Polling

Home Forums Decaffeinated Coffee Polling

Viewing 8 posts - 1 through 8 (of 8 total)
  • Author
    Posts
  • #2329160
    akuperma
    Participant

    Even before 2020, the polling industry had NOT adjusted from the switch from landlines to mobile phones, which undermined how they conduct polls. An accurate poll would require having access to census records, and going door to door, and that would be too expensive. Plus while the voting system has been evolving, the growth of early voting and unlimited (no “need” requirement) postal ballots totally undermined the credibility of exit polling. And going back at least a generation, most people assume that pollsters are from the “mass media” who are increasingly perceived as left-wing, and since people often answering the poll often give the pollster the answer they believe the pollster is looking for, results in the left-wing doing substantially better in pre-election polling than in elections. In addition, it appears that America is evolving with the Democrats becoming the party of the upper class urban elite, while the Republicans are becoming the party of the working class – a reversal from what existed a generation ago – resulting in the pollsters models of the population becoming invalid.

    Anyone claiming that a poll is accurate, is probably lying. Those basing their claim how having done well in 2020, probably were just lucky (someone has to right). Anyone getting worried because of polls should stop worry since the fact your preferred candidate, is winning, or losing, or tied, in the pre-election polls is unknowable (unless you wait until after the election, and then use a time machine to go back a few days and make predictions).

    #2329404
    ujm
    Participant

    The 2020 polls were badly mistaken. They overestimated the Democrats.

    Agreed that polling in general has overestimated Democrats compared to the election results. This has been the case for a long while, already. Especially the non-final polls; IOW all the polls until the last one was really skewered in favor of the Democrats. By time their final poll before election day comes out, and they know that’s what they’ll be judged on, they become slightly more honest. Which is why you’ll usually find in the last polls before the election Republicans usually see improvements.

    BTW, how did polling do any better before telephone landlines became almost universally owned by Americans? As far as today, I understand polls started calling mobile telephones over the last number of election cycles.

    P.S. I believe the polls will, for the third time in a row, prove to have under-counted Trump voters; and he will have a convincing win on election day (especially in the Electoral College, but he’ll most very likely win the popular vote as well.)

    #2329654

    It is not just mobile phones, response rates are now very low. One way to deal with it is to re-weight results using demographics to account for non-random responses. Previous weighting by a small number of factors – party, age, gender, education, etc – was proven not to be sufficient when alliances change, like with Trump. So, this year pollsters seem to be using more weighting. As some researchers point out that because of this increased weighting current polling is abnormally stable over time and, at some point, reflects your weighting assumptions instead of actual voter preferences. So, at this point, there are 2 possible outcomes: or vote will be indeed very close and take long time to sort out, or one of the candidates is way ahead and polls systematically made the same error. In the latter case, it is more likely that T was “misunderestimated” but there is no guarantee that there was not a over-compensation the other way. Note that campaigns have way more internal data and you can try getting clues by where the candidates are campaigning at the end – defending their own or attacking in new areas.

    #2329675
    akuperma
    Participant

    While landlines failed miserably in 1936 (they predicted Alf Landon, since only wealthy people had phones), once phones became almost universal (in the 1950s) you could easily call a random number of people and get a random swath of the electorate. You also had phone books giving the addresses of almost all customers. When most people switched to cell/mobile phones this was no longer possible.

    P.S. My guess is the over estimation of Democrats will hold true, suggesting Trump will win, perhaps comfortably. Also, many people may perceive as Trump winning and decide to vote for him to get the matter settled once and for all.

    #2330026

    There might be over-estimation of Ds, BUT the data is corrected after that – and this year they are correcting using more factors than before. This is in theory more accurate, but can introduce a bias as the correction is based on some assumptions about voting population. Ok, you might reasonably assume that ratio of D and R voters will be same as last time (will it?), but when you also include age, area, income, etc – your assumptions essentially guide your polling correction.

    #2330323
    akuperma
    Participant

    My prediction in the post script of my last post was correct. What is silly (and disturbing) is that the polls have been consistently wrong, and in the same direction (overstating the Democrats) for a generation.

    #2330431
    ujm
    Participant

    Multiple generations.

    #2330462

    as I mentioned, this is because of a biased correction. If they would have simply missed some groups, then all polls would be off by 2% but there would be variation between polls and for same poll over time. As some analysts noticed, the polls stayed suspiciously synchronized and stable over time.

    What can we learn from this? Sometimes, we can apply good thinking but be burdened by unjustified assumptions that bias us towards certain opinion that we hold historically, or because we were raised in a certain hashkofa, or because it is comfortable … We need to always analyze our intellectual biases. It is often hard to do on your own. A lot of analysis in Bavli is about that type of discussion. And, hopefully, discussions here, when done in a right way, help us to see issues from different perspectives.

Viewing 8 posts - 1 through 8 (of 8 total)
  • You must be logged in to reply to this topic.