Almost everyone who wrote about the 2016 and 2020 elections said that Donald Trump appealed to the "white working class." In the Washington Post, Nicholas Carnes and Noam Lupu argue that almost everyone was wrong: the headline is "Trump didn’t bring White working-class voters to the Republican Party. The data suggest he kept them away." A smaller heading under that says "White working-class voters had been moving to the Republican Party for years. Trump stopped the trend." Those headlines are a bit misleading (they were probably written by editors)--according to their figures, Trump's 2016 share of the white working class vote was the highest of any Republican in the period covered by their data (which in 1980), and his share in 2020 was the second highest. But his strong performance was just a continuation of a long-term trend--2016 and 2020 don't stand out as unusual.
I looked at this issue in a couple of posts, and said that there was a long-term trend in the effects of college education on vote, but that 2016 was unusual--Trump's relative performance among less educated voters was much better than any previous Republican. Why were my conclusions different? One difference is that I looked at relative performance--Carnes and Lupu looked at actual share of the "white working class" vote. Another is data sources--they used the American National Election Studies and I used exit polls. The ANES data are generally regarded as of higher quality (they are a national probability sample and have a high response rate) but the samples are smaller, so they are more affected by sampling error. Finally, Carnes and Lupu take account of both education and income: they define the working class as people without a college degree who are below the median income. Taking account of income is good, but cutting the size of the "working class" also means more sampling error. The result is that there's a good deal of noise in their estimates, making it harder to see if any election deviates from the trend.
An alternative way to take about of both income and education is to estimate the effects of both together. I did this using the General Social Survey, partly because the samples are usually larger, but mostly because I'm more familiar with that data set. I did logistic regressions of Democratic vs. Republican votes (omitting third parties) on a dummy variable for college graduates vs. all others and the logarithm of household income (adjusted for inflation). The estimated effects of college degree: