When emotional intelligence measures disagree: discordant profiles in allied health students and why classification uncertainty matters.

Introduction: Emotional intelligence (EI) is widely studied in health professions education, but two methodological habits limit the field: EI is usually treated as a single continuous predictor, and person-centered studies typically assign students to profiles and then test associations as though those assignments were certain. We addressed both issues, and a third that emerged from our own data: the validity of the profile solution itself. Methods: Allied health sciences undergraduates at Kuwait University completed two self-report EI measures: the Schutte Self-Report Emotional Intelligence Test (SSEIT; a self-report measure based on the ability model, indexing self-perceived emotional skill) and the Kuwaiti-Arabic Trait Emotional Intelligence Questionnaire-Short Form (TEIQue-SF; trait model). Of 561 responses, 559 remained after duplicate removal and 489 after excluding 70 invariant (straight-line) responders - a screen motivated by the fact that reverse-scoring maps every constant TEIQue-SF response pattern to a global mean of exactly 4.00, creating duplicate data points that induce degenerate, variance-collapsed solutions in unconstrained mixture models. Latent profile analysis of the two standardized scores was estimated with a variance floor; profile-outcome associations used the bias-adjusted three-step (BCH) approach with bootstrap confidence intervals (1,471 valid resamples of 2,000), compared against naive modal-assignment tests, with Holm and Bonferroni multiplicity control. Results: The two measures correlated only modestly (r = 0.35). A four-profile solution was favored on the Bayesian Information Criterion (entropy = 0.65): High EI (28.6%), Average EI (34.2%), Low EI (32.3%), and a Discordant EI profile (4.9%) combining very low self-perceived skill with near-average trait EI. Profile rank-ordering differed across instruments, indicating qualitatively distinct configurations, and the Discordant profile was stable across random starts, outlier removal, and bootstrap resampling. Two associations were robust: students in the Low EI profile attended voluntary EI workshops substantially more often than High and Average profiles, and stress-management confidence was highest in the Discordant profile and lowest in the Average profile. GPA showed no robust association. Discussion: Self-report EI measures from different theoretical traditions can disagree enough to define distinct student subgroups, and person-centered EI research must attend to solution validity - careless responding and degenerate likelihood spikes - as well as classification uncertainty.