482 



Fishery Bulletin 103(3) 



tion of samples in unsuitable habitat. As a measure of 

 variability, the magnitude of the variance is dependent 

 on the magnitude of the data (Zar, 1996). Thus, the 

 variances of trawl catches decrease as the observed 

 means decrease (Taylor, 1953). A lower variance, there- 

 fore, does not necessarily indicate a better estimator, 

 but instead may reflect lower population abundance. In 

 the six cases in this study where the variance of the 

 unstratified estimate was less than the variance of the 

 estimate poststratified by habitat, the unstratified abun- 

 dance estimate was less than the abundance estimate 

 poststratified by habitat. The low unstratified abundance 

 estimates in these six cases were the result of a dis- 

 proportionately large number of samples in nonhabitat 

 areas in relation to the size of the nonhabitat areas. 

 Therefore, although the unstratified estimate was more 

 precise, it was also likely to be an underestimate of the 



true abundance. Thus, we suggest that the estimate 

 poststratified by habitat is the most desirable estimator 

 in these situations, despite the decrease in precision in 

 relation to the unstratified estimator. 



In many cases, small sample size was likely the rea- 

 son that the estimates poststratified by habitat and fish 

 density were not the most precise of the three estimates. 

 Poststratification produces precise estimates when the 

 overall sample size and the sample size in each stratum 

 are large (Scheaffer et al., 1996). In our study, the esti- 

 mator poststratified by habitat and fish density was the 

 most precise estimator of the three when sample size 

 in the HFD stratum was 20 or greater and the sample 

 size in the LFD stratum was 9 or greater. The number 

 of samples in the HFD stratum appears to have had a 

 larger influence on the precision of estimates stratified 

 by habitat and fish density than the number of samples 



