Small Area Estimation with Winsorization Method for Poverty Alleviation at a Sub-District Level

Abstract

Poverty is one of the serious problems faced in many municipalities in Indonesia. Indicating poverty maps and finding the most influential factor causing poverty at a sub-district (kecamatan) level in these areas are crucial. The data on poverty used for developing these maps or models are usually based on Susenas data. Meanwhile Susenas data has some limitations when we want to generate information at a sub-district (kecamatan) level. The size of the sample would be very small or there might not be any information at all. Hence small area estimation (SAE) can be used to overcome this problem. Common SAE is typically based on a linear mixed model (LMM) assumptions and the effect of outliers are ignored. When the relationship between the interest and the auxiliary variables is not linear in the original scale, than the SAE based on LMM could be inefficient. Another problem faced in developing SAE model is the effect of extreme outliers. Kurnia et all (2013) purposed a SAE winsorization technique in linear mixed model to overcome this problems and based on simulation studies the results was a satisfactory. Hence, this paper will focus on the application of this technique in poverty data to develop a better poverty map and model in order to find the most influential factor causing poverty at a sub-district (kecamatan) level. Results show that the purposed model using EBLUP along with winsorization method can improve the predictive ability on non-sampled area and sampled area. Therefore we can say that Winsorization technique can be an alternative method to overcome extreme outliers in SAE.