Self-compassion may be widely discussed as a single psychological strength, but new evidence suggests that the popular 12-item measure used to assess it may be capturing several closely related traits rather than one unified quality. In a study of 1,389 adults in the United States, researchers tested competing explanations for how the Self-Compassion Scale–Short Form is structured. Their analysis found that a six-factor model—representing self-kindness, self-judgment, common humanity, isolation, mindfulness, and over-identification—provided the strongest overall fit. A simpler two-factor model, separating positive and negative dimensions, also performed adequately. By contrast, both a higher-order model and a single-factor model showed poor fit, raising questions about whether a single total score can fully represent how people relate to themselves during difficult experiences.
The findings, published in Mindfulness, are important because the Self-Compassion Scale is used extensively in psychological research, clinical studies, and intervention trials. Self-compassion is generally defined as responding to personal suffering with understanding rather than harsh self-criticism, recognizing that hardship is part of the shared human condition, and maintaining balanced awareness of painful thoughts and emotions. The concept has been associated with lower depression, anxiety, perceived stress, and self-harm, as well as stronger emotional regulation and psychological well-being. Yet the measurement debate surrounding the scale has persisted for years. Researchers have disagreed over whether its positive and negative items should be combined, whether the six components are truly distinct, and whether the resulting total score can be interpreted as a single latent psychological trait.
The short-form scale was developed by Raes and colleagues in 2011 as a less burdensome alternative to Neff’s original 26-item instrument. It contains 12 questions, with two items assigned to each of six components. The positive dimensions are self-kindness, common humanity, and mindfulness. The negative dimensions are self-judgment, isolation, and over-identification; these items are reverse-scored when calculating an overall self-compassion score. For example, the positive items reflect treating oneself with care, remembering that personal difficulties are not unique, and maintaining perspective during distress. The negative items assess harsh self-evaluation, feeling cut off from others, and becoming overwhelmed or absorbed by painful thoughts and emotions. Because each component is represented by only two items, the scale is efficient but statistically challenging to evaluate.
To examine its structure, the researchers conducted confirmatory factor analysis, a theory-driven statistical method that tests whether observed responses conform to predefined models. The study compared four possibilities. The first treated the six components as correlated but separate factors. The second placed those six factors beneath one general self-compassion factor in a higher-order structure. The third grouped the items into two correlated factors—positive and negative self-compassion. The fourth assumed that all 12 items measured one general factor. Analyses used robust maximum likelihood estimation to address non-normality in the responses, with weighted least squares estimation used as a sensitivity check because the items were measured on a five-point ordinal scale. Both analytical approaches produced a similar ranking of the models.
The six-factor correlated model showed excellent descriptive fit under the primary analysis, with a robust comparative fit index of 0.99, a root mean square error of approximation of 0.04, and a standardized root mean square residual of 0.03. The chi-square test was statistically significant, but the researchers noted that chi-square is highly sensitive to large samples and can indicate misfit even when other indices suggest that a model describes the data well. The standardized item loadings were generally strong, ranging from 0.62 to 0.86 across the six components. However, the apparent success of the six-factor model came with an important warning: some latent factors were so highly correlated that their practical distinctiveness became uncertain. Isolation and over-identification correlated at 0.99, while self-kindness correlated at 0.94 with mindfulness and at 0.91 with common humanity.
These high correlations suggest that the six-factor solution may fit the data mathematically without proving that the six dimensions represent fully independent psychological processes. In psychometric terms, factors that correlate above approximately 0.80 may have limited discriminant validity, meaning they may not be sufficiently distinct from one another. The researchers therefore describe the six components as closely related facets rather than isolated traits. The two-factor model offered a more compact alternative. It produced a robust comparative fit index of 0.95, a root mean square error of approximation of 0.07, and a standardized root mean square residual of 0.05. The positive and negative factors correlated at 0.40, indicating that they were related but not excessively overlapping. Its relative simplicity may make it attractive for studies focused on broad self-compassionate versus self-critical tendencies, although some fit indices were weaker than those of the six-factor model.
The higher-order model performed poorly, despite its previous use in research on both the short and long versions of the scale. Its comparative fit index was 0.82, its root mean square error of approximation was 0.14, and its standardized root mean square residual was 0.14. The one-factor model performed even worse, with a comparative fit index of 0.68, a root mean square error of approximation of 0.18, and a standardized root mean square residual of 0.16. These results indicate that the 12 responses could not be adequately explained by one common underlying factor in this sample. The distinction is crucial: a scale can have high internal consistency while still failing to be unidimensional. In this study, the total score had strong reliability, with a McDonald’s omega of 0.93, but hierarchical omega was only 0.51, suggesting that much of the reliable variance came from the component factors rather than a dominant general self-compassion factor.
