Zhou el al. • NEST-SITE SELECTION AND NESTING SUCCESS OF THRUSHES 495 
TABLE 2. Spearman’s correlation matrix for variables included in models for nest-site selection of Grey-backed 
Thrushes in Dagang Forestry Farm, China, in 2008. 
Distance 
to path 
Canopy 
cover 
Ground 
Heighi of ground 
cover 
Density of 
shrubs 
Basal area of 
small trees 
Basal area of 
large trees 
Distance to edge 
0.090 
0.312** 
-0.241** 
-0.013 
-0.017 
-0.279** 
0.470** 
Distance to path 
1.000 
0.136 
-0.079 
0.010 
-0.033 
-0.008 
0.245** 
Canopy cover 
1.000 
-0.327** 
-0.147 
0.097 
0.163* 
0.428** 
Ground cover 
1.000 
0.363** 
-0.332** 
-0.252** 
-0.246** 
Height of ground cover 
1.000 
-0.237** 
-0.323** 
-0.100 
Density' of shrubs 
1.000 
0.292** 
-0.064 
Basal area of small trees 
1.000 
-0.316** 
* Correlation significant at the 0.05 level (2-tailed). 
’* Correlation significant at the 0.01 level (2-uulcdl. 
cover (Bibby et al. 2000), ground cover, and 
height of ground cover. Species, height, and DBH 
of trees (DBH > 3.0 cm), and number of shrub 
main stems were measured, and transformed to 
the remaining nest-patch variables, including 
density of shrubs, basal area of small trees (3 cm 
s DBH < 8 cm), and large trees (DBH > 8 cm). 
Basal area was calculated with the equation: 
Basal area= Y!!- , n(DBIf,/2) 2 . All variables se¬ 
lected were based on studies of nest-site selection 
for other species of Turdidae (Table I). 
We used a random numbers table to select one 
random plot by pacing for comparison against the 
nest-patch data from the cardinal directions of east, 
south, west, and north, which were —50 m from the 
nests (Martin and Roper 1988). The random plots 
had the same shape and area as ncst-site plots. We 
measured the same nest-patch characteristics in the 
random plots as in the nest-site plots. 
Data Analyses .—We estimated daily survival 
rate (DSR) of Grey-backed Thrushes using 
Program MARK Version 6.1 (While and Bum- 
ham 1999, Cooch and White 2010). We evaluated 
the variation of DSR across the entire breeding 
season by building a trend model, reflecting the 
relationship between DSR and the increasing days 
of the breeding season. We assumed a complete 
nestling cycle of 29 days including four periods: 
laying (5 days), incubation (12 days), brooding 
(days 1-6 alter hatching), and late nestling (days 
7-12 after hatching). We estimated the DSR of 
each nest age. and compared them among 
different nestling periods. 
We used original data for the analyses of nest- 
site selection. We compared habitat characteris¬ 
tics of nest sites and random plots by binomial 
logistic regression with used or unused (1 or 0) as 
the categorical dependent variable. Continuous 
explanatory variables were distance to edge. 
distance to path, canopy cover, ground cover, 
height of ground cover, density of shrubs, basal 
area of small trees, and basal area of large trees. 
Correlations among variables could compromise 
the results of multiple regressions. However, 
Spearman’s correlation matrix for variables in¬ 
cluded in ncst-site selection models did not 
indicate strong correlation (r 4 > 0.6 and P < 
0.05: Hosmcr and Lemeshow 1989) between any 
two explanatory variables (Table 2). Thus, we did 
not analyze the effects of interactions among 
variables on ncst-site selection. 
The best subset of models was selected from all 
possible combinations using Akaike’s Information 
Criterion (AIC) to evaluate the relative effect of 
different habitat variables on nest-site selection. The 
AIC, (second-order Akaike's Information Criterion 
for small sample sizes), AA1C ( . (the difference in 
AIC,. between each candidate model and the model 
with the lowest AIC,.), and Akaike weights (W,) 
were used to rank the models. We also conducted 
goodness-of-fit tests on all models using a log- 
likelihood ratio x 2 statistic to assess their closeness 
of fit. Relative importance of each variable was 
assessed by sum of model weights containing the 
variable (Burnham and Anderson 2002). 
We incorporated six nest-location and eight 
nest-patch habitat variables between successful 
and depredated nests to evaluate the effects of 
nest-site habitat characteristics on nesting success. 
We screened variables by significance tests 
(Independent-sample r-tests for normally distrib¬ 
uted data and Mann-Whitney U-tests for abnor¬ 
mally distributed data) to increase the statistical 
power under the circumstance of relative small 
sample siz.e, and included variables when they had 
significant differences at the 0.25 level but did not 
correlate with each other strongly (r s > 0.6 and 
P < 0.05). Models comprised of intercept and 
