Volume 25: Back to the House
A bigger and better log-loss comparative analysis of quantitative and qualitative projections for the 2020 U.S. House of Representatives elections
We’re back at it.
After 2018’s well-received analysis of the accuracy of House race raters and modelers, we’ll make another attempt for 2020. But we’re including more participants, gathering more data, and producing more analysis.
Especially in years with Presidential elections, Congress tends to get the short end of national attention. This is especially true for the House of Representatives, and it is unfortunate. Although its members are often little known and its behavior is occasionally unruly, the House is the closest body we have to a representative and responsive federal democracy. It is not separated from the voters by an electoral college. It does not have widely disparate representation depending on which side of a state line you live. It is not appointed and confirmed—it’s elected directly, by all of us.
It is, also, fascinating for a statistician. Presidential and Senate races are plagued by small sample sizes. A set of 435 samples is much more fun. So, it’s in the House where we’ve been gathering data, and in the House where we will carefully consider the results.
What's this project again?
In addition, I have created my own qualitative ratings for each district.
The significant difference from our previous analysis is that we do not have only the current ratings for each seat. We have, in fact, been capturing all ratings changes since each forecaster’s launch. This collection of data will allow for an analysis of how the national environment has changed over the election cycle—at least, as it is represented by these forecasters’ predictions. This also allows us to run other interesting analyses, for example, a time series of the accuracy of each forecaster by day. Hopefully, we’ll see them getting more accurate as the election gets closer, demonstrating that the professionals are understanding and incorporating news and data as it becomes available.
What are the experts saying?
We’ve undertaken this project despite the fact that control of the House of Representatives after the upcoming election is not really in doubt.
Recall that in the 2018 election, the House was retaken by the Democrats, gaining 40 seats and winning the overall House vote by about 8% nationally. Democrats won seats that had been traditionally Republican, or had been won by Donald Trump in 2016, or both. Because of this, at the start of the current election cycle, most expected a rebound in GOP fortunes. The question being asked was whether Nancy Pelosi would be able to hold on to a slim majority. Now, the quantitative and qualitative experts unanimously agree that the Democrats are more likely than not to gain seats on November 3rd, building on their gains from the 2018 elections.
There are several reasons that views about this cycle have changed. First, court-ordered redistricting in North Carolina has likely moved two seats to the Democrats, right off the bat. Second, several popular GOP incumbents in swing seats are not running (for example, TX-23, GA-07, and IN-05). Third, several GOP incumbents have lost primaries that unnecessarily put seats in play (CO-03 and VA-05). Then, the GOP had recruitment failures in several districts ripe to be retaken (ME-02, NY-19, and KS-03, among others). Finally, the unpopularity of Donald Trump has put more traditional Republican seats in play (IL-13, NY-02, and MO-02).[1] Also, several freshman Democrats are running noticeably effective campaigns in otherwise tough seats (SC-01, CA-45, and UT-04).
The Republicans do have a few bright spots. David Valadao, a popular incumbent who lost a close race in CA-21 in 2018, returned for a rematch. Freshman Debbie Mucarsel-Powell drew a strong opponent, Carlos Gimenez, the popular, long-term GOP mayor of Miami-Dade County. Dave Loesback, a Democrat from the swingy IA-02, hung up his spikes after seven terms.[2]
But on net, the Democrats are viewed to hold the stronger hand, irrespective of their large existing majority. We see this by looking at each rater’s expected number of Democratic seats.[3] Recall that the Democrats currently hold 233 seats to the Republicans’ 202.[4]
As you can see, all the raters in our project now expect the Democrats to gain seats.[5] The spread between the forecasters is small; only seven seats from low to high estimate. This is a high level of agreement and causes me to wonder if there is some amount of herding going on.[6]
Looking at ratings as a time series, we can see that the forecasters’ views have been moving steadily toward an expanded Democratic majority throughout the summer and fall:
While a move from 232 to 239 seats may represent less than 2% of the chamber, it is more significant than it appears. Among the qualitative raters, almost all of the rating changes have been in favor of Democrats over the past three months. These professionals have access to data and analysis that we do not; if they have been seeing a move this consistent over several months, it is highly indicative of the overall state of the race.
