19
egalitarianism rejected was not.
QCA's and logistic regression's accounts of the antecedents of party preference does
not show a unified picture. With the number and combinations of antecedents available, a
complete overlap can not be expected. But the degree of convergence on significance,
complexity, and specific antecedents nevertheless shows that the two methods' outcome is
different in degree, not in kind.
Grid-group theory's four political orientation constructs are capable of accounting for
party preference. The model fits better for some parties than for others. Nothing in the present
analyses adjudicates which camp is right in interpreting the theory, either in its construction or
in its application, as best suited for complexity or simplicity. In addition, a theory is in the
long run only satisfactorily tested against other theories.
7. DISCUSSION
The point of departure for this paper was the claim by analysts that QCA is a better method to
account for complexity since the method pursues an open ended combination of antecedents.
The method is rooted in small-N comparativist research tradition. By contrast, logistic
regression is perceived by the same analysts to represent a penchant for parsimony and power,
in that analysts often focus on a limited number of independent variables that can be added or
averaged across a large number of cases. However, QCA dovetails with logistic regression in
that both methods' dependent variable is the analysis of an event. From this nexus, small but
important adjustments in the application of both methods made their procedures sufficiently
similar as to allow for a meaningful comparison of their outcomes. Specifically, QCA can
simply be run as a cross-tabulation in order to overcome case restrictions and to extract tests
of significance; with logistic regression, analysts must heed coding procedures when
accounting for causal complexity by way of interaction terms.
The general conclusions of the anlysis are a) that the two methods were in agreement
as to the statistical significance of party models (which is not surprising given identical
antecedents and number of cases), b) that the methods converged as to which antecedents
24
Such tests are under way in various disciplines, eg, political science (Ellis and Thompson,
1997a; Grendstad, 2003a; Grendstad, 2003b; Grendstad and Selle, 1999) and risk analysis
broadly defined (Dietz, Stern, and Guagnano, 1998; Marris, Langford, and O'Riordan, 1998;
Peters and Slovic, 1996; Sjöberg, 2003; Wildavsky and Dake, 1990; Zwick, 2002).
Convention