Re:Why I won’t give talks about being a woman in tech
by DannyBee on Wednesday July 20, 2016 @12:03PM
(#12130080)
Assume for a second these tests are valid ways to prove what you are trying to prove[1].
First, you can't simply reslice the data and think you have shown anything valid. Simpson's paradox and all that. It could be that women do better than men in all but one race, for example, and that race "brings the average down" to the point where the overall gender scores look like they do. In fact, if you look at the wikipedia article on simpson's paradox, it gives the great example of .. a gender bias study :)
Whoever is creating these numbers by confounding variables and grouping data from other studies is just making a mess of the original data :(
(This is sadly not uncommon, and happens a ton in race and crime statistics, etc. The folks who make the original data often go to great pains to explain their methodologies, etc, and then some random org just throws it all in excel so they can produce charts that support whatever, and acts like anything they do to the data makes it still valid)
Second, ignoring that serious issue for a second, The only thing this gives you is "a snapshot of the world today". This is a test of actual current performance trying to predict success, not trying to predict potential performance. The SAT gives no data on what innate ability is, nor does it try to. That means it can change, and you can improve the numbers.
[1] i'm pretty sure the main correlation the SAT shoots for, and succeeds at, is predicting whether folks will be "successful the first year of college". There are a number of studies that show this. There are also newer, but less studies, that show other predictors may be just as good.
First, you can't simply reslice the data and think you have shown anything valid. Simpson's paradox and all that. It could be that women do better than men in all but one race, for example, and that race "brings the average down" to the point where the overall gender scores look like they do. In fact, if you look at the wikipedia article on simpson's paradox, it gives the great example of .. a gender bias study :)
Whoever is creating these numbers by confounding variables and grouping data from other studies is just making a mess of the original data :(
(This is sadly not uncommon, and happens a ton in race and crime statistics, etc. The folks who make the original data often go to great pains to explain their methodologies, etc, and then some random org just throws it all in excel so they can produce charts that support whatever, and acts like anything they do to the data makes it still valid)
Second, ignoring that serious issue for a second, The only thing this gives you is "a snapshot of the world today". This is a test of actual current performance trying to predict success, not trying to predict potential performance. The SAT gives no data on what innate ability is, nor does it try to. That means it can change, and you can improve the numbers.
[1] i'm pretty sure the main correlation the SAT shoots for, and succeeds at, is predicting whether folks will be "successful the first year of college". There are a number of studies that show this. There are also newer, but less studies, that show other predictors may be just as good.