SMABS 2004 Jena University
SMABS 2004 Home Organization About Jena Sponsors Links Imprint / Contact SMABS Home

European Association of Methodology

Department of methodology and evaluation research

Jena University

Contributions: Abstract

Modeling linear dynamic bipolarity: formal and ontological aspects

Stéphane Vautier
Université de Toulouse Le Mirail
Rolf Steyer
Friedrich Schiller University Jena
Saïd Jmel       Eric Raufaste
Université de Toulouse Le Mirail

How is affective change rated with positive adjectives such as good related to change rated with negative adjectives such as bad? Steyer et al. (1997, 2000) showed how individual differences in change can be modeled using SEM. In an empirical study dedicated to positive and negative activation, Vautier and Raufaste (2003) drew on this work to connect two latent change score models, and to assess the linear relationship between both kinds of change.

Here, linear dynamic bipolarity is further studied from two perspectives. First, nested perfect and imperfect forms of dynamic bipolarity are distinguished. Perfect bipolarity means that latent change scores correlate to -1, whereas imperfect bipolarity means that they correlate negatively, but not extremely. Second, the issue of building the latent variables is addressed. Analogue rating scales provide data with complex and not well-known distributional properties. SEM analyses of self-rated affect may require the fitting of conventional polychoric correlations based on data collected using Likert-type formats. But the drawback of relying on polychoric correlations is that the estimated models may only account for extrapolated structures, which questions the ontological status of such models. What psychological phenomena can they really capture?

Four-wave models for perfect and imperfect dynamic bipolarity were specified and applied to six data sets from Steyer and Riedl (2003). Polychoric correlation structures were built using tetrads of affective items related to three constructs related to core affect (Russell, 2003), namely valence, energy, and tension.

Results suggested that perfect bipolarity is compatible with valence self-ratings, whereas imperfect bipolarity is compatible with tension and energy self-ratings. The paradox of defining change scores for extrapolated latent variables and then, of assessing whether they correlate perfectly or not, is discussed, in the perspective of evaluating their utility for substantive research in core affect.