> The meta-analysis that showed that studies don't show a significant affect on depression from SSRIs ... has very few serious disputes or alternative ways to v

> The meta-analysis that showed that studies don't show a significant affect on depression from SSRIs ... has very few serious disputes or alternative ways to view the numbers. Here's one: http://slatestarcodex.com/2014/07/07/ssris-much-more-than-you-wanted-to-know/ http://slatestarcodex.com/2014/07/07/ssris-much-more-than-yo... A couple relevant bits include: "Effect size is a hard statistic to work with (albeit extremely fun). The guy who invented effect size suggested that 0.2 be called “small”, 0.5 be called “medium”, and 0.8 be called “large”. NICE, a UK health research group, somewhat randomly declared that effect sizes greater than 0.5 be called “clinically significant” and effect sizes less than 0.5 be called “not clinically significant”, but their reasoning was basically that 0.5 was a nice round number, and a few years later they changed their mind and admitted they had no reason behind their decision. Despite these somewhat haphazard standards, some people have decided that antidepressants’ effect size of 0.3 means they are “clinically insignificant”." ...and... "They also note that Kirsch’s study lumps all antidepressants together. This isn’t necessarily wrong. But it isn’t necessarily right, either. For example, his study used both Serzone (believed to be a weak antidepressant, rarely used) and Paxil (believed to be a stronger antidepressant, commonly used). And in fact, by his study, Paxil showed an effect size of 0.47, compared to Serzone’s 0.21. But since the difference was not statistically significant, he averaged them together and said that “antidepressants are ineffective”. In fact, his study showed that Paxil was effective, but when you average it together with a very ineffective drug, the effect disappears. He can get away with this because of the arcana of statistical significance, but by the same arcana I can get away with not doing that. So right now we have three different effect sizes. 1.2 for placebo + drug, 0.5 for drug alone if we’re being statistically merciful, 0.3 for drug alone if we’re being harsh and letting the harshest critic of antidepressants pull out all his statistical tricks."