Abstract

Non-communicable diseases (NCDs) represent a growing challenge for developing countries, as they are among the leading causes of morbidity and mortality within the working-age population. Beyond their effects on health outcomes, NCDs are associated with reductions in labor productivity, which may influence long-term development dynamics through human capital. This study estimates the conditional labor productivity penalty associated with Non-Communicable Diseases (NCDs) among employed workers in Mexico using an observational high-dimensional framework. By addressing high-dimensional confounding under the Conditional Independence Assumption (CIA), the analysis isolates the relationship between NCD status and hourly labor earnings. Using microdata from the Mexican National Survey of Household Income and Expenditure, labor productivity is defined as the ratio of labor income to hours worked. To obtain robust causal estimates, a Double Machine Learning (DML) framework is implemented, integrating regularized regression and machine learning algorithms to control for high-dimensional confounding factors. These covariates include socioeconomic, demographic, and health-related variables, allowing for flexible adjustment of observable heterogeneity across individuals. The empirical findings demonstrate a statistically significant conditional negative penalty of −11.44% (p = 1.03 × 10−11, 95% CI: [−14.59, −8.40]) on hourly labor earnings under the cross-fitted DML specification using XGBoost and Random Forest. To resolve an observational measurement artifact where chronic individuals report artificially inflated hourly wages due to severe hours contraction, a presenteeism adjustment factor (θ*=0.8125) was calibrated. Incorporating this presenteeism correction, the aggregate national economic burden among the employed population is estimated at approximately 496.82 billion pesos annually. The findings suggest that NCDs constitute a relevant constraint on labor productivity at the microeconomic level. The study highlights the importance of prevention-oriented health strategies, workplace health promotion, and early disease management. From a methodological perspective, the results illustrate the usefulness of Double Machine Learning for purging high-dimensional confounding in cross-sectional survey data, providing robust conditional empirical benchmarks for public health policy.

IPC Classification

Keywords