Longitudinal Measurement Invariance of Likert-Type Learning Strategy Scales

Abstract

Whether or not learning strategies change during the course of higher education is an important topic in the Student Approaches to Learning field. However, there is a dearth of any empirical evaluations in the literature as to whether or not the instruments in this research domain measure equivalently over time. Therefore, this study details the procedure of longitudinal measurement invariance testing of self-report Likert-type scales, using the case of learning strategies. The sample consists of 245 University College students who filled out the Inventory of Learning Styles-Short Version three times. Using the WLSMV estimator to take into account the ordinal nature of the data, a series of models with progressively more stringent constraints were estimated using Mplus 6.1. The results indicate that longitudinal measurement invariance holds for all but two learning strategy scales. The implications for longitudinal analysis using scales with varying degrees of measurement invariance are discussed.

Longitudinal measurement equivalence of subjective language brokering experiences scale in Mexican American adolescents.

“…The fit for each model was evaluated using the comparative fit index (CFI) and the root mean square error of approximation (RMSEA). Good model fit is represented by CFI values of 0.95 or above, and RMSEA values of 0.08 or below (Coertjens et al, 2012). …”

“…To test this, the chi-square difference test (Δχ 2 ) was conducted by using the DIFFTEST command provided by Mplus, and the change in Comparative Fit Index criterion (ΔCFI) was relied upon. Hypothesis of invariance is rejected when Δχ 2 has a probability lower than .05 ( p <.05) and the ΔCFI decreases by .01 or more ( ΔCFI ≥ .01) (Coertjens et al, 2012; Pendergast et al, 2015). The ΔCFI value was weighted more heavily than the Δχ 2 in evaluating change in model fit because chi-square values are more sensitive to sample size and the chi-square difference test tends to be too stringent (Pendergast et al, 2015).…”

“…First, for invariance analyses across parent gender and waves, cross-group reports were not independent from each other. Therefore, instead of using multi-group CFA, we modeled data for translating for mothers and fathers or across waves within a single covariance matrix (Coertjens et al, 2012; Kim et al, 2014). Second, responses to some items were collapsed if they were not endorsed in a certain group.…”

“…If so, adolescents may have more positive attitudes towards language brokering as a long-term effect of frequent translating (Tilghman-Osborne, Bámaca-Colbert, Witherspoon, Wadsworth, & Hecht, 2015), and may interpret items in the language brokering scale differently across time. Establishing measurement equivalence across time ensures that the underlying structure of the scale remains stable longitudinally (Coertjens, Donche, De Maeyer, Vanthournout, & Van Petegem, 2012). Having measures with longitudinal measurement equivalence is an important first step in hearkening to the call by Kam and Lazarevic (2014a) for future longitudinal investigations of language brokering.…”

Group, team, or something in between? Conceptualising and measuring team entitativity

“…For each factor it was tested whether factor loadings and intercepts are equal over time (Coertjens, Donche, De Maeyer, Vanthournout, & Van Petegem, 2012). This was done by means of testing and comparing three models: (1) a baseline model (i.e., the basic model structure is invariant, factor loadings can differ over time), (2) a model including invariant loadings over time (i.e., metric invariance, items are interpreted in a similar way), and (3) a model including invariant intercepts over time (i.e., scalar invariance).…”