702 



Fishery Bulletin 103(4) 



incorporating seasonal growth and measurement error 

 estimates (Table 5). Estimates of proportion of outliers in 

 the data set (p) greater than zero were due to lack of fit 

 and dropped to zero in model 5. Preliminary bootstrap- 



ping showed that fitting p regularly produced spurious 

 model fits. Because the full data sets were estimated 

 to have no outliers, it was considered reasonable to fit 

 model 4 (equivalent to model 5, but with p held equal to 

 zero) to all bootstrapped data sets for site-specific growth 

 estimates and comparisons. 



Estimates of p also dropped to zero in model 5 when 

 this model was fitted to the sex-specific data sets, ex- 

 cept for females at Lord's Bluff. Holding p = in model 

 4 for females at Lord's Bluff resulted in a less good fit 

 compared to that of model 5 and also produced slightly 

 different parameter estimates than those of model 5, 

 namely increasing growth (g., andg 28 ), growth vari- 

 ability (v), and measurement error (m) (Table 6). Vi- 

 sual inspection of residuals showed an obvious outlier 

 in the data set. When this was removed and model 5 

 was refitted, p fell to zero and the other parameters 

 estimates were very close to the values estimated from 

 fitting model 5 to the original data set, and there was 

 a large improvement in likelihood. Therefore the model 

 for females at Lord's Bluff was based on the data set 

 with the outlier excluded, and model 4 with p held at 

 zero was fitted to all bootstrap data sets for sex-specific 

 growth estimates and comparisons. 



Site comparisons With the exception of s at Lord's 

 Bluff, the proportion of bootstrap parameter estimates 



