The Logic of Outcome Research
How can we find out whether a psychological or biomedical treatment truly provides relief for someone with a particular mental disorder? As a way of tackling this issue, let us suppose that we want to find out whether a given drug—say, Wellbutrin—reduces the symptoms of depression. For a first step, we could administer Wellbutrin to a group of patients diagnosed with depression, and assess the patients’ status before and afterreceiving the medication. This assessment would provide useful information, but a lit-tle reflection shows that relying only on this procedure would be a mistake.
One problem with the before-and-after assessment is that it ignores the possibility of spontaneous improvement. With some disorders, many patients get better on their own, whether they receive treatment or not. Therefore, if patients are indeed better off after taking Wellbutrin than they were before, their improvement might simply reflect spontaneous improvement and have nothing to do with the medication. In addition, even without spontaneous improvement, most disorders fluctuate in their severity, and it seems likely that patients will seek treatment when they are feeling particularly bad. If so, the odds are good that the normal ups and downs of the disorder will make the patients look better a few weeks (or months) later. If they have been taking medication during those weeks, we might be inclined to assume that the drugs had caused the improvement.
To control for these factors, we need to compare two groups of patients, with one group taking Wellbutrin for, say, 6 weeks, while the other group receives no treatment. We also need to make certain that these two groups are matched at the start of the study—with similar symptoms, similar backgrounds, and so on. As we discussed, our best means of achieving this matching would be via random assignment— with a coin toss (or the equivalent) determining whether each participant in the study ends up in the treatment group or the no-treatment group. Then, if the participants in the two groups differ at the end of the procedure, we can be reasonably sure that the dif-ference arose during the study—presumably as a result of the treatment.
In addition, we would want to ensure that the participants in our study are reason-ably uniform in their disorder. Thus, we would want to make certain that all of the par-ticipants were truly depressed, and that their depression was not complicated by other problems that might mask the effects of the treatment. Likewise, we would probably want to make sure the participants were all similar to one another in the duration and severity of their depression. Otherwise, variations from one participant to the next might prevent us from detecting the therapy’s effects.