Setting the Record Straight About the Self-Compassion Scale

Recently, Muris and colleagues published a letter to the editor of the journal Mindfulness with the colorful title “Stripping the forest from the rotten trees: Compassionate self-responding is a way of coping, but reduced uncompassionate self-responding mainly reflects psychopathology.” (Muris et al. 2018). It was written in response to a journal article I wrote with colleagues (Neff et al. 2018a) titled “The forest and the trees: Examining the association of self-compassion and its positive and negative components with psychological functioning,” and contains some seriously erroneous claims. The Self-Compassion Scale (SCS; Neff 2003) measures self-compassion as a system-level balance between compassionate self-responding (kindness, common humanity, and mindfulness) and reduced uncompassionate self-responding (reduced self-judgment, isolation, and over-identification). Neff et al. (2018a) examined the link between scores on the SCS and well-being in a variety of domains: psychopathology, positive psychological health, emotional intelligence, self-concept, body image, motivation, interpersonal functioning, and markers of sympathetic nervous system activity after stress. The study used a total SCS score, six subscale scores, and also a mean score of the three positive and three negative components to examine the association of self-compassion with outcomes. Although my research does not find that two separate positive and negative factors of self-compassion are supported psychometrically and in fact the two cannot be clearly distinguished because they have so much overlap (see below; Neff 2018b), we calculated mean scores for the three positive and three negative subscales in order to better investigate claims that inclusion of the negative components in the SCS inflates the link between self-compassion and psychopathology (Muris and Petrocchi 2017). Given the deep intertwining of the various components in the operation and measurement of self-compassion, however, we used zero-order correlations rather than regressions or partial correlations in our analyses because to separate out their shared variance would change the meaning of components in a way that would render findings uninterpretable. …