272 
THE WILSON JOURNAL OF ORNITHOLOGY • Vol. 123, No. 2, June 2011 
(A) 
(B) 
1 
r r | 1 r 
1 ' r 1 
(C) 
P)j 
1 
, 1 Li 
T 'T- 
1 
O 
c 
o 
(E) 
i2 o 
a> 
6 
u n 
o 3 
u 
< 
,1 . 
[ ^rrr 
0 2 4 6 8 10 
Time lag (yrs) 
FIG. 3. Autocorrelation function plots (1993-2004) for: 
(A) Canada Goose nest success, (B) Canada Goose nest 
density, (C) median hatch date, (D) relative abundance of 
collared lemmings, and (E) proportion of occupied arctic 
fox dens near Cape Churchill, Manitoba. Lag values are in 
years (0 = current year, thus autocorrelation = 1). Dashed 
lines represent a = 0.05 significance level. 
surveyed in each year ranged from four to 17 
resulting in substantial variation in width of the 
estimated 95% Cl for the proportion of occupied 
dens (Fig. 2E). The autocorrelation function 
indicated no cycling between 1993 and 2004 
(Fig. 3E). There was no significant association 
between the proportion of occupied fox dens and 
L P (F..io = 0.28, P = 0.61) or LT (F uw =1.18, 
P = 0.30) and no significant association between 
occupied arctic fox dens and nest density (rho = 
0.29, P = 0.35). Our data suggested that arctic fox 
den occupancy was influenced by fox removal 
efforts. There was a lower (F uo = 12.5, P < 
0.005) proportion of fox dens occupied in years 
with (1994-1997) than in years without fox 
removal (1998-2004). 
Model Summary .—We considered 17 models in 
our analysis (Table 2) and all models contained an 
intercept term (p„). The model including nest 
density and fox den occupancy received substan¬ 
tial support as the best among those evaluated 
(Table 2). None of the other models was <2 AIC r 
TABLE 2. Regression analysis for nest success of 
Eastern Prairie Population Canada Geese in northern 
Manitoba. Canada. Models were ranked using Akaikc's 
Information Criterion corrected for small sample sue 
(AIC,) and compared using differences in A1C, (A,I. and 
Akaike weights (w,). k = number of model parameters. All 
models included an intercept term. Variable definitions are 
in Table 1. 
Model 
AIC.. 
A 
w, 
1 
D + FD 
-2.27 
0.00 
0.61 
4 
D + T 
0.28 
2.55 
0.17 
4 
HATCH + T 
2.95 
5.22 
0.04 
4 
FD 
3.79 
6.06 
0.03 
3 
D 
3.79 
6.06 
0.03 
3 
T 
3.98 
6.25 
0.03 
3 
HATCH + FD 
4.05 
6.33 
0.03 
4 
D + LT 
4.56 
6.83 
0.02 
4 
D + LP 
5.34 
7.61 
0.01 
4 
FD+ LT 
5.55 
7.83 
0.01 
4 
Intercept only 
5.93 
8.20 
0.01 
2 
LT 
6.91 
9.18 
0.01 
3 
HATCH 
7.26 
9.53 
0.01 
3 
HATCH + LT 
7.77 
10.05 
0.00 
4 
LP 
7.84 
10.11 
0.00 
3 
HATCH + LP 
9.25 
11.52 
0.00 
4 
units from the top model. The top model received 
0.61 of the Akaike weight (tv*), over three limes 
more support than the next-best-supported model. 
The top model had a pseudo-/?* of 0.64. The lop 
six models, which included the top model, a 
model with the fox trapping effect and nest 
density, a model with median hatch date and the 
fox trapping effect, and the single-factor models 
of nest density, lox den occupancy, and the fo* 
trapping effect had a cumulative AIC, weight of 
0.91 (Table 2). Coefficient estimates from the top 
model indicated a positive relationship between 
nest density and nest success (p D = 0.05:95$ Cl: 
0.01. 0.10); an increase of one goose nest/10fiha. 
on average, increased nest success by 0.05. The 
proportion of occupied fox dens was negatively 
associated with nest success (Pro = -0.64: 95$ 
Cl: - 1.14, —0.15). An increase of 0.10 in the 
proportion of occupied dens (~l additional 
occupied den) decreased nest success by 0.06 
Fox trapping, based on the second best-supporteu 
model, had a positive effect on nest success. Nest 
success in years when fox were trapped increased 
by 0.28 (95% Cl: 0.04. 0.56). FD was included in 
the best-supported model (A f = 0) or top model 
set (A, < 2) in 87% (n = 173) of iterations in out 
post-hoc simulation. This model was usually the 
same as the top model from the original analyst 
