Participation in Online Mindfulness was Associated with Improved Self-Compassion and Prosocial Attitudes

Participants were recruited through a free 4-week online mindfulness program delivered by Monash University through the UK-based online course provider, FutureLearn. Of the 12,095 participants who registered to participate in this run of the online mindfulness program (in early 2021), 1043 (8.6%) completed more than 90% of the program 6 weeks after it had commenced. A total of 2006 participants completed the pre-program survey and 320 completed the post-program surveys. A total of 111 cases that had post-program data but were not able to be matched with pre-program responses were excluded. The analysis included 209 participants whose post-program responses could be confidently matched with pre-program responses. The age of these participants ranged from 18 to 85 years, and most participants were aged 56–65 years (see Table 1). Of the 209 participants, 171 (81%) identified as female and 38 (19%) identified as male (Table 1). The nationality of participants varied, with 34 different nationalities represented in the study population, with the majority being British (111 participants), followed by Australian (23 participants), Canadian (8 participants), and American (7 participants). Prior mindfulness experience also varied among the study population, with 63% stating they had “some experience” (131 participants) prior to starting the program, and 34% stating they had “no past experience” (71 participants).

The Mindfulness for Wellbeing and Peak Performance open online course contained an information page at the beginning and at the end of the program, inviting participants to complete pre- and post-program surveys and to choose whether or not to provide consent for the use of their data in research. Participation in this study was not required for entry or completion of the online mindfulness program. A plain language statement was provided to participants detailing the aims of the study, processes to ensure anonymity, and the voluntary nature of the study. The statement explained that the study would examine the impact of mindfulness on self-compassion and prosocial behavior and provided a definition for each concept. The pre- and post-program surveys were identical (details below). Participants were asked to generate anonymous participant identification codes (based on birth month and first three letters of their mother’s maiden name) to enable us to match pre-program responses with post-program responses, while preserving anonymity. After excluding incomplete and ambiguous codes, some cases with non-unique codes were further identified by age, nationality, and/or gender. Cases were included if the combination of these variables provided a unique match for pre- and post-responses. Participants’ responses to individual questionnaires were excluded from analyses involving that questionnaire if < 80% of items were completed.

The online mindfulness program was coordinated by medical and psychological professionals who have extensive knowledge and skills in mindfulness and mindfulness training. It was designed and run by two of the authors of this paper (CH and RC). The program focused on a different mindfulness theme each week, mostly building upon the skills and teachings of the previous week. The themes were as follows: (1) Introduction to Mindfulness, (2) Mindfulness and Mindful Stress Reduction, (3) Mindfulness to Improve Work and Study Performance, and (4) Mindfulness as a Way of Life. Program work was asynchronous (i.e., delivered via prepared resources accessible to participants at their convenience, without real-time teacher-led interaction) and comprised a combination of short explanatory videos, text, links to articles and online resources, downloadable guided mindfulness meditations, weekly feedback videos, quizzes to test knowledge, and forum discussions, all of which took most learners approximately 3 hr per week to complete. The forum was facilitated by a psychology-trained mentor (SC) who provided support, feedback, answers, and guidance to participants daily. The discussions, questions, and themes generated in the forum were the basis for weekly feedback videos generated by the program authors (CH and RC). Participants were encouraged to progress through the program 1 week at a time over 4 weeks; however, participants had access to all program materials for 2 additional weeks. Survey responses were collected pre- and post-program using an online Qualtrics survey via links embedded in the program materials at the beginning of week 1 and the end of Week 4 of the online mindfulness program. There was no explicit self-compassion or prosociality training, beyond encouraging an attitude of gentleness during one of the meditation practices.

All measures were assessed both pre- and post-program. Age, nationality, gender identity, and past mindfulness experience were recorded at pre-program. Age was assessed categorically by using age brackets starting with 18–25 years and increasing by increments of 10 years thereafter. Past mindfulness experience was assessed categorically by asking participants to select from 3 possible responses (No past experience, Some past experience, and Very experienced). These measures were also assessed at post-program to provide additional identification items to assist with matching pre- and post-program responses if ambiguity in ID code matching occurred.

