An extended End-User Computing Satisfaction model for learning systems in higher education

This is the authors' abstract. We don't add key points for this paper.

Background This study reports an exploratory, single-institution scale-development pilot of a learner-centered extension to the End-User Computing Satisfaction (EUCS) framework, hereafter EUCS-LS, that adds three candidate dimensions, user participation, technical support, and multimodal accessibility, to the five classic EUCS dimensions for use with higher-education learning systems. Methods A cross-sectional pilot survey was administered to undergraduate students at one private university using stratified sampling with a proportional allocation target across ten faculties; after deduplicating by student ID, the analytic sample was N = 394 . Each of the eight constructs was operationalized as a single-item indicator. Joint covariation with overall satisfaction was assessed using hierarchical OLS regression with an intercept, principal component analysis, single-factor maximum-likelihood exploratory factor analysis, partial least squares regression with Variable Importance in Projection (VIP) scores, and post-stratification weighted least squares as a sensitivity check. Variance inflation factors, the Kaiser-Meyer-Olkin measure, Bartlett’s test of sphericity, and scale-level Cronbach’s alpha were used as diagnostics, and the OLS assumptions (linearity, normality, homoscedasticity, and independence of residuals across faculties) were tested formally, with heteroscedasticity-consistent (HC3) and faculty-clustered standard errors estimated as robustness checks. Results The eight items jointly accounted for 62.1% of the variance in overall satisfaction (adjusted R 2 ; F ( 8,385 ) = 81.50 , p < .001 ), and adding the three candidate learner-centered items to the classic block produced a small block-level gain ( Δ R 2 = 0.041 , F -change ( 3,385 ) = 14.24 , p < .001 ). Seven of the eight items reached the p < .05 threshold; timeliness did not ( p = .093 ). Collinearity diagnostics were unproblematic (all VIF between 1.75 and 2.24 ), yet the predictors shared a single dominant covariance pattern: one principal component captured 59.0 % of their joint variance, PLS VIP scores fell within a narrow band (0.928–1.091, spread 0.16 ), and treating the eight constructs as items of one general scale yielded Cronbach’s alpha = 0.900 . Formal assumption tests supported linearity and independence, detected mild non-normality and heteroscedasticity, and the substantive conclusions were preserved under robust standard errors. Together, these diagnostics indicate that the predictor set behaves as a single general perceived-quality factor with several correlated facets and that the rank order of individual coefficients is not statistically identifiable at this sample size.