• Don Piano@feddit.org
      link
      fedilink
      English
      arrow-up
      12
      arrow-down
      1
      ·
      16 hours ago

      That’s how a standard error with normal-ish data works. The more data points for the estimation of a conditional mean you have, the fewer of the data point will be within it. For a normal distribution, the SE=SD/√N . Heck, you can even just calculate which proportion of the distribution you can expect to be within the 95% CI as a function of sample size. (Its a bit more complicated because of how probabilities factor into this, but for a large enough N it’s fine)

      For N=9, you’d expect 26% of data points within the 95% CI of the mean For N=16, 19% For 25, 16% For 100, 8% For 400, 4% Etc

      Out of curiosity: What issue did you take with the error margin not including most data points?

      • Don Piano@feddit.org
        link
        fedilink
        English
        arrow-up
        7
        ·
        16 hours ago

        Oops, should have multiplied those intervals with 1.96, ao here again:

        9 - 49%

        16 - 38%

        25 - 30%

        100 -16%

        400 - 8%