Expectancy-Value-Cost Profiles: Exploring the Affordances and Drawbacks of the Bifactor Measurement Model

• In Event: Bifactor Models to Improve Theoretical and Analytical Precision and the Opportunity to Study Motivational Processes

Abstract

Objective and Perspective
Bifactor measurement models are useful in motivation research. For example, they can separate specific and general forms of motivation considered in complex motivational theories such as Situated Expectancy-Value Theory (Eccles & Wigfield, 2020). However, bifactor models do not always improve model fit with younger learners (i.e., Tolan et al., 2020) and the extent to which they develop more differentiated latent profile solutions is unexplored. Thus, the goal of this work is to critically consider the following questions: 1) What is the most appropriate factor structure for adolescents’ expectancies, values, and cost in afterschool spaces? 2) How do afterschool expectancies, values, and cost combine into motivational profiles for adolescent students? 3) What are the affordances and drawbacks of the bifactor measurement model when used for latent profile analysis?
Method and Materials
Our sample includes 197 middle school students (37% sixth, 38% seventh, and 23% eighth) who attended a school-based afterschool program in an urban school district (54% male, 81% Black). Students reported their expectancies and perceptions of values, and cost related to afterschool program activities using a 5-point likert scale (Jiang et al., 2018). We obtained student demographic data from the partnering school district.
Results
A series of Confirmatory Factor Analyses (CFA; see Table 1) suggested good fit for both a traditional CFA and a bifactor model, though slightly better fit for the traditional CFA. Thus, we retained factor scores from these two models and conducted latent profile analyses by estimating two- through seven-class models for each. Using a combination of fit statistics, entropy, and theoretical interpretability, we confirmed that a 4-class model most appropriately fit the bifactor data, and a 3-class model most appropriately fit the traditional CFA data (see table 1). In the traditional CFA solution, about half of the students reported moderate cost and value (n = 97), 40% reported low cost and moderate value (n = 78), and 11% reported high cost and moderate value (n = 22). In the bifactor profile solution, the largest profile of moderate cost and value was replicated (n = 75), suggesting that across modeling approaches, most students reported average expectancies and values in the afterschool program. Similar to the low cost and moderate profile above, the bifactor model presented a low cost and high value profile (n = 31), with a moderate global factor representing task value. The final two profiles in the bifactor LPAs represent a low cost and value profile (n = 42) in which students reported moderately low levels of cost, value and expectancies across the board, and a rather large profile with more shape effects (n = 49). This final profile demonstrates students’ moderately low effort and opportunity cost, but above average ego and emotional cost.
Significance
Taken together, results suggest that the bifactor model may not always present a better fit than traditional CFA models when working with samples of youth. However, when used in profile analysis, the bifactor model supports an additional profile with shape effects, highlighting how bifactor models may support more differentiated profile solutions.

Authors

• Ashlee Lester Sjogren, University of Virginia

• Kristy A. Robinson, McGill University

• Alison C. Koenka, Virginia Commonwealth University