PS2

Toolkit

Terms

endogeneity causal-diagram ols-estimator instrumental-variables two-stage-least-squares wu-hausman-test instrument-relevance weak-instruments instrument-validity heckman-selection-model sample-selection-bias inverse-mills-ratio

Lectures

Instrumental Variables Logit & Probit Models Sample Selection & Heckman Correction

setup

Women's Fertility & Education. Does education causally reduce the number of children a woman has?
1

Endogeneity of Education
2

Causal diagram (DAG)
3

OLS regression
4

Instrumental variables (2SLS)
5

Why OLS and IV differ
6

Instrument relevance
7

Instrument validity (exclusion restriction)
8

Heckman selection correction

What this PS demonstrates

Three layers of bias in a naive fertility regression: endogeneity (Q1–Q5, OLS understates the negative effect because reverse causality cancels part of it), instrument quality (Q6–Q7, parents' education is highly relevant but its validity is debatable), and selection bias (Q8, Heckman shows education's effect operates primarily through the decision to have any children, not family size).

Related: Lec_04-Instrumental Variables · Lec_03-Logit & Probit Models (probit step in Heckman) · Lec_05-Sample Selection & Heckman Correction