Often cited is the Przeworski & Teune (1970: 30) thesis that: “the role of comparative research in the process of theory-building and theory-testing consists of replacing proper names of social systems by the relevant variables.” However, in this paper, we do not single-handedly end up on the classic Przeworski & Teune (1970) position which would be to always aim for more cross-national variables. Instead, we side with another modern classic, Grofman’s (1989) note on cross-national results with pooled data: namely that it is possible that the actual statistical relationship in sub-samples or individual countries can be different from the ones found in the full pooled sample and that this is worth exploring in more detail.
In this paper we investigate three Institutional factors, namely the investiture vote rules that is known under the heading of positive and negative parliamentarism (Bergman 1993). We also look at two other “usual suspects” in the discussions on the impact of institutions, namely semi-presidentialism and bicameralism. These three variables take different configurations in different countries (sometimes one or more is absent), but there exists very few “natural experiments” where it is possible to study the impact of rule change in one country. In the paper we explore the researcher’s dilemma when the average effect is systematically different over countries that make up the panel aspect of the pooled analysis. We also analyse the problem of how near time-invariant (institutional) variables can be important to study, but often not very suitable for statistical analysis simply because the theoretically and conceptually important variables are constant within countries.
