306 
Fishery Bulletin 114(3) 
Table 1 
Number of surveys, by season and tide cycle, conducted over a 7-year period (from 
summer 2005 through summer 2012) at Point Heyer Artificial Reef, Puget Sound, 
Washington. 
Transect 
Flood tide 
Slack water 
before 
Ebb tide flood tide 
Slack water 
before 
ebb tide 
Total surveys 
by transect 
Spring 
T-1 
4 
4 
1 
3 
12 
T-2 
2 
5 
0 
1 
8 
T-3 
3 
4 
1 
0 
8 
Summer 
T-1 
0 
6 
1 
0 
7 
T-2 
2 
6 
0 
0 
8 
T-3 
3 
7 
0 
0 
10 
Fall 
T-1 
3 
3 
1 
0 
7 
T-2 
7 
0 
0 
1 
8 
T-3 
6 
1 
0 
0 
7 
Winter 
T-1 
2 
3 
0 
1 
6 
T-2 
2 
4 
0 
0 
6 
T-3 
1 
5 
0 
0 
6 
Grand totals 
35 
48 
4 
6 
93 
participating in this study, and they periodically used 
hand-held graduated staffs to calibrate their visual es- 
timates of fish length. 
The genetic data were compiled from a study in 
which molecular markers had been used to estimate 
genotyping error rates from brown rockfish cap- 
tured at PHAR (Hess et al. 2012). In that study, 718 
brown rockfish ranging in length from 10 to 37 cm (M 
[mean]=22; SD [standard deviation]=7.1) were sampled 
and returned alive to their point of capture during all 
seasons between spring of 2004 and summer of 2009. 
The genetic data were used to identify individuals that 
were recaptured in multiple years. 
Results 
A total of 93 survey dives were conducted between 30 
June 2005 and 18 September 2012. All surveys com- 
menced between 1.5 and 7 hr after sunrise (M=4 hr, 
SD=1.2), and the mean survey time per transect was 
28 minutes (SD=7.8). In no case was diver-estimated 
visibility less than twice the width of a survey lane. 
The numbers of surveys by tide cycle are presented in 
Table 1. The total numbers of brown and copper rock- 
fish observed by length class summed over all 3 tran- 
sects are presented in Figure 2. A list of all species 
recorded on transect over the course of the study (a 
subjective appraisal of how often the species were ob- 
served) and length classes for species from the 3 visu- 
ally dominant taxonomic families are presented in the 
Supplementary Table [Online]. 
We fitted a generalized linear mixed model 
(GLMM) to the data by maximum likelihood by using 
the Laplace approximation and a Poisson link with 
the lme4 package, vers. 1.1-8 (Bates et al., 2015) and 
statistical software R, vers. 3.1.1 (R Core Team, 2014). 
For each of the 2 species, we ran a random effects 
only (null) model with count as the response variable, 
and year and transect as random effects. We then 
added season as the explanatory variable to produce 
a full model. We used AN OVA to compare the null and 
full models and for both species the results were sig- 
nificant (x^=386.64 and 214.09 [3 dfj for brown and 
copper rockfish, respectively, P<0.001). The full model 
was selected over the null model by both Akaike in- 
formation criteria (Akaike, 1974) and Bayesian infor- 
mation criteria (Schwarz 1978). The log-likelihood in- 
creased, and the deviance, which in linear models is 
equal to the residual sum of squares, decreased with 
the full model, further indicating that the full model 
provided a better fit to the data (Table 2). The GLMM 
back-transformed seasonal mean counts with 95% con- 
fidence intervals (confidence intervals were computed 
before back-transformation) are presented in Figure 3. 
We conducted a multiple means comparison of counts 
between seasons with Bonferroni corrected alpha 
(0.0042 from 0.05) to test the null hypothesis of no 
difference in mean counts between seasons. For both 
species, spring was not significantly different from 
