1.5 Study Design

Once we have a sample selected, data can be collected in one of two general study designs:

  1. Observational Study: Data is collected in such a way such that the researcher does not manipulate or intervene in the characteristics of the individuals. Researchers observe or record characteristics of the sample through direct measurement or through a questionnaire or survey.

  2. Experiment: Data is collected in such a way such that the researcher does manipulate or intervene in characteristics of the individuals by randomly assigning individuals to treatment or control groups. Researchers then record characteristics of the individuals in the sample within the treatment and control groups.

The main reason for doing an experiment is to estimate a relationship between a treatment and a response.

For example, imagine that we want to know if taking a daily multivitamin reduces systolic blood pressure. If we performed an observational study, we would select a sample (hopefully a representative one) from a population of interest and then ask whether an individual takes a daily multivitamin and then measure their blood pressure. Would this data provide enough evidence to conclude that vitamin use causes a reduction in blood pressure?

Individuals that take daily multivitamins may be more health-conscious, eating more fruits and vegetables and exercising more, which may be related to blood pressure. The diet and exercise would be acting as confounding variables, making it impossible to say for certain if vitamin use has a direct impact on blood pressure.

Confounding Variables: A third variable that is a common cause for the “treatment” (e.g. multivitamin) and the “response” (e.g. blood pressure). In this case, a healthy lifestyle could be a variable that confounds the relationship between multivitamin use and blood pressure.

For example, say we note that on days with higher ice cream sales, there are typically a higher number of pool drownings. What variable could be confounding the relationship between ice cream sales and pool drownings?

In an experiment, we “manipulate” the characteristics for an individual by randomly assigning them to a treatment group. This random assignment is intended to break the relationship between any common causes and the treatment so as to try to reduce the impact of confounding. It is impossible to entirely remove the possibility of confounding, but the random assignment to a treatment helps. (Note: things can get complicated if individuals don’t comply with the treatment such as take the multivitamin every single day…)