Benefits and Challenges of Delivering Meditation Instruction Live Online: Lessons from the COVID-19 Pandemic Regarding Accessibility and Connection

The focus of this investigation was to understand what participants liked or disliked about online meditation programs that predominated as a result of COVID-related social distancing measures. Participants were asked to indicate whether they had engaged in any online meditation programs since the onset of the pandemic by selecting all that applied from the following options: (a) no, have not participated in any online meditation experiences; (b) yes, participated in one-time meditation events; (c) yes, meeting regularly with a teacher or group; (d) yes, participated in an online retreat; (e) other, provide own response. The 318 participants who indicated that they had participated in an online meditation program (options b–e) constitute the sample analyzed in this paper. These participants were asked two follow-up questions: “What have you liked or found helpful about these online formats?” and “What have you disliked or found unhelpful about these online formats?”.

We also asked participants to rate the degree to which pandemic-related changes (1) “improved or strengthened” and (2) “hindered or disrupted” their meditation practice on a sliding scale that ranged from 0 to 6 with the following anchors: (0) none, (2) a little, (4) somewhat, (6) a lot. As a measure of practice frequency, participants reported the number of days per week they practiced meditation before and since the onset of the pandemic, with the options of 0 to 7 (days/week) and an additional option for less than once per week, which was scored as 0.5 in the current analyses. We also asked them to report whether the frequency of their practice changed since the onset of the pandemic on a sliding scale with the following anchors: (0) decreased a lot, (3) stayed the same, (6) increased a lot. Participants indicated how often they practiced with a group before and since the onset of the pandemic, with the following options: (1) never or rarely, (2) once per month or less, (3) a few times a month, (4) once per week, and (5) more than once per week. Lastly, participants rated how diligent they felt they had been in their meditation practice over the last month using the following options: (1) not at all, (2) not very, (3) somewhat, (4) very, (5) extremely. To reduce complexity, a non-hierarchical cluster technique (described in further detail in the “Statistical Analyses” section and Supplementary Material) was used to consolidate these variables into two clusters to be used in further analyses.

Participants were asked to indicate how many individuals they lived with in each of the following categories: (a) children under 18 years of age; (b) spouse/partner under 55; (c) spouse/partner over 55; (d) family under 55; (e) family over 55; (f) housemate, friend, guest under 55; (g) housemate, friend, guest over 55; (h) pet. For analyses, responses were binned into three categories (children under 18 [option a], other adults [options b–g], and pets [option h]), and summed for each category. Participants also indicated the type of geographic community they lived in (a) large city, (b) suburb near a large city, (c) small city or town, or (d) rural area.

Participants reported their current working arrangements by selecting all that applied from the following options: (a) full-time, (b) part-time, (c) occasional/casual, (d) self-employed/independent worker, (e) stood down/furloughed/laid off, (f) unemployed, (g) not working by choice, (h) student, (i) retired, (j) prefer not to respond, (k) other—write in response. An additional option, (l) multiple part-time jobs, was added part way through baseline data collection. For analysis, participants were categorized as “working” if they selected option a, b, c, d, h, or l; or “not working” if they selected option e, f, g, or i, but not option a, b, c, d, h, or l (i.e., the “working” options). Participants who selected only “prefer not to respond” (Option j) were categorized as such. Those who selected only “other” (Option k) were removed from analyses involving working status. Participants were also asked whether they were working from home (yes/no), and if so, whether they were able to perform the tasks expected of them, responding on a slider that ranged from 1 to 7 with the anchors (1) not able to meet expectations, (4) can meet expectations adequately, and (7) able to surpass expectations. A separate checkbox was provided for participants to respond with “NA” if this was not applicable for their situation.

Participants also reported whether they were mostly staying at home (e.g., in isolation, quarantine, following stay-at-home orders, working from home) or if their lifestyle involved going out (e.g., to work, eating at restaurants). The options included (select 1) (a) in mandatory self-isolation/quarantine (due to COVID-19 infection or possible exposure); (b) voluntarily staying home and working or studying from home (though there are no orders or guidelines to do so); (c) working or studying from home because there are government orders or guidelines to do so; (d) going about daily business as usual (going to work, school, etc.) but don’t visit restaurants, cinemas, etc.; (e) going about daily business as usual (going to work, school, etc.) and visiting restaurants, cinemas, etc.; (f) work requires me to go out (e.g., medical professional, delivery, public transport) while others are recommended/mandated to stay at home; and (g) am (temporarily or permanently) homeless and finding difficulty accessing stable housing, sheltering, or staying in quarantine at all. We categorized participants as “at home” if they selected options a, b, or c, or as “not home” if they selected options d, e, f, or g.

