liulinxmu / soep-unemployment-health-spillovers

Replication code for Everding and Marcus (2020, Health Economics), combining DiD estimation, Lasso regressions, and entropy balancing

soep-unemployment-health-spillovers

Replication code for Everding and Marcus (2020, Health Economics).

Specifically, this study examines the causal effect of unemployment on spousal smoking behavior. For this purpose, we focus on involuntary entries into unemployment, combining difference-in-differences (DiD) estimation with a matching strategy based on entropy balancing (EB, Hainmueller 2012). For selecting control variables, we complement our econometric approach with two different procedures for control variable selection; a conventional approach based on previous studies and economic intuition as well as a machine learning approach based on Lasso regressions, the post-double-selection method (PDS, Belloni et al. 2014).

Main do-file

master_spousal_ue.do defines all relevant macros, the folder structure, and executes all files sequentially

Additional do-files

The actual steps of data pre-processing and analyses are divided into several additional do-files ("script" files in Stata), as described in the following:

gen_spousal_ue.do pulls the data

trans_spousal_ue.do transforms the data and generates the relevant variables

candidatevar_spousal_ue.do constructs and adds leads, lags, transformed variables (polynomials and log. trans.) and imputation flags

pds_spousal_ue.do fits Lasso regressions (PDS, see Belloni et al. 2014) on candidate variables

pdsmech_spousal_ue.do fits Lasso regressions and selects controls for analysis of mechanisms (i.e. alternative outcomes)

main-analysis1_spousal_ue.do runs main analysis (with EB, see Hainmueller 2012), part 1 (without post-double selection)

main-analysis2_spousal_ue.do runs main analysis (with EB), part 2 (only post-double selection)

desc-stats1_spousal_ue.do generates table for descriptive statistics and matching quality, part 1

rob-analysis-pds_spousal_ue.do performs all post-double selection robustness analyses

het-analysis1_spousal_ue.do investigates treatment effect heterogeneity by smoking status at baseline, part 1

het-analysis2_spousal_ue.do investigates treatment effect heterogeneity by smoking status at baseline, part 2

mech_spousal_ue.do runs analysis of mechanisms (with PDS), part 1

mech2_spousal_ue.do runs analysis of mechanisms (without PDS), part 2

desc-stats2_spousal_ue.do generates table for descriptive statistics and matching quality, part 2

Data

The main data source is the German Socio-Economic Panel (SOEP, version 33).

See https://www.diw.de/soep for detailed information on data access options.

References

Belloni, A., V. Chernozhukov, and C. Hansen. 2014. Inference on treatment effects after selection among high-dimensional controls. The Review of Economic Studies, 81(2), 608-650.

Everding, J. and J. Marcus. 2020. The effect of unemployment on the smoking behavior of couples. Health Economics, 29(2), 154-170.

Hainmueller, J. 2012. Entropy balancing for causal effects: A multivariate reweighting method to produce balanced samples in observational studies. Political Analysis, 20(1), 25-46.