I learned from a completely reliable source that the letter to the editor I published in the Journal of Theoretical Biology was largely in error.
I have to admire the thousands of anonymous Wikipedians who catch the mistakes of poseurs such as myself, and I’m looking forward to the forthcoming correction in the journal. The editors of JTB must be pretty embarrassed to have published a letter that was so wrong–but I guess it makes sense that they could be bamboozled by a statistician from a fancy Ivy League college.
Oh well, at least it’s better that I learn the error of my ways now than that I live the rest of my life under the illusion that I knew what I was doing all this time!
P.S. I wish the anonymous Wikipedia editor had contacted me directly regarding my mistakes. As it is, I may never learn exactly how my criticisms have been “already addressed and corrected.”
P.P.S. Due to the dynamic nature of Wikipedia, the above is now out of date. (Latest version is here.) Take a look at the comments below.
Wikpedia hath spoken.
The wiki references this article as adressing your critique.
Kanazawa, S., & Reyniers, D. J. (2009). The role of height in the sex difference in intelligence. American Journal of Psychology, 122, 527-536.
I'm unable to get a copy of it and so cannot comment on its contents, but interestingly its published in 2009 long after your letter.
Sarcasm just doesn't work on the internet.
Steve: Yes, I saw this link and read the article but it did not explain how my criticisms were refuted.
Dan: You're probably right. I probably should've either played this one straight or waited until April 1 to post it. Actually, usually I'm pretty good at avoiding sarcasm in my blogging.
One can examine the revision history on Wikipedia. The edits came from JStephanieApril.
In earlier times, someone with an a broadband account at Sky.com in UK repeatedly tried to delete the link to Andrew's article.
I undid the changes of JStephanieApril. All of that user's edits happened over the course of a couple days, and all were related to evolutionary psychologists. The user seems to have a vested interest in protecting the claims of evolutionary psychology.
We'll see whether the changes stick or whether someone starts a fight over them…
I edited these slightly to capture the "controversy" – I'm convincedthat Andrew is 100% right, but as long as Kanazawa gets that stuff published in serious journals the best Wiki can do is point to the opposing views.
The self-organizing nature of Wikipedia in action!
For some smiles, do check out the changes made by 90.206.101.224 and other IP addresses of some broadband customer of Sky.com in UK over several months.
I slightly edited Sebastian's changes, because he was actually giving Kanazawa too much credit: In claiming to have refuted the endogeneity problem and the multiple comparisons problem (in Kanazawa and Reyniers 2009), all he cites is the initial article–the one that AG is responding to! So I made that clear, rather than suggesting that Kanazawa had subsequently responded to anything.
Is Kanazawa really notable enough for Wikipedia? Apparently he's more notable than Columbia statistician Andrew Gelman.
And much more important than me.
Bob – well, since he says a lot of crazy things he gets quite a bit of exposure.
Andrew could absolutely have a Wikipedia entry, I don't think anyone would oppose, but someone would have to write it.
(Thanks Cauchy, I hadn't caught that).
Inserting an attack on the anonymous into an overall good criticism of the incompetent or deceptive is bad social epistemological form, in my opinion.
My class of anonymous social epistemologists can and do play a useful role in our overall work to create better models of reality, it seems to me.
I think you'd be more on the side of angels if your posture was how to to optimize the participation of anonymous social epistemology participants, rather than to try to attach stigma to us in posts like this.
By the way, the sarcasm in your post worked just fine. And this comment is honestly sarcasm free.
Perhaps move the hyperlink in "completely reliable source" down to the PPS, and replace it with this link to the article's history: http://en.wikipedia.org/w/index.php?title=Satoshi…
it took me a long time to understand what you were getting at :(
… "already addressed and corrected" in a peer-reviewed journal, no less (see pg 534)
Thanks, but I have a correction to your correction. You wrote "Gelman further argues that Kanazawa's analysis does not convincingly show causality." That's part of it, but really I'm arguing something more basic, which is that I see no convincing evidence that there is a positive correlation between child's sex and parental attractiveness, and I certainly don't believe that whatever differences are anything close to the 8 percentage points claimed in that cited article.
Here it is: http://en.wikipedia.org/wiki/Andrew_Gelman
Very short, but at least it exists now :-) Andrew, if you know good (citable and independant) sources that can be used to improve this article, I am sure some of the commenters here will be glad to use them.
Still believe the only end to this will be (for some credible group to convicingly) to show the claims do not replicate (and some may intentionally or un-intentionally delay that for a very long time)
Unfortunately, some or even many just do not get the logic of statistics and the high risks invloved in trying to learn things even carefully from very small data (single) sets.
Keith
Unrelated possibly, but meant to add that it is always good style to cc authors on corrections or letters to the editor that you write – even if just concurrent to it going to the editor or online.
Keith
Keith: I don't know that this ever needs to end. From a scientific perspective, effects of 1% or less on the sex ratio of births aren't so important; it's not like there's a need for a blue-ribbon commission to investigate it. And, from the political or headline-grabbing perspective, it's perfectly fine to claim effects of 8% (or 36%) and few will care. But I'm hoping that my American Scientist article with Weakliem will cause people to think twice about such issues when they arise in important settings.
P.S. And, yes, I did email the author of that article at some point, I can't remember if it was before or after submitting my letter to the editor. He never replied. He also was offered to appear on a radio show with me but declined to do so. (Which was a relief to me, as I was pretty scared about the prospect of arguing in a public forum with someone who much have much more media experience than I do.)
Cool! My sister has a Wikipedia page, so it's only fair that I have one too.
Andrew: agree that this is not a particularly important finding (say compared to a wide spread adoption of a new medical treatment or social policy) – but seeing it fail to replicate may signal to those who currently don't quite get the statistical arguments, that they are at some peril and should try harder to get the issues.
To quote Rubin – "smart people don't like being repeatedly wrong"
Or maybe not.
Keith
p.s. the thing about sending the letter was prinarily meant for those making the changes to wikipeadea
Please correct me if I'm wrong. In your comment to Mr. Kanazawa's paper you say something like this:
1. Total number of children of the other sex can be causally dependent on parent's occupation.
2. Intermediate outcome is a variable that can be causally influenced by the predictor of interest.
3. Controling for intermediate outcomes can lead to biased estimates.
One other example you give is "The model … controls for income, which is also an intermediate outcome … so it does not make sense to compare two people with different occupational classes and the same income.''
What I don't understand is this. Let the variables X, Y and Z be causally dependend in such a way, that X influences both Z and Y, and Z influences Y, but not X, disturbances for all three variables being independent. It seems to me that controling for Z, which is an intermediate outcome in this situation, makes perfect sense. This is exactly what one would do to estimate the strength of causal influence of X on Y which is not mediated by Z. Controling for Z would be problematic if Z was influenced both by X and Y, which is exactly the case when X is occupation, Z is number of children of the other sex and Y are the biological factors responsible for the sex of the child. The model X->Y, X->ZY, X->Z->Y doesn't. So it is not controling for intermediate outcome as such that can give biased estimate. Surely, I must be wrong here, right?
Andrew: My last comment to this post got srcambled. The third sentence from the bottom should say: The model X infuelcing Y, X influencing Z and Y influencing Z makes controling for Z problematic, the model X influencing Z, Z influencing Y doesn't. I suppose my ascii left arrow was interpreted as html tag. Still curious of the answer.
Sarcasm just doesn't work on the internet.
Only because many people reading the internet have autism spectrum disorders.
I smarted at Gelman's statement that Kanazawa is doing something admirable. The man is practicing a cargo cult imitation of science, in service of his crackpot right wing ideology.