Psychometric evaluation of the problematic internet use questionnaire in Chinese emerging adults: an exploratory alternative three-factor structure and unidimensional short forms

Introduction Problematic internet use (PIU) is a growing concern among university students. The 18-item Problematic Internet Use Questionnaire (PIUQ-18) is widely used, yet its factor structure remains debated across cultures. This study examined the psychometric properties of the PIUQ-18 and its short forms, the PIUQ-SF-9 and PIUQ-SF-6, among Chinese emerging adults. Methods Participants were 511 Chinese emerging adults (85.50% female). A split-sample design was used, with n = 260 n = 260 for exploratory factor analysis (EFA) and n = 251 n = 251 for confirmatory factor analysis (CFA). We evaluated internal consistency, factor structure, and construct validity, including associations with well-being indicators. Results The PIUQ-18 demonstrated good internal consistency (α = 0.87 α = 0.87, ω = 0.87 ω = 0.87). The original three-factor model produced latent correlations exceeding 1.00 (Obsession–Control = 1.05; Neglect–Control = 1.09), rendering it statistically inadmissible in this sample. An alternative EFA-derived three-factor model was the best-fitting admissible PIUQ-18 model (CFI = 0.91, TLI = 0.90, RMSEA = 0.07, SRMR = 0.07), although its incremental fit remained below conventional benchmarks. HTMT values were below conventional thresholds, but confidence intervals for the latent correlations indicated that the dimensions remained substantially related. For the short forms, the PIUQ-SF-6 demonstrated acceptable unidimensional fit (CFI = 0.97, TLI = 0.95, RMSEA = 0.08, SRMR = 0.05), and the PIUQ-SF-9 showed comparable but weaker fit (CFI = 0.91, TLI = 0.88, RMSEA = 0.11, SRMR = 0.07). The Control Disorder subscale exhibited weaker reliability and validity and requires refinement. Discussion The original three-factor structure of the PIUQ-18 was not statistically admissible in this sample. An alternative EFA-derived three-factor structure showed promise, but its incremental fit was below conventional benchmarks, and the dimensions remained substantially related. The Control Disorder subscale showed weaker reliability and validity, indicating a need for refinement. The PIUQ-SF-6 showed preliminary support for a unidimensional factor structure and theoretically consistent associations with well-being indicators, whereas the PIUQ-SF-9 showed comparable but weaker fit. Overall, these findings should be interpreted as preliminary evidence rather than as evidence establishing screening utility.