Automation Exposure and Older Worker Labor Force Nonparticipation: A Methodological Demonstration of Doubly Robust Estimation

Abstract

This paper demonstrates the application of doubly robust estimation methods to study the relationship between occupational automation exposure and labor force nonparticipation among older workers. Using synthetic microdata calibrated to American Community Survey population characteristics, I illustrate how augmented inverse probability weighting (AIPW) can be applied to labor market questions involving selection on observables. Using only pre-determined covariates (demographics and education) to avoid post-treatment bias, the synthetic analysis suggests an association of approximately 0.9 percentage points between high-automation occupations and labor force nonparticipation, concentrated among workers aged 61–65. The paper demonstrates key methodological components: propensity score estimation and diagnostics, covariate balance assessment, and calibrated sensitivity analysis. Because actual ACS data does not provide occupation for individuals not in the labor force, this specific research design requires panel data (e.g., HRS, SIPP) for implementation with real data. The contribution is methodological: illustrating how doubly robust methods and sensitivity analyses can be combined in policy-relevant applications.