Folks, reading the paper, Table 2 basically proves the exact opposite of what the paper claims. The link is not at all proven. They cherry-pick one significant result out of 36 statistical tests. The level of significance is not specified but, the way it's reported, is probably between 0.01 and 0.05 (wrong between once in a 100 trials or 1 in 20), while Table 2 reports 36 statistical tests. In other words, significance of a test at this level of alpha (type I error) is not at all established. Moreover, there's no dose dependence, it would seem, whereas typically there would be a log-linear dependence on dose. The study is somewhere between inconclusive and (based on their loose understanding of statistics) junk.
Slashdot Mirror
Folks, reading the paper, Table 2 basically proves the exact opposite of what the paper claims. The link is not at all proven. They cherry-pick one significant result out of 36 statistical tests. The level of significance is not specified but, the way it's reported, is probably between 0.01 and 0.05 (wrong between once in a 100 trials or 1 in 20), while Table 2 reports 36 statistical tests. In other words, significance of a test at this level of alpha (type I error) is not at all established. Moreover, there's no dose dependence, it would seem, whereas typically there would be a log-linear dependence on dose. The study is somewhere between inconclusive and (based on their loose understanding of statistics) junk.