Ary Santoso, Anang Kurnia*, Bagus Sartono and Siti Muchlisoh
Departemen of Statistics Bogor Agricultural University,
Jalan Pajajaran, Kampus IPB Baranangsiang, Bogor 1615, Indonesia.
*corresponding author: Anang Kurnia
Corresponding author name: Anang Kurnia
Meranti wing 22 level 4 IPB Dramaga, Bogor, West Java
Statistics Indonesia (a.k.a. BPS) utilizes the generalized least squares (GLS) approach to determine sample weights in their survey data, including National Socio-Economics Survey (Susenas). The weakness of this approach is the possibility of the appearance of weird weights such as negative values and extremely large values. This present paper discusses evaluating techniques of GLS to weight data. Two weigthing methods approacheswhich are examined are weighting by the sampling design and weighting by the use of generalized least squares.A simulation studywas conducted to explore the characteristics of weight design and GLS as well asto solve the problem of negative weights of GLS. The study revealed that the negative values of GLS can be solved by adjustment matrix of auxiliary variable (X).
Keywords: Sample Weights, Weighting by the sampling design, Weighting by Generalized Least Square (GLS)
Read More Full Article