Vom 22. bis 24. Juni 2023 fand die jährliche European Political Science Association Conference in Glasgow statt. Dort hat Martin Elff Vorträge zu den Themen "Models for multicategorical responses - Testing the hypothesis of constant probability" und "The RiLe Index of Party Ideology - A Reductio" gehalten.
Titel: Models for multicategorical responses - Testing the hypothesis of constant probability
Categorical responses are ubiquitous in social and political research. Binary logit and probit models already appear elusive to may researchers, not the least because of their intrinsic non-linearity. Even more elusive are models of responses with more than two categories. Here the values and, possibly, the statistical significance of coefficients depend on the choice of the baseline category of the response. To address this, a recent paper published in Political Analysis recommends to look at the statistical significance of differences in
probabilities rather than the statistical significance of individual coefficients. I argue that such a recommendation does not help with practitioners’ confusion about these models as it conflates inference with interpretation. I further show that there are established techniques of inferences in the statistician’s toolbox that avoid such conflation. Third, I show that probability changes always occur if any of the coefficients in a multinomial logit model are non-zero. Finally, I present a two-stage logit model that allows for some of the response probabilities to remain constant while others vary with the values of independent variables.
Titel: The RiLe Index of Party Ideology - A Reductio
The RiLe Index based on the data from the Manifesto Project is one of the most
widespread measures of parties’ political or ideological positions. Despite being
subject to criticism from various authors, it does not seem to have lost in popu-
larity or reputation. There may be a variety of reasons for the continued popu-
larity of this index: (1) that it gives the ”right” results, (2) that its construction is
conveniently simple and easy to understand, (3) alternatives often involve com-
plicated substantial and methodological considerations that seem intransparent
to most applied researchers, and (4) criticisms usually are buried in appendices of methodological papers that propose alternatives. The paper addresses this situation as it highlights the (implicit) assumptions that justify the construction of the RiLe index and demonstrates some paradoxical implications of these assumptions: The better the fit between these assumptions and reality, the more the RiLe index both exaggerates and occludes the actual movements of political parties. The consequences of these distortions are illustrated by a reanalysis of recent publications. While the MarPor data are an invaluable and irreplaceable source of information about parties’ political positions, the RILE index is clearly not best way to make use of them.