490 
THE WILSON JOURNAL OF ORNITHOLOGY • Vol. 124. No. 3. September 2012 
available area (i.e., >300 m from the waterline or 
inside the inundated area of the lake). 
Cover types were categorized as: clay, vegeta¬ 
tion. gravel (<5 cm), cobble (>5 cm), sand, or 
wrack (organic material washed up from wave 
action). Percent cover in a I-nr quadrat was 
measured at 5-. 15-, 30-. and 45-m distances from 
the center of the territory. These quadrats were 
arranged in a spiral pattern. 90 from one another 
and from the territory center. Distance to lake- 
shore was measured to the highest waterline of the 
lake, distance to vegetation was measured to the 
nearest location with >50% vegetation cover 
within I nr, and distance to nearest channel was 
measured to the nearest location where freshwater 
flowed into the lake (formed by overland flow 
during precipitation or from a groundwater 
spring). 
The available area for nest placement was 
delineated as an area -200 m in diameter 
centered on a nest site (i.e., within the presumed 
territory of nesting birds). Potentially available 
sites at this scale were selected at a random 
orientation (between 0 and 359 ) and distance 
(between 0 and 100 m) from a nest using two 
random number lists. A digital photograph of each 
microsite, including a I-nr quadrat marked at Id¬ 
em intervals centered on the nest, was taken from 
1.7 m above the ground. Microsite photographs 
were digitally overlayed with a 10 X 10-cin grid 
in Program Jasc Paintshop Pro© by Corel 
Corporation (www.corel.com). One of the six 
cover types used in the mcsoscale analyses within 
each grid cell was recorded and summed; the total 
of all cover types equaled 100%. 
Statistical Analyses.— We tested for an associ¬ 
ation between salinity and occupancy using 
Fisher’s exact test. Multiple logistic regression 
was used to test candidate models for the separate 
and combined effects of lake perimeter, aeolian 
lunette size, and elevation on the binary response 
variable of lake occupancy (unused = 0 , occupied 
- 1). Akaike’s Information Criteria (AIC), 
r C n1^ ed fo I dis P ersion and small sample size 
i 1 u and Akaike s weights (H7) were used to 
select the most parsimonious model (Burnham 
and Anderson 2002). The most important param¬ 
eter was identified by summing Akaike’s weights 
‘ hC pilrame,er °f in.eresi 
(Burnham and Anderson 2002). 
ofir;:^ ? d r est c ° vcr ^ p^m m <i 0 % 
a " quadrats (i.e., sand \9%\ and wrack I 7 % 1 ) 
w= rc excluded f rom further a „l lys e 5 
ing percent cover variables were transformed using 
arcsin(Jx|) to minimize non-normal distributions 
and multi-collinearity. The probability of territory 
use (0 -- available. I = used) was modeled as a 
response variable to the percent cover of four cover 
types (clay, vegetation, gravel, and cobble) at four 
distances from the territory centerpoint (total of 16 
variables) using a generalized linear mixed-effects 
model (GLMM) with penalized quasi-likelihood 
(PQL) estimation (MeCullagh and Searle 2000). 
I his method incorporates the repeated measures of 
four quadrats measured within 100 m of the center 
ol the territory. Several sites at the territory scale 
were measured on the shorelines of the same lakes, 
and lake was included as a random effect in the 
model in the analysis of cover types using a 
backward stepwise approach. We inspected the 
parameter estimates of each model and selectively 
removed non-significant covariates until :i final 
model was reached (MeCullagh and Searle 2000). 
I he territory-scale variables distance-to-lake- 
sliore. distance-to-freshwater channels, and dis- 
tance-to-vegetation w'ere compared among lakes 
using one-way ANOVAs. No significant differ¬ 
ences in these distance variables between lakes 
were found (P value range = 0.35-0.48), and we 
did not include lake in subsequent analyses. No 
significant correlations were found between the 
variables distance-to-lakeshore, vegetation or 
channel. Wc used multiple logistic regressions to 
construct seven candidate models that explained 
the etfects of these variables on territory use. 
Variables were included in models separately and 
in all combinations with (he binary dependent 
variable available (0) and occupied (I). AIC 
model selection corrected for dispersion and small 
sample sizes (QAIC,) was used to identify the 
most parsimonious of all candidate models. Those 
with a AQAIC f value <4 were considered the best 
subset (Burnham and Anderson 2002). Model 
averaging ol this subset was used to calculate 
parameter estimates. The most important param¬ 
eters were identified by summing Akaike's 
weights, in' from models (in the subset) including 
the parameter ot interest (Burnham and Anderson 
2002). 
Used and available nest sites within territories 
wore not independent as they were observed 
within the same putative territory. Thus, the 
probability of use at this scale (available = 0. 
used = I) was modeled as a response variable to 
the percent cover ot clay, vegetation, gravel, and 
cobble using conditional (paired), multiple logis- 