The Mindfulness Adherence Questionnaire (MAQ) (Hassed et al., 2021) consists of 12 questions that assess the quality and quantity of participants mindfulness practice over the preceding week. The questionnaire assesses participants’ formal and informal practice independently. The first two questions provide information on the frequency and duration of formal meditation practice. The remaining 10 questions use a 6-point Likert scale from 0 (never) to 6 (always) to assess the quality of both formal and informal mindfulness practice. For the formal scale, an example item was: “When meditating, how much of the time was your attention focused on what you intended to focus on (body, breath, sounds, etc.)?”, and for the informal scale an example item was: “In your daily life, how much of the time were you practicing being aware of your thoughts, emotions and reactions?” The scores are summed to provide scores for quality of formal practice, quality of informal practice, and quality of total mindfulness practice. The scale has demonstrated good validity and reliability: in Week 4 of a mindfulness course for the Formal subscale Cronbach’s α = 0.87, and for the informal subscale Cronbach’s α = 0.93 (Hassed et al., 2021).

The 24-item version of the Five Facet Mindfulness Questionnaire (FFMQ-24; Bohlmeijer et al., 2011) assesses trait mindfulness using a 5-point Likert scale ranging from 1 (never or rarely true) to 5 (very often or always true). The FFMQ-24 consists of 24 statements describing everyday experiences of mindfulness, such as “I pay attention to physical experiences, such as the wind in my hair or the sun on my face.” The FFMQ-24 score is calculated by reverse-scoring 14 items that describe mindless behavior, for example “I find myself doing things without paying attention” and then summing responses. Higher scores indicate a higher level of trait mindfulness. The FFMQ-24 has demonstrated adequate internal consistency and convergent validity, with Cronbach’s α ranging from 0.75 for the non-reactivity subscale to 0.87 for the describing subscale (Bohlmeijer et al., 2011; Brady et al., 2019).

The Self-Compassion Scale (SCS; Neff, 2003) assesses how people act towards themselves during different situations through statements such as “I try to be loving towards myself when I’m feeling emotional pain.” This 26-item questionnaire uses a 5-point Likert scale ranging from 1 (almost never) to 5 (almost always). The statements are designed to address the three components of self-compassion, and their counterpart dimensions (self-kindness vs self-judgement, common humanity vs isolation, mindfulness vs over-identification). The negative traits (self-judgement, isolation, and over-identification) are reverse-scored before computing subscale means for each, which are summed with the means from the remaining subscales to create a grand mean. The higher the grand mean, the greater the level of self-compassion. The SCS has demonstrated good validity and reliability, with Cronbach’s α = 0.93 (Neff, 2003; Neff & Pommier, 2013; Neff et al., 2007). Since its original development, the SCS has also had its factor structure confirmed as well as its psychometrical equivalence across different age, cultural, gender, and language groups (Neff et al., 2019; Tóth-Király & Neff, 2021).

The Prosociality Scale for Adults (PSA; Caprara et al., 2005) assesses participants’ levels of prosocial behavior in response to a variety of everyday situations, for example, “I am willing to make my knowledge and abilities available to others.” This 16-item scale uses a 5-point Likert scale ranging from 1 (never/almost never) to 5 (always/almost always) and was selected for its brevity in comparison with other prosociality scales. The scores are summed to create a total score, with higher scores indicating greater prosociality. The English version of the PSA scale has demonstrated good validity, reliability, and sensitivity across both a large range of prosociality and to small differences between individuals (Caprara et al., 2005), with a Cronbach’s α = 0.93 (Kanacri et al., 2021).

Imputation for missing data was conducted only if ≥ 80% of the data from a specific questionnaire were available; if less than 80% responses were recorded, the cases were excluded from analyses. The case mean substitution method was applied for imputing missing scores for individual items within a scale at the pre-program timepoint, which has been demonstrated to improve statistical power in a study while not biasing the values of the measures (Bono et al., 2007; Roth et al., 1999). Applying this method of imputation for missing values has been demonstrated to provide good estimates of measure reliability for scales similar to those used in this study, and the method has been used before in large-scale online mindfulness research (Bailey et al., 2018; Bono et al., 2007; Downey & King, 1998). The baseline observation carried forward (BOCF) method was applied when there was a missing item in post-program responses that were case matched with pre-program responses. The BOCF method is a conservative approach that assumes no change from baseline for the missing data. This method avoids potential inflation of results due to imputation of missing data by assuming missing items have not changed from baseline (even if other items in the scale have all changed in a similar direction) and thus reduces the potential impact of imputation as a confound. Missing items from the MAQ questionnaire were not imputed because both formal and informal MAQ sub-scales contain too few items to obtain a reasonable estimate from other scale items in the case of missing items. For participants who had missed an item in the MAQ questionnaire, the subscale which this item corresponded to was excluded from the analysis. In total, 30 items were imputed across three of the scales at the pre-program timepoint, and 38 items were imputed across four scales at the post-program timepoint. Participants with missing items on more than 20% of any particular scale were excluded from analyses that involved that scale (4 participants for the MAQ, 1 participant for the FFMQ-24, 6 participants for the PSA, 6 participants for the self-compassion scale).

