Originally Posted by claymisher
I doubt I can be much help. Shalizi's discussion of factor analysis is right at the edge of my comprehension. What Shalizi's getting at is that what you're assuming to be true ("Studies which DO," "Studies ... tend to find high correlations," "Studies ... tend to find low correlations") may not actually show what people think the show.
If you're curious about the mechanics of twin studies, I got another link for you
OK. I read that. But it is statements like this from Shalizi that prompted my original question: "If identical twins experience more similar environmental influences on accent than fraternal twins, then even it [the study] will conclude that accent is, in fact, heritable." This is true, but the reason for this possible finding is that the study is only comparing identical and fraternal twins, who have a large amount of genetic similarity to begin with, and which therefore does not isolate the influence of genes versus environment. For such a finding to have any meaning, the same things would have to be studied in individuals with fewer and with no shared genes and then compared to the results among twins. In a well controlled and designed study, these same correlations of accent would probably also show up among regular siblings and even among unrelated strangers, which would undermine any heritability conclusions one could draw from the twin data. Shalizi seems to argue that the problem could be solved by a more rigorous twin study, which no-one has ever done apparently, but the problem actually cannot be solved unless the study also involves individuals with no shared genes. The key is not to compare results between twins groups but to compare results among all the relatedness and environmental combinations. That's why I am curious as to whether studies that examine all of the groups I listed in my original post, and thereby do a better job at finding and omitting erroneous data about environmental and heritable correlations, are in fact not as vulnerable to these statistical fallacies and errors that Shalizi is concerned about. I guess I just get the feeling that what Shalizi is criticizing is really an artifact of poor study design and sloppy thinking, both of which are certainly deserving of criticism, but don't really say anything about the invalidity of data that is collected and analyzed more rigorously in whatever field of study. If, however, this factor analysis problem exists regardless of study design and analysis, that would be worth understanding better, though it would seem to ultimately leaves us unable to know anything about anything in the social sciences.