Finally, levels of social interaction were measured with 12 questions asking about the frequency of interactions in the last month across three different groups (family, friends, colleagues) and across four different modalities (in-person, messages [e.g., text, WhatsApp], phone, or video chat [e.g., Skype, Zoom]). For example, “In the past month, how often have you exchanged messages with friends (e.g. text, WhatsApp, Facebook).” Response options were (0) never/NA, (1) a few times a month, (2) once a week, (3) a few times a week, (4) once a day, and (5) at least a few times a day.

As with the meditation variables, we used a non-hierarchical cluster technique (described in the “Statistical Analyses” and Supplementary Information) to consolidate these variables into eight clusters for use in further analyses.

We used the 30-item Big Five Inventory short-form (BFI-2-S; Soto & John, 2017) to assess five core aspects of personality: open-mindedness, conscientiousness, extraversion, agreeableness, and negative emotionality. Participants indicated the extent to which they agreed or disagreed with statements reflecting aspects of their personality using a 7-point scale ranging from (1) strongly disagree to (7) strongly agree. Possible scores for each subscale range from 6 to 42, with higher scores indicating more of that personality trait.

Qualitative responses to the questions “What have you liked or found helpful about these online formats?” and “What have you disliked or found unhelpful about these online formats?” were coded in Dedoose, a qualitative analysis software. Coding occurred between March 2021 and November 2021.

Responses were first “subject coded,” meaning the content of each response was either coded verbatim or closely paraphrased without the raters adding additional meaning or categorization. For instance, one participant responded that they liked “The sense of community and [t]he teachings from the facilitator.” Two subject codes were applied to this sentence: “sense of community” was applied to the first half of the response and “teachings from the facilitator” to the second half of the response. This coding was conducted by three independent raters (S.V., S.D., and C.P.) under the supervision of J.P. At least two raters coded each response.

In the next phase of coding, J.P., S.V., and S.D. collectively created a codebook of themes by grouping together subject codes with common topics. Themes were defined in the codebook to guide coders in applying codes systematically. Coders first applied these themes to each response independently and then met to discuss any discrepancies before determining the final codes. During this process, if a participant’s response did not align with any already existing themes, a new theme was created and applied to that response. In total, 24 themes were identified for the question “What have you liked or found helpful about these online formats?” and 21 themes were identified in response to the question “What have you disliked or found unhelpful about these online formats?” Because some practitioners in Study 2 were also meditation teachers, one additional theme was created and applied to responses that were about respondents’ experiences of teaching meditation online (rather than attending as a participant). For a full list of the theme codes, see Appendix 1 in the Supplementary Information.

Responses to each question had at least one theme code applied to them, with most having more than one theme code applied. Multiple themes were sometimes applicable for the same segment of a response, allowing for the overlapping application of theme codes. For example, themes of “accessibility/convenience,” “no commute,” and “logistics” were applied to the excerpt “Not having to commute has made these a lot easier to attend on a regular basis.”

To use the theme codes in analyses, we used the measure of code presence: if a theme code was applied to an individual’s response, they were assigned a 1 for that code; if it was not applied, they were assigned a 0 for that code. Since the responses were typically short, it was rare that the same theme code was applied to an individual’s response more than once. As such, code presence (a categorical binary response of 1 or 0) was nearly identical to code application (total number of times a code was applied to a response).

Statistical analyses were performed in R version 4.1.3 (R Core Team, 2022). Summary statistics of variables were conducted using the psych package version 2.2.5 (Revelle, 2022). Visualizations were created using ggplot2 version 3.4.0 (Wickham, 2016). Effect sizes were calculated with the effsize package version 0.8.1 (Torchiano, 2020).

As we did not have many specific predictions regarding what participants would like or dislike about their online meditation experiences, we examined participants’ overall patterns of likes and dislikes separately, and then examined whether those patterns of likes or dislikes were predicted by participants’ personality traits, demographic characteristics, current meditation practice, or their living situation, working arrangements, and interactions with others. To do this, we generated distance matrices of the like and dislike theme codes and then used multivariate distance matrix regression (MDMR) to assess whether our variables of interest predicted inter-individual variations in these multivariate outcomes (Anderson, 2001; McArdle & Anderson, 2001). To consolidate the current meditation variables and the living, working, and interaction variables, we first used cluster analyses to group participants along these different dimensions, essentially creating profiles for each set of variables. We then used participants’ cluster assignments as predictors in MDMR analyses to assess whether these profiles predicted patterns of likes or dislikes.

