Statistical Analysis of Maize Postharvest Loss Across Farmers, Processors, and Marketers in Ghana

Overview This registration documents the statistical analysis of maize postharvest losses across three groups of actors in the maize value chain in Ghana: farmers, processors, and marketers. The study examines the magnitude and distribution of reported postharvest losses, differences across districts and actor characteristics, and factors associated with cumulative estimated postharvest loss. Study population and data The analysis uses survey data collected from maize farmers, processors, and marketers across selected districts in Ghana. The farmer dataset initially contained 160 responses, of which 149 provided sufficiently complete and unambiguous information for construction of the primary six-stage cumulative loss outcome. The processor dataset contained 50 responses, of which 47 provided exactly five identifiable stage-specific loss values and were retained for the primary processor analysis. The marketer dataset contained 50 responses and was retained for analysis. Outcome construction For farmers, the primary outcome was cumulative estimated postharvest loss calculated as the sum of respondent-reported percentages for harvesting, drying, shelling, storage, transportation, and processing. For processors, cumulative estimated loss was calculated from drying, storage, transportation, processing, and marketing stages. Marketers reported an overall average loss percentage rather than the same stage-specific structure; therefore, marketer losses were analysed separately. Statistical analysis Descriptive statistics were calculated for the overall and stage-specific loss measures, including means, standard deviations, medians, interquartile ranges, minima, and maxima. District-level summaries were also produced. Because the farmer cumulative loss outcome was positive and right-skewed, distributional diagnostics were conducted and non-parametric methods were used for group comparisons. Wilcoxon rank-sum tests were used for binary group comparisons, while Kruskal-Wallis rank-sum tests were used for comparisons involving three or more groups. Pairwise district comparisons following the farmer Kruskal-Wallis test were adjusted using the Holm procedure. The primary multivariable farmer analysis used a generalized linear model with a Gamma distribution and log link. Predictors included district, postharvest training, farmer-group membership, sex, storage facility, and drying technique. Reference categories were specified for interpretation. Heteroskedasticity-consistent HC3 standard errors were used for statistical inference, and exponentiated coefficients were interpreted as adjusted mean ratios. Sensitivity analyses included a linear regression model using the natural logarithm of cumulative estimated loss as the outcome, leave-one-out re-estimation of the farmer Gamma model, and a reduced Gamma model containing district and storage facility. These analyses were used to assess whether the principal associations were sensitive to outcome transformation, individual observations, or model specification. Processor and marketer analyses were conducted separately because their questionnaires and outcome definitions differed from those of farmers. Descriptive statistics and non-parametric group comparisons were used to examine district and selected management characteristics. Interpretation The analyses are observational and therefore identify statistical associations rather than causal effects. In particular, the cumulative loss measures are based on respondents' reported percentages and should not be interpreted as physical mass-balance estimates of grain loss. Results are intended to identify patterns of postharvest loss and potential factors associated with those patterns across the maize value chain. The registration is intended to provide a transparent record of the statistical workflow, analytical decisions, results, tables, and supporting analysis materials.