished or rewarded based on prevailing conditions is hardly revolutionary.  Where my model departs from most others, though, is that it incorporates incumbency as a conditioning variable, with the expectation that national conditions are stronger predictors of election outcomes when an incumbent president is running.  Note that I'm not saying conditions don't matter in open seat contests; just that they matter more in incumbent contests.   National conditions are measured here with an index that takes into account the level of presidential approval (Gallup) and aggregate levels of satisfaction with personal finances (Survey of Consumers), both averaged over the summer months of the election year.   You can access a copy of my paper here.

The key Figure is posted below:

Figure 1. National Conditions and US Presidential Elections, 1952-2008

As this figure illustrates, national conditions track much better with election outcomes in years when the incumbent president is running than in open-seat contests.

So what does all of this mean for the 2012 election?  Right now the national conditions index (explained in greater detail in the paper) stands at 63.3.   This is the fifth lowest reading from the fifteen elections presented above, and the third lowest of the nine elections involving an incumbent. Based on this value for national conditions, the forecast for the 2012 election is 47.9% for President Obama and 52.1% for Governor Romney.  

I'd like to be clear that the national conditions index for 2012 points a less-than-ideal situation for a sitting incumbent, but it does not suggest that conditions are so bad that the challenging party can coast to victory.   Consider the years in the sample for which the national conditions index had a lower value than the current year: 1952, 1980, 1992, and 2008.  These were all really tough years for the incumbent party, and it is easy to see why when considering the prevailing conditions in those years.  I see 2012 as somewhat different.  As I say in the paper, given the small sample of elections, and the size of the out-of sample prediction error, this points to a tight race but one where Governor Romney is the favorite.

Bump Time

With the major party conventions kicking off next week it's time to return to one of my favorite topics, conventions bumps.  Four years ago, I posted a summary of convention bumps and I'm updating that information here so we can get a sense of what to expect this year.

I measure convention bumps as the percentage point change in the convening party's share of the two-party vote, comparing polls taken between six days and two-weeks prior to the convention with polls taken during the seven days following the convention.  Note that this is a short-term measure of the convention bump and does not say anything about the rate of decay in the weeks following the convention.  The figure posted below summarizes the convention bumps for both the Republican and Democratic nominees from 1964 to 2008.

The first thing to note is that there is a lot of variation in convention bumps. Fortunately, as I showed in Do Campaigns Matter?, there is a systematic component to that variation. Two things in particular seem to drive the size of the bumps. First, candidates who are running ahead of where they "should" be (based on the expected election outcome) tend to get smaller bumps, and those running behind their expected level of support get larger bumps. In this way, the conventions help bring the public closer to the expected outcome and help to make elections more predictable. A perfect example of this phenomenon is the 1964 conventions. Goldwater got a huge bump, in part because he was running 16 points behind his expected vote share, and Johnson got no bump, in part because he was running 6 points above his expected vote share prior to the Democratic convention.  For the years presented above, the correlation between how far ahead of the expected outcome (based on an election forecast) a candidate is running in pre-convention polls and the size of the convention bump is -.42.  While all candidates may want to get a big bump from their convention, big bumps are not always a good thing; they could signal that the campaign is not doing as well as expected.

Timing also plays a role in explaining convention bumps.  When I first started looking into this (1996) there was a tendency for the party holding the earliest convention to get a larger bump.  This made sense for a couple of reasons. First, it is the out-party holding the earliest convention and people usually know less about the out-party nominee, since the incumbent party almost always runs an incumbent president or vice-president (John McCain being the most recent exception).  In addition, historically the first convention had been held sometime in late July or early August, a time period when there might be more undecided voters.  Importantly, the tradition at the time was that the in-party convention would usually be held three to four weeks later, which gave time for the message from the first convention to resonate.  However, in 1996 and 2000 the conventions were held fairly late and only two weeks apart.  And, more recently, the 2008 conventions were held in back-to-back weeks in late August and early September, creating a situation in which the messages of both campaigns no doubt interfered with each other.  This, of course, is also the timing of the 2012 conventions. When taking these changes in timing into account, it turns out that the number of days a convention is held before or after¹  the other party's convention is a stronger predictor of convention bump (r=.37) than simply going first (r=.22).  Of course, the "days between conventions" measure captures both how early the first party convenes and how close together the conventions are.

Finally, it also appears that there is a general trend toward smaller bumps in the last few elections cycles, though this could simply reflect the changes in timing of the conventions.  Truly large bumps were somewhat common prior to 1996 but have not made an appearance since then.

Taking all of these factors into account, and adding a dummy variable to control for the disastrous Democratic conventions of 1968 and 1972, we can see that there is a predictable element  to convention bumps:

                      

This is not to say that there aren't interesting and unique aspects of every convention that might explain why the convening party over- or under-performs the expected outcome--just that, by and large, the pattern of convention bumps is explained by a few simple variables.  The correspondence between predicted (by the model shown above) and actual convention bumps illustrates this point quite well:

               

Generally, conventions expected to produce large bumps tend to do so, and those predicted to produce small bumps also fit the pattern.  What's important to note, though, is the the pattern is far from perfect and some of the larger deviations from the from the predicted outcome make sense, given what we know about those specific conventions.  For instance, the Democratic convention of 1984--a convention plagued by in-party fighting--underperformed by a little over five percentage points, while the Democratic convention of 1992--one that benefit from Ross Perot dropping out of the race and near-endorsing the party's nominee, Bill Clinton--exceeded expectations by almost six percentage points.

