Abstract
BACKGROUND: Although the 12-item Short Form Health Survey version 2 (SF-12v2) is suitable for measuring health status in the general Thai population, it has been evaluated using classical test theory. Rasch analysis, however, offers a psychometric testing method that converts ordinal scales to interval-level data without breaching parametric assumptions. Thus, this study aimed to assess the measurement properties of Thai SF-12v2 and SF-6D items derived from it among the general Thai population. METHODS: Rasch analysis was performed on 300 participants randomly selected from a pool of 1200 general Thai population. The initial SF-12v2 items, physical component summary (PCS), mental component summary (MCS), and SF-6D items were evaluated for overall fit to the Rasch model using the Chi-square test and several key elements of Rasch measurement properties, including unidimensionality, reliability (internal consistency), local independence, ordering thresholds, targeting, and item invariance (differential item functioning [DIF]). RESULTS: The two-domain subtests of the SF-12v2 showed an acceptable fit to the Rasch model (χ(2)[12] = 20.16, p = 0.06), unidimensionality, satisfactory reliability (PSI = 0.72), local independence, and no significant DIF. Although five items initially misfitted the Rasch model, these issues were resolved in the subtest analysis. Individual PCS and MCS were not recommended for measuring health status due to inadequate reliability. The SF-6D items demonstrated a good overall model fit (χ(2)[20] = 23.18, p = 0.28), unidimensionality, local independence, and good internal consistency (PSI = 0.73). However, all scales showed poor targeting for the person-item threshold distribution. CONCLUSIONS: Rasch analysis supports the use of two-domain subtests of the SF-12v2 questionnaire for between-group analysis in the general Thai population. However, it does not support the use of individual PCS and MCS due to poor reliability. The SF-6D, achieving satisfactory reliability, was endorsed for eliciting utility scores for between-group analysis.