442 
Fishery Bulletin 106(4) 
laid photographs with a transparency containing a grid, 
in which each grid cell represented a 5 cm x 5 cm area of 
the seafloor. From 1996 through 2000, the percent cover 
of hydroids, bushy and encrusting bryozoans, sponges, 
and F. implexa was recorded in each grid cell. Data were 
summed across grid cells to calculate the total percent 
cover of each type of colonial organism per photograph. 
In 1994, the percent frequency of several types of colo- 
nial epifauna was measured, instead of percent cover. 
Results of the 1994 surveys were presented in Collie et 
al. (2000a). In addition, noncolonial, megafaunal species 
were enumerated in each grid cell and identified to the 
lowest possible taxonomic level. Sediment type and the 
number of pieces of cultch (i.e., broken bivalve shells) 
were also recorded during analyses of the photographs. 
Like colonial epifauna, some fish species use the three- 
dimensional structure generated by cultch to obtain 
shelter from predators (Auster et al., 1995). 
Because hard bottom (e.g., cobble and gravel) and 
soft bottom (e.g., sand, silt, and mud) environments 
support fundamentally different benthic communities, 
we decided to remove from our data set those transects 
where sand constituted a large percentage of the sub- 
stratum. Transects where the majority of photographs 
contained greater than 50% sand cover were usually 
clustered around a distinct area or were located at the 
far edge of our study sites (Fig. IB). To guarantee that 
only gravel habitat was examined, 34 photographs from 
sandy transects were removed from the data set. Once 
photographs that were sandy, blurry, or overexposed 
had been selected and removed, 386 photographs re- 
mained, covering an area of 100.1 m 2 . 
Throughout this study, benthic photographs were 
analyzed by five observers. To evaluate the extent to 
which between year and within-year fluctuations in 
epifauna abundance might reflect observer bias, calibra- 
tion tests were performed during which two observers 
examined the same photograph(s). T-tests, in which 
photographs analyzed by different observers were treat- 
ed as matched pairs, were used to determine whether 
observer bias affected estimates of epifaunal abun- 
dance. Results indicated that, with the exception of the 
coiled worm ( Spirorbis spp.), patterns of observer bias 
did not match the direction and magnitude of between- 
year and within-year variations in epifaunal cover or 
megafaunal abundance (Asch, 2006). Data on Spirorbis 
spp. are not presented here because of concerns about 
observer bias. 
Statistical analyses 
Colonial epifauna A series of two-way analysis of vari- 
ance (ANOVA) tests were conducted to investigate 
which taxa of colonial epifauna exhibited significant 
differences in percent cover between disturbance cat- 
egories and years. Because large variations in organ- 
ismal abundance with depth may overshadow subtler 
fluctuations related to bottom fishing, each depth stra- 
tum was considered separately in the ensuing analyses. 
Response variables included six measures of the cover 
of colonial epifauna taxa and cultch. A previous exami- 
nation of spatial patterns indicated that autocorrelation 
existed between photos from the same transect (Asch, 
2006). Therefore, percent cover from photos were aver- 
aged across transects, allowing us to use transects as 
our primary sampling unit when conducting parametric 
tests. All response variables were arcsine square-root 
transformed in order to ensure that percent cover data 
would conform to the normal probability density func- 
tion. Bottom-fishing disturbance classifications and 
year were used as factors in the ANOVAs. At shallow 
sites, the interaction term {disturbance x year) was used 
to evaluate whether differences between sites inside 
and outside of CA-II increased over time because of the 
continued recovery of organisms in CA-II. Since no data 
from a deep, disturbed site were collected in 1999, this 
particular year was removed from the data set when 
we considered the deeper depth stratum, so that the 
experimental design would be balanced. Because of the 
nonorthogonal design of these ANOVAs, type-III sums 
of squares were used. When interannual differences 
were indicated to be an important factor affecting a 
particular type of colonial epifauna, Tukey-Kramer 
multiple comparisons tests were applied to determine 
which years differed significantly (S-Plus 6, MathSoft, 
Inc., Seattle, WA). 
Noncolonial organisms Although many noncolonial 
organisms were identified to the species level, other 
organisms could only be identified to genus or family 
because of the limited resolution of photos or the need 
for microscopic examination of distinguishing features. 
In cases where some organisms belonging to a par- 
ticular taxonomic group could be identified to species, 
but others could not, all members of the taxonomic 
group in question were lumped together in order to 
guarantee that classifications were mutually exclu- 
sive. Ninety-eight percent of the noncolonial organisms 
identified in photographs belonged to five extremely 
abundant taxa: Spirorbis spp., the tubeworm Protula 
tubularia, the jingle shells Anomia spp., the barnacles 
Balanus spp., and an unidentified species of burrowing 
anemones (order Ceriantharia). For each depth and 
disturbance category, the R statistical package (Free 
Software Foundation, Boston, MA) was used to fit the 
negative binomial distribution to the data on the abun- 
dance of these five species. Next, a U test was applied 
to evaluate the goodness-of-fit of the negative binomial 
distribution (Krebs, 1999). Because in all cases but one 
(i.e., Anomia spp. at deep, undisturbed sites) the nega- 
tive binomial distribution provided an adequate fit, we 
determined that these five highly abundant species have 
an aggregated spatial distribution, whereby they were 
absent from many photographs but obtained very high 
densities in a few areas. For example, Anomia spp. and 
Balanus spp. were not identified in two-thirds of the 
photographs sampled but were occasionally observed at 
concentrations as high as 1328 per m 2 and 760 per m 2 , 
respectively. Because random sampling of such dense 
aggregations could result in the detection of artificial 
