Quick Recap: Experiments

Confounding Variables

• When trying to understand the causal effect of X on Y, one of our biggest worries are confounding variables
• Confounding variables are:
  • Correlated with X (our independent variable), and
  • Affect Y (our dependent variable)
• Not accounting for confounding variables could lead us to make incorrectly conclusions about the effect of X on Y

Experiments

• Experiments allow researchers to mitigate the problem of confounding variables
• An experiment is defined by a researcher having control over the treatment assignment
  • Treatment assignment determines who gets the treatment and who does not
• Randomized treatment assignment
  • In experiments we ideally want to randomly assign some units to get the treatment and not others
    • This assures that assignment to treatment is not correlated with confounding variables
  • The group that gets the treatment is called the treatment group, while the group that does not is called the control group
  • Randomized treatment assignment creates two (or more) groups that are very similar, on average, across observable and unobservable characteristics
• Causal effects
  • Since random treatment assignment creates treatment and control groups that are very similar on average, and assures that assignment to treatment is not correlated with confounding variables, we can be fairly certain that any difference in the outcomes that we observe between the treatment and control groups is due to the treatment (causal effect!)
• Experiments are not always feasible, practical, or ethical