Choat et al.: Comparison of ichthyoplankton sampling methods 



199 



less likely to involve Type-I error and is more robust 

 to heterogeneity of variance than comparable tests 

 (Green 1979). All analyses were performed using SAS 

 Version 6 (SAS 1987). 



A more subjective procedure was used to select taxa 

 for size-frequency measures. For meaningful compari- 

 sons, it was necessary to select taxa that were well 

 represented in the collecting devices and that covered 

 a reasonable size-range (>8mm) within each method. 

 Apogonids, gobiids, lutjanids, and pomacentrids met 

 these criteria and also accounted for over 95% of the 

 variation in the main data set from net sampling. 

 Catches for nets and aggregation devices 

 were analyzed separately. For net catches, 

 density was expressed as mean number/ 

 1000 m ! within 2 mm size-classes among the 

 different methods and compared by one-way 

 ANOVAs. With aggregation devices, the 

 variable was the number of fish per sample 

 and comparisons were made by <-tests. 



Results 



The 83 samples contained a total of 57,701 

 fishes of 70 families, excluding clupeoids 

 (Table 1). Table 2 lists families which con- 

 stituted at least 1% of the individuals taken 

 by any sampling method and records their 



size-ranges by method. We refer to these as 'abundant 

 families'. 



Taxonomic composition and size structure 

 of the samples 



There were marked differences in taxonomic com- 

 position of the samples among methods. The bongo 

 net collected the largest number of families overall 

 (Table 1), including all of the abundant families and a 

 wide size-range within most families (Table 2). The 

 light-trap collected the fewest families overall and only 



