Yet the trends may not be real. Instead, according to Patrick Ruffini, a pollster at the Republican-aligned firm Echelon Insights, they mainly reflect a change in how the Journal conducts its polling. In the past it had joined forces with nbc News to field a nationally representative poll by telephone, a practice that grew expensive and prone to error as response rates plunged. So the newspaper recently turned to norc, which over the past decade has developed polls that collect responses over the internet from a representative sample of Americans recruited by mail.
The problem arises when the types of people contacted by phone and the internet can be representative of Americans demographically, but statistical weirdos in other ways. Studying the differences in populations from telephone and online surveys—which industry insiders call “mode” effects—the Pew Research Centre in 2019 pointed out that its own online panellists may have been more honest about key metrics than its telephone respondents and less subject to “social desirability bias”, a phenomenon where people are more honest about things like their financial situation and height when asked anonymously than they are over the phone. Pew found that on the same question, among the same respondents, it got different responses by mode.