16 



Fishery Bulletin 92(1), 1994 



where Y(t) is the estimated length or weight at age 

 t, andy 1 and v., represent size at two ages t x and t.„ 

 which are typically the youngest and oldest indi- 

 viduals in the sample. The estimated parameters a 

 and b describe how the model connects y ; and y 2 . 

 Values of a and b and their 95% confidence inter- 

 vals lead to the selection of other submodels. 



The Schnute model (written in Microsoft 

 Quickbasic) was fit to the size-at-age data (Fig. 5) 

 by nonlinear regression on an IBM-compatable mi- 

 crocomputer. Growth modelling was restricted to 

 individuals from the central North Pacific samples, 

 because of inadequate age representation from the 

 western and eastern North Pacific samples. 

 Paralarval size-at-age estimates were included in 

 the growth models for males and females, because 



size-at-age results were similar for juvenile (66-83 

 mm ML) males and females. 



Model comparison If we assume that the Schnute 

 model exactly predicts the size of an individual, then 

 the residual sum of squares (RSS) of this full model 

 is an estimate of measurement error. To ascertain 

 if a reduced model with fewer parameters (e.g., 2- 

 parameter exponential) adequately describes the 

 data, the RSS's from the reduced model and full 

 model were compared using an F test statistic: 



( RSS R - RSS F )/( DF h - DF F ) 

 RSS f /DF F 



f- 





with DF R - DF F ,DF F degrees of freedom. 



