The Relationships between Individualism, Nationalism, Ethnocentrism, and Authoritarianism in Flanders: A Continuous Time-Structural Equation Modeling Approach

Abstract

This article analyzes the relationships among nationalism (N), individualism (I), ethnocentrism (E), and authoritarianism (A) in continuous time (CT), estimated as a structural equation model. The analysis is based on the General Election Study for Flanders, Belgium, for 1991, 1995, and 1999. We find reciprocal effects between A and E and between E and I as well as a unidirectional effect from A on I. We furthermore find relatively small, but significant, effects from both I and E on N but no effect from A on N or from N on any of the other variables. Because of its central role in the N-I-E-A complex, mitigation of authoritarianism has the largest potential to reduce the spread of nationalism, ethnocentrism, and racism in Flanders.

Notes

An alternative to the EDM is the Approximate Discrete Model (ADM). The ADM utilizes only simple linear constraints to approximate the differential equation model to estimate the DT parameters. It can be estimated by means of standard SEM software programs like LISREL that do not allow imposing the nonlinear constraints in Equation (4) required by the EDM (see Oud & Delsing, 2010).

Note that the 1995 and 1999 samples were supplemented with new random samples of new eligible voters to compensate for deceased voters. These cases are not included in this study.

Note that there are two ways in longitudinal studies to get reliable parameter estimates: increasing the number of participants (N) or increasing the number of time points (T). As explained by Voelkle, Oud, von Oertzen, and Lindenberger (2012), when using SEM for longitudinal model estimation, N and T are arbitrary. In the present study, the number of time points is small (T = 3), but the number of participants is large (N = 1,274) such that the total number of observations is 3,822. The parameter estimates are thus based on a large number of observations, which show up in the large z values of inter alia the autoregressions and the loadings. It should also be noted that the intervals between observation time points are quite large (ISPO, 1991, 1995, 1999). Hence, the analysis covers a time span of almost a decade.

The Mx script is available at blogs.unpad.ac.id/toni/publications

The augmented moment matrix (Jöreskog & Sörbom, 1996) is the moment matrix combined (augmented) with the vector of means (including the unit moment 1). Formally, the augmented moment matrix reads and the augmented sample moment matrix S reads . ML estimation can be done by fitting the model-implied means vector and covariance matrix separately to the sample means vector m and covariance matrix , respectively, or by fitting the augmented moment matrix to the augmented sample moment matrix . The two procedures are equivalent and give equal estimates of model parameters, means , and (co)variances (Browne & Arminger, 1995, p. 193). As emphasized by Browne and Arminger (1995, p. 193), the augmented moment matrix procedure is more generally applicable than the means/(co)variance procedure because all SEM software programs accept the augmented moment matrix as input but not all accept the separate means vector and covariance matrix as input. That is, software programs that accept covariance matrices as input also accept the augmented moment matrix. In the augmented moment matrix, the number of distinct elements in (as well as in S) is one more than in and (m and ) combined in that it contains the unit moment 1. However, this additional element 1 in is estimated by the corresponding element in . Hence, the degrees of freedom df for both estimation procedures are equal.