Some Common Problems in the Use of Classification Procedures in Social Sciences

Abstract

This paper deals with some common problems in connection with classification procedures in social research. First, an abstract model is shown on which most surveys are based. A number of classification tasks can be deduced from this model. However, classification procedures are rarely applied. The reason for this is discussed in the second part of the paper. The paradigma of structural equation models shows why some well known problems of numerical classification are evaluated negatively and hence classification procedures are rarely used. However, it is too easy to explain the rare use of classification by this paradigma only. Another important reason is the negligence of measurement errors. The standard statistical packages, like BMDP, SAS and SPSS-X, do not contain classification procedures that allow an estimation of measurement errors. The third part shows that latent class analysis and multivariate mixture models can be interpreted as classification procedures which make allowance for measurement errors.