Tree based statistical models can be used to investigate data structur
e and predict future observations. We used nonparametric and nonlinear
models to reexamine the data sets on tree growth used by Bechtold et
al. (1991) and Ruark et al. (1991). The growth data were collected by
Forest Inventory and Analysis (FIA) teams from 1962 to 1972 (4th cycle
) and 1972 to 1982 (5th cycle). We used tree based models to group obs
ervations into clusters that were specified by covariate values. Next,
we performed a permutation test on the grouped data to test for a cha
nge in tree growth rates from the 4th cycle to the 5th cycle. Our tech
niques differed from those used by Bechtold et al. (1991) and Ruark et
al. (1991). The data was not assumed to follow any parametric distrib
ution, the relation between response and covariates was not assumed to
be linear, and the test for a change in tree growth did not require a
ny parametric assumptions. The methodology presented here is general a
nd applicable to other situations where the significance of a specific
covariate is in question. Despite these relaxed constraints of analys
is, our results generally agreed with those of Bechtold et al. and Rua
rk et al.