484 



Fishery Bulletin 103(3) 



fies an important drawback to using the poststratifica- 

 tion method. Because strata criteria are unknown when 

 sampling, it is not possible to insure that there will be 

 sufficient samples in each poststratified stratum. When 

 resulting sample sizes in some strata are small, post- 

 stratification may be ineffective at increasing precision. 

 If the resulting sample size in one or more strata is one, 

 the poststratification variance will be inestimable. If 

 the resulting sample size in one or more strata is zero, 

 poststratification may not be possible. 



Because sample size is a limiting factor for increased 

 precision with poststratification, there are strong impli- 

 cations for survey design. Many multispecies surveys 

 are conducted by using a stratified random sampling de- 

 sign. There are two ways to apply poststratification to a 



stratified survey. First, for an unbiased estimator, each 

 stratum of the stratified survey can be poststratified 

 individually (Cochran, 1977). For the poststratification 

 estimator to have increased precision beyond that of 

 stratified random sampling, each of the original strata 

 must have a large number of samples to allow suffi- 

 cient samples in each poststratified stratum. Therefore, 

 investigators who intend to poststratify data within a 

 stratified random survey for unbiased estimates need 

 to construct large strata with many samples in the 

 original sampling design. Second, if poststratification 

 is applied to data that were not collected under a prob- 

 ability sampling design, the estimator may be more 

 precise, but may be biased. For the analysis of data 

 that were not collected under a probability sampling 



