Elk
Train speed and relative abundance explained the variability in elk strikes (Log
likelihood = −113.0 with 2 df, v
2 = 26.3, p < 0.001). Elk strikes increased on
average e
0.07 = 1.07 per segment, with each 1.0 mph increase in maximum posted
train speed when elk relative abundance was held constant (Table 9.3). Likewise
when the SPEED max was held constant strikes increased e
0.03 = 1.03 for each one
unit increase in elk RA rail corridor . The variable SPEED max averaged 37.5 ± 1.85
mph (60.3 kmph) and ranged from 20 to 50 mph.
Deer
The variables deer relative abundance (RA rail corridor ), SPEED max and ROW mean
best explained the variability in deer strikes using the model selection process (Log
likelihood = −78.46 with 5 df, v
2 = 48.89, p < 0.001). The parameter coefficients
from a maximum likelihood fit indicated that e
0.06 = 1.06 additional deer strikes
were observed on average with an 1.0 mph increase in the posted train speed limit
when deer abundance and ROW width were held constant. Likewise, e
0.009 = 1.009
additional deer strikes were observed for each 1.0 m increase in ROW width when
speed and deer abundance were held constant. The variable ROW mean was on
average 79.78 ± 7.68 m across all segments.
Bears
A single predictor model, including the variable BARRIER c , best explained bear
strikes (Log likelihood = −57.233 on 4 df, v
2 = 14.686, p < 0.001). The parameter
estimated for BARRIER c (b barrier = e
0.458 = 1.58) revealed that for each additional
barrier feature per segment bear strikes increased on average 1.58 (95% CI 1.13 to
2.20) (Table 9.3). A second single predictor model has almost equal explanatory
power. This model included the variable BRIDGE c . However, BRIDGE c and
BARRIER c were not correlated (r = 0.23, t = 1.1968, df = 26, p = 0.24). The
variable BRIDGE c was positively correlated to bear strikes (r = 0.44, p < 0.02).
For bears, neither estimate of relative abundance was significant; therefore, the
final model did not include either variable describing bears relative abundance. This
changes the modeling results from assessing risk to incidence rate. For this reason,
the model selection procedure was repeated with the variable RA rail bed log transformed and held in the model as an offset to assess relative risk (Zuur et al. 2009).
This model selection procedure resulted in a single predictor model that included
SPEED max , indicating that as posted train speed increased, so did the “risk” of bear
strikes (b = e
0.053 = 1.054). However, train speed explained only 17% of the
deviance, compared to 82% explained by the BARRIER c model (Table 9.3).
9 Relative Risk and Variables Associated with Bear and Ungulate …
149
Train speed and relative abundance explained the variability in elk strikes (Log
likelihood = −113.0 with 2 df, v
2 = 26.3, p < 0.001). Elk strikes increased on
average e
0.07 = 1.07 per segment, with each 1.0 mph increase in maximum posted
train speed when elk relative abundance was held constant (Table 9.3). Likewise
when the SPEED max was held constant strikes increased e
0.03 = 1.03 for each one
unit increase in elk RA rail corridor . The variable SPEED max averaged 37.5 ± 1.85
mph (60.3 kmph) and ranged from 20 to 50 mph.
Deer
The variables deer relative abundance (RA rail corridor ), SPEED max and ROW mean
best explained the variability in deer strikes using the model selection process (Log
likelihood = −78.46 with 5 df, v
2 = 48.89, p < 0.001). The parameter coefficients
from a maximum likelihood fit indicated that e
0.06 = 1.06 additional deer strikes
were observed on average with an 1.0 mph increase in the posted train speed limit
when deer abundance and ROW width were held constant. Likewise, e
0.009 = 1.009
additional deer strikes were observed for each 1.0 m increase in ROW width when
speed and deer abundance were held constant. The variable ROW mean was on
average 79.78 ± 7.68 m across all segments.
Bears
A single predictor model, including the variable BARRIER c , best explained bear
strikes (Log likelihood = −57.233 on 4 df, v
2 = 14.686, p < 0.001). The parameter
estimated for BARRIER c (b barrier = e
0.458 = 1.58) revealed that for each additional
barrier feature per segment bear strikes increased on average 1.58 (95% CI 1.13 to
2.20) (Table 9.3). A second single predictor model has almost equal explanatory
power. This model included the variable BRIDGE c . However, BRIDGE c and
BARRIER c were not correlated (r = 0.23, t = 1.1968, df = 26, p = 0.24). The
variable BRIDGE c was positively correlated to bear strikes (r = 0.44, p < 0.02).
For bears, neither estimate of relative abundance was significant; therefore, the
final model did not include either variable describing bears relative abundance. This
changes the modeling results from assessing risk to incidence rate. For this reason,
the model selection procedure was repeated with the variable RA rail bed log transformed and held in the model as an offset to assess relative risk (Zuur et al. 2009).
This model selection procedure resulted in a single predictor model that included
SPEED max , indicating that as posted train speed increased, so did the “risk” of bear
strikes (b = e
0.053 = 1.054). However, train speed explained only 17% of the
deviance, compared to 82% explained by the BARRIER c model (Table 9.3).
9 Relative Risk and Variables Associated with Bear and Ungulate …
149