About seven months ago, something good happened on Twitter (I know: a shocking and rare event!). My former colleague (and all-around-great-person) Sean Trende posed a question to the wonkier side of the elections analysis world:
Imagine you’re placing a large bet on the outcome of the House, but you can know who will win any one House race beforehand (but not the margin). Which do you pick?I might say TX-32. @Redistrict @nathanlgonzales @NateSilver538 @Nate_Cohn @geoffreyvs @kkondik @databyler @DavidNir
— Sean T at RCP (@SeanTrende) February 28, 2018
The basic idea here is to figure take a look at which districts the data world collectively thought would be bellwethers earlier in the cycle (when we had much less district-by-district data, fewer special election results, incomplete retirement data and more) and see if we can learn anything. And my fellow data nerds responded! So I decided to look through Twitter (and other sources) to find responses from people our readers might be familiar with and summarize the results.
| District | Nominator | FiveThirtyEight GOP Win Probability | CNN Forecasted Margin (GOP) | Latest Upshot Poll GOP Margin |
| ME-02 | Dave Wasserman, David Byler, Nathaniel Rakich | 38.50% | 1 | 0 |
| CA-45 | Nate Cohn, Dave Wasserman, Clare Malone | 30.30% | -2 | -5 |
| TX-32 | Sean Trende, Henry Olsen | 67.60% | -1 | 1 |
| VA-02 | Amy Walter, Geoff Skelley | 68.90% | 3 | 3 |
| CA-21 | Nathan Gonzales | 78.40% | 7 | NA |
| IA-3 | Robert Wheel | 37.50% | -1 | -1 |
| IL-06 | Nathaniel Rakich | 54.50% | -2 | 1 |
| IL-12 | Amy Walter | 70.20% | 2 | 9 |
| KS-2 | Henry Olsen | 46.80% | -5 | -1 |
| MI-08 | Nate Silver | 48.10% | 1 | 3 |
| NJ-07 | Dave Wasserman | 35.60% | -1 | 1 |
| TX-7 | Henry Olsen | 49.00% | -1 | 3 |
| VA-10 | Perry Bacon Jr. | 10.80% | -6 | -7 |
| WA-08 | Kyle Kondik | 48.50% | 2 | -1 |
This table should (hopefully) be relatively straighforward. The first column is the district (state abbreviation plus number), and the second column is the journalist/analyst(s) that picked the district (I looked around on Twitter and in articles for this: If I missed you or wrongly attributed a pick to you, I genuinely apologize. Email me your tweet and I’ll put in a correction). People often offered more than one pick, so some names come up more than once. The last three columns show the probabilities for GOP victories in the FiveThirtyEight House Forecast, the CNN House Forecast and the latest Upshot Live poll as of Tuesday midafternoon. Larger numbers are better for the GOP in all three of these columns.
Overall, I think we did a decent good job of picking competitive districts. The FiveThirtyEight projections suggests that they’re all at least somewhat competitive, and CNN forecasts a single-digit margin in each race. Collectively, these picks are basically a toss-up. I did a weighted average (weights based on how many people nominated each district) of the FiveThirtyEight win probability and the CNN forecasted margin for each district, and on average these races were toss-ups. And it’s encouraging that we weren’t a total hive mind: The districts in that list have different demographic features, are in different regions, and lean in different directions in the upcoming election.
But to me, the overall closeness of these races is interesting. It means that Republicans are doing somewhat better in this group of bellwethers than they’re doing overall. Major quantitative models suggest that the race for the House should be rated as at least “Likely Democratic,” which essentially indicates that Democrats have a solid advantage but a Republican upset is still definitely on the table.
You could argue that the collective closeness of these races is a reason for increased Republican optimism. But I wouldn’t recommend coming to any big conclusions based on this table. I don’t know why other analysts made the picks they did, but it’s possible that they were going for something other than finding the one seat that would tip the balance of power (e.g. someone might have thought they had a good idea of where the popular vote was heading and then picked a district that’d tell them something about a broader demographic group). Moreover, it might be hard for anyone (people who do this for a living included) to pick the right districts roughly eight months before an election.
To me, this is mostly a fun exercise. It’s a way to look back at what we knew eight months ago, take stock of what has changed and what hasn’t and do elections analysis in a different, game-like way. And after the election has wrapped up, we can argue again about whose pick was the best.
