Comments are presented on an article by H. R. Knol on the use of statistics for simple random samples when, in fact, the sample is not of the simple random type (see Mens en maatschappij, 1976, 51, 2 [no title given]). The use of the terms "strata" & "stratify" & the purpose of stratification are discussed. Stratification is used as a means of attaining more homogeneity, at least within the strata, to arrive at closer confidence; but whether or not these closer confidence limits & greater homogeneity are arrived at depends on the relation between the invested phenomena & the criterion of stratification. Knol describes the stratification procedure but does not mention the criterion of stratification. In fact, he describes seven strata & an eighth stratum comprised of implicity stratified clusters. The implicit stratification had been done along province & degree of urbanization. But in the arithmetic he neglects the consequences of stratification in the eighth stratum. Due to this neglect, Knol arrives at "method-effects" which are overestimations of the real "method effects." The effects he computed are most strong for religion & level of education. These phenomena are (at least in the Netherlands) strongly related with degree of urbanization, the stratification criterion in the eighth stratum. If Knol had decided to account for the stratifications within his eighth stratum, the computed method effects would be smaller than those he found now. Knol overestimated the underestimating of variance because of the use of simple random statistics in nonsimple random samples. In Nawoord Bij de kanttekeningen van Cozijn (Reply to a Note by Cozijn) Hans R. Knol claims that Cozijn's criticisms are based on a misunderstanding. He took a note that was meant to apply to the 1974 sample to apply to the 1970 sample as well. Hence, his conclusion that the method-effect is overestimated does not apply. Cozijn's remark that stratification is used to obtain greater homogeneity within strata is discussed. In answer to Cozijn's doubts, it is claimed that stratified sample statistics are always justified in the case of a stratified design. The rule of thumb of multiplying by two the simple random variance estimate is discussed. The value of such a rule is doubted, & it is pointed out that Cozijn should not have applied it to a result coming from the whole sample, but rather to the stratified part of the sample only. A. Orianne.
