Decomposition of Latent Class Regression Model Estimates: An Application to Hunting Land Lease Prices

Prior studies in agricultural and forestry economics that employ Blinder decomposition methods assume that each of the samples compared is characterized by a single regression function. The present study is the first to relax this assumption in order to allow for the existence of latent classes represented by different regression functions hidden within each sample. More specifically, we introduce the use of a latent-class or finite-mixture model to serve as a basis for Blinder-Oaxaca decompositions. Our method is applied to hunting land lease data from Mississippi (U.S.), and our results indicate that both the east and west regions of this state are best characterized by three regression functions, resulting in a total of fifteen possible pairwise decompositions. This finding allows us to decompose differences between two different east regimes or two different west regimes, in addition to the usual east-west decompositions. Our results reveal that a traditional prior analysis of these data characterizes (fails to characterize) about 35 percent (65 percent) of the sample. Our latent- class approach to decompositions can be applied to any circumstance where decompositions are needed and where model specification concerns loom large.