Meßniveau und Analyseverfahren – Das Problem „strittiger Intervallskalen”

Abstract

Abstract Many variables used in social research are measured „by fiat”. The „undermeasurement controversy” focusses on the question whether such variables may be treated as interval scales in the process of data analysis. To resolve this issue, the original prescriptions by STEVENS (1946) are critically examined. From a philosophy of science perspective, the problem of the undermeasurement controversy needs redefinition. The traditional issue „Should ordinal scales be treated as interval” is shown to be strictly nonsensical. A strict comparison of statements about relationships between variables assumed to be „measured” at the ordinal or interval level is not possible. In particular, linearity of relationships is a meaningful concept only if interval scales are assumed. However, statements about 1. existence, 2. direction and 3. comparative strength of relationships are possible at both ordinal and interval levels of measurement. Thus, a meaningful question is whether treatment of an assumed interval scale as merely ordinal yields different or identical statements. It is empirically demonstrated that for typical social survey data of an attitudinal nature, the use of an ordinal measure of association and the use of product-moment correlation coefficients lead to virtually identical statements (to the extent to which their comparison is meaningful). The generalizibility of the result is discussed. An argument for the parametric strategy is made. The coding of variables measured by fiat is discussed. Progress of theory-building in sociology requires that theories state more than just the direction of relationships. Thus ratio scales are needed. It is pointed out that metric concepts are both a condition and consequence of development in theory as well as empirical research. A general adoption of a conservative ordinal strategy would effectively prevent such empirical and theoretical progress.