A Mahidol University team translated the Self-Compassion Scale through two forward translations, synthesis, two back-translations, review by the translators plus three additional experts, and a 20-person focus group before analysing 428 Thai adults recruited online. Item 26 loaded only .31 and was removed, leaving 25 items. The four-factor model returned χ²/df=3.40, RMSEA=.075, CFI=.943, TLI=.936 and SRMR=.057; the six-factor model returned χ²/df=3.48, RMSEA=.076, CFI=.943, TLI=.934 and SRMR=.056. Overall α was .92 and single-measure test–retest ICC .901. This supports initial structure and reliability, not diagnostic or predictive validity for mental health.
Key findings
- Item 26 had a standardised loading of .31 and was removed for statistical and Thai-interpretation reasons, leaving 25 items.
- After removal, the four-factor model gave χ²/df=3.40, RMSEA=.075, CFI=.943, TLI=.936 and SRMR=.057, with loadings .43–.86.
- The six-factor model gave χ²/df=3.48, RMSEA=.076, CFI=.943, TLI=.934 and SRMR=.056, with loadings .43–.87; the authors preferred it for consistency with the original scale.
- The 25-item scale had α=.92, subscale α=.71–.84 and overall single-measure ICC=.901 (95% CI .870–.925), supporting internal consistency and short-term stability.
- LCA selected three profiles: low 40 (9.4%), moderate 147 (34.3%) and high 241 (56.3%); these sample-dependent proportions are not population diagnostic cut-offs.
Why this matters globally
A systematic translation enables cross-cultural research and evaluation of Thai wellbeing programmes, while the removed item illustrates that self-kindness does not transfer literally across languages. International comparison requires measurement invariance; clinical use needs convergent, divergent, criterion and responsiveness evidence. One-sample LCA profiles should not classify individuals as diagnostic risk groups.
Thai researcher contribution
Mahidol University's Faculty of Graduate Studies, Ramathibodi and Siriraj psychiatry teams combined child-and-family psychology with clinical psychiatry to produce a transparent Thai adaptation and its 25 items. This gives Thai self-compassion research a more inspectable measurement foundation.
Limitations to consider
Voluntary online convenience recruitment may overrepresent educated and higher-socioeconomic participants; nearly 95% held at least a bachelor's degree, 79.9% were female and ages clustered in mid-adulthood despite a range of 18–73. Religion was not collected and male subgroup power was inadequate. External, convergent, divergent and criterion validity, measurement invariance, DIF and bifactor models remain untested. ICC does not establish long-term stability, and an LCA with entropy .651 from one sample requires replication before cut-offs are used.