The 3 Types of Districts that Could Give Democrats the House

The 3 Types of Districts that Could Give Democrats the House



A series of political handicappers, prognosticators, and pundits are all aboard the “2018 as a Democratic wave” train. While their conclusions are certainly backed up by my 2018 forecast model, at least to the extent that a wave has a 50-ish percent chance of happening, it’s worth it to take a deeper dive into where that wave could come from.

To that end, I’ve come up with three categories of House seats that could provide Democrats a majority in 2018. The first group of congressional districts is obvious: Republican-held seats that Hillary Clinton won in 2016. The second group of seats we can call “tipping points” — they’re the districts that most often provide the 218th seat for Democrats. Finally, the third group is based on a newly-developed measurement of importance to House strategy. Based on my modeling, these districts have a high “Majority Power Indicator.” These are seats that, when won, disproportionately improve the Democrats’ chances of taking back the House.

But before we start, a note on what this blog post is not doing. It is not a “Five Seats Democrats have to win for a House majority” take. Most of these articles are a bit silly; they assume that only certain seats will be competitive, rather than a type of seat being able to be flipped. Sure, there are some characteristics of a race that are particular to a seat — like candidate quality or fundraising — but for the vast majority of the time if Republicans are weak in one well-educated, diverse district they’re going to be underperforming in others as well.

This blog post is not meant to say that Democrats should focus all their efforts on a few seats and win the election. This would ignore other seats in which they have a golden opportunity to gain ground. In other words, when I say “Democrats should focus on PA-06 because it will disproportionately help them win the House” I don’t mean they should only try to win PA-06. They’d be ignoring the other 23 districts they have to pick up to gain power. Rather, they should focus on all the districts where they have a better chance of winning up to and including PA-06. With that in mind, let’s dive in!


1. Republican districts that Clinton won (23)


Split-ticket districts are the heart of any competitive ground game. In 2014, for example, eight of the 15 districts that Republicans flipped from Republican to Democratic control were districts that leaned towards Obama in the 2008 and 2012 presidential elections. In our 2016 election forecast, a larger share (roughly 70%) of projected Republican-to-Democratic turnovers are these split-ticket districts.

Those districts are listed here, sorted from most Clinton-leaning to most Trump-leaning:

District Dem 2016/14 (%) Clinton Incumb. Dem Win Prob.
Republican Districts that Clinton Won
FL-27 45.1 60.1 OPEN 99.0
FL-26 43.7 58.3 R 71.5
CA-21 43.3 58.2 R 69.9
VA-10 47.1 55.3 R 72.9
MN-03 43.1 55.1 R 59.2
CO-06 45.6 54.9 R 66.4
CA-39 42.8 54.6 R 56.6
CA-49 49.7 54.0 R 76.7
IL-06 40.8 53.7 R 46.4
CA-25 46.9 53.6 R 67.1
CA-45 41.4 52.9 R 45.7
AZ-02 43.0 52.6 R 50.5
NY-24 39.4 51.9 R 35.4
TX-23 49.3 51.8 R 69.2
CA-10 48.3 51.6 R 65.1
WA-08 39.8 51.6 OPEN 83.9
PA-07 40.5 51.2 R 36.5
TX-32 36.4 51.0 R 23.5
CA-48 41.7 50.9 R 40.1
TX-07 43.8 50.7 R 47.0
KS-03 44.2 50.6 R 47.6
NJ-07 44.3 50.6 R 48.3
PA-06 42.7 50.3 R 41.1

We can see that this table provides quite a wide range of districts. There’s FL-27, for instance, which most people know now as the best Democratic pickup opportunity in the nation. Its Republican Representative, Congresswoman Ileana Ros-Lehtinen, announced her retirement in early 2017 and the district immediately shifted to the Lean or Likely Democratic column in most ratings.

There’s also TX-32, however, which voted for its Republican congressman 64 to 36% in an uncontested race last year. Rep. Pete Sessions is looking to continue his reign there, and it’s uncertain whether any Democrat has yet come forward that can pose a formidable challenge to this 11th term Republican.

