198 



Fishery Bulletin 91(2), 1993 



trap captures few larvae <6mm, and we wished to 

 compare density estimates among gears for the sizes 

 of fishes captured by the light-trap. Damaged fish (-3% 

 of total) were excluded from the length analysis. 



The terminology of early-life-history stages of fishes 

 is complex and ultimately arbitrary, whether based on 

 morphological or ecological criteria (Kendall et al. 1984, 

 Kingsford 1988, Leis 1991b). We were primarily inter- 

 ested in taxa of which the adults are benthic on coral 

 reefs, but did not want to exclude semipelagic reef- 

 associated taxa by use of an ecological term like 

 'presettlement', nor did we wish to exclude partially- 

 or fully-transformed but still pelagic individuals of 

 benthic taxa by the use of a morphological term like 

 'larva'. Therefore, we use the terms 'larvae' and 'pe- 

 lagic juveniles' for the fishes collected during this study, 

 or refer to them collectively as 'pelagic fishes'. 



Larval, transforming, juvenile, and adult clupeoid 

 fishes of several types (including Spratelloides spp., 

 Dussumeria sp., Stolephorus sp., and probably Her- 

 klotsichthys sp.) were captured in large numbers, 

 mainly by light attraction. These clupeoid fishes rep- 

 resented a distinct assemblage of fishes with a differ- 

 ent age and size structure and adult habitat than the 

 reef species of primary interest to us. These clupeoids 

 are not considered here, but will be dealt with in a 

 separate publication. 



Reduction of data sets and analytical 

 procedures 



Sampling produced a data set comprising 70 families 

 of fishes (exclusive of the Clupeidae and Engraulidae) 

 collected from the sampling nights of 3, 5, and 6 De- 

 cember by six methods. For ease of analysis and un- 

 ambiguous interpretation, it was necessary to reduce 

 the number of families treated. We initially removed 

 from consideration any family which did not consti- 

 tute at least 1% of the catch of at least one method. 

 The removal of taxa of this level of rarity would be 

 unlikely to influence the outcome of the analyses (Green 

 1979). This excluded 51 families, leaving 19 (referred 

 to as 'abundant families') for analysis beyond simple 

 listing of numbers of families sampled (e.g., Table 1). 

 Relative-abundance information obtained by all six 

 sampling methods for the 19 abundant families was 

 subjected to Principal Component Analysis (PCA) us- 

 ing the variance-covariance matrix. As a check, the 

 same analysis was run incorporating the next 10 most- 

 abundant families; this generated identical patterns. 

 Reducing the data set from 29 to 19 families did not 

 change the resulting pattern. 



The PCA analysis identified patterns in the complex 

 data set of 19 families sampled by six methods. Many 

 of these 19 families were relatively rare and contrib- 



uted little to the variation in the data set. A detailed 

 examination of the factors contributing to these pat- 

 terns required factorial analyses such as multivariate 

 analysis-of-variance (MANOVA). These procedures are 

 best carried out with a reduced number of variables, 

 which allows a clearer interpretation of trends in the 

 data. This called for a further reduction in the number 

 of families analyzed. 



To achieve this reduction, the data set of 19 families 

 collected by nets was subjected to a PCA, which iden- 

 tified the taxa that contributed most substantially to 

 the variation in the data set. This PCA identified 

 apogonids, atherinids, gobiids, lethrinids, mullids, and 

 pomacentrids as major contributors (95.2%) to the 

 variation in the data set. These six taxa were used in 

 a MANOVA. This design provided sufficient degrees of 

 freedom for testing and interpreting the significance 

 of method and night of sampling. The analysis was 

 carried out on samples from nets only. 



For graphic display of trends in sampling by nets, 

 the eight most-important taxa from the PCA were de- 

 picted. These were apogonids, atherinids, gobiids, 

 lethrinids, lutjanids, mullids, pomacentrids, and 

 labrids. Labrids were included in this group at the 

 expense of schindleriids, as they were an abundant 

 reef-associated taxon of considerable interest to reef 

 fish biologists. This substitution did not affect the cu- 

 mulative variance accounted for by the eight families. 



Unlike nets, aggregation devices did not allow for 

 adjustment offish densities to a common volume. More- 

 over, aggregation devices collected a different set of 

 fishes. An additional PCA run on light-trap and light- 

 seine data identified atherinids, gobiids, labrids, 

 lethrinids, mullids, and pomacentrids as taxa, which 

 explained over 90% of the variability in the data set. 

 The families selected showed a strong relationship to 

 the overall abundance ranking, although two relatively 

 rare taxa (lethrinids and mullids) were included. 



Aggregation devices sample an unknown volume of 

 water. Because catches by aggregation devices could 

 not be standardized to number offish per unit volume, 

 we made separate comparisons of nets and aggrega- 

 tion devices. The variables used were mean number/ 

 1000 m a for nets, and mean number/sample for aggre- 

 gation devices. A factorial analysis was designed to 

 test for differences in sampling method (fixed) and time 

 (random). For factorial analyses, residual analysis was 

 performed (Snedecor & Cochran 1980) to check assump- 

 tions of normality and homogeneity of variance. Taylor's 

 Power Law (Taylor 1961) was used to determine the 

 appropriate transformation. 



Canonical Discriminant Analysis and Tukey's Stu- 

 dentized Range Test (HSD) were used to display the 

 differences detected. For MANOVA, the multivariate 

 test statistic (Pillai's Trace) was used because it is 