Statistical analyses were performed in JASP v0.13 (Love et al., 2019). To test hypotheses H1, H2, and H3, separate repeated-measures ANOVAs were conducted to assess pre- to post-program changes in self-compassion, prosociality, and trait mindfulness. Age and gender were included as between-subject factors (since age was measured using a categorical approach). An additional ANCOVA with the same design, but with previous mindfulness experience included as a covariate, was conducted to assess whether including participants’ previous mindfulness experience in the analysis altered the pattern of results. To address hypothesis H4, separate univariate linear regression models were utilized in a stepwise manner to test whether the mindfulness adherence measures reported at post-program (MAQ quantity of formal practice, quality of formal practice, and quality of informal practice) predicted changes in trait mindfulness, self-compassion, and prosociality. The Benjamini and Hochberg (1995) False Discovery Rate (FDR) method was used to control for multiple comparisons across all planned primary hypotheses and exploratory hypotheses (the main effects of time for the ANOVAs assessing trait mindfulness, prosociality, and self-compassion, and the overall regression model for predicting changes in prosociality, trait mindfulness, and self-compassion, the interaction between gender and time for prosociality, and the initial mediation analysis). In order to enable comparison with other research, both corrected and non-corrected p-values are reported (FDR-p and p, respectively). A priori power analyses were not performed, as the online study of an existing mindfulness program did not provide us with the ability to control the number of participants that were recruited. Instead, Bayesian analyses were also performed for the repeated-measures ANOVA and regression design tests to provide an indication of the degree of evidence the data provides to suggest we should update our beliefs from the null hypothesis (of no effect) to the alternative hypothesis (that participating in the mindfulness course changed scores from pre to post). BF10 values are reported for the most likely model, and BFincl values are reported, which reflect the ratio of evidence for the alternative hypothesis within a specific main effect or interaction to the evidence supporting the null hypothesis, when comparing the evidence for models that include a main effect or interaction of interest against the models that do not include that main effect or interaction of interest. BFincl values of 1–3 can be interpreted as weak evidence, BFincl values of 3 to 20 reflect positive evidence, BFincl values of 20 to 150 reflect strong evidence, and BFincl values > 150 reflect extreme evidence in support of the alternative hypothesis (Raftery, 1995).

To test our exploratory hypotheses: (E1) we assessed the interaction between timepoints and gender within the repeated-measures ANOVA that tested the effect of the program on prosociality, and (E2) we assessed the interaction between timepoints and age in the repeated-measures ANOVA that tested the effect of the program on prosociality. For exploratory hypotheses E1 and E2, post hoc t-tests were planned to explore the individual comparisons between specific gender or age groups and the timepoint factor to determine the cause of significant interactions between timepoint and gender. The Bonferroni-Holm method was used as a multiple comparison control across these t-tests (reported as p-holm). Finally, to test exploratory hypothesis E3, two separate mediation analyses were conducted using the bootstrap method (10,000 replications) in the lavaan package in JASP to examine whether the relationships between the quantity and quality of mindfulness practice and changes observed in self-compassion and prosociality outcome measures would be mediated by observed changes in trait mindfulness. Effect sizes are reported using the generalized eta squared \(\left(\eta^2G\right)\) and interpreted following Cohen (1988), so that small = 0.01, medium = 0.06, and large = 0.14. Partial eta-squared \(\left(\eta^{P^2}\right)\)  values are also reported to enable comparison with other research.

Table 2 presents results from analyses of variance used to address our first aim. Regarding the analysis of trait mindfulness, in support of hypothesis H1, there was a small (bordering on medium) but significant main effect of time on mindfulness (FFMQ-24), indicating that mindfulness increased from pre- to post-program (Fig. 1): F_(1,194) = 56.208, p < 0.001, FDR-p = 0.002, η²G = 0.057, ηP² = 0.225, BFincl = 2.177e + 30 (extreme evidence). The model including only the main effect of time was the most likely model, with BF10 = 1.782e + 30.

As illustrated in Fig. 2, there was a small main effect of previous mindfulness experience. Participants who reported they were very experienced prior to the program reported higher levels of trait mindfulness than participants who reported some or no prior mindfulness experience: F_(1,193) = 4.595, p = 0.033, η²G = 0.018, ηP² = 0.023. Inclusion of experience as a covariate reduced the effect size of the main effect of time, but the main effect was still significant F_(1,193) = 13.253, p < 0.001, η²G = 0.014, ηP² = 0.064. There were no other significant main effects or interactions involving time, age, or gender for trait mindfulness or for past mindfulness experience and trait mindfulness (all p > 0.05, all BFincl < 0.54).

