4 ms·

> given participation in a set (1, 1, 2) When the set is of 12, that's 4 of 12, which is exactly one third, assuming independent participation (which obviously

by runako 4y ago

> given participation in a set (1, 1, 2) When the set is of 12, that's 4 of 12, which is exactly one third, assuming independent participation (which obviously is what makes the number optimistic/ a ceiling). > Given a sample size of one pipeline Fair! If the OP only ever hires one person, you are correct. At Microsoft's scale, they are more like "the house" at a casino, and so their ratios should more closely approximate the population. I am making the larger point that yes, in one instance, this hiring manager could have experienced a challenge building a diverse pipeline, but that this experience is not generalizable to Microsoft or the industry as a whole. And the related takeaway that if your pipeline is routinely not presenting you with the significant plurality represented by "diverse" candidates, then you have a solvable process problem.

strix_varius 4y ago
> When the set is of 12, that's 4 of 12, which is exactly one third, assuming independent participation (which obviously is what makes the number optimistic/ a ceiling). I think this misunderstanding of statistics may be at the core of our disagreement. Yes, 4 is one-third of 12. No, you cannot expect overlapping, non-mutually-exclusive sub-populations of 9% and 18% to equal 27% of a total population. If you're interested in learning why, read (1), but the TL;DR is: given three people, A: black female, B: white male, C: white male, if you sum sub-populations as you're trying to do, you'll get (1 black + 1 female = 2) from a population of 3, and you'll extrapolate (incorrectly) that 2/3rds of this population meets your criteria. I hope that helps explain why your expectation of the gender & racial diversity of CS graduates is inaccurate. 1. https://stats.libretexts.org/Bookshelves/Introductory_Statistics/Book%3A_Introductory_Statistics_(OpenStax)/03%3A_Probability_Topics/3.03%3A_Independent_and_Mutually_Exclusive_Events https://stats.libretexts.org/Bookshelves/Introductory_Statis...
- runako 4y ago
  > No, you cannot expect overlapping, non-mutually-exclusive sub-populations of 9% and 18% to equal 27% of a total population. Obviously, this is true. I don't know why you continue to exclude the Latino population, but again Latino CS grads (to focus on that narrow slice of tech jobs) are also ~7%. > I hope that helps explain why your expectation of the gender & racial diversity of CS graduates is inaccurate. But (as you have pointed out) we do know more about the sub-populations that let us infer that the populations are less overlapping than we might prefer. If you're interested, there are published demographic data that go into more detail than I have time to do here. But yes, I understand the concept that a Black Latina fits into three categories. > you'll extrapolate (incorrectly) that 2/3rds of this population meets your criteria. The basic idea here is simple: at Microsoft scale, their pipelines broadly should look like the collegiate exit pipeline. If you think that < 30% of programmers are women and/or Black/Latino, I think that assumption is the problem. No matter what assumptions you make around the numbers, my core point still holds. Which is that if your pipeline is routinely missing X% of qualified applicants, your sourcing is not good enough to know whether you're even seeing the best candidates. That is as true if X% is 15% or 30%. Using an analogy: if your company isn't seeing any applicants from states representing 15% of the population (say: Texas and New York), that says more about your sourcing than the talent available.

by runako 4y ago

strix_varius 4y ago

> When the set is of 12, that's 4 of 12, which is exactly one third, assuming independent participation (which obviously is what makes the number optimistic/ a ceiling). I think this misunderstanding of statistics may be at the core of our disagreement. Yes, 4 is one-third of 12. No, you cannot expect overlapping, non-mutually-exclusive sub-populations of 9% and 18% to equal 27% of a total population. If you're interested in learning why, read (1), but the TL;DR is: given three people, A: black female, B: white male, C: white male, if you sum sub-populations as you're trying to do, you'll get (1 black + 1 female = 2) from a population of 3, and you'll extrapolate (incorrectly) that 2/3rds of this population meets your criteria. I hope that helps explain why your expectation of the gender & racial diversity of CS graduates is inaccurate. 1. https://stats.libretexts.org/Bookshelves/Introductory_Statistics/Book%3A_Introductory_Statistics_(OpenStax)/03%3A_Probability_Topics/3.03%3A_Independent_and_Mutually_Exclusive_Events https://stats.libretexts.org/Bookshelves/Introductory_Statis...

runako 4y ago
> No, you cannot expect overlapping, non-mutually-exclusive sub-populations of 9% and 18% to equal 27% of a total population. Obviously, this is true. I don't know why you continue to exclude the Latino population, but again Latino CS grads (to focus on that narrow slice of tech jobs) are also ~7%. > I hope that helps explain why your expectation of the gender & racial diversity of CS graduates is inaccurate. But (as you have pointed out) we do know more about the sub-populations that let us infer that the populations are less overlapping than we might prefer. If you're interested, there are published demographic data that go into more detail than I have time to do here. But yes, I understand the concept that a Black Latina fits into three categories. > you'll extrapolate (incorrectly) that 2/3rds of this population meets your criteria. The basic idea here is simple: at Microsoft scale, their pipelines broadly should look like the collegiate exit pipeline. If you think that < 30% of programmers are women and/or Black/Latino, I think that assumption is the problem. No matter what assumptions you make around the numbers, my core point still holds. Which is that if your pipeline is routinely missing X% of qualified applicants, your sourcing is not good enough to know whether you're even seeing the best candidates. That is as true if X% is 15% or 30%. Using an analogy: if your company isn't seeing any applicants from states representing 15% of the population (say: Texas and New York), that says more about your sourcing than the talent available.