Comparing the measurement precision of three online addiction scales in Chinese primary school students: a bifactor IRT study

Brief measures of problematic smartphone application use, social media use, and gaming assess related but behavior-specific constructs. Because instruments calibrated separately do not share a latent metric, their measurement precision cannot be compared directly. This study examined whether three widely used scales could be placed on a common general-factor metric so that their conditional precision could be compared among Chinese primary school students. A total of 1,108 students completed the Smartphone Application-Based Addiction Scale (SABAS), Bergen Social Media Addiction Scale (BSMAS), and Internet Gaming Disorder Scale–Short Form (IGDS-SF9). Single-scale and cross-scale confirmatory factor analyses were conducted. A joint bifactor graded response model then placed all 21 items on one general factor, with orthogonal scale-specific factors retained. We compared general-factor average item information (AII), relative efficiency (RE), conditional reliability, and standard error of measurement. The cross-scale bifactor model fit better than the one-factor and correlated three-factor alternatives; the general factor accounted for 72.3% of the common variance (omega hierarchical = 0.842), while 47.9% of the BSMAS’s within-scale common variance remained scale specific. The joint bifactor GRM showed good fit (CFI = 0.983, TLI = 0.979, RMSEA = 0.046, SRMSR = 0.054). On the common metric, per-item information peaked at a general-factor theta of 1.50 for the SABAS (1.224), 2.20 for the IGDS-SF9 (1.409), and 1.95 for the BSMAS (0.591). The SABAS was more efficient per item than the IGDS-SF9 below a general-factor theta of 1.30, whereas the IGDS-SF9 was more efficient above this point. The IGDS-SF9 nevertheless showed poor fit as a single-factor scale. Under the fitted bifactor model, the three scales shared a model-dependent general response dimension while retaining behavior-specific variance. The AII and RE results are presented as a psychometric demonstration of how the scales conditionally measure shared variance, not as guidance for substituting one behavior-specific instrument for another. Instrument choice should remain determined by the target behavior. Without external criteria, the findings do not support generic screening, diagnostic accuracy, clinical cutoffs, or case classification.