Everytime you-know-who publishes a paper, I seem to get an email about it. But this time I’m not responding.
12 thoughts on “Self-restraint”
Is it just me or is that way too cryptic a remark?
Voldemort? I didn't realize he was still writing.
Seriously, I have no idea who "you-know-who" is.
Voldemort? ;)
Lol. I was about to send you this exact paper. I re-ran his findings using the GSS data and at least that model seems robust to different specifications (e.g., education and time as dummies, including age-squared, etc). But in no way is the number of questions correct on a vocabulary test a measure of innate intelligence, as you-know-what implies.
Oh, come on!
Ha ha, SK is making controversial, poorly backed up claims again?
Satoshi Kanazawa.
Thanks Ben – until your comment I hadn't realized he was questionable. Too bad I didn't know that before I posted on his paper Friday.
One of my biggest regrets at Canterbury is that I didn't know Kanazawa's work for the three months that we overlapped here – he left for England shortly before I started reading him. I regularly kick folks from our Psych department in the shins for having let him go. Would have loved to have had a beer with him.
Actually, Andrew, you would do a good community service if you were to analyze the paper and point out what (if anything) is wrong with the latest paper in terms of statistical methods. (Obviously, the interpretation is well beyond any science, so let's just concentrate on how good the data is). Problem is, a lot of folks seem to believe everything as long as it is stated that p is [some very small value].
E.g., in the Methods, is it really enough to state "because both of these dependent variables
are measured on an ordinal scale, I use the
ordinal regression (McCullagh 1980) to estimate
these models" to make the work reproducible by others? Or, what do you think about presenting error bars as SEM instead of CI or SD? Or maybe what you think of statements like "The comparison of standardized coefficients shows that adolescent intelligence has more than four times as strong an effect on the value for sexual exclusivity for men as it does for women (0.0465 vs. 0.0110)". That close to a noise, can one seriously and meaningfully compare the b values? I would have thought that any hidden or unaccounted for variable has a potential to completely throw off anything like these no matter what the formal p is – but I'd love to hear a voice of professional statistician.
I'm pretty sure wordsum is a fairly valid proxy measure for intelligence, so I don't agree with E.F.'s remarks. The problem, as I see it, is making all sorts of causal arguments with this data and models. His (SK's) models are quite poorly specified–it seems he didn't think about other reasons why liberalism and atheism and intelligence are correlated.
Nanonymous: I'd do this, but only if you'd pay me a lot. I'll do lots of things for free, but reading anything more by this dude isn't one of them!
Is it just me or is that way too cryptic a remark?
Voldemort? I didn't realize he was still writing.
Seriously, I have no idea who "you-know-who" is.
Voldemort? ;)
Lol. I was about to send you this exact paper. I re-ran his findings using the GSS data and at least that model seems robust to different specifications (e.g., education and time as dummies, including age-squared, etc). But in no way is the number of questions correct on a vocabulary test a measure of innate intelligence, as you-know-what implies.
Oh, come on!
Ha ha, SK is making controversial, poorly backed up claims again?
Satoshi Kanazawa.
Thanks Ben – until your comment I hadn't realized he was questionable. Too bad I didn't know that before I posted on his paper Friday.
One of my biggest regrets at Canterbury is that I didn't know Kanazawa's work for the three months that we overlapped here – he left for England shortly before I started reading him. I regularly kick folks from our Psych department in the shins for having let him go. Would have loved to have had a beer with him.
Actually, Andrew, you would do a good community service if you were to analyze the paper and point out what (if anything) is wrong with the latest paper in terms of statistical methods. (Obviously, the interpretation is well beyond any science, so let's just concentrate on how good the data is). Problem is, a lot of folks seem to believe everything as long as it is stated that p is [some very small value].
E.g., in the Methods, is it really enough to state "because both of these dependent variables
are measured on an ordinal scale, I use the
ordinal regression (McCullagh 1980) to estimate
these models" to make the work reproducible by others? Or, what do you think about presenting error bars as SEM instead of CI or SD? Or maybe what you think of statements like "The comparison of standardized coefficients shows that adolescent intelligence has more than four times as strong an effect on the value for sexual exclusivity for men as it does for women (0.0465 vs. 0.0110)". That close to a noise, can one seriously and meaningfully compare the b values? I would have thought that any hidden or unaccounted for variable has a potential to completely throw off anything like these no matter what the formal p is – but I'd love to hear a voice of professional statistician.
I'm pretty sure wordsum is a fairly valid proxy measure for intelligence, so I don't agree with E.F.'s remarks. The problem, as I see it, is making all sorts of causal arguments with this data and models. His (SK's) models are quite poorly specified–it seems he didn't think about other reasons why liberalism and atheism and intelligence are correlated.
Nanonymous: I'd do this, but only if you'd pay me a lot. I'll do lots of things for free, but reading anything more by this dude isn't one of them!