Back to search

University of San Francisco

SPATIAL ANALYSIS OF MAIZE YIELDS AND HOUSEHOLD DIETARY DIVERSITY IN SUB-SAHARAN AFRICA: CAUSAL EVIDENCE FROM A MACHINE LEARNING INSTRUMENT

Abstract

dc:description.abstract

MAIZE YIELDS AND HOUSEHOLD DIETARY DIVERSITY IN SUB-SAHARAN AFRICA: EVIDENCE ON THE CAUSAL EFFECT FROM A MACHINE LEARNING INSTRUMENT Addi Joof, University of San Francisco, May 2026 This paper provides estimates of the effect of maize yield on household dietary diversity across seven Sub-Saharan African countries, using data from 36,474 maize-growing households in the Living Standards Measurement Study Integrated Surveys on Agriculture (LSMS-ISA) panels for the period 2008--2021. OLS estimates which include district and country-by-wave fixed effects yield a positive and significant coefficient of 0.070 (SE = 0.006, p < 0.001), implying that a 10 percent increase in maize yield raises the Household Dietary Diversity Score (HDDS) by approximately 0.007 food groups. Because self-reported yields are subject to classical measurement error and unobserved household characteristics confound the OLS estimate, we construct an instrumental variable using a Random Forest model trained on satellite-derived agro-environmental features including the Enhanced Vegetation Index, Normalized Difference Vegetation Index, Growing Degree Days, precipitation, and soil moisture obtained from Google Earth Engine at the GPS-coordinate-by-year level. Our preferred 2SLS specification, which adds region fixed effects to absorb persistent sub-national geographic variation, returns a positive point estimate of 0.217 (SE = 0.206; 95% CI: -0.19 to +0.63, p = 0.294) larger than OLS and consistent with downward attenuation bias, but imprecisely estimated and statistically indistinguishable from zero at conventional levels. We interpret the OLS estimate of 0.070 as a credible lower bound on the true effect, with the IV evidence pointing to a moderately larger but uncertain magnitude. Heterogeneity analysis reveals a declining gradient across consumption quintiles the effect is strongest among the poorest households (Q1: 0.055, p < 0.01) and diminishes with wealth (Q5: 0.015, not significant) while showing no gradient across market-integration levels. These patterns are consistent with own-consumption, rather than market income from surplus sales, as the primary pathway from yield to dietary diversity. OLS estimates are positive and significant in five of six estimable countries. The findings suggest that maize yield improvements can serve as a nutrition co-investment, particularly for subsistence-oriented smallholders, though the available data do not permit a sharp causal quantification of the effect size. Keywords: dietary diversity, maize yields, instrumental variables, Sub-Saharan Africa, LSMS-ISA, satellite remote sensing, food security JEL Codes: O13, Q12, Q18, I15, C26

Degree

thesis:*
Name thesis:degree_name
Master of Science in International and Development Economics (MSIDEC)
Level thesis:degree_level
Thesis
Discipline thesis:degree_discipline
Economics

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Joof, Addi
Contributors dc:contributor
  • Andrew Hobbs

Subjects

dc:subject × 10

Identifiers

dc:identifier.*
Repository record dc:identifier
https://repository.usfca.edu/thes/1644

Chain of custody

source
Harvested from
University of San Francisco
Base URL
repository.usfca.edu/do/oai/
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Joof, Addi. SPATIAL ANALYSIS OF MAIZE YIELDS AND HOUSEHOLD DIETARY DIVERSITY IN SUB-SAHARAN AFRICA: CAUSAL EVIDENCE FROM A MACHINE LEARNING INSTRUMENT. Thesis thesis, https://repository.usfca.edu/thes/1644