258 
Fishery Bulletin 106(3) 
were applied to data from recreational and commercial 
fisheries, and to accomplish the second, indices of abun- 
dance were computed and tested for correlation. 
Materials and methods 
Data used in multivariate analyses 
To identify assemblages of species within fishery land- 
ings, statistical grouping techniques were applied to 
two fishery data sets, one recreational (headboat) and 
one commercial. Both data sets encompassed areas from 
Cape Hatteras, North Carolina to Key West, Florida. 
These data were chosen because of their importance 
for stock assessment of species in the snapper-grouper 
complex. 
The recreational sector was represented by logbook 
data reported by headboat operators and verified by 
port samplers. Headboats are large, for-hire vessels 
that typically accommodate 20-60 anglers on half- or 
full-day trips. Data collection began in 1972 with a 
focus on coastal waters off North and South Carolina. 
The area of collection was extended in 1976 to include 
the coastal waters of Georgia and northern Florida, 
and again in 1978 to include those of southern Florida. 
We used 1972-2005 headboat data. Records from each 
trip contained information on number of anglers, trip 
duration, date, geographic area, and landings (number 
fish) of each species. 
The commercial sector was represented by logbook 
data reported by commercial anglers with snapper- 
grouper permits. We used 1992-2006 commercial data; 
however, 2006 was a partial year (data through Sep- 
tember). Records contained information similar to those 
in the headboat data set, but landings were reported 
in weight (pounds). Excluded were nonsensical records 
suspected to be misreported or misrecorded. Analyses 
of commercial data were restricted to trips with han- 
dline gear (~87% of records) to avoid the possibility 
of confounding estimated assemblages with effects of 
gear. Furthermore, these analyses included only trips 
of one-day duration (~50% of records) to minimize the 
possibility that catch in a trip was taken from widely 
separated geographic areas with potentially different 
assemblages. 
Species assemblages 
Following Lee and Sampson (2000), we used more than 
one statistical technique to identify species assemblages. 
We applied three techniques: ordination and two types of 
cluster analysis. For all three techniques, the Sprenson 
(also called Bray-Curtis) measure of dissimilarity (dis- 
tance) between species was used (McCune and Grace, 
2002). In comparison to other measures, Sprenson dis- 
tance has been found more robust in ecological studies 
(Field et ah, 1982; Faith et ah, 1987) and provides more 
ecologically interpretable results (Beals, 1973). Perhaps 
for these reasons, it has been considered appropriate in 
studies of fish assemblages (e.g., Mueter and Norcross, 
2000; Gomes et ah, 2001; Williams and Ralston, 2002). 
To compute dissimilarities, we formatted each data 
set as a matrix, with rows representing species and 
columns representing vessel-months. That is, each ele- 
ment (Cy) of the matrix quantified the amount (in units 
of number fish for headboat or pounds for commercial) 
of a species (i) landed by each vessel pooled over one 
month (vessel-month j). The duration of a month was 
chosen as a reasonable compromise between maximiz- 
ing the variety of species landed (longer duration) and 
minimizing the number of different locations fished 
(shorter duration). Locations fished per vessel were 
generally consistent within a month, but could have 
changed on the time scale of seasons (perhaps follow- 
ing fish migrations, for example). Species were removed 
if they appeared in fewer than 1% of all trips because 
rare species may distort inferred patterns (Koch, 1987; 
Mueter and Norcross, 2000). This restriction left 25,293 
records of vessel-month-species in the headboat data set 
and 143,426 in the commercial data set. 
Before computing dissimilarities, data were trans- 
formed with the root-root transformation to moderate 
the influence of abundant species: 
This transformation has been preferred for density and 
biomass data, particularly when used in connection 
with the Sprenson measure of distance (Field et al., 
1982). After transformation, a matrix of dissimilari- 
ties between species was computed with the Sprenson 
measure of distance: 
\ C ij C 'hj\ _ 
( C ij +C hj) 
( 2 ) 
where D lh - the distance between species i and h\ and 
J = the number of columns (vessel-months). 
To identify species assemblages, the ordination method 
of nonmetric multidimensional scaling (NMDS) was ap- 
plied to the matrix of dissimilarities (Kruskal, 1964). As 
stated by McCune and Grace (2002), “Nonmetric multi- 
dimensional scaling is the most generally effective ordi- 
nation method for ecological community data and should 
be the method of choice, unless a specific analytical goal 
demands another method.” NMDS searches for positions 
of n objects (here, n species) in d dimensions such that 
dissimilarities in ordination space are close to those of 
the original space. We extracted the first two dimensions 
of ordination space (d= 2) for graphical presentation. 
In addition to ordination, we applied nonhierarchical 
and agglomerative hierarchical cluster analyses. The 
nonhierarchical cluster analysis was used to partition 
species into groups, based on the method of £-medoids, a 
more robust version of the classical method of Lmeans 
(Kaufman and Rousseeuw, 1990). The /e-medoids ap- 
proach attempts to identify k objects from the data set 
