The Edwin Diaz effect on the midterms
Losing Pythag, winning games

I’m having a good Facebook conversation with SABRMatt regarding 538’s predictive model on the upcoming midterms. Conventional wisdom suggests the Democats will win the house - but a recent poll shows that although the Democrats have a big party preference nationally, in the contested races the Republicans hold a slim lead.

I am calling this the “Edwin Diaz“ effect, that suggests a team like the Mariners can overachieve by winning the tight games, and breaking even or even losing the blowouts.

Matt has even more interesting observations regarding the polling data and the assumptions - the data being used to predict house races is basically bad. Perhaps he will weigh in here - I hope. It reminds me of how poor umps can hide what is really going on with a hot pitching prospect.

Hopefully, this little post will generate some good discussion while we enjoy the Seattle Pilots march to the WORLD SERIES!

P.S., I am terrible at sizing and placing artwork here, so if it loooks terrible, forgive me and you’ll know why I deleted this after you see it.



So there are several problems with polling for the House races that don't become an issue for Senate and Gubernatorial (and Presidential) races, even after you do things like give stronger weight to more recent polling, give more weight to polls with larger samples, and give more weight to polls with better historical accuracy.

1) National polling and media outlets routinely poll contested state-wide races...they do it a few times a week (as a group) for every race that matters. ABC, YouGov, Rasmussen, NBC/Marist, PPP, etc...they're constantly hammering away at the people trying to figure out how Senate and Gubernatorial races will go. But House races commonly see just a few polls the entire race. The main reason they don't get polled as often is a lack of funding. The parties actually pay money to many of the most active polling houses to give them data on statewide races, but it's not worth it to either party to do that for house races. The cost of doing the polling is too large relative to the amount of money spent on those races.

2) The few polls that do get done for any given house race are, most often, done by young start-up polling agencies that charge less, or by colleges out of academic curiosity, or by local newspapers. The lack of premium compensation buys you agencies that don't have a record of accuracy and accomplishment on which to base your analysis of the results. We don't know what methods they use, they likely aren't up to date on accurate sampling, and they probably won't sample a very large group.

3) The models used by analysts like fivethirtyeight to parse the polling data and make predictions from it are tuned on statewide races. They use a Bayesian-like estimate for the likely accuracy and bias of any given polling firm, for example, that starts with the assumption that an unknown polling company will have the average accuracy/error margin of any 'new' company in the sample of agencies from which they have data. This is likely wrong in House races, meaning their probability of victory odds will be overconfident. Also, they use things like political fundaments (state of the economy, average results of elections with X,Y,Z factors in place like who's the President and what's his job approval, etc), the R/D split of the region germane to the vote, and things like the national generic congressional ballot. Those fundamentals really are good things to look at and do help the models do better than simple polling statewide races where the models are tuned. But, in House races, the fundamentals don't translate as well. Regions with shared cultural identities would be more useful to look at, and, as the OP pointed out, in this case, the swing districts are thinking differently than the rest of the nation.

4) In this election season, in particular, there has been a major disruptive event that has changed the landscape in Senate races and in a few gubernatorial races. The September 30th hearing wherein Brett Kavanaugh and Christine Blasey Ford gave their emotional testimonies has drastically shifted the mindset of conservative voters and galvanized fence-sitting right-leaning moderates against Democrats. You can see the sharp break in polling results with populations entirely sampled after 9/30 very clearly. Since those hearings, Senate races in Nevada, Arizona, North Dakota, Tennessee, Missouri, and Montana have shifted by between 3 and 10 points toward the Republican column (interestingly, Florida hasn't moved...that Senate race seems to be divided along different lines for whatever reason). Before the hearings, Democrats had a near 50/50 shot to even up the Senate or take control, and they now have a miniscule chance of doing so as crucial flip chances have all but vanished. The problem is...if you go to and look through the House races, you'll have a pretty hard time finding any of them where polling was done after September 30th.

All of which is to say...I believe the results of this house election will be tilted more toward Republicans than the conventional wisdom currently has it...but how much more is unclear because the data is nearly useless.


There's no real way to "pick" who becomes a zillionaire, but of the Gates, Ellison, Brin, Jobs, Zuckerman, Bezos, et al, group, I think Allen has done the most to benefit those he grew up with - for him, the people of Seattle and the Pacific Northwest. If you used the Suzzallo library before his grant and the Suzzallo-Allen after, you have seen his contribution to UW. Whether or not you like the architecture, his museum of popular culture is a wonderful asset for Seattle. Anyway, his parents can be proud of the son they raised. He will be missed.


This is an interesting line of argument, but I'm not sure it's got as much traction as you may think. According to 538's Deluxe Model, there are 15 Toss Ups, 15 Lean R states, and 15 Lean D. Stated as percentages, that's a sample of 37.5% neutral races (dead heat), 37.5% republican-leaning races, and 25% democrat leaning races. So the sample is substantially skewed towards republican voters, since they make up 12.5% more of the races. As such, a one-point republican advantage in this forty-race sample is probably to be expected, even in a democrat-friendly environment.

Also per this model, the democrats have 193 "Solid D" seats, or ones where it would take a freaking miracle for them to lose, versus just 141 for the republicans. That leaves the aforementioned 40 close races, as well as 15 "Likely D" and 46 "Likely R". If you give each side all of their Solid and Likely seats (which is probably generous to republicans, since they have 3x the number of likely-but-not-solid seats) the races is at 208 D -- 187 R, with 40 seats in the middle. The majority hinges at 218, so the democrats only need to win 10 of the 40 close races to seal the deal. Republicans may have a 1-point "Edwin Diaz" margin in these races... but it'll be hard to translate that into a 75% win rate in contested games.

Just ask the Mariners. They'll be easy to find: they're sitting on their couches, watching the Red Sox/Astros/Dodgers/Brewers vie for the World Series. Y’know, the favorites.

Not to say this race is a sure thing: far from it. There's still polling error to consider, and voter turnout, and any last minute Comey-esque developments. And nothing I said reflects on Matt's points about regional identities, which he surely knows far more about than I do. I'd be fascinated to hear Matt and Nate Silver argue it out on the 538 podcast, which I've been listening to a lot over the last few months. I'm sure he's considered the same factors Matt has, and stands by their models anyway, which is interesting. Love to hear two subject-matter experts (in this case statistics) debate a topic where they've both looked at the same data and reached different conclusions. Always good stuff.

At any rate, we'll see how it goes in a few weeks.


...that I didn't say I thought republicans would keep the house. I think polling models are overestimating democrat numbers, but I still think they'll probably take the house.

Add comment

Filtered HTML

  • Web page addresses and e-mail addresses turn into links automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <blockquote> <code> <ul> <ol> <li> <dl> <dt> <dd><p><br>
  • Lines and paragraphs break automatically.

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.


  • Allowed HTML tags: <a> <em> <strong> <cite> <blockquote> <code> <ul> <ol> <li> <dl> <dt> <dd>
  • Lines and paragraphs break automatically.
  • Web page addresses and e-mail addresses turn into links automatically.