Beware Isolated Demands For Rigor
slatestarcodex.com
slatestarcodex.com
>But when other people are totally happy to talk about speed and blood pressure and comas and the crime rate, and then suddenly switch to a position that we can’t talk about IQ at all unless we have a perfect factor-analytical proof of its obeying certain statistical rules, then I worry they’re just out to steal cows.... But when people never even begin to question the idea of different cultures but make exacting demands of anyone before they can talk about different races – even though the two ideas are statistically isomorphic – then I think they’re just out to steal cows.
There is of course, the opposite problem, which is that people take things like IQ and infer wild conclusions, usually in the service of white supremacy. An example of this is the MAO "Warrior Gene." There are several studies showing a weak correlation between a variant of the gene and violence. Black Americans have this gene more often.
There is a specific, very bad, study that is often quoted. It cherry picks its data, and still can't achieve a significance of p > .05.[1] Yet it is often brought up as indisputable proof of the savage nature of black people. It's ridiculous, and IMHO, much worse than the tendencies criticized in the blog post.
Funny thing about white supremacy. If you go on raw IQ statistics alone, then whites are certainly not the master race. That title would belong to Asians and Ashkenazi Jews.
But then again I'm a skeptic of IQ = everything.
By context, I mean that IQ has meaning for the way life is for humans right now, but given a return to an earlier context (hunter-gatherer) or a transition to a new context (say some high-tech utopia where computers do the heavy thinking), it becomes functionally meaningless.
http://neoacademic.com/2011/10/05/gladwell-was-wrong-high-an... http://pss.sagepub.com/content/22/10/1336
The nature of the relationship between ability and performance is of critical importance for admission decisions in the context of higher education and for personnel selection. Although previous research has supported the more-is-better hypothesis by documenting linearity of ability-performance relationships, such research has not been sensitive enough to detect deviations at the top ends of the score distributions. An alternative position receiving considerable attention is the good-enough hypothesis, which suggests that although higher levels of ability may result in better performance up to a threshold, above this threshold greater ability does not translate to better performance. In this study, the nature of the relationship between cognitive ability and performance was examined throughout the score range in four large-scale data sets. Monotonicity was maintained in all instances. Contrary to the good-enough hypothesis, the ability-performance relationship was commonly stronger at the top end of the score distribution than at the bottom end.
[1] http://slatestarcodex.com/2013/08/30/military-strikes-are-an...
[2] http://slatestarcodex.com/2014/08/11/does-the-glasgow-coma-s...
[3] http://slatestarcodex.com/2014/08/12/does-race-exist-does-cu...
Remember that Scott was able to write his (justly famed) Neoreactionary FAQ because he knows a pile of the neoreactionaries personally.
[1] c.f. the recent case of http://whyevolutionistrue.wordpress.com/2014/08/09/our-lette... in which 139 geneticists told Nicholas Wade to stop misreading and quote-mining their own results in the service of scientific racism.
The question of how we trade off different methods of achieving broadly humanitarian ends is a different question than how we trade off broadly humanitarian ends against different ends. If the only case being made for the air strikes is humanitarian, then it makes sense to see if it's reasonable humanitarian bang for the buck. Of course, there are other ends that the air-strikes might serve, though I have no sufficiently informed opinion on their effectiveness (or counter-productiveness).
Case in point: a tech company releases diversity numbers. A reasonable point is to ask for a breakdown into tech, admin, support, etc. Often in these companies the diversity is shoved into sales and marketing.
But then another commenter will ask about the diversity of the candidates, and speculate that they are identical to the company's hires, and then "prove" that the company's practices are unbiased.
So then you point out that a company is not a blushing maid, that they choose the people they recruit, project an image that influences self-selection, and heavily encourage people to refer their friends. And then you get attacked for demanding too much rigor.
What the argument is really about is whether there is a problem to fix. The diversity numbers clearly indicate that there is. But people who like to believe there is not alternately dig trenches and build walls of rigor to block the debate.
You can't draw a normative conclusion from a positive claim alone. You additionally need some normative axiom. No fact can tell you what you should do.
In the discussion you cite, the disagreeing parties have different normative axioms. Some people believe in statistical equality as an end goal, whereas others have equal treatment. If you believe in statistical equality (as I think you do), then the facebook numbers are a problem to fix. If you believe in equality of treatment (as I suspect the others do), then you can draw no conclusion - the numbers do not directly imply anything about equal treatment.
Similarly, consider a debate between a "meat is murder" activist and a carnivore. No one disagrees that animals are killed - the only disagreement is over whether that's a bad thing.
If you agree that women are not somehow less mentally capable than men, then the fact that technical jobs are 85% dominated by men should be a concern. Not just for "normative" reasons or "statistical equity", but simply that there are a lot of talented people being excluded for one reason or another. If not, you should explain WHY not. Anomalies demand explanation.