It may suit us to categorize these districts further. To determine where the groups fit best, I used a set of statistical techniques collectively called ‘cluster analysis’ to identify which groups have characteristics that set them apart. After taking into account the partisanship, race, education, and income of the 23 Republican-Clinton districts I saw that they fit comfortably into 3 categories:

  1. Diverse Democratic Partisanship:
    • AZ-02, CA-10, CA-21, FL-26, FL-27, NY-24, TX-23/li>
    • These districts voted Democratic at the presidential level since at least as far back as 2008. They have a more diverse Demographic profile (47% hispanic, 44% white 5% black) and swung 25% away from their Republican representatives since 2014.
  2. Moderate Flip-Flop:
    • CA-25, CA-39, CA-48, CA-49, CO-06, KS-03, MN-03, PA-06, PA-07, TX-07, TX-32, WA-08/li>
    • These are districts that, on average, voted by a hair for John McCain in 2008, handily for Mitt Romney in 2012, then flipped back to Democratic in 2016 to vote for Hillary Clinton over Donald Trump. The Demographic profile of these districts is mixed: 67% white, 6% black, 16% latino, 42% have a college degree, and average yearly income is $80K.
  3. Emerging Democratic:
    • CA-45, IL-06, NJ-07, VA-10
    • These are your craft beer districts. Your suburban college graduate, white, wealthy districts. They are full of small businesses and entrepreuneurs. In a nutshell, this district is the millennial D.C. district. With an average income above 6 figures, majority college educated and 71% white these seats are the ideal areas to launch a demographic wave. In the recent Virginia gubernatorial election VA-10 swung 3% toward the Democrats (which was about par for the course).

Roughly sorted in descending order from most to least Democratic, these districts provide a range of benchmarks for a Democratic wave. If challengers are running ahead in highly educated and highly diverse districts, they’re probably keeping pace with a slight loss in 2018. If they’re also looking at winning the mixed-majority and above-average white districts, we can be relatively certain that a House majority is in competitive play.

But we can do more than this demographic/partisan grouping. Utilizing the power of my 2018 forecasting model, we can see which districts “tip” the election towards Democrats most often.


2. “Tipping point” districts


Districts that usually fall in the middle of the pack are “tipping point” districts. They tell us that, in a tied election, these districts most likely land the 218th seat for the winning party. For Democrats, the “tipping point” district is the one that gives them the 24th seat they need to win the election given that they’ve already won 23 other seats (most likely the ones detailed above).

To figure out which districts are tipping points, I calculate the percentage of time it lands in that 218th spot. For each of the 50,000 simulations in my House forecast, I rank the seats from most to least Democratic. Then I go down that list and, for the winning party, move each seat they win into a list for that party. Once that list has 217 districts, the seat that they win next is the “tipping point” district, because it tips the election in their favor. Over the course of the simulations, the district with the higher tipping point value will be the one that most often tips the election to either the Democrats or the Republicans.

We can also split up the tipping point values by party. By this measurement, Democrats will have their own list of the districts that most likely tip the election, and Republicans will have their own list.

These congressional districts with the highest 20 tipping point values are listed below, arranged from highest Democratic tipping value to lowest.

District Dem 2016/14 (%) Dem. 2018 Forecast (%) Tipping Point (% of trials) D. Tipping Point R. Tipping Point
Top 20 2018 House Tipping Point Districts
MI-08 41.2 46.6 1.4 2.1 0.8
PA-07 40.5 48.3 1.8 2.0 1.6
IL-13 40.3 46.5 1.2 2.0 0.4
PA-16 44.4 47.2 1.4 1.9 1.0
FL-18 44.6 46.8 1.4 1.9 1.0
OH-01 40.8 46.5 1.4 1.8 1.0
VA-07 42.3 47.1 1.5 1.8 1.2
PA-06 42.7 48.9 1.8 1.7 1.9
KS-03 44.2 49.7 2.1 1.7 2.4
IL-14 40.7 47.0 1.2 1.7 0.7
NC-02 43.3 46.9 1.1 1.7 0.6
PA-08 45.5 49.2 1.8 1.7 1.9
TX-10 40.1 45.6 1.4 1.7 1.2
TX-24 41.2 46.7 1.8 1.6 2.0
IL-06 40.8 49.5 2.0 1.6 2.4
UT-04 43.4 47.1 1.1 1.6 0.7
NE-02 49.4 50.5 1.9 1.6 2.2
NC-13 43.9 46.5 1.4 1.5 1.2
TX-22 40.5 46.1 1.8 1.5 2.1
GA-06 38.3 45.8 1.6 1.5 1.7

