Micronutrient deficiencies impair the health and productivity of over 2 billion
people worldwide, yet reliable estimates of these “hidden” deficiencies are
scarce. Rather than measuring micronutrient status directly, we often rely
on food recall data and food composition tables (FCTs) to calculate vitamin
or mineral intake for a given population. FCTs report sample-specific mean
nutrient content by food, ignoring that nutrient content is a distribution and
shifts over space and time. For this reason micronutrient intake estimated via
FCTs will tend to under-estimate deficiency prevalence, and will fail to detect
key vulnerable populations. In rural Uganda, crops sampled at market are
far lower in nutrient status than newly harvested crops sampled from homes;
children reliant on those market-purchased crops are differentially vulnerable
to zinc deficiency. This paper models crop zinc content for farms across
Uganda using a unique dataset containing crop nutrient content and a larger,
nationally representative panel dataset. It explores the spatial, householdspecific,
and time-varying factors that drive selection into the staple market,
and therefore influence crop zinc content at market in any given season. The
low zinc content of market crops is explained primarily by regional selection
into market, but zinc content at market also shifts over time. Temperature
variation, in particular, drives selection to market and shapes the nutrient
content of market crops.