p. 37 of this document: https://docs.cdn.yougov.com/cv6lbp28dm/scotus_poll_results.pdf

  • save_vs_death [they/them]
    ·
    edit-2
    1 year ago

    what you really want is having a representative sample e.g. making sure that out of those 2000 people, they are, roughly, a good reflection across ethnicity, gender, religion, occupation, state, so on, so forth; if that were the case, (which I doubt) then 2000 would not be out of question as being a decent survey

    a lot of pretty important psych papers have sample sizes way below 2000, for example, in fact, the bigger the sample size, for some statistical tests (not pertinent for the current subject of discussion) the more likely that you'll get false positive results (type II errors)

    that's just how psychometrics is sometimes

    more to the point, a poll that someone else points out disagrees with this poll (https://hexbear.net/post/271974/comment/3521696) was done on 1000 respondents; technically, as long as a sample is representative, its number would not matter unless it was under, let's say, 100 respondents

    • D61 [any]
      ·
      edit-2
      1 year ago

      I didn't say it was completely unusable, just leaning towards "grain of salt" more than "evidence of future trend".

      Or at least that was my intent.

      • save_vs_death [they/them]
        ·
        1 year ago

        oh, i see, i see, for what it's worth i completely agree with you, sorry for the unwarranted mansplain

        • D61 [any]
          ·
          1 year ago

          No worries, its the internet, about half of out intent when trying to communicate with each other winds up failing to be conveyed. :solidarity: