Republicans need a polling error to win the House. Fortunately for them, polling errors happen all the time. That's why they have a ~15-20% chance. 53eig.ht/2RyJvu6
Trying to thread a needle here:
1—Small-to-medium sized polling errors happen a lot
2—The errors are often correlated, affecting many districts at once
SO
3—The GOP has a shot at the House
BUT
4—The House polls are really quite poor for the GOP, maybe more than people realize
1—Small-to-medium sized polling errors happen a lot
2—The errors are often correlated, affecting many districts at once
SO
3—The GOP has a shot at the House
BUT
4—The House polls are really quite poor for the GOP, maybe more than people realize
Per our Deluxe forecast, the GOP has to go 24-10 in the 34 most competitive districts. That's really hard to do *unless* polling errors are correlated. There's also a LONG list of "Likely R" (but not "Safe R") districts where Democrats are competitive and could cause problems.
If you're a skeptic on the importance of the "Likely R" districts, scroll through our list of the 105 (!) most competitive races. I think you'll get pretty far down the list before you reach districts that the GOP can really feel great about based on polling and other indicators. 
—If there's a systematic polling error in the GOP's favor, those Likely R districts won't be a big problem
—If polls are about right overall, they'll produce headaches that could offset GOP wins in tossups
—If there's polling error AGAINST the GOP, they'd yield major D seat gains
—If polls are about right overall, they'll produce headaches that could offset GOP wins in tossups
—If there's polling error AGAINST the GOP, they'd yield major D seat gains
But to bring this back: When people look at 538 and ask "what if the polls are wrong?", that's *the question our model is trying to answer*. The polls say Ds will win the House. Our model says, based on how accurate polls have been in the past, there's a 15-20% chance they won't.
