680 



Fishery Bulletin 91(4), 1993 



assessment methods that also use the assumption of 

 constant selectivity. I analyzed the simulated catch-at- 

 age data with the CAGEAN program (CAGEAN-PC, 

 version 4, release 2) of Deriso et al. (1985, 1989) and 

 with the multiplicative catch-at-age model of Shep- 

 herd and Nicholson ( 1986. 1991 ). 



The Stock Synthesis program and the CAGEAN pro- 

 gram use similar approaches for modeling catch-at- 

 age, but they differ in their assumptions about vari- 

 ability in the observed data. Stock Synthesis assumes 



a multinomial error structure for the catch-at-age data. 

 If p a is the predicted proportion of age class (a) cap- 

 tured, then the variance associated with a random ob- 

 servation ofp„ is proportional to p„- (1-pJ. CAGEAN, 

 however, assumes a lognormal error structure. If c a is 

 the predicted number of fish caught of age class (a ), 

 then log, ,(c a /c a ) is normally distributed with a zero mean 

 and a constant variance. 



The multiplicative catch-at-age model is essentially 

 an approximation to the catch model that underlies 



