Multivariate Adaptive Inverse Gaussian Regression Spline Modeling for Estimation of Household per Capita Expenditure
Ifah Durrotun Nisa (a), Bambang Widjanarko Otok (a*), Sutikno (a)

a) Departement of Statistics, Faculty of Science and Data Analytics, Institut Teknologi Sepuluh Nopember, Surabaya, 60111, Indonesia
*dr.otok.bw[at]gmail.com

Abstract

Multivariate Adaptive Regression Spline (MARS) is used for high-dimensional data modeling. It is able to allow the additive and interactions effects among predictor variables. For continuous responses variable, sometimes, the data has highly skewness to right. An alternative method that can handle it is inverse Gaussian regression (IGR). Multivariate Adaptive Inverse Gaussian Regression Spline (MAIGRS) model is a combination of MARS and IGR. In this modeling, the estimation of basis function parameters obtained by Weighted Least Square (WLS) method. In this study, MAIGRS model is applied for prediction household per capita expenditure. Secondary data from the Pohuwato Regency National Socioeconomic Survey (SUSENAS) for 2020-2021 was used with an observation unit of 1061 households. There are nineteen predictor variables used in the modeling. The result show that demographic composition, housing conditions, and household asset ownership have a role in predicting household expenditure. Four variables that have a major role in predicting household per capita expenditure are car ownership, percentage of paid worker of household members, percentage of household members with high school education and above, and type of floor, which variable importance level more than 75 percent.

Keywords: Expenditure- IGR- MAIGRS- MARS- WLS

Topic: Mathematics and Statistics