As of today, Michigan’s eighth Congressional district is the seat with the highest Democratic tipping value. In just over two percent of our simulations, MI-08 provides Democrats with their House majority. MI-08 is a classic midwestern district; it had voted for Obama in 2008 only to fall into Republican hands in 2012. It became even redder in 2016 when President Trump and GOP Rep. Mike Bishop won by 7 and 17 points, respectively. In terms of the distribution of House seats — you can imagine a straight line where each seat gets places from left to right according to its Democratic lean — MI-08 is the 232nd district. The district is also 86% white, more than 10% higher than the nation as a whole, and similarly more educated than the median Congressional district.

MI-08 as a “classic midwestern district” is important. Indulge my model talk momentarily:

A good tipping point district is a seat that not only can conceivable flip parties but also has a strong relationship to other districts that Democrats need. MI-08 is important as a tipping point because it also has a strong relationship to other Lean Republican/tossup districts. Since our simulations take into account similarity between House seats (based on the partisanship, demographics, and region) when generating errors, hypothetical strong performance in MI-08 also means strong performance in its sister seats. Below I provide a graph of MI-08’s relationship with a random selection of districts, just to illustrate how this works.

By this measure of similarity (Pearson’s correlation coefficient), the congressional districts most like Michigan’s eighth are OH-14, OH-12, PA-08, IN-05, OH-15, PA-06, IA-03, NH-01, MN-02, and OK-05. These seats have an average partisanship of R+8. So, when Democrats win MI-08, they also win or come close in these other 10 districts. And since our calculations of the national environment tell us that Democrats need somewhere near an 8-9% lead in the national popular vote to win the Majority, MI-08 as a tipping point makes a lot of sense!

Circling back to my earlier point, this helps make the case that if Democrats are winning MI-08 they’re also likely winning the House majority. That’s the power of a tipping point district.

However, the tipping point measure has its weaknesses. For one thing, there are 435 seats in the House of representatives. Roughly 200 of them tip the election in at least one of our simulations. This has the effect of making the highest tipping point value roughly 2.5% for any given day. On top of that, the districts that do become tipping points all look to be cut from similar cloth: what’s the real difference between Democrats being competitive in MI-08 and PA-06, for example?


3. The “Majority Power Indicator” (MPI)


To provide some more direction to people who want to hone in on the important House districts, I created a measurement that quantifies the relationship between the Democrats (or Republicans) winning a particular seat and winning the election. I call it the Majority Power Indicator (MPI).

The Majority Power Indicator (MPI) is simply a measure of the increase in the probability that a given party wins the House majority given that they win a given seat. Mathematically, MPI is equal to (1) difference between (A) the number of trials a party wins a given seat and wins the House majority minus the number of trials they win that seat but lose the majority and (B) the number of trials that a party loses a given seat and wins the House majority minus number of trials they lose that seat but lose the majority, (2) all divided by the number of trials/simulations in our forecast model.

The subtraction allows us to account for safe Democratic/Republican districts that don’t correlate at all with winning a House majority. Republicans may win the House 30% of the time when they win TX-05 — an uber-conservative district the rural northwest of the state — but they also lose their majority in 25% of simulations where they also win TX-05. In other words, and to use another district, Republicans are going to win Wyoming’s at-large congressional district whether they win or lose the House.

In short, the MPI is a measurement of utility. Specifically it’s a measure of gained utility. It answers the question: “Which seats increase the likelihood that my party wins the election?”