You dismiss the numbers because of an assumed equality of treatment and saying, well, I don't see anything therefore it's not a problem... and if it is, it's somebody else's problem. That's exactly the mismatch of rigor I'm talking about. You put the burden on others to "prove" that something is amiss, then when they do, demand more.
As for the undue rigor involved, it's you who seems to be demanding a lot: you demand proof that men and women are statistically unequal merely to admit it as a possibility.
More precisely, let A = "women are unequally treated" and B = "women are not statistically equal to men" (at the point of application for Facebook employment). The diversity numbers imply A || B. You are assuming that B is false because there is no proof it is true, and therefore concluding A must be true (since A || B and !B => A). But strangely, you are not asserting that A is false without a rigorous proof that A is true (which would imply B).
That's the fundamental asymmetry that Scott Alexander is talking about. The people you disagree with are merely saying "A || B, we don't know which."
[edit] For fuck sake - if you had noticed a 50% imbalance in CPU usage on your server farm, surely you'd spend some time trying to understand it. What the hell is wrong with so many people who are busy inventing facile excuses and dredging up mathy reasons why to NOT recognize the problem as a problem and looking into why it might exist?
for {
x <- f()
y <- g()
} yield (x+y)
And if someone suggests g() might be the problem, you are assuming they are wrong and demanding an excess level of rigor from them. That's exactly what Scott Alexander is warning against.
[edit: I'm sticking to epistemology. I'm not hijacking the thread to discuss gender.]
I frankly don't care whether it's a pipeline problem or a filter problem, or both. I have more than a decade of performance work under my belt; I'm familiar with the usual arguments. What makes me take notice is that so much effort is put into arguing why there isn't a problem at all, or that people who say there is a problem are somehow mistaken.
Seriously. Do you think that women are intellectually flawed? Yes or no? If yes, I'd love to hear why. If no, what is your best guess why they are underrepresented? No tricks, no shifting of levels. Why do you think it's so? "I don't know, let's find out together" is a perfectly good answer.
Christina Hoff Sommers' guesses: https://www.youtube.com/watch?v=l-6usiN4uoA, http://www.theatlantic.com/sexes/archive/2013/03/what-lean-i...
My best guess is that relevant male math-type skills have a higher variance than female math-type skills. That means that even if the average ability is about the same, there are more men than women at the very top of the skill distribution and ALSO at the very bottom of the skill distribution. If you look for math morons, you'll find more men; if you look for math geniuses, you'll find more men. If you look for people with roughly average math ability, you'll find more women. If the variance is even a LITTLE larger, it can explain an arbitrarily large imbalance at the high end, and we have both (some) statistical evidence that it IS larger and plausible biological reasons to expect that it MIGHT be larger.
Women have two X chromosomes; men only have one. All the important stuff that makes us human is on the X chromosome. So for the sake of intuition, let us imagine that there were only THREE available genes for math ability, all equally prevalent in the population, and any X chromosome you inherit includes one (and only one!) of these three "math ability" genes. The genes are:
(1) "moron" (recessive)
(2) "genius" (recessive)
(3) "average" (dominant)
In that world, men would have a 1 in 3 chance of being a genius and women would have a 1 in 9 chance of being a genius.
Similarly, men would have a 1 in 3 chance of being a moron and women would have a 1 in 9 chance of being a moron.
In that world, having two X chromosomes makes women "more average" than men, which means that if some career requires the "math genius" gene, that career will be 75% male.
Does that all make sense?
Okay, now let's suppose that "genius" gene is a bit more rare. Let's say that only 10% of X chromosomes have the "potential math genius" gene. In that case, men would outnumber women 10-to-1 instead of 3-to-1 in professions requiring it. The rarer that gene is, so long as that gene is either recessive or to some degree averages out with its counterpart on the other X chromosome, it will be hugely rarer that women fully express the properties that gene codes for.
If this analysis is valid, it suggests that women are less likely to be at the very TOP of math-related fields (or will have to work a lot harder to get there) for essentially the same reason women are far less likely than men are to be colorblind.
(relevant study: http://www.ncbi.nlm.nih.gov/pubmed/7604277 )
1. Hypothesis: Women are discriminated in tech.
2. Argument: There are less women in tech. (shows number of women in tech)
3. Counter argument: There are less applicants.(shows number of applicants, number of uni students)
4. Counter counter argument: They self-select out of tech because of the discrimination.
Afaik, there is solid evidence for points 2&3. What is the evidence for 4?
In isolation, any two peers are either going to fail to negotiate entirely or converge in short order.
There's little hope for convergence of vocabulary or context much less rigor among the chatter of a hundreds, thousands or millions of nodes trying to dictate the protocol. More likely, you end up with a split brain where each island establishes its own consensus totally apart from the other.
What does this idiom mean? I've never heard of it before.