StackOverflow to becoming more welcoming. 👏
For me (cis het dude with monster-sized Asian tech privilege), learning to StackOverflow properly was certainly a learning experience, and I have the closed questions to prove it. I've also gotten enough useful feedback through it (and other StackExchanges, like GIS and statistics) that I try to give back by "mentoring" & showing how to improve the question.
Downvoting unhelpful comments is a great start. I see that too much.
Now I am taking these Implicit Association Test (IAT, a bias test):
https://implicit.harvard.edu/implicit/index.jsp
'Your data suggest a slight automatic association for Male with Science and Female with Liberal Arts.' 😢 it was a *lot* easier for me to hit the right key for
- "male OR science"/"female OR liberal arts" than
- "male OR liberal arts"/"female OR science".
I personally live by Vera Rubin's principles and this is something to think about.
@22 I’m rather skeptical of these results, and would bet they would be very different if they switched the order of the last two tests.
@22 Well, to test whether that was true (that they randomize the test order), I just did it again, and got a different order.
And guess what? My personal result changed significantly, I went from "moderate bias", to "little or no automatic association between Female and Male with Science and Liberal Arts", as I expected.
I also noticed they asked me about twice as many exit questions with this result for some reason, but maybe that's because I did the first on my phone, and this on laptop?
@22 However, this is still a useful/cool tool/test, and the fact that the results are so skewed towards male+science and away from female+science, shows that there probably is a bias 😕
@taoeffect Sorry just saw this.
Ya, I don't have reason to doubt that ordering at least doesn't move the results much, and as a population study, it's pretty definitive that there's awful bias.
But it's frustrating that people (including the authors) present it as a good way to measure *your personal N=1 bias*, which, if your experience generalizes, it isn't.
(In general, population results are useful—"% risk of cancer"—but most people want a personalized answer—"MY risk of cancer".)
@taoeffect Oh how pernicious! So on the population-scale ordering might not make a big difference but on an N=1 level, treating the test as a personalized result (which the site definitely suggest you do, right?, the StackOverflow blogger was shocked with his results, and one of the FAQs is "What can I do about an implicit preference that I don’t want?"), ordering might make the results too noisy to be of any use 😩. They probably shouldn't market a population research tool for personal uses…
@taoeffect They had a section about that on the results page—basically, order does matter but only a bit: https://implicit.harvard.edu/implicit/faqs.html#faq6
"One very common question is about the order of the parts of the IAT. The answer is yes, the order in which you take the test can influence on your overall results. But, the effect is very small."
I don't usually have a lot of hope, but I'd hope that they'd control for this obvious issue and I was glad they had.