312 
THE WILSON JOURNAL OF ORNITHOLOGY • Vo/. 124. No. 2. June 2012 
TABLE 2. Models exploring the relationship between daily survival rate (DSR) of Sprague's Pipil and distance tohahitat 
edges at Bowdoin National Wildlife Refuge. Montana using Akaike's Information Criterion, corrected for small sample sizes 
(A 1C,.). AAIC, l A1C, model / - AIC, minimum), Akaike weight (w,|. and the number of parameters (k) are included for each 
model (*S = DSR; 7 = date of nest on initiation; Age = age of nest; ORIENT = direction of nest opening). Minimum 
distances to roads and edges are the potential thresholds beyond which there is no effect of the road or edge 
Model 
AIC r 
AAIC, 
H’, 
Model likelihood 
k 
Dcvuttt 
{5(7 + Age + Clutch Size)} 
496.97 
0 
0.119 
1 
4 
488.95 
1 S(T+ Age)} 
497.47 
0,50 
0.093 
0.78 
3 
491.45 
fS<Age)| 
498.02 
1.05 
0.070 
0.59 
2 
494(0 
{5(7 + Age + Clutch Size + MIN(Fdgesn) 
498.29 
1.32 
0.062 
0.52 
5 
488.15 
{5(7 + Age + Clutch Size + MIN(Edges.300))| 
498.34 
1.36 
0.060 
0.51 
5 
488.29 
{5(7 + Age + Clutch Size + MIN(Edges.200))} 
498.36 
1.39 
0.060 
0.50 
5 
488.32 
(5(7 + Age + Clutch Size + MIN(Edges.50))| 
498.64 
1.67 
0.052 
0.43 
5 
488.60 
[S(T + Age + Clutch Size + MIN(Roads,200))} 
498.71 
1.74 
0.050 
0.42 
5 
488.67 
(S(7 f Age + Clutch Size + Conceal)} 
498,80 
1.83 
0.050 
0.40 
5 
488.76 
1 S(T + Age + Clutch Size + MIN(RoadsJOO))} 
498.82 
1.85 
0.047 
0.40 
5 
488.78 
|.S(7 + Age + Clutch Size + MIN(Road.s.50))) 
498.84 
1.87 
0.047 
0.39 
5 
488.80 
{5(7 + Age + Clutch Size + ORIENT) | 
498.87 
1.89 
0.046 
0.39 
5 
488.82 
[S{T + Age + Clutch Size + MIN(Edges.lOO))} 
498.94 
1.96 
0.045 
0.38 
5 
488.89 
{S(T + Age + Clutch Size + MIN(Roads))} 
498.95 
1.98 
0.044 
0.37 
5 
488.91 
(Sir + Age + Clutch Size + MIN(Rnads.lOO))} 
499.00 
1.99 
0.044 
0.37 
5 
488.93 
{5(7*2 4 Age)) 
499.48 
2.50 
0.034 
0.27 
4 
491.45 
{5(.)} 
{5(7 + Age + Clutch Size + SecRoad + PriRoad 
499.71 
2.73 
0.030 
0.26 
1 
497.70 
4 RailRoad + ShoreLinc 4 AgField)} 
400.00 
3.02 
0.026 
0.22 
9 
481.87 
{5(7 4 Age 4 Clutch Size 4 ORIENT A 2)} 
500.23 
3.25 
0.023 
0.20 
6 
488.17 
et al. 2002. Rotella ct al. 2004). Program MARK 
uses a generalized linear model approach to 
modeling DSR and maximum likelihood estimation 
to estimate model parameters and sampling vari¬ 
ances (White and Burnham 1999, Dinsmore el al. 
2002). Program MARK calculates Akaike’s Infor¬ 
mation Criterion corrected for small sample sizi 
(AIC,,) (Burnham and Anderson 2002), ranking U 
fit of each model in ascending order of AIC, value 
The intercept-only {|S(.)|| model (where a 
daily survival estimates arc assumed to be equa 
was included in the model set to provide 
baseline to evaluate the importance of covaria 
models. Five non-distance variables were indue 
ed in the analysis prior to evaluating the road an 
edge variables to avoid unwarranted conclusior 
about the distance variables and to avoid Simj 
son s paradox (i.e., an apparent paradox in whic 
a correlation [trend| present in different groups 
reversed when the groups are combined) (Agres 
2007). The non-distancc variables included: (I 
T -trend in DSR across the nesting season (dat 
o nest initiation). (2) age - age of the nest, n 
dutch size, (4, CONCEAL - the arithmetic mea 
of the percent cover taken from directly abov 
nest and in the four cardinal directions fo 
each nes, (Jones and Dieni 2007)> "°” s ° 
ORIF.NT - the direction the nest opening was 
facing. We used the minimum of the distances to 
roads and edges as the predictive variable, and 
included a threshold lo examine if there was an 
effect within 50, 100. 200, or 300 in. We ran the 
model separately for each distance threshold and 
compared all possible models including or 
excluding all variables. There may be an effect ot 
distance on nest survival within some threshold- 
hut we assumed there would be no effect on the in¬ 
dividual nest beyond this threshold. Thus, the 
distance variable was used in the model up to the 
threshold with all distances > threshold replaced 
with the threshold value. These models exfttninci- 
diHerein thresholds to assess effects immediate o 
edges. We did not include any models with vi> 
effects, because the sample size was smaJI wheo 
partitioned into years. We also did not examine 
models with interactions because of small sanipk 
sizes. The likelihood of detecting spurious ellec v 
increases with the number of models evaluated 
f Burnham and Anderson 2002). particularly "i'll 
small sample sizes as in this study. 
RESULTS 
The overall DSR (n = 125) was 0.95 ± W® 5 ' 
(SE) with a 95% confidence interval of 0 . 94 -t'9b 
