Where typologies reach their limit: A critical discussion about operationalizing welfare policies as independent variables

The paper critically discusses existing strategies of operationalizing welfare policies as independent variables in quantitative comparative research. It focusses particularly on conceptual issues which arise from the common application of welfare typologies and concludes by discussing alternative strategies. Two central problems constitute the focal point of the discussion. The first problem addresses definitional inconsistencies of welfare typologies. This becomes apparent in the improper distinction between real and ideal types which are frequently confused in theoretical discussions. Furthermore, most typologies suffer from the fact that there is no agreement on which indicators to choose, which areas of the welfare state to focus on, and which methods to use for the empirical test. Both aspects lead to a lack of comparability. The second problem arises from the selection of countries. It is common practice to employ typologies based on a sample that is neither random nor systematic. However, since the types are usually construed by the proximity between the examined countries (e.g. when performing hierarchical cluster analysis), the results are strongly influenced by the selection made in the first place. This problem is increased by the fact that Central and Eastern European countries are often excluded from the sample or treated only superficially. Regarding the application of such typologies as an independent variable, we are left with a profound problem: the lack of comparability makes it very difficult to determine which typology might be suited best for explaining the dependent variable at hand. In addition, the classifications rarely cover all countries one wishes to examine. As a result, researchers have to either (1) combine approaches which are not equivalent, (2) group unclassified states on instinct or (3) exclude undefined countries from the analysis. Neither of these strategies seems satisfactory. Simulations in this paper show that when regressing the same dependent variable on slightly varying typologies the explanatory contribution differs remarkably. As a result of these conceptual shortcomings, the paper advises against the application of welfare typologies as independent variables. Concluding, possible conceptual starting points for an alternative measurement taken both from the existing literature and additional considerations of the author are discussed. The work is part of the authors PhD thesis.