Skip to main content

Returns to education with instrumental variables (IV/2SLS)

Following the Mincer OLS example: because educ may be endogenous (unobserved ability), we use IV/2SLS for a consistent estimate. Figures are illustrative.

Summary: use an instrument for schooling (e.g. distance to school, an education reform, regional tuition) to isolate the exogenous part and estimate an unbiased return to education.


Step 1 — Ideation

  • Question: what is the true return to education after removing endogeneity?

Step 2 — Literature Review

Returns-to-schooling literature using IV (Card 1995; Angrist–Krueger 1991 — quarter of birth); discuss instrument validity.

Step 3 — Data Collection

RoleVariableExample
Dependentlnwagelog wage
Endogenouseducyears of schooling
Instrument ZZdistance_to_schooldistance to the nearest school
Exogenousexper, gender, regioncontrols

Step 4 — Modeling

Choose the IV & simultaneous equations family → IV/2SLS; declare the endogenous educ and the instrument ZZ:

Illustrative results (format — not real results):

OLSIV/2SLS
educ0.082***0.104***
First-stage F24.5 (> 10 ⇒ strong instrument)
Hansen J (p)0.31 (instruments valid)
Durbin-Wu-Hausman (p)0.04 ⇒ endogeneity present, IV needed

Sample interpretation: IV gives a higher return than OLS (0.104 vs 0.082) — consistent with much of the literature; F > 10 and a non-rejected Hansen ⇒ the instrument is acceptable.

* ── IV/2SLS: returns to education ─────────────────
* Endogenous: educ | Instruments: near_college, parent_educ
ivregress 2sls lnwage exper exper2 ///
(educ = near_college parent_educ), first robust

* ── Diagnostics ───────────────────────────────────
estat firststage // First-stage F-statistic
estat endogenous // Durbin-Wu-Hausman test
estat overid // Sargan / Hansen J test

Step 5 — Reporting

Export a report + replication code; report the first-stage F, the endogeneity test and overidentification.

Video tutorial

Video Tutorial: Guide to running IV/2SLS in EcoLab

See also