The summary is as follows. The number of people who are willing to answer random phone calls by pollsters declined seriously over time to the point where almost all calls fail. As a consequence pollsters now rely heavily on panel polling, where people who are willing to answer polls are recorded and repeatedly re-polled over and over in return for trivial sums of money.

This has created an enormous and unfixable bias problem, called pro-social or volunteering bias. It's exactly what it sounds like - the sort of people who are now being regularly polled are drastically more likely to be volunteers than normal people. This can be measured by polling panels on how many of them volunteer for things in their community and then comparing the answers against census results [2]:

"While the level of voter registration is the same in the survey as in the Current Population Survey (75% among citizens in both surveys), the more difficult participatory act of contacting a public official to express one’s views is significantly overstated in the survey (31% vs. 10% in the Current Population Survey).

Similarly, the survey finds 55% saying that they did some type of volunteer work for or through an organization in the past year, compared with 27% who report doing this in the Current Population Survey."

This level of bias is extreme and unfortunately completely uncorrectable. It's inherent to polling the sort of people who are willing to regularly spend time filling out long forms in return for basically no compensation - you're getting a population of Ned Flanders style good neighbours who are, relatively speaking, obsessed with civic duty. That's why they're doing the polls - they think it's for the general good!

Making things worse, panels get more dominated by such types over time because less volunteering-oriented people drop out at a higher rate. Thus a panel that might start out somewhat representative will become less so over time, and they don't really have a good way to control for that or even measure it.

Pollsters argue that this doesn't matter for the types of polls they normally run (commercial/marketing surveys), and for elections and other votes where they have regular consistent reference points they can compute adjustment models. But it's an open secret in the polling community that any poll question that has a "pro-social" or "civic duty" component to the answers will get wildly disproportionate results to what the public actually thinks. And worse, such questions tend not to have any obvious reference points to which the results can be re-calibrated.

The political scientist being interviewed in the article probably either doesn't know about these problems, or doesn't care, because he seems very willing to 'correct' politicians about what 'the public' really believes. But he has no real way to know that. He can only know what the sort of public that is willing to spend time talking to political scientists believes, and there's good reason to think they're not really representative.

[1] https://dailysceptic.org/archive/do-online-opinion-polls-ove...

[2] https://www.pewresearch.org/politics/2012/05/15/assessing-th...