• Doc Avid Mornington
    link
    fedilink
    English
    11 year ago

    I mean, pollsters actually do account for how elections work in their models. There are all sorts of actual reasons polls have failed to be reliable lately, but if you think it’s because they just count total responses across the country, that isn’t the case.

    • Jordan Lund
      link
      English
      71 year ago

      Not really, case in point is this very poll:

      “In the national survey of 910 voters, 47% of voters said they would definitely or probably support Biden, while just 40% said they would back Trump.”

      Which is meaningless, because unless 47% of voters flip the correct states, it won’t matter how much Biden wins.

      Remember, Clinton won the popular vote. Gore won the popular vote AND Florida. It didn’t matter.

      • Doc Avid Mornington
        link
        fedilink
        English
        21 year ago

        So, I think you’re probably right, in this case. But you’re just quoting the reporting on the poll, which is very misleading. It makes it sound like there is no statistical model involved at all. From the methodology on the linked full poll results: “The full sample is weighted for region, age, education, gender and race based on US Census information”. Like I said, I think you’re right - I doubt if they mean weighting for “region” to imply they did an electoral college analysis - but until you look at the actual poll and it’s methodology, you can’t just assume that an article reporting on the poll is giving an accurate impression. There are polls that do account for state breakdown, and the reporting in an article on such a poll would probably be just the same as here.

        It seems the focus of this poll was to get some initial idea what kind of impact a third-party run with Manchin and some Republican running mate would have, and looking at weighted national numbers is probably “good enough” for that purpose, at this time. Definitely not a basis to conclude Biden has it in the bag, and the poll itself doesn’t seem to be trying to claim that.

        Sorry I’m going on, but yeah, big picture, you are correct, at least in this case.

        • Jordan Lund
          link
          English
          11 year ago

          Oh, there’s no doubt a statistical model to represent the entire country. The problem with popularity contest polling like this is the election isn’t a popularity contest.

          Now, a similar survey running down each contested state and calling out the electoral college votes, that would be useful.

          Anything that leads with “a national poll…” can be safely disregarded.