686 



Fishery Bulletin 91|4). 1993 



merit errors and to randomness in the catch process. A 

 complete analysis of the problem would measure how 

 the assessment program transforms variability in the 

 catch-at-age data into variability in the resulting esti- 

 mates (e.g., Kimura, 1989). 



Because the catch model with constant selectivity 

 has fewer unknown parameters, when applied to noisy 

 catch-at-age data, the assessment program's estimates 

 could obtain greater precision (but not accuracy) by 

 assuming constant selectivity, even though the assump- 

 tion was incorrect 3 . However, it seems unlikely that 

 noise in the data could ever reduce the bias resulting 

 from a structural deficiency in the underlying catch- 

 at-age model. 



'.John Shepherd, Ministry of Agriculture, Fisheries, and Food, Fish- 

 eries Laboratory, Lowestoft, Suffolk, NR.S3 0HT, U.K., pers. commun. 

 April 1992. 



In the experiments with the simulated catch-at-age 

 data, the year-to-year changes in selectivity were not 

 particularly drastic, but I know of no studies to sup- 

 port my conjecture that they are realistic for the stock 

 of widow rockfish. The simulated changes in selectiv- 

 ity were comparable to those observed by Houghton 

 and Flatman (1981) for North Sea cod and by Gordoa 

 and Hightower (1991) for Cape hake. The fact that 

 experiments with "random" changes in selectivity 

 produced results similar to those from experiments 

 with trends in selectivity confirm that the biased esti- 

 mates were not just artifacts of having a simple trend 

 in selectivity, rather than a more complex type of 

 variation. 



One surprising result of the experiments with dif- 

 ferent assessment methods was the large discrepancy 

 between the estimates from Stock Synthesis and 

 CAGEAN when selectivity decreased (Fig. 6). The two 

 programs differ primarily in how they account for vari- 



