House-keeping: What is the path to a Republican House majority?
At this point, we know almost everything we’re going to know about the national environment in 2018. There is a slight chance that polls can change in the final 9 days of the campaign — see: 2016 election — but likely no more than a point or so, according to the distribution of historical error. That means that Democrats are going to lead in the polls by as few as 7 and as many as 9 percentage points on November 6th. Moreover, combined with other indicators, the actual chance in forecast national votes will likely fluctuate no more than 0.5 points in the next week and a half. So, the race is basically set where it currently lies.
Of course, Republicans are still holding onto hope that the latest cultural issues — Trump is campaigning
on a caravan of migrants that are “approaching” the US-border (they’re 1000 miles away) and promise to invade our borders (most will legally apply for asylum at a port of entry, and ~80%
of them will be denied) — can spur voter turnout at the last minute, rescuing close suburban and ancestrally Democratic seats from the “blue wave.” Credit where credit is due: Trump and Republicans are campaigning on an issue that stokes fears and resentment among their mostly white base, possibly increasing turnout. But with historical data saying that polls are unlikely to change this late, what they’re really relying on is the chance that those polls are off again. Maybe they don’t catch the late movement. Maybe they underestimate the share of non-college educated white voters. But there’s no way to know.
What remains now of Republican prospects to hold onto their House majority is almost solely correlated polling error. That is, if the generic ballot and 500+ district-level polls are consistently overestimating Democrats, then we’re in for a wild election night. GOP operatives might have hoped to be in a better situation right now, but with little room for shifts in the national environment and observations for 90% of plausibly competitive House districts, this is all that remains — “this” being a roughly 20% chance that Democrats pick up fewer than 23 seats on election night.
This path runs through Republican-held districts in the suburban northwests, Orange County, CA districts, and the whiter Obama-Trump districts in the midwest. If we are overestimating a shift toward Democrats among whites by putting too much weight on the college-educated, the tossup districts in these areas could end up staying red.
So, this is the threat, but how likely is the chance of correlated polling error this extreme? The easy answers are (A) that we can’t know until election day or (B) about 20%, given that that’s the chance our models say the Republicans have of holding the House. But by my best estimate, the polls that *are* properly weighting samples to avoid such an error — high quality live-caller national polls like those conducted for Pew Research Center, NBC/WSJ, CBS News, Economist/YouGov etc. and district-level polls from the NYT Upshot/Siena College — are about as Democratic-leaning as those that aren’t — small outfits fielding district-level polls and internet pollsters practicing bad weighting of national samples. But in both, Democrats have big enough leads — about 6 to 9 percentage points in any given average — to be favored to win the majority of House seats.
TL;DR This has been a long, in-depth, slightly rambling opening to this week’s newsletter. Here’s the takeaway: Democrats have imposing leads in both the high- and low-quality polls of the 2018 House midterms. The chance that Republicans hold the House has come down to the chance that these polls are wrong. If so, we currently have no clues as to what that would look like. Even so, in House elections, prediction error isn’t nearly as correlated as it is in presidential elections. The bottom line: a lot of things have to go right for the GOP to pull off a House majority on November 6th. That could very well happen.