it seems like they're trying to fit national data for the house races, and using local data for the gubernatorial ones. they probably should have used the same methodology for the house (and senate.) that they're using for the governor races. and, given that this data probably exists here - as opposed to in a presidential election, where it might not - they probably have less of an argument about a deficit of data.

this is the same argument i've been making for years.

and i have no understanding at all about what affect it may have on the outcome.

one possibility that you could see is that millennial turnout may end up very high in already-leaning democratic districts. if that happens, the election results will be spitting on a fish, and not a lot of seats will flip. the popular vote will end up as a mirage. that is probably the most extreme type of error, but how unlikely is it, really? i mean, if what's driving this is saturday night live, the polls may be broadcasting a kind of circle jerk in the easily led "still watches tv in 2018" demographic - even if these are the same people annoying you on facebook with dubious articles from vox.

again: i have no understanding of what the numbers say. i'm not suggesting that this will affect the accuracy of the outcome. but, you should expect a model like this to be off by quite a bit - it can't be anything better than a crude guess.

https://fivethirtyeight.com/features/our-final-forecast-in-the-senate-house-and-gubernatorial-races/