MSI 





/r. 



4*7 1> 





k 4 



o j 



BROWNS 



INO LA KAtt 



■ BANKS 







' / A 



r?i. 





=a 





' 





N 





•ESTERIf-'p 

 '\NOVA SCOYlA'' 

 J BANKS G 







p""'-- . 



__^ 



t= — i 





-t=i — i=i — 





-3 G H 



rw^_ western; - - 



" GE0R6E 9 



---,--& 





=, ba- 









EAST 



GEO 



ERN . 

 GtV 

 HI* 







. f- ' 





4 



, — 1= — 1= — t=, — 1= 



=1 1= 



■*>— - 



1 1 





Figure 1.— Sampling areas. 



category were sufficient for analysis. Commer- 

 cial catches were sorted into scrod (those fish 

 under approximately 2.5 pounds) and large size 

 categories at sea by the fishermen. Fish of each 

 size category were unloaded from the vessels in 

 carts of about 500-pound capacity. A sample 

 was composed of varying numbers of fish taken 

 from one or more of these carts from a single 

 vessel's trip. 



There were 82 samples collected over the 

 years for a total of 7,774 measurements. The 

 distribution of these samples among the various 

 factors is presented in Table 1. The geograph- 

 ical areas are outlined in Figure 1. 



Samples were not taken in strictly random 

 fashion. In order to treat these data statis- 

 tically, we must assume the samples taken from 

 each boat's catch to be representative of the 

 total catch and the boats sampled were repre- 

 sentative of all boats fishing. 



To study the relation of dressed to round 

 weights, lengths and weights of individual fish 

 were recorded at sea while fresh and at the 

 dock after the fish had been dressed and stored 

 aboard commercial vessels for periods up to 10 

 days. In one case both sets of measurements 

 were made at dock side. There were nine 

 samples of fish with measurements of gutted 

 and round weights, and two samples with 

 gutted and gilled, and round weights (Table 4). 



For the length-weight regressions, an 

 equation of the form W = clP was assumed, 

 where: 



W = weight in pounds, to the nearest 



tenth, 

 L = fork length in centimeters, and 

 c and b are constants to be estimated. 



Regressions were fitted by the least squares 

 method to the equation Y = a + bX , where: 



Y = log e W 

 X = log e L 

 a = logg c 



It is realized that the least squares fit to this 

 equation is not the same as the least squares fit 

 to the untransformed equation; however, it is 

 convenient to deal with the linear form The 

 regression statistics for each sample are given in 

 Appendix Al. Notations for regressions and 

 covariance analyses throughout this report fol- 

 low Snedecor (1956). The term significant 

 refers to a probability level less than 0.05. 



Inadequate distribution of samples pre- 

 vented the use of a factorial analysis to 

 determine the existence and significance of 

 interactions among the factors. Therefore, 

 where data permitted, a separate analysis of 

 covariance among the levels of a given factor 

 (e.g., among years) was run within each of the 

 other factor combinations, and the series of 

 analyses thus obtained were pooled to yield a 

 single result. 



An approximate F test was used to take 

 subsample variation into account when tests 

 were made using samples from a single trip. 

 The mean squares for the differences in regres- 

 sion coefficients and adjusted means were 

 divided by the corresponding mean squares for 

 differences among subsamples taken from 

 Appendix Table A2 (see Appendix Table A3). 



Since many of the sample cells (Table 1) 

 contain only one or two samples, comparisons 

 among them would not provide for adequate 

 estimates of error variance. It seemed best to 

 pool all the available estimates of sample-to- 

 sample variation to provide a single denom- 

 inator for all tests. In these cases the denom- 

 inators in the F tests were the estimates of 

 variations among samples taken from Appendix 

 Table A3 (see Appendix Table A5). 



In this paper, the term Approximate F Test 

 refers to either of the aforementioned ratios. 

 Because of the variable sample numbers, the 

 probability levels are not exact, and thus the 

 use of term approximate. 



