Bifactor Modeling and Latent Profiling of Biology Undergraduates' General and Specific Achievement Goals and Task Values

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

Abstract

Objectives and Theoretical Background
Our study aligns to the theoretical background and methodological approaches described in the synopsis. Our specific objectives were to compare the construct and predictive validity of achievement goal (Harackiewicz et al., 2002) and expectancy value (Eccles & Wigfield, 2020) theory data modeled with specific motivation factors (e.g., mastery approach, attainment value) versus including specific and general bifactors (e.g., general and specific achievement goal factors; Figure 1).
Methods
Sample, Context, and Data Collection
Week 2 motivation survey and final exam data were collected from participants enrolled in two Fall 2019 (n=399) and three Fall 2020 (n=367) biology sections that employed the same syllabus and pedagogy for introductory biology coursework at a large southeast US university (Tables 1-3).
Results
Confirmatory Factor Analyses
First, we conducted a confirmatory factor analysis (CFA) with only specific motivation factors (e.g., effort cost; Tables 4 and 5). Subsequently, we tested a series of models adding a single bifactor (i.e., either achievement goal, cost, or value). Next, we estimated a CFA with all three general bifactors, allowing each to correlate with each other, but not any specific factor (Reise, 2012). Given estimation problems, we investigated whether data-model fit and convergence would improve with a model where the achievement goal theory bifactor was indicated by either “performance-only” items or “approach-only” items. The latter model had the best data-model fit, therefore we chose this model, which included all three general bifactors (i.e., value, cost, and the “approach-only” achievement goal general bifactor), as our final model (see Table 6 for standardized factor loadings). There was remarkable consistency in factor loadings across semesters and latent factor reliability values were generally strong (Tables 7 and 8).
Predicting Exam Performance
We conducted structural equation models with latent factors predicting the four exam scores and found models including the general bifactors resulted in R-squared values greater than models with specific factors only. The difference was larger for Fall 2019 data.
Latent Profile Analysis
Finally, we used general and specific factor scores as measured variables in latent profile analyses (LPAs) for each semester, separately. Information criteria and bLRT values for the LPA supported a 5-class solution (Tables 9 and 10). Attempts to estimate models with a higher number of classes resulted in errors. Across both semesters, Classes 1 and 2 comprised maladaptive motivational profiles, including low achievement goal endorsement, high perceived cost, and low value (Table 11, Figures 2 and 3) with Class 2 comprising more extreme ipsative shapes. Class 3 was a moderate profile across both semesters, and Classes 4 and 5 were the most adaptive profiles in each semester, with high approach goals, low perceived cost, and high value scores with Class 5 being less extreme than Class 4. LPAs including general factors had more predictive validity than those without (Tables 12-14).
Scholarly Significance
Our work showed strong construct and predictive validity evidence for general bifactor models of motivation, exceeding levels achieved by specific-factor-only models, and indicating a need for more research on bifactor motivation models.

Authors

• Jeff A. Greene, University of North Carolina - Chapel Hill

• Matthew L. Bernacki, University of North Carolina Chapel Hill

• Robert D Plumley, University of North Carolina - Chapel Hill

• Shelbi Laura Kuhlmann, University of North Carolina - Chapel Hill

• Kelly Hogan, University of North Carolina - Chapel Hill

• Mara Evans, UNC Department of Biology

• Kathleen M. Gates, University of North Carolina - Chapel Hill

• A. T. Panter, University of North Carolina - Chapel Hill