Quick Recap: Observational Research Design

Observational Research Designs

• When we want to study the causal effect of X on Y, but we cannot conduct an experiment, we rely on observational research designs
• In observational research, treatment assignment is not under the control of the researcher
  • This means treatment and control groups not necessarily similar, on average, and therefore have to worry about confounding variables
• We can use statistical methods, like linear regression, to "control" for confounding variables

Controlling for confounding variables

• The first step in controlling for confounding variables is thinking about all possible confounders we can think of
• We then need to measure these confounding variables so that we can control for them using statistical methods
• However!
  • Some confounding variables are likely not measurable, available, or observable (such as a dictator’s willingness to let their country democratize)
  • We could also overlook confounds, either because we don't think of it or because we didn't know it was a confound
• The difficulty in controlling for confounding variables makes observational research harder to sell when trying to measure a causal effect between an independent variable and a dependent variable

Dig Deeper

• Galderisi, Peter. Understanding Political Science Statistics: Observations and Expectations in Political Analysis. Routledge, 2015.
  • Chapter 10: "Research Design and the Use of Control Variables"