Yesterday, I posted the link to Yair Ghitza's Catalist analysis of voter composition in 2018. Here are some comments on the data, based on scrutinizing the spreadsheet linked to in the article.
First, it must be said there are some very important differences between these data and the exit poll data. This is true of both levels and trend. That is, not only does the Catalist data differ from the exits in terms of reported 2018 composition--for example, the levels Catalist reports for young voters and for Latinos are far lower than in the exits--but the differences between 2014 levels and 2018 levels reported by Catalist differ as well.
To deal with an obvious issue first: are the Catalist data the "correct" data? Should we just rely on the Catalist data and disregard everything else? Perhaps not. Their methodology, while sound, has a lot of moving parts and is almost certainly not getting everything exactly right. Besides which, as Ghitza points out, they will be revising their 2018 estimates over time as more data becomes available, particularly state voter files.
That said, I do believe these data deserve close attention. The exit polls have well-documented problems and it seems quite plausible to me that the Catalist data are "righter" than the exit polls even if not exactly right.
I'll concentrate here on what the Catalist data indicate about trend rather than levels. Among the more interesting findings are the following (all comparisons between 2018 and 2014).
* White vote share declined by 3 points, which agrees with the exit polls.
* Latino vote share did go up, just as the exits suggested, though not as much (1 percentage point vs. 3 in the exits).
* Youth (18-29) vote actually went up by a percentage point rather than remaining stable as the exits indicated.
* Black vote share went up a percentage point rather than declining by a point as the exits showed.
* White noncollege vote share declined by a whopping 5 points (the exits were useless on this trend because of methodology changes).
* White college vote share went up a point.
* White college women went up a point while white college men were stable.
* Suburban vote share went up 3 points, while rural vote share declined by 2 points.
* Suburban white college vote share went up 4 points, which is consistent of course with the conventional wisdom about the election.
* Suburban white noncollege vote share went down 3 points and rural white noncollege vote share went down 2 points.
Lots to chew on here and I do think it helps illuminate what happened in 2018.