Regarding the analysis of the self-compassion measure, in support of hypothesis H2, there was a small but significant main effect of time on self-compassion, indicating that self-compassion increased from pre- to post-program: F_(1,189) = 41.229, p < 0.001, FDR-p = 0.002, η²G = 0.031, ηP² = 0.179, BFincl = 2.727e + 24 (extreme evidence) (Fig. 1). The model including only the main effect of time and main effect of age was the most likely model, with BF10 = 4.286e + 24. There was also a medium-sized significant main effect of age, with older age groups showing progressively higher scores in the self-compassion scale F_(6,189) = 2.400, p = 0.029, η²G = 0.061, ηP² = 0.071, BFincl = 2.04. However, this effect was not explored further, as we had no hypotheses related to this finding and only report it for the potential interest of readers. There were no other significant main effects or interactions between time, age, and gender for trait mindfulness or for past mindfulness experience and self-compassion (all p > 0.05, all BFincl < 0.35, with the exception of an interaction between age and gender, BFincl = 1.24 and p = 0.073).

Regarding the analysis of the prosociality measure, in support of hypothesis H3, there was a small but significant main effect of time on prosociality, indicating that prosociality increased from pre- to post-program: F_(1,189) = 8.212, p = 0.005, FDR-p = 0.008, η²G = 0.005, ηP² = 0.042, BFincl = 73.888 (strong evidence) (Fig. 1). The Wilcoxon rank test indicated this result was still present when the violation of the assumption of normality did not influence the analysis: Wilcoxon test statistic = 5232.5, p < 0.001, matched rank biserial correlation effect size = 0.378. In support of exploratory hypothesis E1, there was also a small but significant interaction between time and gender for prosociality: F_(1,189) = 4.829, p = 0.029, FDR-p = 0.0331, η²G = 0.003, ηP² = 0.025, BFincl = 1.372 (weak evidence). The model including the main effect of time, main effect of gender, and interaction between time and gender was the most likely model, with BF10 = 246.428. Post hoc t-tests indicated that the significant interaction was due to a significant difference between pre-test scores and post-test scores for males t =  − 2.860, p-holm = 0.028, Wilcoxon test statistic = 57.50, p < 0.001, matched rank biserial correlation effect size = 0.795, but no statistically significant difference between pre and post scores for females (p-holm = 1.00), and no difference between the genders at the pre-program time-point (p-holm = 0.670) or post-program (p-holm = 1.00) (Fig. 3).

However, while the exploratory post hoc t-tests were controlled for multiple comparisons, the interaction analysis within the repeated-measures ANOVA was not included in controls for multiple comparisons, so this result should be considered tentative. No other main effects or interactions were significant; all p > 0.05, all BFincl < 0.52, with the exception of the main effect of gender, which provided a BFincl = 2.188 (weak evidence), although the p-value was non-significant (p = 0.484), and the influence of previous mindfulness experience when included as a covariate was not significant (p = 0.456).

In partial support of hypothesis H4a, the quality of formal practice in the final week of the mindfulness program was a significant predictor of changes in trait mindfulness scores. The overall model showed significance: F_(1,201) = 15.793, p < 0.001, FDR-p = 0.002, and the contribution of quality of formal practice as a covariate showed significance: t = 3.974, p < 0.001 (Fig. 4), BF10 = 204.806 (extreme evidence). Trait mindfulness increased by 0.594 (95% confidence intervals = 0.299 to 0.889) for each unit increase in quality of formal mindfulness practice. The 7% of the variance in change in trait mindfulness scores was accounted for by formal practice quality (adjusted R² = 0.069). No other MAQ predictors significantly predicted changes in trait mindfulness scores (including quality of informal practice and quantity of practice).

In partial support of hypothesis H4b, the quality of formal practice in the final week (MAQ) was also a significant predictor of changes in self-compassion scores. The overall model showed significance: F_(1,197) = 7.371, p = 0.007, FDR-p = 0.008, and the contribution of the quality of formal practice as a covariate showed significance: t = 2.715, p = 0.007, BF10 = 4.661 (positive evidence) (Fig. 4). Self-compassion scores increased by 0.023 (95% confidence intervals = 0.006 to 0.039) for each unit increase in quality of formal mindfulness practice. Just 3% of the variance in changes in self-compassion score was accounted for by formal practice quality (adjusted R² = 0.031). No other MAQ predictors showed significant prediction of changes in self-compassion scores (including quality of informal practice and quantity of practice).

