4 ms·

>"So when a team lacks diversity, that’s a bad sign. What are the odds that the decisions that were made to create that team were really meritocratic? " By wha

>"So when a team lacks diversity, that’s a bad sign. What are the odds that the decisions that were made to create that team were really meritocratic? " By what kind of a priori logic does the author presume that any given meritocratic filter will happen to have an equal pass rate along arbitrarily chosen orthogonal characteristics such as race and gender? For any given test for any kind of ability, I would be shocked if the population selected happened to perfectly fit the hiring-brochure rainbow. It's just not very likely. It is very possible that a more meritocratic selection procedure results in less diversity. In fact, that's precisely what happened when California passed a referendum in the late 90s to forbid considering race in admissions to the University of California. The process became less biased and the results became less diverse. My apologies to the author's preconceived ideas. >"Demographic diversity is an indicator. It’s a reasonable inference that a group that is homogeneous in appearance was probably chosen by a biased selector. Even if men have an innate advantage at software development, the gap would have to be massive in order to explain why startup after startup has an all-male team." Innate biological gender differences are not the only cause of differential average programming ability between gender populations. Sociological factors matter too. While some of these sociological forces may be unjust and we might want to address them, that does not change the fact that by the time they reach adulthood the population of qualified programmers has many more males than females. Thus, a lack of diversity can emerge from a just meritocratic process. The thinking here is just sloppy.

eries 17y ago
> By what kind of a priori logic does the author presume > that any given meritocratic filter will happen to have > an equal pass rate along arbitrarily chosen orthogonal > characteristics such as race and gender? I don't think that's a correct reading of the argument. The right question to ask is "What would have to be true about the underlying population to support the observed result that a particular startup is 100% male?" Let's assume that there is significant "differential average programming ability between gender populations" and that this has a combination of biological and sociological causes. Even so, I think we can agree that the two populations are probably normally distributed around each average, right? Now, is it safe to assume a roughly equal standard deviation for both curves? I'm not aware of any data to suggest otherwise. So now we can be specific: how large would that differential have to be before we find ourselves in the part of the curve that has many men but basically no women? I haven't done the math, as it seems intuitive to me that this is going to yield an absurd result, namely, that men would have to have an overwhelming advantage. At that point, it seems more reasonable to me to infer that we have a selection bias in our filter, than that the curves are so far apart in reality. My personal, albeit anecdotal, experience has given me no reason to doubt that conclusion, either. So, I don't think the argument depends on an equal pass rate, and I think it stands up even if you believe in substantial gender differences, regardless of cause.
- plinkplonk 17y ago
  "Now, is it safe to assume a roughly equal standard deviation for both curves?" No, not really. Why? Which bit of statistical theory tells you this would be the case? "I'm not aware of any data to suggest otherwise." Well you don't have any data going the other way either do you?
  - eries 17y ago
    I simply can't imagine what relevant factor would cause a difference in the width of these curves. I know that, in general, men have a higher standard deviation for most activities and attributes than women, but these effects are minor. So, my question is, is there anything relating to the SD that you think affects the soundness of the underlying argument? If so, I'd actually be curious to learn more about it. My argument here is simply the best reasoning I can muster given the data that I'm aware of. You are welcome to poke holes in the reasoning or present alternate data. I'm eager to learn more in either case.
    - plinkplonk 17y ago
      "So, my question is, is there anything relating to the SD that you think affects the soundness of the underlying argument? " What underlying argument? You are presenting a political position unsubstantiated by any facts. Arguing about the distribution of standard deviations about two imaginary curves is like arguing about something in Alice in Wonderland. (Note that i didn't make an argument either way about the std dev in my comment). Speaking against political correctness will probably get me downvoted, but here goes anyway, From your blog post, I gather you think "diversity" of genders/races makes better dev teams. All I am saying is "prove it". With data, not hypotheses. I have yet to see any real evidence for this, whether in terms of sustained commercial success or even in a statistical sense. Show us these supposed benefits from race/gender diversity as applied to software development and startups. Is that too much to ask?
    - btilly 17y ago
      So because you can't imagine it, it must not exist? Even though you can't imagine how the standard deviations could differ, the evidence of a difference is extremely strong. And a 10% higher standard deviation is much less minor than you think if you are looking at an ability cutoff. If you have two populations of equal size with equal medians, one of which has a 10% higher standard deviation, then about 99.5% of the top 1% across both groups belongs to the group with the higher standard deviation. It is that extreme because the standard bell curve has tails that drop off very rapidly, so one curve has pretty much ended while the other curve is still going on.

eries 17y ago

> By what kind of a priori logic does the author presume > that any given meritocratic filter will happen to have > an equal pass rate along arbitrarily chosen orthogonal > characteristics such as race and gender? I don't think that's a correct reading of the argument. The right question to ask is "What would have to be true about the underlying population to support the observed result that a particular startup is 100% male?" Let's assume that there is significant "differential average programming ability between gender populations" and that this has a combination of biological and sociological causes. Even so, I think we can agree that the two populations are probably normally distributed around each average, right? Now, is it safe to assume a roughly equal standard deviation for both curves? I'm not aware of any data to suggest otherwise. So now we can be specific: how large would that differential have to be before we find ourselves in the part of the curve that has many men but basically no women? I haven't done the math, as it seems intuitive to me that this is going to yield an absurd result, namely, that men would have to have an overwhelming advantage. At that point, it seems more reasonable to me to infer that we have a selection bias in our filter, than that the curves are so far apart in reality. My personal, albeit anecdotal, experience has given me no reason to doubt that conclusion, either. So, I don't think the argument depends on an equal pass rate, and I think it stands up even if you believe in substantial gender differences, regardless of cause.

plinkplonk 17y ago
"Now, is it safe to assume a roughly equal standard deviation for both curves?" No, not really. Why? Which bit of statistical theory tells you this would be the case? "I'm not aware of any data to suggest otherwise." Well you don't have any data going the other way either do you?
- eries 17y ago
  I simply can't imagine what relevant factor would cause a difference in the width of these curves. I know that, in general, men have a higher standard deviation for most activities and attributes than women, but these effects are minor. So, my question is, is there anything relating to the SD that you think affects the soundness of the underlying argument? If so, I'd actually be curious to learn more about it. My argument here is simply the best reasoning I can muster given the data that I'm aware of. You are welcome to poke holes in the reasoning or present alternate data. I'm eager to learn more in either case.
  - plinkplonk 17y ago
    "So, my question is, is there anything relating to the SD that you think affects the soundness of the underlying argument? " What underlying argument? You are presenting a political position unsubstantiated by any facts. Arguing about the distribution of standard deviations about two imaginary curves is like arguing about something in Alice in Wonderland. (Note that i didn't make an argument either way about the std dev in my comment). Speaking against political correctness will probably get me downvoted, but here goes anyway, From your blog post, I gather you think "diversity" of genders/races makes better dev teams. All I am saying is "prove it". With data, not hypotheses. I have yet to see any real evidence for this, whether in terms of sustained commercial success or even in a statistical sense. Show us these supposed benefits from race/gender diversity as applied to software development and startups. Is that too much to ask?
  - btilly 17y ago
    So because you can't imagine it, it must not exist? Even though you can't imagine how the standard deviations could differ, the evidence of a difference is extremely strong. And a 10% higher standard deviation is much less minor than you think if you are looking at an ability cutoff. If you have two populations of equal size with equal medians, one of which has a 10% higher standard deviation, then about 99.5% of the top 1% across both groups belongs to the group with the higher standard deviation. It is that extreme because the standard bell curve has tails that drop off very rapidly, so one curve has pretty much ended while the other curve is still going on.