Urbanization Effect on Depression Prevalence via Double Machine Learning
Urbanization Effect on Depression Prevalence via Double Machine Learning
Embedded Files
Investigator: Dr. Qingyan Xiang, Vanderbilt University Medical Center
Investigator: Dr. Qingyan Xiang, Vanderbilt University Medical Center
Team: Fan Yang, Zhixia Ren
Team: Fan Yang, Zhixia Ren
Depression is a serious mental disorder that affects how you feel, think, and act. The relationship between rurality and depression prevalence is marked by conflicting evidence, with studies identifying rural life as both a risk and a protective factor. The goal of this research is to investigate the causal effect of rurality on adult depression prevalence. It helps inform decisions by care providers, policymakers, and public health officials about allocating resources for depression.
We conducted a cross-sectional ecological study using data from 2,609 U.S. counties. The exposure was county-level rurality, defined by the 2013 USDA Rural-Urban Continuum Codes (RUCC), operationalized as both a binary (non-metro vs. metro) and a continuous variable. The outcome was the 2019 county-level prevalence of adult depression from the CDC PLACES database. We employed a Double/Debiased Machine Learning (DML) framework with Random Forest learners to estimate the causal effect after adjusting for 19 structural confounders spanning socioeconomic, demographic, healthcare, and environmental domains.
In this project, I led the collection / process of geospatial data on county level regarding socio-economical elements, and used census Python package. Then spatial data analysis of depression prevalence, RUCC, and a series of related confounders are done in ArcGIS Pro, visualized in the following maps.
Our primary analysis identified an average protective causal effect: rurality reduced depression prevalence by 0.435 percentage points when comparing non-metro counties to metro counties. Also, with each level of increase in RUCC, which means the county is more rural, the depression prevalence is on average reduced by 0.103 percentage points.
Average Effects of Rurality on Adult Depression in the Primary Analysis
Model Type Average Effects 95% Confidence Interval P-value
Binary (Non-metro v. Metro) -0.435 [-0.587, -0.283] < 0.001
Continuous (Urban Index) -0.103 [-0.150, -0.057] < 0.001
We also explored the effect heterogeneity through Conditional Average Treatment Effect (CATEs). It reveals that the causal impact of rurality effect's direction reversed based on county-level context; the protective effect observed in lower-income counties became a significant risk factor in counties with median incomes above approximately $50,000. Similar reversals were observed based on higher educational attainment and a larger percentage of Black residents.
Data Source:
CDC Places Dataset: https://www.cdc.gov/places/index.html
United States Census Bureau Open Data
U.S. Department of Agriculture Economic Service Research: https://www.ers.usda.gov/data-products/rural-urban-continuum-codes
Page updated
Google Sites
Report abuse