In contrast to the results for trait mindfulness and self-compassion, hypothesis H4c was not supported—no MAQ variables of quantity or quality of mindfulness practice significantly predicted changes in prosociality scores (p > 0.05, all BF10 < 0.50). To assess whether the lack of association between quality of informal practice/quantity of practice and trait mindfulness/self-compassion and prosociality was due to distributions that did not represent a sufficient spread of these variables, violin plots were generated (Fig. 5). These plots suggested that a sufficient spread was present for associations to be detected if present. Additionally, in order to assess a proposed post hoc explanation that practice quality increased trait mindfulness, and then might have affected prosociality indirectly changes in trait mindfulness, we performed a Spearman’s correlation between changes in FFMQ-24 scores and changes in prosociality. This showed no significant relationship (Spearman’s rho = 0.105, p = 0.134).

The quality of formal mindfulness practice in the final week of the program predicted changes in self-compassion, so a mediation analysis was implemented to examine whether changes in trait mindfulness mediated this relationship. Results partially supported exploratory hypothesis E3. An indirect effect was identified, whereby changes in trait mindfulness mediated the relationship between quality of formal practice and changes in self-compassion (z = 3.702, p < 0.001, FDR-p = 0.002, 95% CIs = 0.010 to 0.031), and the total effect showed significance (z = 2.742, p = 0.006, 95% CIs = 0.006 to 0.040) (Fig. 6). However, with changes in trait mindfulness included as a mediator, the direct effect of quality of formal mindfulness on self-compassion was not significant (z = 0.564, p = 0.572, 95% CIs = –0.008 to 0.017). Other orderings of the mediation model (with changes in self-compassion specified as the mediator rather than the outcome, and changes in trait mindfulness specified as the outcome) were not significant, with all indirect p-values > 0.05. It is worth noting that the indirect effect often has more statistical power than the total effect, and while there is disagreement in the research, some experts have suggested it is valid to conduct tests of the indirect effect in the absence of a direct effect when a logically justified hypothesis exists with regard to the indirect effect, particularly when a total effect is also significant (Agler & De Boeck, 2017). However, it is also worth noting this hypothesis was exploratory and not controlled for multiple comparisons, so should be viewed tentatively. No measures of mindfulness practice quantity or quality predicted changes in prosociality in the regression analysis, so a mediation analysis was not conducted for this outcome variable.

The results of this study contribute to the growing body of evidence supporting the efficacy of online mindfulness programs for improving trait mindfulness. Spijkerman et al. (2016) conducted a meta-analysis of 15 RCTs and demonstrated that online mindfulness programs produced a small but significant effect on mindfulness (Hedge’s g = 0.32), with the greatest effect observed when online mindfulness programs involved interactions between mindfulness providers and participants (where teachers could offer specific guidance for participants, as per the program examined in the current study). This is in keeping with other studies that demonstrated beneficial psychological effects after short online mindfulness programs (Bailey et al., 2018; Cavanagh et al., 2013; Mak et al., 2015). Our results provide further support for the implementation of online mindfulness interventions for mental health, as increased trait mindfulness has been proposed as a primary mechanism of action underpinning the mental health benefits of mindfulness interventions, and changes in trait mindfulness have also been demonstrated to predict decreased depression in meta-regression (Linardon, 2020).

Our results also demonstrated that self-compassion increased during an online mindfulness program (with a small effect size). These results are consistent with a meta-analysis of online mindfulness interventions, which also report small effect sizes (Hedge’s g = 0.29) (Linardon, 2020). Our results also suggest that online mindfulness interventions are associated with improved self-compassion, even without explicit self-compassion teaching or practice, congruent with a meta-analysis by Berry et al. (2020). There is, however, uncertainty surrounding the sustained effects of online mindfulness programs on self-compassion, with some research showing self-compassion levels returned to baseline at 2 months post intervention (Halamová et al., 2018). As such, the results from this study support the use of online mindfulness programs to improve self-compassion; however, there is a need for further research into the lasting effects of mindfulness interventions with implicit self-compassion training.

In addition to the effects on trait mindfulness and self-compassion, the online mindfulness program demonstrated a beneficial effect on prosociality with a small effect size. The improvement in prosociality associated with participating in the mindfulness program is of particular interest because it occurred in an online mindfulness program which did not provide the possibility of direct face-to-face interactions with other participants and facilitators, in contrast with traditional face-to-face mindfulness group interventions. Despite the lack of direct face-to-face contact in the program, our results demonstrated that online mindfulness training can be associated with changes in prosociality. This study’s results are in line with meta-analyses of in-person mindfulness interventions that have demonstrated mindfulness interventions produced similar improvements in prosocial behavior when compared to programs specifically aimed at improving prosociality, although in-person studies showed a larger effect size than this study (meta-analysis reported d = 0.51) (Donald et al., 2019).

