Treatments

In a one-factor study, the treatments are the same as the levels of the factor. In a multifactor study, each unique combination of factors is a treatment. For example, a two-factor study of three kinds of deodorant and four levels of exercise has twelve treatments.

Choosing which treatments (which factors, how many levels, etc.) to study is not easy. This is part of experimental design, which we will discuss more in a couple of weeks. For now, here are some main points to keep in mind:

• There are generally two goals of a study or series of studies: one is to understand a process, and the other is to make a prediction or comparison.

  Studies that are intended to help in understanding a process may start out small: the initial investigation may involve only a few factors, or a few levels of the factors; and then after the main factors of interest have been identified a more complete study may be undertaken, to provide a more detailed picture of the process.

  Studies that are intended to help answer a specific practical question tend to have fewer factors, but even here including some extra levels to help provide some understanding of the findings is useful. For example if your company is considering changing to a new manufacturing process that is expected to allow you to increase the number of units per day that you produce, and you want to study defects in the manufactured products, you may want to try include not only the ``old'' method under the current production schedule, the ``new'' method under the current production schedule, and the ``new'' method under the accellerated production schedule. This way you can disentangle the effect of the change in process from the effect of speeding up the production schedule, on the number of defects you see.

• The definition of ``What is a treatment'' is not always obvious. This is especially clear in educational studies. Suppose two teaching methods were being compared, and different teachers were using the different methods. Since the teacher has an effect on how well a method is implemented, it's may not be clear whether the treatment is just the teaching method, or a combination of teacher and method.

• A control treatment consists of withholding all active treatments but still applying the same experiemental procedures as would be needed for the active treatments. Controls are needed, for example,
  • when the general effectiveness of all the active treatments is not known or not consistent; or
  • in studies where ``halo'', ``placebo'', or ``experimenter bias'' effects could make an ineffective active treatment look effective.
  Some studies, such as comparisons of a new treatment for a disease vs. a current, already effective treatment for the disease, do not need a control since there is an effective treatment standard, the current treatment, to which we may compare. In such studies it may also be unethical to include a control treatment (withholding treatment from those in need).