Modelling the prevalence of stunting toddlers using spatial autoregressive with instrument variable and S-estimator

Abstract

Stunting is caused by multidimensional factors and not only caused by chronic malnutrition. Paying attention to toddler nutrition and other socio-economic variables are solutions to reduce stunting prevalence. This study uses seven explanatory variables which are indicated to have an effect on increasing or decreasing stunting prevalence. The unit analysis in this study is all districts/cities in Java island. The spatial autoregressive model (SAR) was considered. The presence of outliers can cause inaccurate parameter estimation results. Removing outliers in spatial data can change the composition of spatial effects on the data. We use the instrument variables with S-estimator to overcome the presence of these outliers. The R shiny program was developed to estimate model parameters. The results showed that the underweight variable, expenditure variable, education variable, and the variable of household waste management had a significant effect on the prevalence of stunting in Java. The results of this study also found that three other variables, namely defecation behavior, the level of difficulty in accessing the public health center, and access to safe drinking water had no significant effect on the prevalence of stunting in Java. The result of the model evaluation show that instrument variable with S-estimator had a lower residual standard error and higher R-squared than the instrument variable without S-estimator.