Double-Blind
From SkepticWiki
[edit] Definition
A double-blind experiment is an experiment (typically a clinical test) performed under conditions where neither the experimental subject(s) nor the experimenter running the experiment knows whether or not any given subject is part of the treatment group or part of the control group. Such an experiment, particularly in conjunction with placebo control, has become standard practice in medical trials to eliminate sources of bias and error.
For example, suppose a researcher believes (for whatever reasons) that a diet supplemented by dilithium will reduce hair loss. Rather than giving some patients dilithium supplements (and other patients nothing), the researcher will instead ask a pharmacist to prepare two bottles of supplements, some containing dilithium, and others containing nothing beyond the stocks sugar pill, without telling the researcher which is which. Not knowing which is which, the doctor will give some patients (randomly chosen) pills from the red bottle and some patients pills from the blue bottle. Since neither he nor the patients know whether the red or the blue bottles contain dilithium, he can compare the differences between the two groups without letting his biases interfere.
So-called "randomized double-blind placebo-controlled trials" are the gold standard against which clinical experiments are judged, as they provide the least opportunity for fraud or self-deception on the part of the researchers. Most forms of pseudoscience and alternative medicine base their claims, at best, on studies that are not performed to this standard. When they are examined using this type of experiment, the results tend to evaporate. This suggests that the claimed successes of many forms of alternative medicine such as homeopathy, reiki, and iridology are, in fact, merely examples of fraud and self-deception and that the treatments have no clinical effectiveness beyond placebo.
[edit] Discussion
As discussed in the related Placebo article, a significant number of patients will report improvement when they receive any treatment at all, even a treatment (like giving a Tic-tac and claiming it's a pill) known to be ineffective. For this reason, it is important when studying the effects of a medicine, to control for this "placebo effect" by giving all subjects something. If the patients given hard candy report as much improvements as the patients given "real" medicine, then the medicine obviously doesn't do anything beyond what the candy does. Alternatively, only if the patients given real medicine show more improvement than the patients given candy, can the medicine be shown to be effective.
In conditions of actual clinical practice, however, patient reports are rarely clear-cut. What should a doctor do when a patient reports, "Yeah, maybe I'm feeling a little better. I guess so"? If she wishes to tabulate the number of patients who self-report "improvement," such an ambiguous report could easily be argued to be either a "Yes," or a "No." Since doctors are human themselves, a doctor who really wants to show that this treatment works might be tempted to put this patient in the "yes" category if the patient received the real drug, but "no" if the patient was in the control group. (Such bias may even occur at the subconscious level and doesn't really reflect badly on the doctor herself.)
Similarly, if the doctor actually knows whether or not the patient is receiving the drug, the doctor's expectations of how (or whether) the patient will improve will change. Patients can often pick up these expectations and will sometimes report "what the doctor wants to hear" (again, often unconsciously).
Therefore, in order to reduce the possibility of this kind of systematic bias, it is important that the doctor not know (for each patient) whether the patient has been given the control or the experimental treatment. Some authorities will extend this idea even further, to a so-called "triple blind" experiment, where neither patient nor provider nor outcome assessor is aware of the treatment received by the patient. (In this case, one doctor would provide the pill from the red or blue bottle, not knowing which is which -- while another, separate, doctor would interview the patients, not even knowing whether each patient was in the red or blue group.)
Although this sort of double-blind (or even triple-blind) experiment is considered to be best possible practice in medical experimentation, there are often situations where it is impractical to carry out. For example, if the experimental treatment were surgical (e.g. a hysterectomy), most patients are capable of figuring out for themselves whether or not they received surgery, most doctors are capable of figuring out whether or not they actually performed a useful surgery -- and few ethical surgeons would be willing to expose patients to the risks of an unnecessary "sham" surgery. For some alternative medical treatments, such as acupuncture, there are similar difficulties. Most patients can easily tell if they have been jabbed by a needle. Most (Western) patients have sufficiently little knowledge of the theory of acupuncture that they may not know whether the jabs were in the correct spot. It may thus be possible to blind the patient to the experiment by poking the experimental group in the "correct" spots, while poking the control group in unrelated and ineffective spots as a placebo.
Unfortunately, as with the surgeons, it is almost impossible to blind the acupuncturist himself about whether the locations in which he is placing the needles are supposed to be effective. For this reason, it is very difficult to run a double- (or triple-) blind experiment testing the effectiveness of acupuncture. This difficulty in running tightly controlled experiments is one reason that many of the claims of alternative medicine persist. Proponents declare on the basis of badly controlled experiments that their treatments work, and at the same time declare that running correctly controlled experiments is a difficult if not impossible task.
