4 ms·

There was a confusingly-written part that I've seen a lot of people misinterpret in that particular way. My understanding of what he was trying to argue was th

by Natsu 9y ago

There was a confusingly-written part that I've seen a lot of people misinterpret in that particular way. My understanding of what he was trying to argue was this: Imagine you have a skill-based RPG, where you gain skill in some task only by doing it and men and women are completely identical in every single way. Now imagine that more women enjoy alchemy, but more men prefer to spend their time mining. In this game, because of how skill gain works, if you look for really talented alchemists, most of your candidates will be women, even though the sexes are every bit as good as each other.

refulgentis 9y ago
From a statistics point of view, this statement is dispositive: "...decrease the false negative rate, not increase the false positive rate." The interaction between type I and type II errors virtually always have an inverse relationship. You cannot decrease one without increasing the other. It is, for example, one of the difficulties in making good medical diagnostic tests.
- dmoy 9y ago
  I'm not clear why this is being down voted, this was drilled pretty hard into my head the first time I had a job that relied on stats.
  - DoreenMichele 9y ago
    I am reasonably mathy and I don't understand the statement. That plus the topic -- sexism in tech generally, especially Damore -- makes me think the motivation for down voting it is probably "political." ie it sounds like it could be supporting Damore's position, much of the world has decided he is an evil sexist pig etc.
    - dmoy 9y ago
      I was reasonably mathy too, but stats was pretty different and I only took a couple semesters of it. It wasn't until I was forced to interact with it in a job setting that the inverse relationship between type I & II errors became really apparent to me. Any time I would crank up settings to reduce type I error, keeping all other things the same, type II error would invariably creep up... and v. versa Then when looking in to it, I saw a lot of the examples given from the medical industry as GP alludes to, where this is a real big problem (e.g. if your test is 99.999% accurate, but has a 0.001% false positive rate, then if your disease actually only affects 1/1000000 people then the overwhelming majority of people who your test select are actually healthy). And then I went back and actually read the relevant sections on the stats textbooks for my courses and learned that (a) it was actually in there, and (b) evidently my stats courses were way, way too easy to pass As for the reply, I think I figured out why it was downvoted. It was a double-post, and the intended GGP was here: https://news.ycombinator.com/item?id=16398560 https://news.ycombinator.com/item?id=16398560 Where it makes much more sense.
      - refulgentis 9y ago
        Alas, this is the 1st time I posted it. There was a heavy bias to down voting all of my comments for the first 4-6 hours, which evened out to the opposite over time. I usually see this pattern in gender-related threads
        Natsu 9y ago
        FWIW, I upvoted your comment for being mathematically correct, even though I was a bit confused as you meant to reply to someone else.

by Natsu 9y ago

refulgentis 9y ago

From a statistics point of view, this statement is dispositive: "...decrease the false negative rate, not increase the false positive rate." The interaction between type I and type II errors virtually always have an inverse relationship. You cannot decrease one without increasing the other. It is, for example, one of the difficulties in making good medical diagnostic tests.

dmoy 9y ago
I'm not clear why this is being down voted, this was drilled pretty hard into my head the first time I had a job that relied on stats.
- DoreenMichele 9y ago
  I am reasonably mathy and I don't understand the statement. That plus the topic -- sexism in tech generally, especially Damore -- makes me think the motivation for down voting it is probably "political." ie it sounds like it could be supporting Damore's position, much of the world has decided he is an evil sexist pig etc.
  - dmoy 9y ago
    I was reasonably mathy too, but stats was pretty different and I only took a couple semesters of it. It wasn't until I was forced to interact with it in a job setting that the inverse relationship between type I & II errors became really apparent to me. Any time I would crank up settings to reduce type I error, keeping all other things the same, type II error would invariably creep up... and v. versa Then when looking in to it, I saw a lot of the examples given from the medical industry as GP alludes to, where this is a real big problem (e.g. if your test is 99.999% accurate, but has a 0.001% false positive rate, then if your disease actually only affects 1/1000000 people then the overwhelming majority of people who your test select are actually healthy). And then I went back and actually read the relevant sections on the stats textbooks for my courses and learned that (a) it was actually in there, and (b) evidently my stats courses were way, way too easy to pass As for the reply, I think I figured out why it was downvoted. It was a double-post, and the intended GGP was here: https://news.ycombinator.com/item?id=16398560 https://news.ycombinator.com/item?id=16398560 Where it makes much more sense.
    - refulgentis 9y ago
      Alas, this is the 1st time I posted it. There was a heavy bias to down voting all of my comments for the first 4-6 hours, which evened out to the opposite over time. I usually see this pattern in gender-related threads
      - Natsu 9y ago
        FWIW, I upvoted your comment for being mathematically correct, even though I was a bit confused as you meant to reply to someone else.