92 



Fishery Bulletin 101(1) 



(Table 1). The errors in these estimates were shghtly 

 greater for the environmental model (more bias and larg- 

 er absolute error) than for the environmental model with 

 process error and traditional model. 



The confidence intervals on R^ were, on average, greater 

 for the traditional and environmental model with process 

 error than for to the environmental model (Table 2), which 

 greatly underestimated the width of the confidence inter- 

 vals. However, the confidence intervals for the traditional 

 and environmental model with process error showed the 

 true value falling below the confidence interval less often 

 than it fell above it. 



As expected, an environmental relationship was more 

 difficult to correctly detect with the traditional model in 

 situations with missing data (e.g. when catch-at-age data 



were missing in the last few years of the series), and, as 

 stated above, using the environmental model is inappro- 

 priate because it has a high probability of detecting a sig- 

 nificant relationship when none exists. The environmental 

 model also has a tendency to under estimate the width of 

 the confidence interval for /?„ and the true value frequently 

 falls outside of thi.s confidence interval. Therefore, the envi- 

 ronmental model with process error is the model of choice. 



Application: relating recruitment in the 



Hauraki Gulf-Bay of Plenty snapper stock to SST 



The environmental model with process error has the 

 lowest negative log-likelihood, but this model has many 

 more parameters than the environmental model (Table 3). 



