Spatial Dependence and Co-Kriging Prediction of Stunting Prevalence Using the Linear Model of Coregionalization

Stunting remains a major public health challenge in Indonesia, marked by substantial regional disparities. This study analyzed the spatial dependence structure of stunting prevalence and its determinants using the Linear Model of Coregionalization (LMC) and co-kriging, which exploit spatial cross-correlation among variables rather than autocorrelation alone. Secondary data for 38 provinces in 2024 from the Indonesian Health Profile comprised stunting prevalence and nine explanatory variables, modeled jointly through empirical semivariograms and cross-semivariograms. The selected model combined a nugget effect with a spherical structure of 600 km effective range (LMC = Nug + Sph(600 km)); the nugget was non-zero for every variable and reported explicitly. Stunting prevalence showed moderate spatial dependence (SDP = 68.78%). Low birth weight showed positive cross-spatial dependence with stunting, whereas exclusive breastfeeding and pediatric health service coverage showed negative dependence. The poverty pattern was weak and mixed in the full sample but turned clearly positive once the two provinces reporting 0% prevalence were excluded. These are co-spatial patterns among provincial aggregates, not causal effects. Co-kriging revealed distinct clustering, with East Nusa Tenggara, parts of Papua, and Sulawesi showing relatively high predicted prevalence. Leave-one-out cross-validation gave a mean error of 0.264 and an RMSE of 6.423 percentage points, close to the between-province standard deviation (7.68), with prediction standard errors of 5.67.7, indicating moderate accuracy. Excluding the 0% values left the dependence class, effective range, and cross-covariance signs unchanged. Multivariate geostatistics thus offers spatial evidence for geographically targeted, ecological-level stunting interventions.