Wild Hog Rooting 
175 
We used PROC STEPDISC (SAS Institute 1985) to select 
remaining variables for construction of a discrimination model. The 
stepwise selection process was used with a significance level of 
variable entry into the model set at 0.15. We selected a moder- 
ate significance level in order to “choose the model that provides 
the best discrimination using the sample estimates” (SAS Institute 
1985:750). We then analyzed variables selected for inclusion in the 
discrimination model using PROC DISCRIM (SAS Institute 1985) 
to test the assumption of homogeneity of the within-group covari- 
ance matrices and to generate a classification outcome. Signifi- 
cance level for the test of homogeneity was set at 0.05. We set 
prior probabilities proportional to sample sizes. 
We used the Kappa statistic to measure the improvement of 
the classification rates over chance assignments. This statistic ranges 
from zero, which indicates no improvement in assignments, to one, 
which indicates perfect assignment (Fleiss 1973, Rexstad et al. 1988) 
Alpha level for this test was set at 0.05. 
We used PROC CANDISC (SAS Institute 1985) on the vari- 
ables selected by PROC STEPDISC to test separation of traps into 
successful and unsuccessful classifications and to obtain standard- 
ized canonical coefficients that suggest directional relationships be- 
tween the selected variables and the two classifications. In an attempt 
to identify important microhabitat variables properly, we validated 
models by analyzing data from both years individually, and then 
compared the results. For each year’s data, the discriminant func- 
tion analyses identified significant variables that were associated with 
successful and unsuccessful traps. 
RESULTS 
Mammals 
Populations of deer mice were low during both 1989 and 1990 
(Table 2). Because population sizes were so small during both years, 
the models selected by CAPTURE did not perform well (Otis et al. 
1978), and the estimates are likely biased with low precision (White 
et al. 1982). The two-way ANOVA performed on deer mouse population 
means pooled from both years revealed no significant differenes: 
(1) between population sizes on rooted and unrooted sites (F = 
0.18, P = 0.68), between the population sizes of 1989 and 1990 ( F 
= 2.99, P = 0.12), or due to interactions between site class and 
year ( F = 0.12, P = 0.74). 
Table 3 summarizes all non -Peromyscus captures. Low capture 
frequencies prevented statistical analysis of any of these data. 
