Monday, September 21, 2009

Too Early for Talk of Democratic Losses?

The last few weeks have seen a number of political pundits begin to speculate about what the current political situation might suggest for the 2010 midterm elections. For instance, in a recent post about 2010, Charlie Cook, a generally spot-on prognosticator, suggested that there are a number of troubling signs for the Democrats:
With 14 months to go before the 2010 midterm election, something could happen to improve the outlook for Democrats. However, wave elections, more often than not, start just like this: The president's ratings plummet; his party loses its advantage on the generic congressional ballot test; the intensity of opposition-party voters skyrockets; his own party's voters become complacent or even depressed; and independent voters move lopsidedly away. These were the early-warning signs of past wave elections. Seeing them now should terrify Democrats.

Given the time that remains between now and the 2010 elections, I was initially skeptical of Cook's warning to the Democrats, so I decided to examine some of the indicators cited by Cook. Although I'm not aware of good time-series data on out-party intensity or in-party complacence or depression, Gallup makes it is fairly easy to find data on presidential approval (for the general public and separately for independents), as well as on the generic congressional ballot. So, the question is, how well do these indicators predict midterm elections 14 months out? In other words, do current conditions necessarily provide clues to next year's congressional elections?

Let's look first at presidential approval. The figure below uses approval data from approximately 14 months (or as close as I could get) prior to midterm elections to illustrate the relationship between approval and seat change for the president's party:

Presidential Approval and Midterm Losses/Gains, 1946-2006

There appears to be positive pattern here but it is weak enough (r=.30) that the relationship is not statistically significant (p=.26). Of course, there is a lot of variation around the regression line, including some data points that are clear outliers; in particular, the 1946 election. In this case, Truman's approval was a sky-high 82% in October, 1945, just a few months after the end of WWII, but the Democrats went on to lose 54 seats in the midterm elections 13 months later. If we assume that Truman's approval was artificially and temporarily inflated and drop 1946, the patterns looks a bit tighter:

Presidential Approval and Midterm Losses/Gains, 1950-2006

Here it appears that early reports of presidential approval are fair predictors of midterm losses (r=.60, p=.02), though there are still a number of data points that are significantly off the regression line: 1958, 1966, and 1994 on the negative side, and 1998 on the positive side. Still, there is a clear relationship here, one that might foreshadow the outcomes of next year's elections. Based on the Obama's current level of approval (52%, Gallup polls in September), the trend in these data predicts a seat loss of 28 seats for the Democrats in 2010.

Should this be cause for Republican jubilation and Democratic hand-wringing? Not quite. The problem with forecasts like this one is that the sample size is small enough(n=15) and the forecasting error is large enough (standard error of forecast=17.9) that a 95% confidence interval around the prediction ranges from a loss of 67 seats to a gain of 11 seats. In other words, the prediction from these data encompasses everything from a complete Democratic collapse to historic gains for the Democrats. The best guess is still substantial Democratic losses, but with plenty of hedging.

What about Obama's standing among independents? Is it particularly important to the Democratic fortunes in the 2010 elections? It turns out that presidential support among independents is no more or less important than the overall level of presidential approval. Note that figure below (the relationship between presidential approval among independents and midterm losses) looks remarkable like the figure that used overall approval (above):

Presidential Approval Among Independents
and Midterm Losses/Gains, 1950-2006

In fact, the two figures are virtually identical, as they should be, given that the correlation between approval among independents and the overall level of approval is .96 for this data series. At latest reading, Obama's approval rating among independents is 49% (is this "lopsided?"), just three points lower than his overall rating.

Finally, we turn to the generic congressional ballot. It is clear from other research that the generic ballot predicts well in the fall of election years (see Abramowitz), but is it really of much use 14 months out? In a word, no. The figure below plots midterm losses against the in-party advantage* on the Gallup generic ballot question, from 1950 to 2006:

Generic Congressional Ballot and Midterm Losses/Gains, 1950-2006
Although there appears to be a slight positive trend in these data, the pattern is not strong enough to generate very meaningful predictions (r=.27, p=.33).

Taken together, these data show that presidential approval 14 months prior to midterm elections has some predictive capacity, though there is enough error in the predictions that you shouldn't bet the farm. Approval among independents does not really add much to the predictions generated by overall approval, and the generic ballot this far out is not very useful at all. So what does this all mean for the 2010 elections? The picture is still fuzzy, but Cook is right to suggest that current conditions do not augur well for the Democrats. But I would still emphasize the fuzziness of that prediction and that it is based on presidential approval and not the generic ballot.

*It is difficult to get generic ballot data exactly 14 months prior to the midterm elections. In eleven cases, I was able to get data from 12 to 15 months out, while the remaining four cases used data measured appreciably close to the midterm elections.