Resumen
Psychological scales play a key role in the assessment, screening, and diagnosis of latent variables, such as emotions, mental health, and well-being. In practice, researchers need shorter scales of psychological traits to save administration time and cost. Thus, a variety of optimization algorithms have been proposed to abbreviate lengthy psychological scales into shorter instruments efficiently. The main goal of this application is to form an abbreviated scale with fewer items while maintaining reliability, relationships among the subscales, and model fit for the full scale. In this study, we use an optimization algorithm (genetic algorithm) and a feature selection algorithm (recursive feature elimination) to abbreviate a psychological scale automatically. Although both algorithms search for an optimal subset of features within a large pool of features, the search mechanism underlying each algorithm is quite different. The genetic algorithm employs a systematic but computationally-expensive sampling process to find the optimal features, whereas recursive feature elimination removes the least important features iteratively until a desired number of features are retained. In this study, we use a 77-item measure of test emotions (Test Emotions Questionnaire) to demonstrate how these algorithms can be used for scale abbreviation. We generate a 40-item short form using each algorithm and compare the quality of the selected items against the full-length scale. The results indicate that both methods can provide researchers and practitioners with a systematic procedure for creating psychometrically sound, shorter versions of lengthy psychological instruments.