The multidimensional, hierarchical model of self-concept by Shavelson, Hubner, and Stanton (1976) is a cornerstone of modern self-concept research. Given the comprehensive research interest in it, it is surprising that one core aspect of this model has yet to be clarified: What is the best way to operationalize the elusive construct of global self-concept as the apex of the hierarchy? Previous research implemented global self-concept by applying reflective modeling procedures (e.g., second-order factor models) that followed a top-down logic, which assumes that global self-concept affects lower order self-concepts. However, theoretical considerations have often equally emphasized bottom-up processes, in which lower order self-concepts form a global self-concept. Yet, a bottom-up approach has not garnered much empirical interest, most likely because the requisite statistical models have not been available. The recently proposed model-based latent composite score can fill this gap. Therefore, we contrasted top-down and bottom-up representations of global self-concept by comparing conventional second-order factors and model-based latent composite scores. Across three independent large-scale studies (Study 1: N = 8,063; Study 2: N = 3,081; Study 3: N = 2,106), the second-order factors reproduced only small amounts of interindividual differences, which boosted the correlations with external criteria (i.e., self-esteem, enjoyment of school, academic outcomes) to theoretically and somewhat empirically implausible levels. By contrast, the composite score showed a more plausible pattern of stabilities and correlations. We discuss the consequences of the two approaches and propose the latent composite score as a new perspective on the apex of the Shavelson model.
Logic Grammars for Diagnosis and Repair
