Cullen and Stevens: Underwater video recordings of Centropristis striata in waters off Maryland 
413 
Table 2 
Results from analysis of variance for the best linear mixed-effects (LME) models, 
determined by using the corrected Akaike information criterion. These results were 
used to compare the influence of habitat type (sand, sand+rock, live bottom) on val- 
ues of the counting metric MeanCount for black sea bass ( Centropristis striata ) ob- 
served on videos collected from 14 June to 4 August 2011 at 6 sampling sites in 
waters off the coast of Maryland. Results are given for 2 categories: all black sea 
bass and nuchal black sea bass (the latter fish were distinguishable from other in- 
dividuals by a darker body coloration, a nuchal hump, and white fin stripes). The 
standard error (SE) for the random effects represents the variance for each sampling 
site around the common intercept. MeanCount data were log transformed (by taking 
a natural logarithm of the variable+1; i.e., \og e [MeanCount+l]) before analysis to 
help meet the assumptions of the LME models. ICC=interclass correlation coefficient, 
which represents the correlation of observations from the same sampling site. 
Category 
Parameter 
df 
F-value 
P-value 
All black sea bass 
Intercept 
1,31 
17.658 
<0.001 
Habitat type 
2,31 
22.364 
<0.001 
Bait method 
1,31 
1.318 
>0.05 
Random effects 
SE 
0.142 
Residuals 
Variance 
0.028 
ICC 
0.838 
Nuchal black sea bass 
Intercept 
1,31 
17.447 
<0.001 
Habitat type 
2,31 
17.973 
<0.001 
Bait method 
1,31 
0.805 
>0.05 
Random effects 
SE 
0.015 
Residuals 
Variance 
0.017 
ICC 
0.469 
All models were first fitted with maximum likeli- 
hood estimation and compared with AICc by using the 
AICcmodavg package, vers. 2.0-3 (Mazerolle, 2016) in 
R. The AICc best models were refitted with restricted 
maximum likelihood, which estimates the variance 
components separately from the fixed effects, thereby 
providing unbiased estimates for the variance compo- 
nents (Zurr et al., 2009). ANOVA, with type-II sums of 
squares for unbalanced data, was used to extract F- 
values and Wald test P-values for the fixed effect habi- 
tat type. Normal quantile-quantile plots, box plots, and 
scatter plots of the residuals were examined for model 
validation. Tukey’s honestly significant difference tests, 
with P-values adjusted by using a Bonferroni correc- 
tion, were conducted for multiple comparisons if the 
ANOVA indicated a significant difference in values of 
MeanCount between habitat types for either category 
of black sea bass. Results were obtained by using the 
multcomp package, vers. 1.4-6 in R, which provides 
multiple comparisons tests for linear mixed-effects 
models (Torsten et al., 2008). 
Additional analyses included Spearman’s rank cor- 
relation analysis to examine the relationship between 
MeanCount for nuchal and non-nuchal black sea bass 
(without nuchal humps) and between MeanCount and 
TFA for both categories of black sea bass with deploy- 
ments as samples (n=4Q). Further, separate correla- 
tions were calculated between MeanCount and TFA for 
both categories of black sea bass for deployments with 
bait {n= 20) in the trap and without (rc=20). Correla- 
tions were obtained by using the stats package in R (R 
Core Team). 
Results 
Habitat appearing in the camera view during deploy- 
ments (n=40) consisted primarily of smooth and coarse 
sand, rock, corals (i.e., sea whips, stony corals), and 
shell. In total, 19 (47.5%) deployments were made in 
sand, 13 (32.5%) in sand+rock, and 8 (20.0%) in live 
bottom habitats. In general, values of MeanCount were 
greatest in the first 5-10 min of video followed by a 
variable decline. Values of MeanCount varied by site 
and date and were highest for both categories of black 
sea bass in sand+rock and live bottom habitats (Table 
1). The proportion of nuchal black sea bass observed in 
the 3 classified habitats were 31.2% in sand, 15.1% in 
sand+rock, and 18.2% in live bottom. A total of 9 black 
sea bass, of which 5 had nuchal humps, were caught in 
the trap, 6 during baited trap deployments and 3 dur- 
ing unbaited trap deployments. 
Weighted linear mixed-effects models (i.e., with the 
random effect for sampling site) that included a con- 
stant variance function that allowed the variance to 
differ for each level of habitat type were identified by 
AICc as the best models for the categories of all black 
sea bass and nuchal black sea bass. The variance for 
