Psicometría: Modelos de medición multidimensionales

Understanding Multidimensional Models in Psychometrics

Overview of Multidimensional Models

• The discussion begins with an introduction to multidimensional models, emphasizing the presence of multiple factors alongside a general factor within psychometric instruments.

• It is noted that these models often involve correlated factors that represent different dimensions of a larger construct, which can be visually represented in a schematic format.

• An example is provided regarding burnout measurement, where three correlated factors are identified without generating a total score from them.

Total Scores and Theoretical Implications

• The ability to compute a total score depends on theoretical frameworks; some colleagues argue for the necessity of evaluating a general factor model to justify this scoring method.

• Statistically, correlated factor models and general factor models yield similar results in confirmatory factor analysis, suggesting they are nearly equivalent despite theoretical differences.

Second Order Factors and Bifactor Models

• A second-order factor model explicitly includes a general factor that explains correlations among first-order factors, distinguishing between their roles.

• In contrast, bifactor models feature independent first-order factors with no correlations among them or with the general factor, highlighting their unique contributions.

Caution in Model Selection

• Care must be taken when applying bifactor models as they define specific factors as independent; this is particularly relevant for constructs like intelligence where various reasoning types do not correlate.

• The independence of specific reasoning types (mathematical, verbal, spatial) suggests that higher performance in one does not predict performance in another.

Statistical Considerations and Misinterpretations

• There is caution against over-relying on statistical fit indices when selecting models; enthusiasm for bifactor models may stem from better fit rather than theoretical justification.

• Researchers should avoid adopting bifactor structures solely based on statistical outcomes without considering the underlying theory and definitions of variables involved.