December 4, 2017

The Crosstab: Data and Democracy | Blog by G. Elliot Morris

Chance of Winning a House Majority: Popular Vote Forecast

In my projection of the two-party election day vote share, based on polls of the generic ballot, The Democratic Party is ahead, earning 54.4% of the two-party vote share on average. The margin of error is roughly 4.1% points, meaning the Democrats could earn as little as 50.3% or as much as 58.5% of the vote.

But, because Democrats are clustered in cities and face harsh gerrymanders, they aren’t expected to win an equivalent share of the seats in Congress. What does electoral geography tell us about the actual outcome? Below are the generic congressional ballot polls used to make that projection: FOLLOW LINK

Outlined seats are the top 20 “tipping point” districts.

Simulated Seats Over Time

Democrats earn a median of 218 seats  in our simulations of the 2018 midterms. This may differ from the strict predictions below because of the larger number of Lean Republican seats than Lean Democratic seats in the current Congress. Effectively we are saying that the below number is an ideal estimate, meant to give you context as to which seats are competitive, but that we expect Democrats to overperform expectations based on the assessment of our error in past predictions.

Individual Seat Projections

Using the average vote share for each district over all of our simulations, we can identify both the districts that have the best chance of flipping parties and the chance that that happens.

Democrats: 211
Republicans: 224

Democrats are likely to pick up 18 seats on November 6, 2018. Republicans are favored to gain 1 seat, for a net gain of 17 seats for the Democratic Party.

Seats Likely to Flip Parties in 2018: FOLLOW LINK

Who’s Vulnerable?

The graph below stacks each House seat on top of eachother above the percentage share of the vote I forecast for the Democratic candidate in the district. The gray shaded area represents a 5% margin of error — roughly what we expect given past error in the national generic ballot polls — identifying vulnerable seats that could be won by either Democrats or Republicans.


Tipping Points and the Majority Power Indicator (MPI)

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).

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.

Together, the tipping point index and MPI tell us which House districts are most instrumental in producing control of the House majority. More information can be found here.

Race Ratings

Each seat is given a vulnerability rating based on the following scale for either party:

  • Tossup: win margin less than 5%
  • Lean Dem./Rep.: win margin betwee 5% and 15%
  • Likely Dem./Rep.: win margin between 15% and 25%
  • Safe Dem./Rep.: win margin greater than 25%

Ratings also take into account qualitative factors like candidate characteristics, contest dynamics, fundraising, etc. And, as a general rule, no open seat forecast to change parties will ever be rated “Safe.”

These ratings are available on the model’s companion post at Decision Desk HQ.

Possible Outcomes

Of course, there is error in our forecasts — measurable error. We can account for that error by simulating the election thousands of times, asking the computer each time to pick random error for national polling that is based on error from past predictions. Then for each “election’s” national error we also generate an error for each congressional district, basing each of those 435 individual errors on what the error is in districts that have performed the same in the past. (So, for example, if a simulated error adds 4% for the Democratic candidate in a district that voted for Romney and Trump by 5%, the error would around 4% for other districts that voted similarly. This is called correlated random error.)

Here’s the distribution of total number of Democratic House seats after engaging in that exercise:

We can take all of those simulations of Democratic seat shares and further ask “what is the chance that Democrats win the election, given the chance that Democrats or Republicans beat expectations?”

Democrats have a 51.8 percent chance of winning a House Majority of November 6, 2018. You can imagine that the rectangle below is a dart board, and if you randomly threw a dart anywhere at the rectangle whichever square you land on corresponds to the party that wins the House majority.