ABSTRACT
This chapter describes the elusive concept of “holding other factors constant” or “ceteris paribus.” It starts with a few case studies, including a simple test for whether adding cinnamon to your chocolate-chip cookie improves the taste, with the idea that you want to design the test to “hold constant” the amount of chocolate, butter, and sugar as the amount of cinnamon varies between two batches. This concept is then translated to a regression, with the concept of making sure other important inputs to an outcome do not vary across people who have different levels of a treatment. The general question is that, as the treatment varies in the sample, what other factors do you want to allow to change with it and what other factors do you not want to change (so that they do not impact the outcome). And so, this includes a discussion of cases in which one would not want to hold a variable constant when estimating causal effects – namely, for mediating factors through which the treatment affects the outcome. Flow charts and simple numerical examples are used to demonstrate how controlling for mediating factors impacts the estimated causal effect.
