Abstract
In recruitment modeling, the response variable is a count and its distribution is often characterized by an excess number of zeros. As a result, standard distributions of probabilities, such as Poisson, are inappropriate. A common approach in forestry consists of using two-part conditional models. These models have two distinct components aimed to predicting the occurrence and abundance of recruitment, respectively. For such data, zero-inflated models might provide a more adequate framework by combining the two components into a joint distribution of probabilities. In this article, a conditional model is compared with two different zero-inflated models, namely, a zero-inflated Poisson (ZIP) and a zero-inflated discrete Weibull (ZIdiW) model. The three models were calibrated using a data set provided by permanent sample plots located in hardwood stands. Parsimony criteria (the Akaike information criterion and the Bayesian information criterion) and diagnostic plots were used to perform the comparison. The results show that the ZIdiW model has the best fit. The flexibility of the Weibull function and the possibility of obtaining a more parsimonious model are two advantages related to the use of a ZIdiW model in recruitment modeling.