Hypothesis
Volunteering should not tell us anything extra about the vote.
Abdullah
An internet panel is a standing pool of people who agree to take surveys. People who join one are not a random slice of the country. The claim is that a short list of traits explains how they differ: age, race, education, party, registration, and a few related traits. Matching pairs each volunteer with a similar person on those traits. Once the traits are matched, knowing that someone volunteered should tell you nothing more about how they voted.
What would count as support
First, the party slope should match across the two surveys. The party slope is how much the vote changes with each step toward Republican on the party scale. It should be the same in the CCES, the online volunteer survey, and the ANES, the face-to-face survey. The article’s numbers are −0.313 and −0.312. Second, state estimates of the presidential vote should miss the certified result, the official count, by a few points, not by ten. Third, vote shares for groups, including Hispanic voters, should sit near the exit poll, a survey of voters leaving the polls. The article says the CCES did that and the ANES Hispanic share did not.
The claim is about the matched and weighted file of 32,800. Weighted means some interviews count more than others so the file looks like the country. The claim is not that the raw volunteer list looks like the country. That list is too white, too old, and too educated. One risk remains. Suppose a reason for joining the panel also predicts the vote, and it is not one of the matching traits. That bias would survive the whole procedure.