• McDropout@lemmy.world
    link
    fedilink
    English
    arrow-up
    16
    ·
    edit-2
    5 months ago

    8000 respondents, given that they’re a fair sample size from each arab-majority country (which I doubt) do not represent what 464 million arabs think.

    That’s 0.0001724138% of the population.

    • Imhotep@lemmy.world
      link
      fedilink
      English
      arrow-up
      17
      ·
      5 months ago

      your comment shows us you don’t know how polls are made

      why not look it up first and only then comment (if relevant)

      • McDropout@lemmy.world
        link
        fedilink
        English
        arrow-up
        15
        ·
        5 months ago

        Maybe include that it is a poll of “~5700 arabs by Qatari institutions” in the title, not “67% of arabs”.

        Misleading title. Sorry to see individuals like you on Lemmy rather than Reddit.

      • hark@lemmy.world
        link
        fedilink
        English
        arrow-up
        2
        ·
        5 months ago

        Just like the polls done by those who “know stats” who predicted a clinton win in 2016.

      • Aceticon@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        ·
        edit-2
        5 months ago

        Whilst it’s quite amazing just how few people you need in your sample to get a +/- 3% error with a 95% confidence for the opinion of hundreds of millions, that narrow error margin with just a few thousand samples only works if the sample is representative of the population in general, which is unclear.

        Plenty of cases of polls out there that are complete total bollocks because they were taken by calling by phone people in relativelly poor countries were only those in the middle class and above have a phone, and at times when most people were out working, so they ended up sampling an atypical subset of people rather than one representative of the whole society.

        The way the question is posed also influences the results, sometimes quite subtly (the mere order or words or order of questions in multiple-question poll can sway the results).

        So whilst what you said makes sense in response to the previous commenter’s point, there’s a lot more to good polling than merelly the number of samples necessary for a certain error margin in the 95% confidence interval when your sampling is random in a uniform distribution (emphasys on the later, as actually making sure the sampling is indeed like that in practice is often difficult and/or costly)