We need a little more vocabulary before we can dive into more details about Regression Discontinuity Design.
Sometimes treatment group assignment is dictated by one continuous variable known as a forcing variable (the company size in our example). The forcing variable, also referred to as a rating variable or running variable, has a cutoff value such that:
- Individuals with a value smaller than the cutoff are in one treatment group.
- Individuals with a value larger than the cutoff are in the other treatment group.
The treatment group is perfectly predicted by the forcing variable. In this scenario, we cannot rely on other causal inference techniques such as matching or weighting methods because there is not a consistent mixture of treatment and controls across different values of the forcing variable.
Take a look at the plot provided in the learning environment. Note how points with a value below 60 on the x-axis belong to the control group, while those with values above 60 belong to the treatment group.