Multinomial logit estimation of a matrix growth model for tropical dry forests of eastern Bolivia

Abstract

Multinomial logistic (MNL) regression was employed to estimate the transition probabilities of a matrix growth model for dry forests of the eastern Bolivian lowlands. Probabilities of mortality, stability, and upgrowth of a size and species group were estimated as a function of tree and stand attributes influencing growth and mortality. Data for model estimation were drawn from logged and undisturbed permanent sample plots (PSPs) measured over 6 and 7 years in Chiquitania forests south of Concepción, Santa Cruz, Bolivia. The estimated transition probabilities of the MNL models are not significantly different from those derived from PSP data by the conventional approach of employing simple mean proportions of observed movements per guild and size class. MNL estimation is advantageous in that it generates a smoother distribution of transition probabilities across size classes, correcting for variance in the data and model estimation errors imposed by limited samples. Moreover, the MNL approach allows deterministic, stochastic, and dynamic prediction of forest evolution, while preserving the simple linear form of matrix models that facilitates their integration into economic optimization studies.