Working Paper: NBER ID: w31005
Authors: John Mullahy
Abstract: With count-valued outcomes y in {0,1,...,M} identification and estimation of average treatment effects raise no special considerations beyond those involved in the continuous-outcome case. If partial identification of the distribution of treatment effects is of interest, however, count-valued outcomes present some subtle yet important considerations beyond those involved in continuous-outcome contexts. This paper derives appropriate bounds on the distribution of treatment effects for count-valued outcomes.
Keywords: No keywords provided
JEL Codes: C10; C25; I1
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| treatment assignment (d) (C90) | observed outcomes (y) (C29) |
| count-valued outcomes necessitate careful consideration of inequalities (C35) | partial identification of treatment effect distributions (C46) |
| max(0, pr(y1=t) - pr(y0=t)) (C29) | best lower bound on treatment effect distribution (C22) |
| similar expressions (Y60) | best upper bound on treatment effect distribution (C22) |