Statistics in a Nutshell

Chapter 5. Research Design

When trying to make sense of the world using statistics, it’s important to consider which type of research design will provide you with the most accurate answer for the type of question you are asking. Typically, the selection of a particular design involves an amount of horse trading between the extent to which a research environment is able to be controlled versus only being able to be observed, and whether the research goal involves determining some underlying cause for a phenomenon (science), or whether the goal is to optimize the yield or output for a specific process while minimizing effort (technology). Various other impediments to the use of certain designs—such as human research ethics committees—may require you to carry out your research in ways that are less than ideal in a statistical sense, but ensure that your work is consistent with community rather than just scientific values.

The two main types of research design are experimental and observational studies. In an experiment, some degree of manipulation is involved, since the intention is that the researcher should maximize control over as many aspects of the environment as possible. All variable measurements and manipulations are under research control, including the allocation of experimental units (or subjects or participants). An observational study implies that no change of the environment is necessary, and that the allocation of experimental units to groups is outside the control of ...