Carpenter et al. • CERULEAN WARBLER MICROHABITAT ASSOCIATIONS 
209 
Wildlife Division (Mirarchi et al. 2004), and 
summed the number of species designated mod¬ 
erate or higher detected at each count. We pooled 
data from both seasons to increase sample size 
because multivariate analysis of variance (MAN- 
OVA) tests suggested there was no interaction or 
the trend was consistent between year and type of 
survey (used vs. unused) lor habitat variables 
(Philai's Trace = 0.11, F = 0.51, df = 18 and 78. 
P = 0.95) and for avian community variables 
(e.g.. abundance, richness and diversity index, 
Philai's Trace = 0.08. F = 2.10, df = 4 and 93, 
P = 0.09). Independent sample /-tests were used 
to test for differences between used and unused 
plots in bird species richness, abundance, diver¬ 
sity, and conservation concern values, as well as 
abundance of Brown-headed Cowbirds ( Molo - 
thrus ater ) and three common nest predators: 
American Crow ( Corvus brachyrhynchos). Blue 
Jay (Cyanocitta cristata ), and Red-bellied Wood¬ 
pecker (Melaneqtes carol inns). 
Analysis of Avian Community Associations.— 
Detrended correspondence analysis (DCA) was 
used with bird abundance to help explain the 
structure of the sampled avian community (Hill 
and Gauch 1980). We chose nonlinear scaling and 
detrended the axes using second-order polynomi¬ 
als following Jongman et al. (1995) to avoid the 
limitations inherent in DCA, including distorted 
gradient structure and a lack of robustness 
(Minchin 1987). Rare species were down-weight¬ 
ed and ordination scores were obtained with biplot 
scaling focused on inter-species distances using 
CANOCO 4.54 (ter Braak and Smilauer 2006). 
We referenced Birds of North America accounts 
(Poole 2005) for general habitat preferences of 
each species to assist in interpretation of DCA 
axis, and excluded species detected at less than 
three locations to reduce the effect of transient or 
accidental species (Wakeley et al. 2007). 
Analyses of Microhabitat Characteristics. —We 
attempted to correct any variables w'ith non-normal 
distributions using Shapiro-Wilk tests and square 
root or logarithmic transformations. Independent 
sample /-tests were used to compare mean mea¬ 
surements of vegetation from used habitat plots with 
unused plots, and two-sample Mann-Whitney 17- 
tests for variables that violated normality or equal 
variance assumptions. Canopy structure complexity 
was estimated with the Shannon-Weiner diversity 
index expressed as a proportion of the maximum 
possible diversity using the number of readings 
assigned to each height interval (Zar 1999). 
Analysis of Microhabitat and Avian Communi¬ 
ty. —We used principal components analysis 
(PCA) to reduce the dimensionality of the original 
microhabitat variables. All components had var¬ 
iance inflation factors <1.1 and were considered 
to be unique contributors to the analysis (Leps and 
Smilaucr 2003). Canonical correspondence anal¬ 
ysis (CCA) was used to expose patterns of 
variation in avian community composition and 
species abundance related to PCA variables (ter 
Braak 1986). and to guide selection of habitat 
characteristics for modeling Cerulean Warbler 
microhabitat. The length of the longest gradient 
(i.e., ordination axis) was 3.68. and we considered 
unimodal ordination methods (e.g.. CCA) more 
appropriate than linear methods (e.g.. redundancy 
analysis) (Leps and Smilauer 2003). We used bird 
abundance and confined the CCA to those species 
detected at three or more locations within 50 m of 
plot center using CANOCO 4.54. This is prefer¬ 
able to analyzing all bird detections, which 
assumes vegetative measures within our plots 
are an adequate representation of habitats used by 
birds that were detected farther away where 
microhabitat characteristics are likely to vary. 
We used randomized Monte Carlo tests (n = 499) 
to evaluate significance of CCA axes. 
Analysis of Miernhabitat Models. —We used 
logistic regression to examine the relationship 
between Cerulean Warbler occurrence and habitat 
variables with the binary dependent variable 
representing used and unused habitat plots. We 
established 20 models a priori and compared 
them using the information-theoretic approach of 
Burnham and Anderson (2004). Variable selection 
was based on Cerulean Warbler literature (Hamel 
2000a), as well as habitat plot comparisons, CCA, 
and field observations from this study. We 
performed a second-order bias correction (A1C ( .) 
because n/K < 40 and calculated evidence ratios 
based on Akaike weights (w,) as an indication of 
model strength in comparison to other models 
considered (Burnham and Anderson 2004). We 
examined a variable's beta coefficient to identify 
its relationship (positive or negative) to Cerulean 
Warbler presence. Variables present in the model 
with the highest tv, were considered the best 
predictors for Cerulean Warbler occurrence. 
An alpha level of 0.1 was selected for all tests 
ot significance due to the conservation status of 
the Cerulean Warbler (Asians et al. 1990). All 
statistical analyses, unless previously described. 
were performed using SPSS ® Version 15.0 (SPSS 
