Higher Order Partial Least Squares Path Modeling Using Binary Data: An Application on Multidimensional Poverty and Social Protection in East Java Province
Rudi Salam *
Department of Statistics, IPB University, Bogor, Indonesia and Statistics Diploma Program, STIS Statistics Polytechnic, Jakarta, Indonesia.
I. Made Sumertajaya
Department of Statistics, IPB University, Bogor, Indonesia.
Hari Wijayanto
Department of Statistics, IPB University, Bogor, Indonesia.
Anang Kurnia
Department of Statistics, IPB University, Bogor, Indonesia.
Timbang Sirait
Statistics Diploma Program, STIS Statistics Polytechnic, Jakarta, Indonesia.
*Author to whom correspondence should be addressed.
Abstract
The standard partial least squares path modeling (PLS-PM) estimation process assumes the observed data as continuous variables. With slight modifications, this estimation algorithm can be used for data on a binary scale and even for more complex models such as higher order constructs. This study aims to determine by simulation and application of real data the performance of the higher-order construct modeling approach, which of the repeated indicator and two-stage approaches provides better results. From simulation study it was found that the repeated indicator approach with binary data (BinPLS) was better than the two-stage approach. Empirical results also show that the BinPLS measurement model with the repeated indicator approach is better than standard PLS. Evaluation of the structural model also shows that BinPLS with a repeated indicator approach is the best because it produces path coefficients and the power to explain multidimensional poverty and social protection models that are better than BinPLS with a two-stage approach and standard PLS.
Keywords: Binary data, higher order PLS-PM, multidimensional poverty