Self-reported mindfulness accounts for the effects of mindfulness interventions and nonmindfulness controls on self-reported mental health: A preregistered systematic review and three-level meta-analysis of 146 randomized controlled trials.

Meta-analyses suggest that mindfulness interventions have positive effects on mental health. Yet, how mindfulness interventions exert their effects is still largely unknown. Self-reported mindfulness may partially mediate the association between mindfulness interventions and change in self-reported mental health. We present the results of a novel application of three-level meta-analysis on the pre–post intervention data of 146 RCTs of mindfulness interventions (total N = 10,979), probing the efficacy of a broad range of mindfulness interventions and meditation training against active, treatment-as-usual (TAU), and wait-list control groups. We found that self-reported mindfulness not only increased in mindfulness interventions (d = 0.54, 95% CI [0.47, 0.61]), but also in active (nonmindfulness) controls (d = 0.27 [0.18, 0.36]) and wait-list controls (d = 0.10 [0.04, 0.17]; but not TAU controls: d = 0.04 [−0.03, 0.12]). In addition, change in mindfulness accounted for change in self-reported mental health (mindfulness interventions: d = 0.65 [0.57, 0.73]; active controls: d = 0.49 [0.36, 0.62]; TAUcontrols: d = 0.20 [0.12, 0.29]; wait-list controls: d = 0.22 [0.14, 0.30]) in all treatment and control groups alike. Thus, self-reported mindfulness apparently is no unique mediator of mindfulness interventions. It may either be more universal, merely a correlate of self-reported mental health, or both. Research should focus on the common denominator of mindfulness interventions and clinically relevant constructs with which self-reported mindfulness shares some of its characteristics. Limitations pertain to the indirect evidence of the three-level meta-analytic approach, the self-report nature of the data, and small-study effects, which suggest the presence of publication bias. The risk of bias may have led to the overestimation of effects and results could further be subjected to effects of shared method variance