What level of significance did they calculate that these results were at?
For the binary questions, they use the usual p<0.05 as the threshold value for reporting; many individual issues have much lower p values quoted. For the ORs mentioned in my previous post they give 95% confidence intervals (many of which are quite wide, yes, but not overlapping 1). Here's the main methodology paragraph in case it's more helpful:
We used chi-squared and analysis of variance to examine the relationship of childhood IQ with drug use in adolescence and adulthood. Analysis of variance was used to examine the mean IQ score at 10 years in those who used each type of drug at 16 and 30 years. We used multivariate logistic regression to estimate the relation of childhood IQ scores with likelihood of ever having used drugs at 16 years and having used drugs in the past 12 months at 30 years. In preliminary regression analysis, we modelled IQ as a linear term with ORs expressed per 1 SD increase in IQ score; to explore non-linear relationships, we repeated all models with tertiles of IQ scores.
Were there roughly equal numbers of subjects considered in each IQ sextile, or were they working from a sample that had a normal distribution of IQ?
The latter. (The former would be decidedly weird!)
It seems that with this sample size, if it was a normally distributed sample, you'd probably only have about 106 people at 98th percentile and above (over 130 IQ). That is few enough that each individual who reported in the affirmative would account for nearly a full percentage point increase in the rate of drug use for that group, and that could make a huge difference in the reported relative chances, since, IIRC, the rates of drug use they found at age 30 were in the single digits.
On the last point, no: in the sample overall, at age 30 they found that 35.4% of men and 15.9% of women had used cannabis in the previous 12 months. They did indeed find lower percentages for other drugs, but cannabis is the only one for which they report results in IQ sextiles as opposed to tertiles, and that's what the ORs in my previous post refer to. Come to think of it, this may be precisely because their sample wasn't big enough to get significant sextile results for the less used drugs, as you suggest.
