How did I miss this one before?
As you may know, an excellent predictor of whether an outcome study favours one over another flavour of psychotherapy is the therapeutic allegiance of the lead investigator. Thus if you fancy CBT, and want to prove it against psychodynamic therapy, then your comparative study is likely to find better results for CBT.
The effect is called 'researcher allegiance', and has been documented now in quite a few studies. These have been meta-analysed in turn. In 2013 Munder et al looked at all these various meta-analyses, and found that those meta-analyses conducted by researchers who believed in the researcher allegiance effect were more likely to find evidence of it.
Munder et al's meta-meta-analysis also found evidence of the allegiance effect in the primary data. I believe we're still waiting for someone with a different set of expectations to do another meta-meta-analysis and come up with a different result.
Empirical psychology, don't you know: it's all lies, damned lies, and statistics - as Twain, it seems wrongly, reported Disraeli as saying.