An Exploratory Analysis of the Internal Structure of Test Through a Multimethods Exploratory Approach of the ASQ:SE in Brazil

Abstract

Abstract Background A wide range of exploratory methods is available in psychometrics as means of gathering insight on existing data and on the process of establishing the number and nature of an internal structure factor of a test. Exploratory factor analysis (EFA) and principal component analysis (PCA) remain well-established techniques despite their different theoretical perspectives. Network analysis (NA) has recently gained popularity together with such algorithms as the Next Eigenvalue Sufficiency Test. These analyses link statistics and psychology, but their results tend to vary, leading to an open methodological debate on statistical assumptions of psychometric analyses and the extent to which results that are generated with these analyses align with the theoretical basis that underlies an instrument. The current study uses a previously published data set from the Ages & Stages Questionnaires: Social-Emotional to explore, show, and discuss several exploratory analyses of its internal structure. To a lesser degree, this study furthers the ongoing debate on the interface between theoretical and methodological perspectives in psychometrics. Methods From a sample of 22,331 sixty-month-old children, 500 participants were randomly selected. Pearson and polychoric correlation matrices were compared and used as inputs in the psychometric analyses. The number of factors was determined via well-known rules of thumb, including the parallel analysis and the Hull method. Multidimensional solutions were rotated via oblique methods. R and Factor software were used, the codes for which are publicly available at https://luisfca.shinyapps.io/psychometrics_asq_se/. Results Solutions from one to eight dimensions were suggested. Polychoric correlation overcame Pearson correlation, but nonconvergence issues were detected. The Hull method achieved a unidimensional structure. PCA and EFA achieved similar results. Conversely, six clusters were suggested via NA. Conclusion The statistical outcomes for determining the factor structure of an assessment diverged, varying from one to eight domains, which allowed for different interpretations of the results. Methodological implications are further discussed.