An experiment is where you, the investigator, control (or attempt to control) some aspect of the study participants’ behavior.
In an experimental study, participants are assigned to either do or don’t do something (e.g. take a drug, adhere to a diet, etc.). This can be contrasted with observational study designs, where study participants choose their own fate, and investigators merely record the results. The prototypical experiment in epidemiology is the randomized controlled trial (RCT).
Hypothesis: Daily use of Palm Pilots by medical students results in atrophy of the hippocampus.
- Measure hippocampi and Palm Pilot usage in a sample of medical students at a single point in time (cross-sectional).
- Find a group of medical students with atrophied hippocampi, a second group without atrophy, and compare past Palm Pilot usage (case-control).
- Find a group of medical students that uses Palm Pilots daily, a group that does not, then measure their hippocampi over time (cohort).
- Randomly assign medical students into two groups. Instruct one to use their Palm Pilots every day, and the other to not. Measurements of their hippocampi will be taken over time (RCT).
Observational studies are subject to many biases that experimental studies are not. In the example above, it is not difficult to imagine that students who depend on Palm Pilots are small brained for other reasons, and while the two are associated, it is not a causal relationship. An experimental study eliminates this possibility by dictating the exposure.
While experimental studies are indisputably the bees' knees, they are not appropriate for all situations. First and foremost, they are very expensive and time-consuming. As with all prospective study designs, you may be old, gray, and broke by the time the outcome of interest occurs. Before undertaking an RCT, your hypothesis should be tested using other study designs. There are also ethical considerations. It is unethical to assign people to an exposure that is known to be harmful. For example, you cannot assign people to smoke. It is also unethical to assign a therapy that is below the current standard of care, e.g. a placebo for persons with AIDS. This can have a huge effect on the sample size needed.