Deviance Residual Moran's I Test and Its Application to Spatial Clusters of Small Manufacturing Firms in Japan

Abstract

It is proposed that when data exhibit local clusters, a logit local association model coupled with deviance residual Moran's I can be an alternative to the global Poisson autoregressive model because the former can explicitly reveal local clusters and remove residual clustering. Because small firms in Japan traditionally exhibit local clusters, they are a good illustration. In this article, the authors introduce the deviance residual Moran's I to capture local cluster tendencies in a set of logit models and then evaluate their performance by simulation and case study. Results show that I_DR can effectively serve as a global measure of a clustering tendency for logit models and can complement other autoregressive logistic regressions for local cluster modeling when a significant I_DR is contributed by local clusters. In addition, ecological covariates identified in the previous literature were sufficient to account for the spatial clustering of small firms in 1990 but not in 2000.