As I mentioned in the models’ methodology section, a lot of people helped make this forecast become a reality. Perhaps no one made it more of a reality than Parker Quinn, a data scientist. He did a lot of the work on the House model. He had previously modeled House elections for his masters’ thesis.
I asked Parker a few questions about the state of play in the House. You might call this getting nerdy with election modeling. This question and answer has been edited.
Harry: Geographically, where are we seeing the 20 closest races in the House?
Parker: Our model had (as of this write-up) 22 toss-up districts (where both candidates have between a 40 and 60% chance of winning). These districts are distributed very evenly across the country – six of them are in the northeast, five in the Midwest, five in the South, and six in the West – suggesting that the battle for the House isn’t confined to a certain part of the country.
Interestingly, the next 20 most competitive districts (where the favored party has between a 60 and 75% chance of winning), we see only two western districts, and zero northeastern districts!
Harry: Do they tend to be in areas that are more or less rural than the national average?
Parker: This is a fight for the suburbs. Toss-up districts are about twice as likely to be sparsely suburban or an urban-suburban mix than the rest of the country. Only two of these districts are completely urban or completely rural, compared to about a quarter of all other districts. Unlike the geographic trends, this is not likely to change significantly – races that lean toward one party (but may become toss-ups by November) are even more likely to be in sparse suburban districts.
Harry: Are they well-educated districts?
Parker: Overall, about 35% of adults in these districts have degrees, compared to 29.7% for the rest of the country. The rates for the races that lean toward one but are close are nearly the same. It seems that no matter what way you looked at it, the tightest races this November will end up being about 20% more educated than the country overall.
Harry: Are there any districts that you’re focusing on?
Parker: Comparing two toss-up districts – Illinois 6 and New Jersey 7 – is a great way to understand the national trends. Both cover the suburban portions of major metropolitan areas (Chicago and New York) and are low or medium in population density. These districts are about 80% white and among the most educated and wealthy districts in the country – both have median incomes of around $100,000 and 50% of the population are college educated. That said, they are slightly different types of suburbs and are in completely different regions of the country.
Harry: What’s the effect of adding district polling on our model? A lot of it is partisan that needs to be adjusted.
Parker: Removing the polls from our model means relying more heavily on the fundamentals and expert ratings. Luckily, the fundamentals are already fairly predictive themselves, and the experts usually take into account polls when rating districts. So, predictably, taking the polls out does not have a huge affect on the “top line” results – the number of seats we expect each party to win and how likely they are to control the house.
The effect of removing polls is more obvious at the district level, where polling often captures the characteristics of this particular election better than the fundamentals. For example, in Texas 23, our model without polling data would show something close to a toss-up race. However, the polls that have come out tell a different story with Republican Rep. Will Hurd heavily favored, which is what our forecast shows.
Harry: Finally, let me ask you a question that’s been asked of me on Twitter. We present a margin of error, but some nerds want to know the “probability” of a Democratic takeover.
Parker: Our model produces both top line and district-level probabilities, but we prefer to focus on margins since those are a little more intuitive. If presented as a probability, our topline results would show Democrats at about 3-in-4 favorites to take the House.
This leaves plenty of room for Republicans to hold the House, as our current topline margin of error suggests.