On Monday we offered an estimate based on the fundamentals of House elections that predicted Democrats would pick up only one additional seat, with a 1 in 4 chance of regaining the majority in that chamber. How certain is this estimate? And what does that tell us about the value of this whole enterprise?
The first question is easy enough to answer. The standard error for the vote share estimate is 5.6%; for seat share, 8.7%. That’s a lot of uncertainty. It means there is at least a little probability of some pretty crazy outcomes. It explains why there is still a 1 in 4 chance that the Democrats will get the 25 seats they need to retake the House, when our own median prediction is only one seat.
The uncertainty is especially large because we don’t know exactly what the districts are like post-redistricting. Under the old redistricting plans in each state, some districts were consistently more competitive than others, beyond what the presidential vote in the district would already imply. There are a lot of reasons why that might be true. Perhaps the incumbent in that seat was particularly vulnerable, or faced a tough challenge year after year. Perhaps the seat contained a lot of wishy-washy partisans. Whatever the reason, even though we can’t explain the difference, we know roughly how large it was. Thus, any prediction involving those districts can include that extra information.
When it comes to the new districts, we don’t have that information. All we know is how the voters in each district voted in the 2008 presidential election. As a result, we’re forced to push the remaining uncertainty into our estimates, and the range of possible outcomes increases accordingly. If we pretended instead that the old districts were being used, the standard error of our seat prediction would be a lean and mean 1.9%, rather than the 8.7% we currently have.
The size of our error might reflect the economy of our model. We wanted to see how far the fundamentals could get us in House elections, and the answer may be, “pretty far, but not as far as we might like.” As I said in the original post, including things like spending or handicapper predictions might reduce the otherwise unexplained differences between districts and make us more certain. Or it may be we’re as close as we can get, and more intangible factors need to take over from here.
Regardless, I would argue that this is exactly the point of such a model. It takes some very important factors that people talk about a lot with respect to House elections, and it says both how important each one seems to be and how confident we can be about using that information to make forecasts. We can expand the model and play with different possibilities, and every result tells us something about the dynamics of House elections and what they mean for this particular year. That makes it easier to have a conversation about what matters, how much, and even why.