The seats with the top 20 MPI are shown here:

District Dem 2016/14 (%) Dem. 2018 Forecast (%) Dem. Win Prob. MPI
Top 20 Districts With the Highest MPI
PA-06 42.7 48.9 41.5 58.3
NE-02 49.4 50.5 54.2 58.1
PA-08 45.5 49.2 44.0 56.9
PA-07 40.5 48.3 36.9 55.3
KS-03 44.2 49.7 47.9 54.1
IL-06 40.8 49.6 46.4 50.1
PA-16 44.4 47.3 29.0 49.2
VA-07 42.3 47.2 28.3 48.2
UT-04 43.4 47.1 28.0 47.1
IL-14 40.7 47.0 27.3 46.7
NC-02 43.3 46.9 26.8 46.1
FL-18 44.6 46.8 26.3 45.9
MN-03 43.1 51.2 59.3 44.6
MI-08 41.2 46.6 24.6 43.8
IL-13 40.3 46.5 23.8 42.6
NC-13 43.9 46.5 24.0 41.5
OH-01 40.8 46.5 23.8 41.5
CO-06 45.6 52.1 66.6 41.4
IA-03 42.6 48.0 34.0 39.7
ME-02 45.2 47.5 30.8 39.3

The district with the highest MPI is PA-06. When Democrats win PA-06, they win the House majority in 34% of simulations. Democrats lose in 3% of simulations where they win PA-06. That makes their residual probability of winning the House, if they win PA-06, 31%. When Democrats don’t win PA-06, the win the House 11% of the time — that’s 27% lower than the percentage of simulations in which they lose PA-06 and lose the House. Therefore, the MPI in the district is 58 points ((34 - 3) - (11 - 38)). This represents the boost in residual probability that Democrats get if the win PA-06.

You’ll notice that most of the GOP-Clinton districts aren’t on this list, just as they are also not generally tipping point districts. MN-03, for example, is at the 13th spot in the list. That’s because MN-03 doesn’t give Democrats as the much extra utility; they win the House in 63% of simulations where they win MN-03, but lose in 23% of simulations – a difference of just 44%. Minnesota’s Third is a good pickup opportunity for Democrats, but it’s also such a close race (one of the first, not last, pickups they would get in a wave) that it doesn’t give us a great hint of how the cycle is unfolding in the aggregate.

Like I said in the discussion on tipping points, this is not to say that Democrats should ignore the GOP-Clinton districts like MN-03 and TX-23. Rather, Democrats should give attention to those districts, where they have a pretty good shot at victory, but they should also focus on these seats that indicate victory in the aggregate. That’s what the MPI is useful for. Obviously, it would make no sense to say that Democrats should put all of their money into PA-06 because it’s the top-ranking MMPI. They’d only have 195 seats on November 6, 2018. That would be foolish. In sum, focus on the “easy” seats, but pay attention to districts that lie in the middle of the pack and are, in most cases, stronger indicators of majority control. If you’re eyeing the list of 23 Republican seats where Hillary Clinton won, don’t forget that extra seat they need to win.


This blog post began with my referencing a growing consensus that Democrats have at least even odds of winning back the House next year. That conclusion is sound — and certainly, our model has been saying something similar for a few weeks now — but the journalists and pundits making these assertions sometimes lack the micro-level data that a forecast model can provide. It’s not enough simply to say “Judging by the Virginia election, Democrats have a good chance of taking back the House.” We have to know the answers to other questions: Where can they pick up seats? Can they pick up enough? Are demographics favorable or unfavorable there? What about turnout? Are certain seats more strategic picks?

I used my forecast to answer many of those questions. I think we have arrived at what I set out to achieve, at least for now: a data-driven, strategic look at where Democrats should be competing in the coming cycle. The answers we’ve explored together look both familiar and unfamiliar — particularly the seats with high MPI. But just because the ratings say PA-06 Leans Republican, for example, doesn’t mean that Democratic efforts there won’t be useful. On the contrary, a win there may be the most useful.


G. Elliott Morris avatar
About G. Elliott Morris
Elliott is a undergraduate government, history, and computer science student at The University of Texas at Austin.


comments powered by Disqus