Predicting women's height from their socioeconomic status: A machine learning approach

Area of Science:

• Public Health
• Socioeconomic Determinants of Health
• Human Welfare Indicators

Background:

• Social determinants of health literature frequently uses socio-economic status (SES) to explain variations in women's height, a key indicator of population welfare.
• Existing research often lacks a systematic evaluation of SES predictive power and potential non-linear relationships with women's height.
• A comprehensive understanding of SES's contribution to health outcomes is crucial for advancing social determinants of health research.

Purpose of the Study:

• To systematically evaluate the predictive power of socio-economic status (SES) on women's height.
• To investigate potential non-linear relationships between SES indicators (education, occupation, material wealth) and women's height.
• To compare the predictive performance of machine learning algorithms against traditional Ordinary Least Squares (OLS) regression.

Main Methods:

• Utilized Demographic and Health Surveys (DHS) data from 1,273,644 women across 66 low- and middle-income countries (1994-2016).
• Trained seven distinct machine learning algorithms and assessed their out-of-sample predictive power.
• Compared machine learning results with Ordinary Least Squares (OLS) regression, controlling for country, community, and sampling year fixed effects.

Main Results:

• In the OLS framework, SES explained only 0.7% of the total variance in women's height (R²).
• The best-performing machine learning model, a Bayesian neural network, offered a negligible 0.3% improvement in explained variance over OLS.
• Findings suggest no significant non-linear relationships between SES and women's height, highlighting the predictive limits of SES.

Conclusions:

• Socio-economic status (SES) has limited predictive power for women's height, with machine learning models showing minimal gains over traditional regression.
• The study indicates a lack of substantial non-linear associations between SES indicators and women's height.
• Scholars should report both the average effect and variance explained by SES to enhance understanding of social determinants of health.