Estimating the effect of food insecurity on depressive symptoms: a double machine learning approach with NHANES data

This study applies double machine learning (DML) to estimate the causal effect of food insecurity on depressive symptoms (PHQ-9 scores) using NHANES 2017-March 2020 pre-pandemic data.Unlike conventional regression approaches, DML can accommodate nonlinear relationships and high-dimensional confounding.Using a partially linear model, we estimated the average treatment effect (ATE) under the conditional independence assumption.The analytic sample included 6,432 U.S. adults, with food insecurity as the treatment and continuous PHQ-9 scores as the outcome.Covariates included age, sex, income-to-poverty ratio, body mass index, education, smoking, alcohol use, diabetes, and race/ethnicity.Although smoking, alcohol use, and BMI were included as potential confounders based on prior literature, their possible roles as mediators and the resulting risk of over-adjustment are acknowledged.Analyses incorporated NHANES complex survey weights.This study aims to strengthen causal evidence on the food insecurity-depression relationship and inform policies addressing food insecurity and mental health.