This simulation study presents Bayes factor testing for β coefficients in the network autocorrelation model using Savage-Dickey ratio and BIC approximation methods. It proposes Bayes factors for two-sided and multiple hypotheses testing procedures
In network autocorrelation model, the classical hypothesis testing procedures for independent variables’ effect on the outcome variable can only be used to falsify a precise null hypothesis of no effect. Classical methods are incompetent for both quantifying evidence for the null and testing multiple hypotheses simultaneously. In order to deal with these limitations, this study presents Bayes factor testing for β coefficients in the network autocorrelation model using Savage-Dickey ratio and BIC approximation methods. We propose Bayes factors for two-sided and multiple hypotheses testing procedures. Simulation results suggest that Bayes factor for the latter shows higher performance and it is the one we recommend. Then, we illustrate the practical use of the proposed Bayes factors with two real data examples and compare the results to those coming from classical tests using p values. Finally, R code used in this study and for computing the proposed Bayes factors is provided.