There is a good debate presented in the September 2009 issue of the Canadian Journal of Psychiatry (pp. 637-643), about the importance of randomized controlled trials in psychiatric research and clinical practice.
Steven Hollon presents a strong case supporting the philosophical foundations of RCT research, while Bruce Wampold presents many good points about the present limitations and weaknesses prevalent in current psychiatric RCT research studies. In particular, Wampold points out that much evidence exists regarding the relevance of the individual therapist (and, I might add, of the individual sense of patient-therapist alliance or connection) in determining therapeutic outcomes, and that this very individual factor may have a stronger influence on outcome than the particular "treatment" being offered (whether it be CBT, psychoanalysis, a medication combination, etc.).
My own view of a lot of the evidence resonates with these ideas. I strongly support the importance of randomized controlled trials in medicine and psychiatry. Yet it often seems to me that many variables are not accounted for. The impact of the individual therapist is one specific factor. If the patient is more comfortable with one therapist than another, than this factor alone may greatly outweigh the effect of the particular style of therapy being offered. Interestingly, this factor may not necessarily depend on the length of experience of the therapist -- sometimes a trainee may have a more positive therapeutic impact than a therapist who has decades of experience. This fact is not surprising to me: a lot of psychotherapy can have a lot to do with the capacity for the therapeutic relationship to grow and be healthy, which may depend substantially on very personal factors in the therapist. This may be humbling to those of us who revere the notion of psychotherapeutic theory being of paramount importance.
The whole of psychiatric theory may, at least in some cases, be less important than the goodness of a single interpersonal connection.
But I do also believe that certain therapeutic techniques are more effective than others. I think that strategies which promote daily long-term psychological work just have to be more effective (along the lines of language learning again). Also I think that strategies which encourage and help a person to face their fears or to move away from destructive habits are more likely to be helpful than strategies which do not look at these issues.
Many other factors are often not controlled (or examined at all) in present psychiatric RCTs, including nutrition, exercise, other self-care activities, supportive relationship involvement, community involvement, altruistic activity, etc.
Another factor that I have considered is the heterogeneity of many studied psychiatric populations. Different individuals with so-called "major depressive disorder" may in fact have different underlying causes for their symptoms; some of these individuals may respond well to one type of treatment, others may respond to something else. I suppose the RCT design remains appropriate in this situation, yet a powerful focus in research, in my opinion, needs to be to examine why some people respond to something, while others don't.
This erratic pattern of response doesn't just happen with individuals in a particular study. There are whole studies in which a well-proven psychiatric treatment (such as an antidepressant) doesn't end up differing from placebo. I don't think such studies show that antidepressants (or other treatments) are ineffective, but I do think it strongly suggests that the current criteria for psychiatric diagnoses are insufficient to predict treatment response as consistently as we need.
Often times, these negative studies are dismissed automatically. In many cases, such studies have been poorly designed, and that is the main problem. But in other cases, I think we need to very carefully examine such negative studies, to understand why they were negative.
This is consistent with another type of scientific rigor (different from the RCT empirical approach): in mathematics, a single counterexample is sufficient to disprove a theorem. If such a counterexample is found, it can be extremely fruitful to examine why it occurred--in this way a new and more valuable theorem can be conceived. The process of generating the disproven theorem was not a waste of time, but could be understood as part of a process to find the accurate theorem. Such examples abound in other fields, such as computer programming--a program or algorithm may work quite well, but generate errors or break down in certain situations. Careful examination of why the errors are taking place is the only way to improve the program, and perhaps also to more deeply understand the problem the program was supposed to solve.