Qualitatively, we can see that all the race raters have the Democrats clearly favored in at least 225 seats (218 are needed for a majority). In addition, there are a relatively small number of Solid or Safe Republican seats, meaning that the Democrats potentially have a “long tail” leading to a caucus of 250 or more members:
Let’s look at some districts!
If you are anything like 58% of Americans, you don’t know the name of your own member of Congress, let alone somebody else’s.[7] This is a bad thing; each of the 434 other members of the House has as much say over policy affecting you as your representative does. Drilling down in individual districts also gives us information about how national elections play out on a smaller scale.
There are a lot of different ways to slice the data.[8] We can look at cohorts of seats, movement over time, or specifics of different race raters. Let’s start by looking at competitive districts where the sitting representative isn’t running.[9]
Among 14 Democrats not running for re-election, only one seat is competitive. However, among 39 Republicans leaving the chamber, there are 17 competitive seats, including 6 that are Tossup or better for Democrats. The simple fact is that 39 Republicans, representing almost 20% of their entire population in the chamber, have decided to exit stage left.
After considering open seats, we can consider which incumbents are endangered. It is unusual for an incumbent to be an underdog, but the following sitting members have at least a 30% chance of losing their seats.
The only incumbent currently not at least even money to be re-elected is Collin Peterson, a 15-term Democrat from western Minnesota. His district has steadily trended right for a long time and was won by Donald Trump by 32% in 2016. In a presidential election year, that may finally be too much partisanship for him to hang on. Otherwise, all the endangered Dems were elected in the Blue Wave of 2018; many were surprise winners in their first race, especially Kendra Horn and Joe Cunningham. We have mentioned that the GOP have drawn a few strong challengers; CA-21 and FL-26 stand out in this respect.
However, despite having suffered such a large setback two years ago, the remaining GOP caucus still has as many endangered incumbents as the Dems. Many of the GOP representatives on this list fit a pattern: moderates representing suburban seats that have seen support for Trump collapse at the same time that their districts have become more racially diverse.
Many of the seats won by Democrats in 2018 were in previously unfriendly territory. How are Democrats doing in red territory?
It’s an interesting group of representatives. There are a lot of freshmen, but also other younger members and a lot of members with military backgrounds. Impressively, 16 of 30 members have managed to make their race only minimally competitive (Likely or Safe D). Every member on this list, save one, voted to impeach Donald Trump.[10] It does not seem to have hurt their electoral prospects.
Using our newfound time series analysis, we can also look at which seats have moved the most over the cycle—in both directions.[11]
Movement toward Democrats:
Movement toward Republicans:
Among the seats that moved toward Democrats, we can again see some patterns. In some districts (CO-03; VA-05), incumbent, likely safe GOP members were defeated in their primaries by fringe candidates,[12] putting the seats in jeopardy. Some Dem incumbents failed to draw significant challengers. Also, some historically GOP seats in the suburbs are moving left quickly.
On the Republican side, CA-25 and FL-26 feature especially strong GOP recruitment.[13] IA-04 saw the primary defeat of the odious Steve King, who will not be missed by his leadership. We also see some Dem weakness among heavily Cuban-American sections of South Florida, as well as some incumbents facing challenges in rural districts.
Finally, I mentioned above the “long tail” of potentially competitive GOP-held seats. We can look at some demographic details of the 20 seats that are currently rated Likely R on average.
Many of these seats have been targeted by Democrats before, including high profile 2018 races in KS-02 and KY-06. But look at North Carolina and Texas. In the former, our oft-mentioned court-ordered redistricting not only gave Democrats two almost certain pickups, but also left three more potentially competitive seats. Texas’s congressional map was a GOP gerrymander in 2011, but growth in the large suburbs—and Democratic gains in the suburbs—have “broken” that map. There are nine potentially competitive GOP-held seats in Texas; one chamber of the gerrymandered state house could also change hands.[14]
Finally, let’s take a look at those all-important districts that are most uncertain. The following 17 districts are currently rated Tossup (in aggregate).
