Isolating The Key Variables For Regression Models In Enterprise Software Acquisition Decisions: A Blocking Technique

Abstract

Often researchers in the field of information systems face problems related to the variable selection for model building; as well as difficulties associated to their data (small sample and/or non normality). The goal of this article is to present an original statistical blocking-technique based on relative variability for screening of variables in multivariate regression models. We applied the blocking-technique and a nonparametric bootstrapping method to the data collected on the USA-South border for a research concerning enterprise software (ES) acquisition contracts. Three mutually exclusive blocks of relative variability for the response variables were formed and their corresponding regression models were built and explained. A conclusion was drawn about the decreasing tendency on the adjusted coefficient of determination (R²_adj) magnitudes when the blocks change from low (L) to high (H) condition of relative variability. The obtained models (via stepwise regression) exhibited significant p-values (0.0001).