Google executives acknowledged this month they need to do a better job surfacing user-generated content after the recent Reddit blackouts.

  • Eidolon@lemmy.world
    link
    fedilink
    English
    arrow-up
    5
    arrow-down
    1
    ·
    1 year ago

    Which isn’t entirely accurate if at all. It extrapolates the dislikes from its own database ie users who have it installed. Compared to the entire user base of Youtube this is an incredibly tiny sample size.

    • tool@r.rosettast0ned.com
      link
      fedilink
      English
      arrow-up
      2
      ·
      1 year ago

      Which isn’t entirely accurate if at all. It extrapolates the dislikes from its own database ie users who have it installed. Compared to the entire user base of Youtube this is an incredibly tiny sample size.

      You need a much, much smaller sample size than you think. Estimates for Youtube’s monthly unique visits range from ~2 billion to about ~2.7 billion. For a 5% margin of error at a 99.9% confidence level, you’d only need to sample 1083 people to get an accurate sample size.

      I’m positive that extension has more than 1000 users.

      • goetzit@lemmy.world
        link
        fedilink
        English
        arrow-up
        3
        ·
        1 year ago

        Don’t you also need to worry about your sample population being biased? You’d only be sampling people who sought out a dislike plugin, these people might be much more likely to dislike a video. Is there any way to account for that?

        • tool@r.rosettast0ned.com
          link
          fedilink
          English
          arrow-up
          1
          ·
          1 year ago

          You’d have to have a separate cohort of non-plugin users & another with a sampling of both, I think. Run some regressions on those data and I think you’d be able to tease out any bias that exists.