Brooks and Stouffer • GRASSLAND BIRDS IN PINE SAVANNAS 
67 
surement within a site and represents heterogene¬ 
ity, or patchiness, within a study site. 
Statistical Analyses— Individual stands were 
treated as the sample unit for all analyses. We 
calculated the number of birds Hushed per hectare 
in each survey and averaged these densities from 
all surveys at each site. Vegetation sampling plots 
also were averaged over each site. We omitted 
one site classified as “other” and four sites that 
were >2 growing seasons since fire for analyses 
comparing bird densities and vegetation variables 
among stand types. Omitting these samples 
removed variation introduced by differences in 
time since fire among sites. Nineteen of the 
remaining 22 sites were one growing season since 
fire and three (1/stand type) were two growing 
seasons since fire. Six of these sites were upland 
stands, six were RCW stands, and 10 were bogs. 
All analyses were performed with SAS Version 
9.2 (SAS Institute 2006). 
We used log-linear generalized models to test 
for differences in densities of each species and 
total grassland birds among stand types. Hen- 
slow’s Sparrows did not occur in one stand type, 
and Bachman’s Sparrows did not occur in 
another; we tested for differences in densities 
only between stand types in which these species 
occurred. We specified a Poisson distribution for 
Sedge Wren and a negative binomial distribution 
for Bachman's Sparrow, Henslow’s Sparrow, and 
total bird densities, both of which are appropriate 
for zero-rich data. Appropriate distributions were 
selected by estimating an overdispersion factor 
(c = Pearson x 2 /df) and choosing the distribution 
with c closest to 1.0. All models had c < 1.02 and 
were not overdispersed (Burnham and Anderson 
2002). We used a significance level of 0.05 for all 
tests. 
We conducted principal components analyses 
(PCA) to reduce the number of correlated 
vegetation structure variables and plant species 
composition guilds to fewer, uncorrelated princi¬ 
pal components (PCs). We performed one PC A of 
vegetation structure variables using the n = 22 
data set and two other PC As with the complete 
data set (n = 27), one on vegetation structure, and 
another on plant species guilds. We used varimax 
rotations to aid in interpretation of the PCs and 
retained all PCs with Eigenvalues >1. We used 
the n = 22 PCs to examine vegetation structure 
differences among stand types; the n = 27 PCs 
were used to model bird occurrence. We used 
MANOVA to test for differences in vegetation 
structure PC scores among stand types (n = 22). 
Pairwise differences between means were tested 
using Tukey-Kramer tests. Residuals were tested 
for normality using Shapiro-Wilks’ tests. 
We used logistic regression to model the 
probability of Henslow’s Sparrow and Sedge 
Wren occurrence based on vegetation variables. 
We used the entire data set (n = 27) and its 
corresponding PCs for these analyses. We used 
the vegetation structure and plant guild PCs as 
independent predictor variables in an information- 
theoretic model selection approach (Burnham and 
Anderson 2002). The third plant PC was highly 
correlated with a vegetation structure PC and was 
not used when constructing candidate models. 
Henslow’s Sparrow and Sedge Wren global 
models were assessed for overdispersion (c = 
1.40 and 1.13, respectively). Models were ranked 
using Akaike’s Information second-order Criteri¬ 
on (AICc) for small sample size (Burnham and 
Anderson 2002). Models with AAICc < 2 were 
considered the best models (Burnham and Ander¬ 
son 2002). We also calculated an r General 
Information Criterion (r G \c) for each model 
(Wright 2001). This is a pseudo r calculated 
from any one of the common information criteria 
and represents the relative proportional variance 
explained by a model. 
RESULTS 
Henslow’s Sparrows occurred at 11 of 27 study 
sites, Sedge Wrens at nine, and Bachman’s 
Sparrows at five sites. Henslow’s Sparrow 
densities were highest across all sites, followed 
by Bachman’s Sparrow and Sedge Wren. Densi¬ 
ties of each species, considering only the sites 
where they occurred, were; Henslow’s Sparrow = 
0.52-13.33, Sedge Wren = 0.42-3.03, and Bach¬ 
man’s Sparrow = 0.42-2.22 birds/ha. The highest 
densities of Henslow’s Sparrows occurred on a 
0.2-ha transect in a small (< 1 ha) bog. Grassland 
birds that could not be identified to species (e.g., 
Ammodramus spp. or other emberizids) accounted 
for 6% (5 birds) of the total bird density across all 
stand types combined. We did not model Bach¬ 
man's Sparrow occurrence because this species 
was detected at so few sites. 
Bird Differences Among Stand Types. —Grass¬ 
land bird densities in upland, RCW, and bog 
stands varied among species (Fig. 1). Bachman’s 
Sparrows were not detected in bog stands, and 
densities did not differ between upland and RCW 
stands (Fj l0 = 0.33, P = 0.58). Henslow’s 
