Rice farming’s link to young women’s joblessness in Africa is fragile, study finds

A new study finds that the share of a country’s farmland devoted to rice is negatively associated with the rate of young women who are not in employment, education, or training (NEET) across 41 African countries, but the authors themselves warn that the association is fragile, specification-dependent, and cannot be read as evidence that growing more rice reduces female youth joblessness. The paper, published in the journal Agricultural and Food Economics, uses an annual unbalanced panel covering 2009 to 2023 and dynamic panel econometrics to probe one of the most persistent puzzles in African labour markets: why so many young women remain disconnected from work, school, and training even as food systems transform around them.

The research, conducted by Babou Sogue and Aminou Arouna, is unusual in how carefully it bounds its own claims. Rather than presenting a headline causal effect, the authors deliver what they call a sensitivity-tested description of conditional dynamic associations. The female youth NEET rate, they show, is highly persistent from one year to the next. In their preferred institutional specification, an alternative specification, and a sensitivity model that controls for male NEET rates, rice-area intensity carries a negative coefficient. Rice import dependence also shows a negative coefficient in those same models. Yet when the researchers impose deeper lags and stricter limits on the number of statistical instruments, the estimates lose precision or change sign altogether.

The NEET indicator itself deserves scrutiny. It measures the share of young women not in employment, education, or training, but the authors stress that it does not fully capture labour-market exclusion. Informal work, unpaid family labour, subsistence farming, and care activities may be imperfectly recorded in the harmonized international statistics the study draws on, which come from the International Labour Organization’s ILOSTAT database. Even though the series is harmonized, cross-country comparability can still be affected by survey coverage, modelled estimates, differing definitions of education, and changes in national statistical systems. The outcome is therefore best understood as a macroeconomic indicator of disconnection from recorded employment, education, or training, not a complete measure of economic activity or empowerment.

The main variable on the explanatory side is equally narrow. Rice-area intensity is defined as the share of harvested rice area in a country’s total agricultural area. It measures the allocation of land to rice, not productivity, not production growth, not the intensity of any public policy, and not the share of workers actually employed in rice. This distinction matters because the theoretical channels the authors discuss operate largely through downstream value-chain activities such as milling, parboiling, aggregation, logistics, retailing, and food services. The dataset does not directly observe any of these segments, so the idea that rice expansion creates accessible jobs for young women remains a plausible interpretation to be tested later with microeconomic or value-chain-specific data, not a mechanism identified by the macroeconomic panel itself.

Methodologically, the study relies on two-step System Generalized Method of Moments estimation with collapsed internal instruments, Windmeijer’s finite-sample correction, and forward orthogonal deviations. This family of estimators, built on the work of Arellano and Bond and extended by Arellano and Bover and Blundell and Bond, is designed for dynamic panels where the lagged dependent variable and time-invariant country heterogeneity complicate ordinary fixed-effects regression. The preferred institutional specification uses 33 instruments for 41 country groups, keeping the instrument count below the number of countries, a standard safeguard against instrument proliferation. The lagged dependent variable is treated as predetermined, while rice specialization, productivity, expenditure, and selected interactions are treated as potentially endogenous.

The diagnostic picture is mixed in ways the authors report transparently. Arellano-Bond tests show first-order autocorrelation but no second-order autocorrelation in all five columns, and the robust Hansen tests do not reject, while instrument counts stay below the number of countries. However, the Sargan test rejects in the two most demanding stress tests, and the persistence coefficient is close to or above unity in several specifications, exceeding unity in the two harshest instrument reductions. That near-unit persistence makes long-run multipliers highly uncertain or inadmissible, so exploratory long-run transformations are reported only in supplementary material with wide confidence intervals. The authors emphasize that failure to reject an overidentification test does not prove instrument validity, and Difference-in-Hansen diagnostics are used to detect manifest failures rather than to establish exogenous identification.

Substantively, the core estimate is modest in its short-run units. In the three closely related specifications, a one-percentage-point increase in the share of agricultural land devoted to rice is associated with a reduction of roughly 0.086 to 0.091 percentage points in the female youth NEET rate, conditional on the included variables and centered moderators, and the coefficient is statistically significant at the five percent level in those models. Rice import dependence also carries a negative coefficient in those columns, but loses significance and changes sign in the strictest stress test. The authors attribute no causal meaning to either pattern, noting that import dependence may reflect market size, urban demand, distribution activity, structural production deficits, or unobserved macroeconomic conditions rather than any effect of imports themselves.

Institutional heterogeneity adds another layer of caution. The study prioritizes a hierarchical specification in which rule of law and its interaction with rice-area intensity enter jointly, with regulatory quality retained only as an alternative sensitivity check. In the preferred specification, the rule-of-law interaction is positive but statistically imprecise, and the authors conclude that the evidence does not support systematic heterogeneity by rule of law. The political-stability interaction, by contrast, is positive in the main models, meaning the estimated short-run slope of rice-area intensity becomes less negative at higher levels of political stability and may cross zero. The authors are careful to note that this model-specific pattern does not imply political stability harms young women, since the interaction may capture correlated differences in formalization, sectoral composition, urbanization, conflict exposure, or measurement.

Descriptive evidence underscores how uneven the landscape is. North Africa shows the highest median female youth NEET rate among African rice economies, while East Africa has comparatively lower medians but marked dispersion, and West and Southern Africa display wide interquartile ranges suggesting substantial variation within regions. Because the panel is unbalanced, year-to-year movements in sample means may reflect both within-country changes and shifts in the set of observed countries, and the post-2020 divergence between rice-area intensity and the NEET rate is explicitly descriptive rather than causal. Country inclusion depends on joint data availability rather than a common threshold of rice dependence, so the sample is not a probability sample, and missing observations may be related to statistical capacity, institutions, or labour-market conditions.

The study’s practical upshot is a monitoring agenda rather than a policy prescription. The authors argue that evaluations of rice-sector transformation should report gender-disaggregated data on employment, education and training transitions, job quality, unpaid work, skills, mobility, and participation in production, processing, logistics, trade, and service activities, rather than inferring inclusion from production growth alone. Future research, they suggest, should combine comparable multi-crop panels with microeconomic or firm-level value-chain data, more clearly identified policy shocks or external instruments, and direct measures of women’s tasks, income, time use, and control over benefits. For now, the bounded conclusion stands: the observed association between rice area and young women’s NEET rates is conditional, specification-sensitive, and unsuitable for a causal policy claim, however tempting the headline numbers may look.

Subject of Research: Dynamic panel analysis of associations between rice-area specialization, import dependence, institutional conditions, and female youth NEET rates in 41 African countries from 2009 to 2023.

Article Title: Rice-area specialization and young women’s NEET rates in African countries: dynamic persistence, institutional heterogeneity, and import dependence

Article References: Sogue, B., & Arouna, A. (2026). Rice-area specialization and young women’s NEET rates in African countries: dynamic persistence, institutional heterogeneity, and import dependence. Agricultural and Food Economics, 14(1), Article 72. https://doi.org/10.1186/s40100-026-00512-8

Image Credits: AI Generated

DOI: 10.1186/s40100-026-00512-8

Keywords: rice, NEET, young women, Africa, System GMM, dynamic panel, import dependence, institutions, value chains, labour market, observational study, econometrics