How accurate are generic ballot polls for Congress?

[boffking]

Each of the 435 members of the House of Representatives is elected individually. A candidate may win, not because of sentiment towards their party, but because they are popular for reasons other than politics. Or their opponent may suffer a scandal. The spoiler effect could also play a role in some races if there are more than two candidates. None of these factors come into play in a poll that simply asks"Republican or Democrat".
Yet we hear over and over the results of generic ballot polls. How closely do they usually get to the results of the election?

[Fretful_Porpentine]

There was a piece about this last year at FiveThirtyEight. Short answer: fairly accurate, although as with any polling there’s a margin of error.

[DSeid]

They are historically moderately predictive of the national House vote total, per that 538 bit averaging 2 points off, but of course may miss one individual race more one way, offset by missing another individual race more the other. And hitting the national number on average missing by two points can mean half the time on the money and half the time off by four.

But

@538 4/24/18

… The bigger question is what to make of the disparity between the overwhelming swing toward Democrats so far in special election results — which would imply a Democratic wave on par with the historic Republican years of 1994 and 2010 — and the considerably more modest one suggested by the generic congressional ballot, which shows Democrats ahead by only 7 points and implies that the battle for House control is roughly a toss-up.1 One plausible answer is that the generic ballot will shift further toward Democrats once voters become more engaged with the campaign in their respective districts and pollsters switch over to likely voter models. Still, both the generic ballot and special election results (when taken in the aggregate) are fairly reliable indicators. Rather than choosing between them, it’s best to consider both. That means entertaining a wide range of scenarios that run between Republicans narrowly holding onto the House and an epic Democratic wave.

538 is more and more big on emphasizing that pretty much anything is possible.

[Chronos]

Hypothesis: Both Generic Democrat and Generic Republican are fantasies. In matchups between Generic Candidate and any named real candidate, Generic Candidate usually has an edge, because Generic Candidate never has any skeletons in his closet, and each individual voter can imagine that Generic Candidate’s views exactly align with their own. And as it happens right now, a lot of voters like the Republican Generic Candidate. But most individual real Republicans fall further short of the ideal of Generic Republican than real Democrats do of Generic Democrat, and so the lean towards the Democrats is greater in actual elections.

Specific (if extreme) example: Generic Republican would have easily beat Doug Jones in Alabama. But Roy Moore was very much not Generic Republican, and so Jones won.

[Tom_Tildrum]

@Fretful_Porpentine

There was a piece about this last year at FiveThirtyEight. Short answer: fairly accurate, although as with any polling there’s a margin of error.

That last part is key. As we saw with the 2016 presidential election, accuracy and precision are two different things, and the polls may be correct within their error range but still unable to reliably indicate a result other than “it’s probably going to be close.”

[Lance_Turbo]

Here’s the crux of the biscuit. Both of these two things are true.

[ol]
[li]It is not that accurate[/li][li]It is the most accurate information available[/li][/ol]