The curious thing about election polls is that they can be wrong by miles, but so long as the candidate with the predicted lead still wins, no one much cares. For example, a survey suggesting Candidate A has a 20-point advantage over Candidate B in an election where A only ends up winning by five points is much less accurate than one which predicts that Candidate A has a three point lead, but goes on to lose to B by one point.
Predicting the wrong winner in a tight contest is simply much more memorable than misjudging the size of victory in a landslide race. So we all have post-traumatic stress disorder following the polling error in the 2016 US presidential election which Hillary Clinton lost (despite winning the popular vote) but forget the even bigger polling error in favour of Joe Biden in 2020, because, well, Biden won anyway.
