All the major meta-analyses come to roughly the same conclusion, which is that anti-depressants have modest positive effects. The paper you linked to finds exactly this, but argues that the effect size they found is insignificant because of the high risk of bias. Other studies have gone to extraordinary lengths to control for potential biases and come up with a very similar effect size. Reasoning about the data is quite challenging, because we're talking about relatively small shifts in a relatively broad statistical distribution.

"Number needed to treat" is a useful epidemiological concept in this case. A perfect drug that cures every single patient has an NNT of one - for each patient you treat, one patient is successfully treated. When you translate the results of SSRI meta-analyses, you get an NNT of about seven - for every seven patients who take SSRIs, one more patient will recover than if those patients were given a placebo.

It's hard to know what to do with that information. Is it worth prescribing antidepressants to 1,000,000 people to treat 142,000 of them? On a pure cost-benefit analysis, the answer is a resounding yes. Antidepressants are relatively cheap and safe. Are you willing to take a pill every day to get a one-in-seven chance of an effective treatment? Maybe. If you're in the throes of a deep depression, you might be willing to give anything a try. If you're just feeling a bit blue, maybe not. By the same token, if you're taking antidepressants and you don't think that they're helping, you're probably right.

>> The bigger problem is “depression” is not reproducibly measurable, because it has no falsifiable or biomedical marker.

That's something I like to point out using SSRIs as an example. The claim is that they increase levels of serotonin and that the cause of the depression is reduced serotonin. The thing is, nobody actually checks your serotonin levels and says "oh that's low take this." Then there's the problem of why ones serotonin is low when it wasn't always that way. There are probably many causes of depression and quite possibly some that don't lower serotonin.

>no more effective than exercise

You say this as if it were a bad thing but if SSRIs are as effective as exercise wouldn't that mean that they are very effective?

If you were to tell me a fat burning pill were no more effective than exercise I would start taking it.

We don't really know how antidepressants work. We don't know how a lot of drugs work. Nobody knows why acetaminophen (paracetamol) cures your headache. Knowing how a drug works is useful but not essential; modern medicine relies on hundreds of drugs with unknown mechanisms of action.

The Cipriani study is a nice meta-analysis: "The random-effects summary SMD for all antidepressants was 0·30" This site explains SMD: http://handbook-5-1.cochrane.org/chapter_9/9_2_3_2_the_stand.... This site shows what 0.3 looks like (actually shows 0.2 and 0.5): http://rpsychologist.com/short-r-script-to-plot-effect-sizes.... Typically, when people are summarizing this line of research saying that there is no effect, they are talking about these kinds of effect sizes which look like they round to pretty much no effect. I think using SMD for reporting by itself is unfortunate: my understanding is that the drugs have large effects for a minority of patients -- rather than that they shift everyone's emotional experience over a little bit.

This is a dangerous opinion and can inspire guilt in folks struggling with mental disorders. Not cool.