Exploratory Likert Scaling as an Alternative to Exploratory Factor Analysis. Methodological Foundation and a Comparative Example Using an Innovative Scaling Procedure

Identifying the dimensional structure of a set of items (e.g., when studying attitudes) is an important and intricate task in empirical social research. In research practice, exploratory factor analysis is usually employed for this purpose. Factor analysis, however, has known problems that may lead to distorted results. One of its central methodological challenges is to select an adequate multidimensional factor space. Purely statistical decision heuristics to determine the number of factors to be extracted are of only limited value. As I will illustrate using an example from lifestyle research, there is a considerable risk of fragmenting a complex unidimensional construct by extracting too many factors (overextraction) and splitting it across several factors. As an alternative to exploratory factor analysis, this paper presents an innovative scaling procedure called exploratory Likert scaling. This methodologically based technique is designed to identify multiple unidimensional scales. It reliably finds even extensive latent dimensions without fragmenting them. To demonstrate this benefit, this paper takes up an example from lifestyle research and analyzes it using a novel R package for exploratory Likert scaling. The unidimensional scales are constructed sequentially by means of bottom-up item selection. Exploratory Likert scaling owes its high analytical potential to the principle of multiple scaling, which is adopted from Mokken scale analysis and transferred to classical test theory.