lpm wrote: ↑Sun Mar 10, 2024 8:56 pm
That's a good piece.
My instinct is that polling errors are related to volatility of support. If there's movement, it's more likely there's opinion poll inaccuracy? Part of the error must be the election campaign and the Tory press going on the attack, which had more impact when voters are uncertain.
In recent years, polling errors have become much more likely, as response rates have plummeted, particularly among specific groups - due to caller ID, call screening, not picking the phone up to people you don't know, better Internet savvy, etc.. With standard phone polling that then weights responses by some mix of age, education, race and income groups, the responses that are received in hard to reach groups are then amplified (weighting also increases the error bars on a sample, and that's not usually included in the stated margin of error in polls).
That necessarily amplifies the responses given by unusual members of those groups (those that choose to answer polls). Where a question is nearly evenly split in that group, this doesn't matter so much, but when that group also has a very strong preference on a question, that is more likely to amplify dissenters from that preference than not (because the distribution is now one-sided). This latter effect is one source of the "crazification" effect in surveys, where you can find support at some level for anything. It also makes polls vulnerable to bad actors who lie to put themselves in the low response rate groups, or who lie about their opinions. Add in a potential real split in opinion between those that choose to answer the phone/that get offered a chance to participate, and then choose to take polls and those that don't (at both steps), and you can get big misses.
A recent tested example of this is the recent headlines where an online opt-in poll found that 30% of young people in the US believe that there wasn't a Holocaust. Recently a more robust poll that didn't have the same age-group-wise response bias as typical political polls found that number to be 3% - the same as found in other age groups.
https://www.pewresearch.org/short-reads ... ic-adults/
Similar effects will occur in phone cold-calling polls, but probably not to the same level (people generally find it easier to lie online than on the phone with a person).
The polling level for Tories is getting low enough that in some groups these effects will be boosting their overall results significantly - similar things are likely happening in the US presidential polls. That can mitigate or exceed the shy Tory effect and lower propensity to vote for some hard to reach groups - and that propensity to vote is explicitly compensated for when US pollsters apply likely voter screens.