Interestingly, in addition to the overall increase in prosociality, the results indicated the pre-post change in prosociality was only significant for males. Despite the stereotypical belief that females are more prosocial than males in the general population, the issue of gender differences and prosociality remains unresolved (Eisenberg et al., 2015). It is understood that males and females are both likely to act in a prosocial manner; however, the situations which result in these behaviors differ greatly depending upon gender (Eisenberg et al., 2015). Males tend to display prosocial behaviors that are more “heroic,” requiring greater risk taking as opposed to females who display altruistic and emotional prosocial behavior (Eagly, 2009; Eagly & Crowley, 1986). These differences in prosociality due to gender may arise from the social expectations based on gender norms (Caprara et al., 2005; Eisenberg et al., 2015). Despite this, the Prosociality Scale for Adults scale used in the current research is thought to be able to detect improvements in males and females equally well (Caprara et al., 2005). However, the interaction between timepoint and gender in our study was not a primary analysis, our sample size of male participants was small, and the Bayesian analysis only suggested weak evidence in support of the interaction. Additionally, the lack of a significant difference between groups at baseline suggests a simple ceiling effect for females does not explain our result, and no other obvious explanation presents itself. As such, this result should be viewed as tentative and in need of further exploration.

The results of this study demonstrated that the quality of formal practice in the final week of the program predicted the improvements in trait mindfulness and self-compassion. These results support preliminary suggestions that the quality of mindfulness meditation predicts improvements in well-being more than the duration (Del Re et al., 2013; Goldberg et al., 2014) (however, see Bailey et al. (2018) and Spijkerman et al. (2016) for results suggesting that the frequency of formal mindfulness practice is an important predictor of changes in well-being). Findings suggesting the quality of practice is a more significant predictor of outcomes are of particular importance because participant time commitments have been highlighted as one of the main barriers to practicing mindfulness (Birtwell et al., 2019; Wyatt et al., 2014). Traditional face-to-face mindfulness interventions require significant time commitment from not only participants but also facilitators, and time demands are cited as a primary cause of non-participation (Hanley et al., 2016). The use of brief online mindfulness programs circumvents the barriers faced by many when contemplating participation in a traditional mindfulness program. Participants could access facilitators and content at a time in their schedule that suits them, and conversely the facilitators can engage with significantly more participants than in face-to-face programs. As such, if further research suggests that quality is more important than quantity of formal mindfulness practice, the case would be strengthened for online mindfulness interventions as a low-intensity and high-reach method to improve mental health.

Although formal mindfulness quality was found to be a predictor of changes in trait mindfulness and self-compassion, informal mindfulness quality was not. These results align with some previous research which has demonstrated formal meditative practice is related to the magnitude of benefits observed in well-being or symptom reduction, while informal practice was not (Crane et al., 2014; Hawley et al., 2014). However, other research has reported that the amount of informal mindfulness practice participants completed was a stronger predictor of improved trait mindfulness when compared to the duration of formal practice (Brown & Ryan, 2003). Our results provide evidence that discriminates between these suggestions, and suggests formal practice is more important. In particular, the questions assessing informal mindfulness practice as measured by the MAQ have conceptual similarities to trait mindfulness measured with the FFMQ-24 (as both refer to mindfulness in everyday life), while the questions in the formal practice scale are restricted to self-reports about formal practice periods. As such, if our results were driven by conceptual similarities alone, it could be expected that the MAQ informal practice quality scale would be more strongly related to changes in trait mindfulness than the measure of formal practice quality. In contrast to this expectation, formal mindfulness quality was related to changes in trait mindfulness, while informal practice quality was not. In this context, the findings of our results suggest that formal practice quality is more important for outcomes, since it was related to changes in trait mindfulness, which are proposed to be a primary mechanism of the effect of mindfulness interventions on measures of well-being.

