Internal structure validity of Utrecht Work Engagement Scale (UWES) in Peruvian workers

The Utrecht Work Engagement Scale (UWES), a measure of work engagement, has had controversies regarding the interpretation of its dimensionality, that is, whether its scores can be one-dimensional, multidimensional or two-level (one-dimensional and multidimensional). The dimensionality of the UWES has been questioned mainly due to the high covariation between its components, an aspect consistently verified in the observed scores and latent variables (De Bruin y Henn 2013), even by the same authors (Schaufeli et al., 2006). This has two implications, one operational and one conceptual. In the operational context, the calculation of scores is also influenced by the statistical differentiation of the factors found in the factor analyzes. This means that if one or more factors are retained and justified, this also determines the same number of observed scores that can be calculated and interpreted. On the other hand, the conceptual implication is that the comprehensive framework of a three-dimensional psychological phenomenon differs from the one-dimensional. For example, the differentiation between dimensions generates the possibility of different trends between dimensions (e.g., one of three is high and the other low) and in this case, the concept of engagement becomes very generic and ambiguous to name and understand the different possibilities. Specific combination of its dimensions or jeopardizes its content validity, implying the need for other more specific concepts, which although related to engagement, can be independent. The objective of the present study was to examine the internal structure (dimensionality, measurement invariance and reliability) of the UWES (two versions: long and short) in a sample of 636 Peruvian workers. Structural equation modeling was applied to evaluate three dimensionality models: a single factor (one-dimensionality), three related factors (multidimensionality) and bifactor (two levels of interpretation: general factor and specific factors). The results indicate that, after applying bifactor modeling, the common variance obtained from a general factor is stronger than the three-dimensional interpretation. This result was repeated in both versions of the UWES (long and short). The measurement invariance was satisfactory at the configurational, metric and scalar levels (in both versions). Regarding internal consistency, satisfactory coefficients (greater than.70) were obtained. Three issues emerge from this study that modify the original theoretical interpretation of the UWES (which consists of using three related factors). The first is the existence of a general factor underlying the items, and which is statistically substantial as a source of variance of the items, independent of specific factors. In this general factor, the items generally contribute to their variance, except for two items whose discriminative capacity is moderately low (11 and 13, both of the specific factor Absorption). The main element of this general factor is the Dedication factor, whose items completely represented this general factor. Second, the differentiation of two specific factors corresponding to Vigor and Absorption does not appear to be psychometrically sufficient to describe nested constructs within the engagement construct, and therefore its independent interpretation of the general factor could be questioned. However, a contrary argument is that both factors show different amount of specific variance, although low compared to the general factor; for example, Absorption shows more divergent validity compared to Vigor, while the latter contains more common variance (general factor). Third, Dedication was completely absorbed by the general factor, and is only relevant insofar as its items are completely related to the general factor. Therefore, this component lost statistical autonomy and, consequently, very poor conceptual differentiation. It is concluded that the UWES - three-dimensional model to obtain the scores must be replaced by a one-dimensional model, represented with a total score. Implications are future research are discussed.