My greatest concern about the 2020 election is the concerted attempt to pre-emptively and spuriously contest the results. This strategy, led by Donald Trump and William Barr, and with the support of too many leaders in the Republican party,[15] is based entirely in bad faith. It takes anecdotal evidence of rare but real issues inherent in a complex process, changes the details, and massively magnifies the potential impact. This game plan is intended to create a vile brew of innuendo, involving nationwide plots of tens of thousands of election officials of both parties, with the only goal being to steal the election from Donald Trump. Every single example cited by this cabal has been thoroughly debunked. On several occasions, the Trump campaign has been forced to admit in court that their claims are based on no evidence. But dismantling a conspiracy theory requires far more time and energy than building one. And fighting back against this has the perverse effect of leaving little time to solve actual problems in our election administration, of which there are many.
Data is an important element in the fight against Trump’s conspiracy to delegitimize representative government in the United States. Surveys of public sentiment and voting intention, as well as the opinions of the professionals included in this piece, are not sacrosanct. But they serve a critical, independent check against the type of widespread fraud that Trump and crew are inventing whole cloth. The election result in 2016, while unexpected, was easily understood through the data ex-post facto. Incorrect assumptions about the likely electorate resulted in a split between popular and electoral vote outcomes. While unexpected, it was easy to see afterward how such a split happened.[16]
We will need to do the same in 2020. We do not know exactly how the data will be wrong. Polls and forecasts have never been perfect before, and they will not be this year. Elections are probabilistic affairs. Each piece of data should affect our priors, but no quantity of data can give us absolute certainty.
But lack of certainty is vastly different from knowing nothing. We do not know for certain if it will rain tomorrow, or who will win this week’s Monday Night Football. But professional meteorologists and sports handicappers are good at their jobs, and their prior probabilities prove out over reasonable sample sizes.
Elections are different; motivated reasoning will play a larger role in stories about politics than about weather. It is important to separate those who know what they are talking about from those who will say anything to get on TV, and to separate those who have a political aim from those who attempt to explain the evidence honestly. It is also important to recognize the limitations and uncertainties of election data, as it is in all data. But understanding the election results while ignoring the data is like planning what to wear without looking at the weather forecast; you will miss a lot of useful information.
Appendix:
The following link includes the latest rating from all raters as of October 27th. I will create and post a new link with final ratings as of November 3rd.
The race raters' websites can be found in the following locations:
[1] Some seats, such as NY-02, fit into multiple categories. [2] A professor of political science at a small Iowa liberal arts college, Congressman Loesback came out of nowhere to win his first term. He’s served in Congress for 14 years longer than anybody ever expected him to. [3] For qualitative forecasters, I’ve decided that Safe or Solid means 99%, Likely means 90%, Lean means 70%, Tilt means 60%, and, as it always has, Tossup means 50%. [4] Actually, it’s 232-197 with 1 Libertarian and 5 vacancies. [5] All ratings are the latest as of October 22nd. I’ll post updated/final ratings as close to the election as practical. [6] To give an example: I’m obviously looking at what others are doing and incorporating their views, consciously or unconsciously. The qualitative forecasts are often included as variables in at least some of the quantitative forecasts. [7] This 58% number is a reasonable estimate based on reviews of several polls on this topic. [8] I’m taking requests. [9] We’ll define competitive as seats where at least one rater has the seat as anything other than Safe or Solid. Note that MI-03 is currently represented by Justin Amash, who left the GOP caucus and is currently a Libertarian. I’ve left off the two seats in North Carolina that the Democrats are virtually certain to gain due to redistricting. [10] Collin Peterson. Jared Golden also voted to impeach on only one of two counts. [11] This won’t include the two R-held seats in North Carolina that are now Safe D due to court-imposed redistricting. [12] It bears noting that the guy who is a bigfoot erotica enthusiast is the “not fringe” candidate in VA-05. [13] In CA-25, the GOP won a 2019 special election after the Democrat resigned due to a sex scandal. [14] If so, it would prevent further gerrymandering; the Texas house will be the most important state-level race on November 3rd. [15] Albeit with some notable exceptions. [16] For what it’s worth, election data analysts worth their salt saw the eventual result as a reasonable likely outcome. It was mostly pundits who viewed the election as a foregone conclusion, and their views were generally not based on evidence.