We saw that our employee contribution example requires a sharp regression discontinuity design: all companies with at least 300 employees have a contribution matching program, and all companies with fewer than 300 employees do NOT have a contribution matching program. Before we decide which companies are similar enough to compare, we must consider some assumptions.
In order to get valid estimates of the treatment effect using a sharp RDD approach, several assumptions have to be met.
- The treatment variable impacts the outcome, but not any of the other variables.
- The treatment assignment happens only at ONE cutpoint value of the forcing variable.
- Treatment assignment is independent of the potential outcomes within a narrow interval around the cutpoint.
- Counterfactual outcomes can be modeled within the interval around the cutpoint.
Take a look at the learning environment and hover over each card to illustrate how each of these assumptions relates to the contribution matching program example.