2012 Bumps

So what does this all mean for the conventions coming up in the next two weeks?  I won't have complete data for the Obama bump prediction until the opening days of the Republican convention, and my general election forecast (used in the model) is preliminary at this point (I will post that model when all the data are in; preliminary data point to narrow Romney win).  But I don't expect any of the data to change dramatically in the next few days, so I will go ahead and make a prediction for the 2012 convention bumps:

Predicted Romney Bump:    3.6 percentage points
Predicted Obama Bump:      1.1 percentage points

Right now, the two candidates are in a tight race, with Obama holding a slight advantage in the polls.  Based on my bump predictions, I expect that the race will continue to be tight after the conventions but that Romney will hold a slight lead.

Huge caveat: Hurricane Isaac.  I have no earthly idea how a hurricane-shortened convention will afect things, though I suspect it would not be good for the Romney campaign

¹So, the first convention gets a score equal to the number of days it begins before the first day of the second convention, and the second convention gets a score equal to the number of -1 times the number of days it begins after the last day of the first convention. 

Ryan Reaction

Here.  (I didn't pick the headline).

It's (Not) the (State) Economy, Stupid!

A couple of weeks ago I saw another story about about how the economies in swing states were improving at a faster clip than the rest of the country, and this might very well be the key to President Obama's re-election prospects.  I understand the appeal of this type of argument; after all, if the public is constantly being exposed to the old refrain, "It's the economy, stupid," it may seem natural to expect that voters living in states with more rapidly improving economies would be more likely than voters who live in states with weaker economies to support incumbent party candidates.  It also makes sense that so much attention has been paid to  the unemployment rates in the states, given the overall high levels of unemployment.   However, I did some work on this a long time ago (dissertation, actually) and found that what appeared to be state-level economic effects in a pooled vote model vanished in the presence of controls for the state of the national economy.  And even though others (see Campbell and Klarner) have found significant state-level economic effects in more recent pooled models that used broad indicators such as change in income and change in economic output, I'm still skeptical that state-to-state variations in economic conditions--especially those based on unemployment--have much to do with presidential election outcomes.

Let's take the 2008 election as a case in point.  The first figure posted below shows the relationship between the September unemployment rate and the Republican percent of the two-party vote in the states in 2008.  As would be expected under the economic voting argument, there is a negative pattern to the data, indicating that John McCain (the incumbent party candidate) did worse in states with higher levels of unemployment.  The relationship is not particularly strong, but it is just barely statistically significant (p=.04, one-tail).   The second graph shows the relationship between the change in unemployment from January to September, and the relationship is of roughly the same magnitude. 

Although these relationships are not particularly strong, they are consistent with the state-level economic voting thesis and might be taken to indicate that my skepticism is uncalled for.  But let's take a closer look.

In the top graph, if you look in the upper left-hand corner--low unemployment and strong support for McCain, you find a collection of mostly Republican states: Wyoming, Oklahoma, Nebraska, North and South Dakota, Utah.  In the lower right-hand corner--high unemployment and low support for McCain, you find a collection of Democratic and competitive states: Michigan, Rhode Island, California, Oregon, and Nevada.  And the same pattern holds for the bottom figure.  In other words, what appears to be a relatively weak, negative relationship between unemployment and support for the incumbent party candidate, likely just reflects the coincidental overlap of unemployment and partisan tendencies.  Indeed, both relationships are spurious and disappear once you take into account the partisan predispositions (Republican share of the 2008 vote) of the states.  In fact, the partial correlation for the September unemployment rate is positive (ryx,z=.20), though not significant.

And it turns out the 2008 is not the only instance of null effects from unemployment.  The Table below provides some findings from simple regression models in which the incumbent party vote share is regressed on either the September unemployment rate or the change in the unemployment rate between January and September (separate models), along with the statewide vote share for the incumbent party candidate in the previous election.  The cell entries are the unstandardized regression coefficients for each year, along with an asterisk to indicate if the slope is significant at the .05 level.

Simply put, there is nary a shred of evidence to suggest that the unemployment outlook in the states has anything to do with state-level outcomes in presidential elections.

In fact, there are only two cases of significant effects--both for the September unemployment rate--and one of those is a positive coefficient, suggesting that the incumbent candidate, Jimmy Carter, benefited from higher unemployment!  While one could make a policy-oriented argument in favor of this, it seems like a bit of a stretch.  Te most obvious conclusion to draw from this table is that state-to-state variations in unemployment are not likely to have much impact on how the states vote this November.

This is not to say that unemployment doesn't have electoral repercussion.  In fact, as I showed last summer, changes in the national unemployment are closely tied to election outcomes.  Simply put, presidential elections are national elections and it is national conditions--not state-level variation around those conditions--that drive them. 

Veepstakes and Candidate Sex

Check it out.