Tutorial for ‘LaLonde (1986) after Nearly Four Decades: Lessons Learned’

Welcome!

This is a Quarto Book (Rbookdown previously) tutorial condensing the analysis from Imbens & Xu (2024).

Note

Replication: For an exact replication, please refer to the source GitHub repository, which contains the data and code necessary to replicate Imbens & Xu (2024): “LaLonde (1986) After Nearly Four Decades: Lessons Learned.” [arXiv].

We reflect on five key lessons from the methodological literature following LaLonde’s 1986 study:

  • The central role of unconfoundedness
  • The importance of overlap in covariate distributions
  • The focus of propensity scores in constructing modern estimators
  • Credibility through validation exercises
  • Investigation of alternative estimands

Revisiting LaLonde’s analysis, we present how modern methods could enhance causal inference using data compiled by Rajeev Dehejia and Sadek Wahba (Dehejia and Wahba 1999, 2002) and survey data of lottery players in Imbens, Rubin, and Sacerdote (2001). Hereafter, we denote the former dataset as the LaLonde-Dehejia-Wahba (LDW) data, and the latter as IRS data. Our findings highlight that causal inference with observational data demands rigorous ( i ) examination of the assignment process, ( ii ) assessment of overlap, and ( iii ) execution of validation checks. This tutorial streamlines the methodology used in Imbens & Xu (2024), making it easier for readers to replicate the study and apply its techniques in their own research. In addition, we supplement a complementary application of our analytics with the original LaLonde (1986) data after the LDW analysis to further demonstrate and validate our methodologies.

Reference: Imbens, Guido and Yiqing Xu (2024). “LaLonde (1986) after Nearly Four Decades: Lessons Learned.” arXiv:2406.00827.

Acknowledgement: We thank Zihan Xie and Jinwen Wu for their excellent research assistance, which makes this tutorial possible.

Data Files

Imbens and Xu (2024) uses the following datasets, which are based on LaLonde (1986), Dehejia and Wahba (1999), Calónico and Smith (2017), and Imbens, Rubin, and Sacerdote (2001).

Data.files Details File_Type Experimental
nsw.dta NSW experimental data, used in LaLonde (1986) Stata Yes
nsw_dw.dta Subset of NSW experimental data, used in Dehejia & Wahba (1999) Stata Yes
cps_controls.dta CPS-SSA-1 controls, used in both papers Stata No
psid_controls.dta PSID-1 controls, used in both papers Stata No
lottery.RData Data of lottery winners, used in Imbens, Rubin & Sacerdote (201) R No
NSW_AFDC_CS.dta Reconstructed NSW AFDC female samples Stata Both