Some observations on sampling tests

Samenvatting

An evaluation of sampling tests and their usefulness in sociological research. Special attention is given to signif tests and statements of sampling errors which perform the important but limited function of estimating the 'error' that results from using a sample of a given size rather than examining the total/sum pop of possibilities. A detailed analysis is made of important misuses of signif tests: (1) They cannot be used to establish the substantive sociological importance of a relationship. (2) They do not measure nonsampling errors. (3) They cannot be used safely if samples have not been selected in such a way as to meet the basic assumptions underlying the test. (4) They do not determine whether generalizations can be extended from one population to another. (5) They do not tell us anything about causality. (6) They are often misused in testing hyp's derived from the data after they are collected. The statement on causal analysis is applied to Union democracy by S. M. Lipset, M. Trow and J. S. Coleman, and it is concluded that even the best studies aiming at causal analysis necessarily leave many factors uncontrolled and are subject to sampling variability within the categories of control and crossclassifications. I. Langnas.