The following article by Philip Bump was posted on the Washington Post website September 6, 2018:
One of the tricky things about assessing midterm elections is that they are much more complicated than, say, a presidential race. In a presidential contest, there are two major-party candidates vying for victory in 50 states. Or, really, in about 10 states, depending on the year. In a midterm, each of the 435 seats in the House are contested, with up to 870 candidates from the two major parties. (That number is always lower, since some candidates run unopposed.)
As is the case with states in a presidential year, not all of those 435 seats are likely to switch hands. But scores will. What’s more, while in a presidential year you can look at polling for the two candidates nationally and get a sense for how states might vote (though this is, uh, not a perfect system), guessing the winner of particular swing House races based solely on national House polling — the generic congressional ballot — is a whole different beast.
Happily there are sites which go through each race on a near-daily basis and assess how likely it is that each seat will end up being held by the same party after voting concludes. One such site is the Cook Political Report, which does regular ratings of House races on a number of different metrics, offering a sense of how the overall national picture might look after Election Day.