A manipulative experiment is one in which the experimenter manipulates the system of study in order to attempt to uncover causal relationships. Such experiments can either be rigidly or loosely controlled. In the former case, investigator may create an artificial system in a laboratory setting and control all confounding variables so as to eliminate ambiguity in the interpretation of results. In the latter case, the investigator may simply modify one or two variables of interest in order to determine how the natural system will respond. The statistical methods used to analyse the data are generally of the analysis of variance type, but there are exceptions (especially in the case of loosely controlled manipulations). Two examples follow, one where rigid control was exercised, and one where loose control was used.

  1. There are many examples of rigidly controlled manipulative experiments, and any undergraduate science student has likely performed one in the course of getting their degree. One such experiment that is particularly well known is the FDA's work with olestra. In initial stages of determining whether or not this food additive is safe for human consumption, the FDA performed a series of trials on rats to determine whether or not increased daily dosages of olestra lead to any adverse health effects. In this experiment, laboratory rats were kept disease free, in identical enclosures with identical diets, water, illumination etc. The only varying factor was the dose of olestra. The conclusion from such a study was that at relatively low doses, olestra was safe for human consumption (That is, if you don't mind a little anal leakage).
  2. A good example of a loosely controlled manipulative experiment in the field of limnology is the classic work of David Schindler, now of the University of Alberta, on the role of phosphorus in determining the productivity of aquatic systems. He divided an unproductive (oligotrophic) lake in Ontario into two sections using a large poly-urethane barrier (essentially, large unfolded garbage bags), and fertilized one half of the lake with phosphorus. After a couple of weeks, the fertilized side of the lake turned green while the unfertilized side remained blue, demonstrating that the limiting nutrient in freshwater systems is indeed phosphorus. This experiment was clearly loosely controlled, as a natural system was used and only a single factor affecting the system of interest (the algal community of the lake) was varied.

Manipulative experiments are probably the kind with which the public is most familiar, through personal experience in educational settings or through the media. They provide researchers with a particularly powerful tool to uncover causal relationships. However, they often suffer from the problem of over-simplification; the systems of interest are often either too large or too complex to be readily simplified in a laboratory setting, and thus researchers must resort to mensurative experiments instead.