It's way too early to discuss the November 2010 Congressional elections. But did that ever stop anyone? Of course not. In fact, with deadlines for filing as a candidate and the necessity of putting together a campaign organization and a warchest, the outlines of the nationwide battle are indeed taking shape.
A number of organizations have lists of who's likely to win re-election and whose not, which seats are open and which seats are safe. One such list can be found at Swing State Project, which categorizes all 435 House races into six categories: incumbent locks (not shown), Likely D, Lean D, Tossup, Lean R and Likely R. Another is Cook, which does a similar categorization.
Perhaps we can do an analysis based on their assessments.
I've written code to produce the expected number of seats gained or lost by each Party -- if we're willing to assign a few plausible probabilities to the above categories. And to run simulations to determine the probability that Republicans will gain control of the House, assuming each of the contested House races is independent of the other ones.
There are 256 House seats that either are currently occupied by Democrats, or were occupied by Democrats and are now vacant (some of which will be filled by special elections before November, but even if so, will have a November election). To win the House, the Republicans must win a net of 39 seats.
I've initially assigned the following probabilities for each category:
Category | P(D->R) | P(R->D) |
Lock | 0.0 | 0.0 |
Likely D | 0.1 | 0.9 |
Lean D | 0.3 | 0.7 |
Tossup D | 0.5 | 0.5 |
Lean R | 0.7 | 0.3 |
Likely R | 0.9 | 0.1 |
And here are Swing State's and Cook's assignments, giving the number of seats they believe are in each category:
| Swing Cur. Ds | Swing Cur. Rs | Cook Cur. Ds | Cook Cur. Rs |
Likely D | 29 | 2 | 35 | 0 |
Lean D | 36 | 0 | 29 | 1 |
Tossup | 27 | 1 | 25 | 2 |
Lean R | 1 | 2 | 4 | 3 |
Likely R | 2 | 13 | 2 | 12 |
So, for example, Swing State says there are 28 races considered to be Tossups,
and of those 28, 27 are in districts which have, or last had, a Democrat as its representative, and 1 is in a district which has, or last had, a Republican.
Here's the results based on the above probabilities and each site's category assignments, with the probability that the Democrats lose the House based on 10,000 simulation runs:
| Exp. Dem losses (seats rem.) | Prob. of 217 Dem seats or less |
Swing State: | 25 (231) | 0.0% |
Cook: | 26 (230) | 0.0% |
What are we to make of these results? The current Intrade contract on the House turning Republican is around 40%, a far cry from the 0% predicted by the above model! Nate Silver thinks the expected number of seats the Democrats will lose is in the low 40's, not around 25 (or at least he did as of mid-February).
Any number of things could be wrong. The assumption of statistical independence for each race is flawed, to be sure, but how to compensate for that is unclear. The probabilities I've assigned to the various categories could be wrong -- e.g., it might be that the average 'Tossup' probability is actually 55% in favor of the Republicans, not the 50% coin toss I assumed. (In other words, for a given race, even if Swing or Cook thought that it slightly favored the Republicans, they would still put it into the 'Tossup' category, and it happens that this is true for the vast majority of those tossups (see below for how this would play out)).
It could also be that the seers are extrapolating to the future, not just looking narrowly to the current state of each race. Perhaps they anticipate that things are not going to get very much better economically in the next seven months, and people are going to be much angrier by then, and directing that anger fully at the Obama administration. And/or that the currently popular 'throw the bums out' sentiment will intensify, and because the Democrats have so many more incumbent seats at stake they will be destroyed by this wave.
In the not too distant past, there have been a couple swings of a sufficient magnitude (39 seats), which, if they occurred in 2010, would cause the Democrats to lose control. The Democrats lost 54 seats in 1994. In 1974, the Democrats won 49 seats. On the other hand the 2006 election, quite a shocker in its own right, only resulted in the Democrats gaining 31 seats. Nate Silver does some analysis of these so-called 'wave' elections here.
In any case it is interesting to see the disparity. Right now, if Swing State and Cook's categories are to be believed, and chance were to decide the outcome of each race, there is almost no possibility the Republicans would take over the House. Yet everyone is talking about how likely it is that Republicans will be in control of the House come 2011.
We can extend this analysis a number of ways. We could ask, say, how many Tossup seats would have to become Likely R, say, for the Dem's to lose the house. Or what happens as we move one race in each category to the next worst category (i.e., move one race from Likely D to Lean D, one race from Lean D to Tossup, one race from Tossup to Lean R, etc). I've chosen a somewhat different way of looking at the issue, which involves instead shifting the probabilities of each category instead.
If we assume, for example, that things get worse for the Dems, we could model this in a simple way by, instead of having a Tossup be a 50-50 coin flip, shifting the odds so that any race in the Tossup category has a 55% chance of going from D->R, and only a 45% chance of going from R->D. Likewise all the other probabilities would shift by identical amounts (never exceeding 1.0, or going below 0.0 of course).
So what do we get as we assume things get worse for the Democrats? It turns out that both Cook and Swing State produce similar results, so I'm just presenting the results based on the Cook assignments, again using 10,000 simulations to determine the probability of the Dems losing the House:
Shifted Prob. | Exp. Dem. loss (rem) | Prob. Dems lose House |
P('Tossup' -> R): 0.50 | 26 (230) | 0.0% |
P('Tossup' -> R): 0.55 | 31 (225) | 5.0% |
P('Tossup' -> R): 0.60 | 37 (219) | 35.7% |
P('Tossup' -> R): 0.65 | 42 (214) | 75.9% |
For example, according to this model, if things got bad enough for the Dems so that their current batch of Tossup candidates only had a 40% shot of winning their race instead of a 50% shot, and other candidates chances shifted accordingly, then the Republicans would be on the verge (but not quite there yet) of winning the House. The Dems would be expected to just hold on with 219 seats, and there would be an almost 36% probability that the Democrats would end up with 217 seats or less.
So even if things get worse for Democrats than they are at present, this model predicts they still have a good shot of holding on to a majority in the House. (Again, the model may be flawed in any number of ways).
Note: We don't really know what it means in terms of political reality for the probabilities to shift by 5% or 10% as I'm doing, but it will likely be highly correlated to how the economy does, Obama's approval and popularity polls, and polls measuring how likely people are to vote for a Dem vs a Rep, and how enthusiastic voters in each category are.
Whatever's really going to happen in November, I hope you enjoyed this analysis.