What?
Treatment effects + structural estimation -> marginal treatment effect (mte). mte strived to be the unifying parameter of treatment effects across different margins (late + ate + att)
Why?
- Absent of heterogeneity in agent-based models:
- Two economists analyzing the same dataset with different valid instrument will esttimate different parameters, have different interpretions.
- Two economists using the same instrument but with different notions about what variables belong in choice equations will interpret the output of iv analysis diffrerently.
- lates solve a specific case of selection bias, but is not always the case of interest
How?
And?
Comparison to structural approach and treatment effects approach:
This post is in the collection of my public reading notes.