What needs to be factored in is how far off each poll was from the 2016 election result. This needs to be done on a state by state basis and in comparison to stated margins of error. A poll predicting Trump would lose Pennsylvania by 5 points with a stated 2 point margin of error has a track record of unreliability. Unless it explains why it was wrong and what it has done to fix its bias or margin of error underestimation, it’s future prognostications should be viewed with skepticism. If it now says Ds should win by 3 points with 2 point margin of error, you might not want to bet the farm. Even worse, since it is systematically missing something, and that something may be changing, any trends from such a poll should be considered even more speculative.
A key problem is that the subpopulation of people responding to polls could well be highly biased. I hardly ever answer my phone for a number I don’t know and neither does anyone else I know. When I do answer for what is billed as a poll, more than half the time it’s a barely disguised campaign pitch. In 1932 telephone polls predicted Roosevelt would be soundly defeated. Were they wrong! Phone polls may have a similar problem today.
The translation of national approval numbers into Congressional election results is also not straightforward. I’ve heard several Hillary voter’s say Trump is terrible but they don’t like the street confrontations, impeachment talk, and other extremes pushed by the Left wing of the Democrats. They don’t approve Trump, but I wonder if they will all show up and pull the D lever like last time. I also think there is a fluctuating percentage who won’t admit to favoring Trump but who did vote for him and who will show up this November.
The distribution of voters is also critical. Democrats may run 2 points better in Deep Blue parts of California, but run 2 points worse in toss up districts in Ohio, overall a negative net electoral result.
Thank you for providing interesting stats and historic perspectives. Next time I’d suggest adding in more caveats about the accuracy, relevance, and reliability of polls.