How dense is too dense? Stocking density, survival, and welfare impairments in Vietnamese shrimp aquaculture

Abstract Stocking density is a key intensification choice in shrimp aquaculture, with implications for productivity and animal welfare. On commercial farms, density is chosen alongside inputs such as aeration and monitoring, and within distinct production systems, making inference from observational data difficult. In this study, we analyse a survey of 210 shrimp farms in Viet Nam covering five systems from extensive to super-intensive production. Welfare is measured with (i) a composite welfare-impairment severity index averaging five observable indicators scored 1–5 and (ii) an ordinal feeding-response severity measure. We assess the index using internal-consistency, correlation-matrix, principal-component, leave-one-out, and component-specific sensitivity checks. The five indicators show acceptable internal consistency ((Cronbach’s alpha = 0.769; standardised alpha = 0.793), all pairwise component correlations are positive, and the first principal component closely tracks the equal-weight mean. Stocking densities vary substantially across systems. In bivariate analyses, higher densities coincide with both higher survival and greater impairment severity. We estimate fractional logit models for survival with system fixed effects and flexible density terms, supplemented by log-density and restricted cubic spline checks, ordinary least squares (OLS) models for the composite index, and an ordered logit model for feeding severity. Density effects are imprecise: log-density specifications are similar to the baseline results, while the survival spline suggests some curvature but no robust overall density gradient. Aeration intensity remains a positive correlate of survival. Trimming extreme densities and estimating quantile regressions also show no systematic survival-density relationship. Alternative weighting schemes and component-specific welfare models do not uncover a density-severity gradient. Species-dominant subsamples suggest diminishing returns, with positive linear density terms and negative quadratic terms, but these estimates should be interpreted cautiously. Overall, the findings indicate that density is better understood as part of a management bundle than as a stand-alone welfare ‘dose’. In these observational data and within this sample, universal density thresholds are difficult to justify unless they are conditioned on production system and complementary inputs.