Wow this post got popular. I got called into work and didnt see the replies, sorry ladies and gentlemen! Trying to catch up tonight.

  • DarthBueller@lemmy.world
    link
    fedilink
    arrow-up
    6
    arrow-down
    1
    ·
    1 year ago

    That study seems to state a conclusion precisely the opposite of what the experimental results were. Based on a small sample set, there’s a high degree of match, far more accurate than random chance, between the observations and the genetic findings.

    • Pipoca@lemmy.world
      link
      fedilink
      arrow-up
      1
      ·
      1 year ago

      Of the 25 dogs identified as pit bull-type dogs by breed signature, 12 were identified by shelter staff as pit bull-type dogs at the time of admission to the shelter (prior to the study visit), including five labeled American Staffordshire terrier mix, four pit bull mix, two pit bull, and one American Staffordshire terrier. During the study, 20/25 dogs were identified by at least one of the four staff assessors as pit bull-type dogs, and five were not identified as pit bull-type dogs by any of the assessors. …

      Of the 95 dogs (79%) that lacked breed signatures for pit bull heritage breeds, six (6%) were identified by shelter staff as pit bull-type dogs at the time of shelter admission, and 36 (38%) were identified as pit bull-type dogs by at least one shelter staff assessor at the time of the study visit

      So, at intake, 18 dogs were identified as pit bulls but only 2/3rds were at least 12% pit bull.

      During the study, 56 dogs were identified as being pit bulls, but only about 1/3rd were in fact at least 12% pit bull.

      This is the classic ‘base rate fallacy’. The false positive rate isn’t that high, and the false negative rate isn’t that high either. But because the true positive rate is pretty low, the ratio of true positives to false positives is much worse than you’d intuitively think.

      Tests for rare diseases and attempts to behaviorally profile terrorists at airports runs into the same problem. Sometimes, a 99.9% accurate test just moves you from searching for a needle on a farm to a needle in only a single haystack.