In addition to the predictive relationship between formal mindfulness practice quality in the final week of the program and changes in trait mindfulness and self-compassion, our mediation analysis suggested a potential causal pathway—that higher quality of formal mindfulness practice led to increases in trait mindfulness, and that the increased trait mindfulness increased participants’ self-compassion. This is aligned with suggestions as to the mechanism of action of mindfulness interventions, research showing that changes in mindfulness predicts changes in self-compassion in face-to-face mindfulness interventions (Birnie et al., 2010), and theoretical perspectives that posit a less reactive thought style implicitly leads to increased more self-compassion (Golden et al., 2021). However, while the mediation analysis suggested the relationship between quality of formal practice and changes in self-compassion was mediated by changes in trait mindfulness, the format of a mediation analysis only provides evidence against the null hypothesis (i.e., that there is no mediating relationship). The statistical test unfortunately does not directly imply support for the alternative hypothesis that changes in mindfulness mediate the relationship, so our results provide only very preliminary support for a hypothesis of a potential causal relationship (Agler & De Boeck, 2017). As such, while the mediation relationship was present, it did not provide good support for causation. However, the finding does provide evidence against the possibility that poorer mindfulness quality (with more struggles to maintain attention during mindfulness practice) could be where the gains are made in terms of mindfulness (with difficult formal mindfulness sessions being the equivalent of lifting heavier weights, leading to stronger effects). It seems this possibility is not a good fit for the current results, and that higher quality of mindfulness practice is at least associated with more gains in self-compassion.

It was also interesting to observe that the quality of formal practice predicted changes in mindfulness and self-compassion, but the quantity (i.e., amount of practice time in the preceding week) did not. One potential explanation we considered was whether the distribution of practice times might be clustered around the low end of the scale, with a majority of participants reporting less than 50 min of practice in the preceding week. Inspection of the distributions (Fig. 5) suggested that there was still a good spread of formal practice time, with a range from 0 to 300 min (5 hr) in the week, and a good representation of practice times up to ~ 150 min (2.5 hr). However, our distribution of practice times does not contain a good representation of durations aligned with typical Mindfulness-Based Stress Reduction (MBSR) or Mindfulness-Based Cognitive Therapy (MBCT) recommendations (45 min per day for 6 days of the week) (Kabat-Zinn, 1982; Segal et al., 2018). MBSR and MBCT interventions are the most well studied and have the most evidence for their efficacy (Galante et al., 2021); however, lower intensity interventions have also demonstrated evidence of changes in self-compassion and mindfulness (Bailey et al., 2018). As such, it may be that longer practice times or a longer program duration would be required before a relationship between practice time and changes in trait mindfulness and self-compassion becomes apparent—as a recent meta-regression has suggested relationships between practice time and outcomes (Parsons et al., 2017), and cross-sectional research from a large sample of experienced meditators provides evidence that higher amounts of mindfulness practice are related to improved well-being (Bowles et al., 2022). Additionally, due to the specifics of the study design, the MAQ only assessed practice times and practice quality from the last week of the intervention. Assessment of practice time and quality across the entire duration of the online mindfulness program may be more likely to detect relationships between practice time and changes in outcome measures if those relationships exist and this approach is recommended for future research.

We also note that the suggestion that formal practice quality is associated with improved mindfulness and self-compassion might not be immediately helpful when considering potential applications of our results. Given that all participants had the same mindfulness teaching materials available to them, we have no indication of why some participants self-reported higher mindfulness practice quality, so can offer no information on how practice quality could be improved. Our suggestion is therefore simply that methods to determine how formal mindfulness practice quality can be improved may be valuable.

Additionally, no measures of mindfulness quality or quantity predicted changes in prosociality. One potential explanation we considered was that the changes in prosociality might have been indirectly affected by practice variables through changes in trait mindfulness. However, despite a previous meta-analysis suggesting trait mindfulness is associated with prosociality (Donald et al., 2019), changes in trait mindfulness and changes in prosociality were not correlated in our data. It may be that while the two measures are correlated across individuals when measured as traits, changes in mindfulness within an individual do not relate to changes in prosociality. Instead, an alternative explanation for our results might be that the changes in prosociality from pre- to post-program were not related to mindfulness practice–specific effects, but rather by non-specific effects of engagement in an intervention (or in a mindfulness intervention, while still not being related to the mindfulness practice).

Lastly, it is worth noting that the mindfulness program took place during the COVID-19 pandemic. As such, the results may be of particular interest when considering the importance of prosocial (as opposed to selfish) behaviors during a pandemic, especially at a time when a large-scale transition to online platforms has been required by restrictions to mitigate the spread of the virus. The current results suggest that online mindfulness programs have the capacity to change participants’ prosocial attitudes, despite minimal interpersonal interaction during the program. Following this line of reasoning, online mindfulness programs may offer an opportunity to increase prosocial behavior. However, our study did not measure examples of prosocial behaviors, so further research is required to determine the effect of online mindfulness programs on prosocial behavior.

