Wild Hog Rooting 
173 
the understory as woody vegetation >2 m in height and <7.5-cm 
dbh. Snags and odd limbs (living or dead) touching the ground 
were included in these measurements because deer mice are known 
to preferentially use snags for refuges (Wolff and Hurlbutt 1982), 
and only small woody trunks are required for escape routes. The 
shrub layer was defined as all vegetation between 0.4 and 2.0 m in 
height and the herb layer as all vascular vegetation in the 0.0-0. 4- 
m range. Our use of the line intercept method followed Hays et al. 
(1981) and consisted of laying out two 2.5-m transects in random 
directions from the center of each trap. 
During the 1990 field season several variables were not 
measured: measurements involving the shrub layer, percentages of 
individual herb species cover (except for greater star chickweed, 
Stellaria pubera, and spring beauty, Claytonia virginica), herb 
richness, percent bare soil, percent leaf litter cover, and soil resis- 
tance were omitted because analysis of the 1989 data indicated these 
variables were not significant. We chose not to use the dbh of the 
closest overstory and understory tree in 1989, but included it in the 
1990 analysis. Barry et al. (1984) found that deer mice oriented 
towards larger trees. 
Data Analysis — We performed a series of discriminant function 
analyses on microhabitat data to determine the most important vari- 
ables. To “identify” truly important microhabitat variables, we be- 
lieved it was important to test the ability of variables to classify 
the success or failure of a trap to capture a small mammal and 
to produce a directional relationship consistent with known animal 
ecology (Tacha et al. 1982). 
To begin the discriminant analysis, we pooled the data from 
the measured variables for all six sites each year. Variables were 
grouped as belonging to successful or unsuccessful live traps. We 
defined a successful trap as any live trap with one or more cap- 
tures and an unsuccessful trap as having no verified captures. All 
variables represented as percentages were arcsine transformed before 
analysis to approximate normal distributions. We performed a corre- 
lation matrix on all combined variables using PROC REG with the 
collinearity diagnostics option (SAS Institute 1985) to eliminate 
intercorrelated variables. Interrelatedness can lead to switching of 
variables in a stepwise discriminant function analysis and to diffi- 
culty in interpreting the importance of predictor variables (Green 
1979). Multicollinearity among regressors also results in unstable 
estimates and high standard errors (SAS Institute 1985). All vari- 
ables with variance inflation factors (VIFs) of greater than 2.0 were 
removed from further analysis. 
