NOTE O'Neili et al : Accuracy of at-sea commercial size grading of Penaeus spp 



397 



these digital scales also makes it possible to pack in 

 more precise grades of prawns. 



Prawns for these small packs are initially sorted 

 by eye by experienced crew, and many are verified 

 by individual weighing. Those prawns that fall out- 

 side a particular size grade are removed and the re- 

 mainder are graded according to corresponding 

 count-per-unit-of-weight tables. 



Large packs Prawns for onboard grading into the 

 large packs are sorted by eye into groups of about 

 the same size (occasionally by counting the number 

 into a unit of weight [often a pound measured on 

 analogue scales] and grouping them accordingly). 

 Very large and very small prawns are removed and 

 regraded. 



Quality control assessment in the factory— source of 

 data for analysis 



A. Raptis and Sons process tiger prawns caught by 

 their own large fleet of trawlers working in the NPF, 

 as well as prawns purchased from other fishermen 

 operating in the same area. The company randomly 

 checks the quality of the packs entering its factory, 

 including the accuracy of the grading. 



Packs for quality-control assessment were selected 

 at random for every vessel and from all consignments 

 entering the factory. All packs were clearly marked 

 with the vessel's name, the prawn species group, the 

 grade of the prawns, and the date caught. 



The selected samples were thawed individually. 

 For the small packs, net weight was recorded, and 

 size grading was checked by counting all prawns from 

 each pack and averaging the count. Large and small 

 prawns that did not fit the grade category were se- 

 lected by eye, weighed, and graded individually, and 

 the percentage by weight and the true grade of these 

 prawns were recorded along with the percentage of 

 those correctly graded. 



With the large packs, a variable 2.5-3 kilogram 

 sample of prawns was randomly taken from each 

 pack, counted, and checked as above. The percent- 

 age by weight and the true grade of incorrectly graded 

 prawns in the sample were recorded. 



Categorical analysis 



The results of all factory quality-control checks on 

 both the small and large packs between mid- 1993 

 and the end of 1994 were examined. Over this time, 

 samples ft-om 51 of the 127 boats that fished different 

 areas in the NPF had been taken. We split the data 

 into three time periods to find out whether the accu- 

 racy of grading early in the year differed from grading 



that took place later in the year when smaller prawns 

 recruited to the fishery. Period one was fi-om July to 

 December 1993, period two fi-om January to June 1994, 

 and period three ft-om July to December 1994. 



Data from the large packs, where the grades used 

 were not the same as those for the small packs, were 

 omitted from the analysis because they could not be 

 compared directly. The commonly used size grades 

 for both pack sizes are shown in Table 1, as well as 

 the equivalent carapace length of the prawns. 



The number of prawns in) contained in the small 

 packs was recovered by converting the net weight of 

 the pack to pounds (weight in kilograms divided by 

 0.45359), and then by multiplying by the count-per- 

 pound derived from the quality-control inspection. 

 The number of misgraded prawns (r) in the small 

 packs was estimated by multiplying the number of 

 prawns in the sample by the percentage misgraded 

 in that pack. The numbers misgraded for each pe- 

 riod and each size grade were analyzed by fitting 

 binary regression models by means of iterative 

 weighted least squares. The number of misgraded 

 prawns was assumed to have a binomial distribution: 



Pr 



n'a-n)" 



(r=0, l...n) 



with P( misgraded) = /rand P( correctly graded) = I- n. 

 The probability ir was modelled in terms of the log- 

 odds or logistic transformation (logiit/il - n)). The 

 computed deviance statistic, approximately distrib- 

 uted as chi-squared, was used in goodness-of-fit tests. 