There are several limitations of this study that need to be highlighted, including self-selection of participants, high attrition, lack of a control group, no follow-up period, and assessing mindfulness practice quality and quantity in only the final week of the program. It has been established that massive open online programs (as per the online mindfulness program in the current study) have high attrition rates, with the highest rates of attrition occurring in the initial weeks of a course (Jordan, 2015). The online mindfulness program reported in this study had 1043 (8.6%) participants complete more than 90% of the content 6 weeks after the program opened, and 209 (1.7%) participants complete both pre- and post-program surveys—similar response rates to previous samples of large online mindfulness programs (Bailey et al., 2018). This completion rate is in keeping with work by Jordan (2015), who examined 221 massive open online courses and determined the median value for completion rates was 12.6%. The relatively low completion rates and potential self-selection bias may have exaggerated the positive effect of the online mindfulness program on all outcomes, as the study population may contain the highly motivated individuals. Further, the lack of a control group prevents us from making claims about causation. However, given that the sample population was recruited from a pool of people who had self-selected to participate in the online mindfulness program, retrospectively applying random group allocation or manipulating the program exposure would have been unethical.

In addition to the limitations provided by the lack of a control group, participants were somewhat informed of the aims of the study (they were informed that we would be examining the effects of an online mindfulness intervention on the outcome measures). This may have influenced the changes from pre- to post-scores through demand characteristics, making our study more likely to show improvements in outcome measures from before to after the mindfulness course. However, it is less likely that demand characteristics would have affected the relationship between the variables and the mediation analysis. If the results were indicative of simply demand characteristics, the variance in the relationships and mediation would be influenced primarily by the participant’s desire to show improvements as a result of the intervention. However, we detected a specific relationship between the quality of mindfulness practice and changes in trait mindfulness and self-compassion, as well as mediating relationships whereby the relationship between quality of practice and self-compassion was mediated by trait mindfulness. These more complex relationships are less likely to be explained by potential demand characteristics, and as such may be interpreted as not influenced by potential demand confounds.

Additionally, to avoid over-burdening our participants (and due to the restrictions of the course delivery platform), we only measured responses at pre- and post-program timepoints. Measurements with the MAQ across each week of the program may be desirable in future research to characterize practice trajectories across the entire program, enable assessments of the relationships between those trajectories and outcome measures, and offer more evidence for potential causal relationships. Alternatively, to avoid the high rates of attrition often found when weekly responses are required, retrospective self-reports of each week at the post-program timepoint may achieve the same goal (although probably with reduced accuracy). Additionally, exact practice time data could be obtained from website access to mindfulness practices in future research. This was unfortunately not possible in the current study due to the constraints of the platform that the mindfulness course was being conducted on. Similarly, due to time and staff constraints, a longer follow-up period was not feasible; and thus, the longevity of the benefits of the online mindfulness program on self-compassion, prosociality, and trait mindfulness remains unclear. Age was also measured using a categorical approach rather than an exact year approach, reducing the variability captured by our analyses, which may have influenced our ability to detect significant effects. It is recommended that future research measure age as a continuous variable with exact years.

It should also be noted that most of the study population were female (81.7%), a common occurrence in mindfulness programs. This limits the transferability of the results to the wider public due to the lack of male and non-binary gendered individuals’ representation within the study population. However, despite the smaller sample size of male participants, interactions between gender and changes in trait mindfulness or self-compassion both provided BFincl values of less than 0.32, suggesting evidence of moderate strength against the hypothesis of an effect for these interactions. This suggests that the smaller male sample size is unlikely to explain the null results in interactions between gender and pre–post changes for the trait mindfulness and self-compassion outcome measures. As well as the unequal gender distribution of our data, almost half of our participants were in the 46–65 age range. While our data did contain a relatively broad distribution of ages outside of this range (Table 1), the large proportion of participants within the 46–65 age range could skew our results towards the effects apparent within that age range, and as such our results could perhaps be viewed as most applicable to that age range. Additionally, our sample was culturally diverse, and reflects the overall trends of participation in the mindfulness course. Future research may be interested in whether the results are influenced by different cultures as well. Our sample also included a large number of participants who had at least some previous mindfulness experience. As might be expected, these participants demonstrated higher levels of mindfulness prior to the course. In order to ensure the inclusion of participants with previous mindfulness experience did not bias our results, we included this variable as a covariate, and found that doing so did not alter the pattern of our results. Future research could address these limitations using an active control condition, more diverse sample, assessing mindfulness quality and quantity across the entire program, and a longer follow-up period.