A Heuristic Bayesian Regression Approach for Causal Explanatory Study: Exemplified by an IS Impact Study

Abstract

Causal explanatory study is a very important research method in empirical research whereof research models are frequently validated by multiple linear regressions (MLR) with significant factors sought. An alternative to MLR is Bayesian regressions where statistical inferences are made with samples drawn from posterior distributions. Efficient simulation algorithms of the Markov chain Monte Carlo type have made Bayesian regressions practical. We propose a heuristic method based on the outputs of MLR to construct informative priors for Bayesian regressions. Data collected from two empirical studies of information systems (IS) impact on performance is used to demonstrate the proposed method. Deviance information criterion shows that this heuristic procedure significantly improves a Bayesian modeling with uninformative priors. When credible intervals are used to locate significant factors, it is found that the heuristic Bayesian approach, capable of finding delicate drivers, can help design better diagnostics for IS problems.