The researchers also examined how the scale’s scores related to perceived stress, depression and anxiety symptoms, and compassion for other people. The total self-compassion score correlated negatively with perceived stress at approximately −0.61 and with symptoms measured by the Patient Health Questionnaire-4 at approximately −0.49. These associations are consistent with the idea that people who report more self-compassion tend to experience less stress and fewer symptoms of depression and anxiety. However, the individual dimensions told a more differentiated story. Over-identification, isolation, and self-judgment showed stronger relationships with perceived stress and mental-health symptoms than self-kindness, common humanity, or mindfulness. Over-identification correlated −0.61 with perceived stress and −0.59 with PHQ-4 scores, while common humanity correlated only −0.15 with stress and −0.04 with the PHQ-4.
This pattern suggests that the absence of self-critical and distress-amplifying tendencies may be more closely connected to mental-health outcomes than the presence of positive self-compassionate qualities alone. That interpretation does not establish causation: the study was primarily cross-sectional for these analyses, and the PHQ-4 was collected at follow-up approximately one month after the baseline self-compassion assessment. Changes in symptoms, recall effects, and other measurement differences could therefore influence the correlations. Compassion for others showed a different pattern. The total self-compassion score had only a small positive correlation with outward compassion, approximately 0.13. Positive self-compassion components such as self-kindness and common humanity correlated around 0.29 with compassion for others, whereas the negative components showed very small or near-zero relationships. The findings imply that being compassionate toward others may overlap with some positive ways of relating to oneself, but it is not simply the same psychological capacity.
The study drew on baseline data from the seventh cohort of a larger longitudinal project conducted during the early months of the SARS-CoV-2 pandemic. Participants were recruited online through Amazon Mechanical Turk and Qualtrics panels in June 2020, a period marked by public-health uncertainty, economic disruption, and heightened national attention to racial injustice after the killing of George Floyd. The sample was intentionally more racially diverse than many earlier validation samples, including 280 Black or African American participants and 315 Hispanic or Latino participants, although it remained majority White and relied on convenience sampling. Exploratory comparisons found small differences in total self-compassion scores across racial groups, with Black participants reporting higher scores than White participants in some comparisons. No significant difference emerged by Hispanic or Latino ethnicity. Because the researchers did not conduct formal measurement-invariance testing, these group comparisons cannot determine whether the scale functioned equivalently across populations.
The authors caution that the results should not be interpreted as evidence that researchers must abandon the total SCS-SF score. Rather, they argue that total scores should be supplemented with subscale or positive-versus-negative composite scores, especially when studying stress, depression, anxiety, or intervention effects. The study also has important limitations, including the exclusion of roughly one-third of consenting participants because of incomplete item responses. Those excluded differed from the analytic sample in age, gender, race, marital status, income, and education, potentially affecting generalizability. The two-item subscales also place an unavoidable ceiling on reliability and precision. Future research using longer item sets, representative samples, formal invariance testing, longitudinal designs, and alternative models such as exploratory structural equation modeling could clarify whether the six dimensions are culturally stable and conceptually separable. For now, the central message is clear: self-compassion may be better understood as a network of related capacities—including both supportive and self-critical tendencies—than as a single number.
Subject of Research: The factor structure, reliability, and construct validity of the 12-item Self-Compassion Scale–Short Form in a racially diverse English-speaking U.S. adult sample.
Article Title: Confirmatory Factor Analysis of the Short Form Self-Compassion Scale in an English-Speaking U.S. Sample
Article References: Manjanatha, D., Karnaze, M. M., Roesch, S., et al. “Confirmatory Factor Analysis of the Short Form Self-Compassion Scale in an English-Speaking U.S. Sample.” Mindfulness (2026). References include Neff (2003), Raes et al. (2011), Muris and Petrocchi (2017), and Muris and Otgaar (2020).
Image Credits: AI Generated
DOI: 10.1007/s12671-026-02962-z
Keywords: Self-compassion; Self-Compassion Scale–Short Form; confirmatory factor analysis; psychometrics; perceived stress; depression; anxiety; mental health; positive psychology; measurement validity
