jackk, the report goes through a lot of stuff and is inconclusive, indeed. Their main supposition is that some of the Trump voters were under-sampled. That is they sampled Republicans in general, but not the most ro-Trump ones that also happened to be distrustful of the government, pollsters, etc. We see similar in vaccination, so this rings true. Here are characteristics that were undersampled in the polls the most when tried to re-match regression of the population data to election results by counties: solid Republican, white, less populated cases, more covid cases.
Another interesting item: a very high number of “new voters” – who did not vote in 2016. Pollsters struggled understanding how many of these voters are and what their preferences are, as they often relied on previous data to stratify voters. These new voters went +14 for Trump in Florida and +10 to 20 for Biden in many other states. There is a lot to unpack here – enthusiasm, groups that started to care, early voting. I imagine this is where suspicion about dead voters and vote harvesting will from. Some of the numbers for new voters and confirmed voters are way abnormal for Georgia and Michigan (see Table 11, for example), I did not try to understand whether this is due to their data collection or something else.