Details regarding the clustering of predictor variables, creation of the like and dislike matrices, and MDMR analyses are included in the Supplementary Materials. In brief, the MDMR analyses report a pseudo-R² value, representing the proportion of variation in the multivariate outcome that is explained by the predictor(s). For results in which there was a significant association between the predictor variable(s) and the pattern of likes or dislikes, we quantified the association between the predictor(s) and each theme code included in the distance matrix (McArtor et al., 2017). This was done by randomly permuting each outcome variable (k) one at a time to calculate a new pseudo-R²_(k) and then subtracting the new pseudo-R²_(k) from the original pseudo-R² to get the difference (i.e., delta [δ]), which represents the effect of that variable. If the predictor(s) had a large effect on a given outcome variable, dissociating that variable should reduce the overall effect, which is reflected in a smaller pseudo-R²_(k) and a larger δ. The theme codes with the top 5 highest δ are reported. We then examine the relationships between the predictors and the theme codes with the highest δ using a series of follow-up tests (see the Supplementary Materials for details). It is important to note, however, that a significant MDMR result reflects the association between predictor(s) and the entire outcome matrix, so significant predictors do not always have a statistically significant relationship with the theme codes with the highest δ on their own.

Demographic characteristics for the subset of participants described in this paper are presented in Table 1 (the full CCC sample can be found in Supplementary Table S3). Participants ranged in age from 21 to 88 years old, with an average age of 53 (SD_age = 15.3). The majority of participants identified as heterosexual (74%), women (73%), of white European heritage (66.7%), and had a graduate or professional degree (60.1%).

On average, participants indicated that pandemic-related changes significantly improved their meditation practice, t(309) = 45.34, p < 0.001, Cohen’s d = 2.58, while hindering their practice to a much smaller degree, t(285) = 14.42, p < 0.001, Cohen’s d = 0.85. Average practice frequency significantly increased during the pandemic, t(307) = 23.58, p < 0.001, Cohen’s d = 1.34, with meditators practicing an average 5.3 days per week during the pandemic compared to 3.7 days per week prior to the pandemic, t(317) = 13.31, p < 0.001, Cohen’s d = 0.80. The frequency of group practice also increased, t(317) = 11.66, p < 0.001, Cohen’s d = 0.88. Only 6.9% of participants reported practicing with a group more than once a week prior to the pandemic, which increased to 44% during the pandemic (Fig. 1). Participants also indicated practicing relatively diligently over the past month, t(317) = 50.14, p < 0.001, Cohen’s d = 2.81. Descriptives for these measures are reported in Table 2 (descriptives for the full CCC study sample can be found in Supplementary Table S5).

Characteristics of the participants’ living and working arrangements and measures of their social interactions can be found in Table 3 (data for the full CCC sample can be found in Supplementary Table S4). Most participants were working (67.3%); lived in a large city (34%), suburb (33%), or small town (27%); and were staying at home (83%). Participants were most often interacting with others by phone or messaging, with video interactions being next most common, and in-person interactions being least common. Interactions with family and friends were more frequent than interactions with colleagues.

Descriptive statistics for the personality variables are reported in Table 4 (descriptives for the full CCC sample can be found in Supplementary Table S6).

First, we assessed whether participants’ demographic characteristics were associated with their patterns of likes or dislikes, assessing each demographic characteristic individually (Supplementary Table S7). There were significant associations between the pattern of likes and participants’ age (pseudo-R² = 0.02, p < 0.001) and sexual orientation (pseudo-R² = 0.03, p < 0.001). For age, the five like theme codes with the highest δ were (1) accessibility/convenience (δ = 6.92), which was negatively correlated with age (⍴ =  − 0.212, p < 0.001); (2) reference to a specific like (δ = 3.00), which was positively correlated with age (⍴ = 0.112, p = 0.047); (3) comfort and safety of own home (δ = 2.22), which was negatively correlated with age (⍴ =  − 0.191, p = 0.001); (4) sense of community (δ = 2.21), which was positively correlated with age (⍴ = 0.137, p = 0.015); and (5) support/solidarity (δ = 1.64), which was not significantly correlated with age (⍴ = 0.081, p = 0.154) (Supplementary Figure S1). For sexual orientation, the five like codes with highest δ were (1) logistics (δ = 5.78), (2) global community (δ = 4.17), (3) support/solidarity (δ = 3.76), (4) accessibility/convenience (δ = 3.68), and (5) reference to specific like (δ = 2.9) (Supplementary Figure S1). Logistics was significantly associated with orientation (χ² = 13.78, df = 2, p = 0.001, Monte Carlo simulated p = 0.001, Cramer’s V = 0.209), with LGBQA individuals more likely to mention logistics than heterosexual individuals (post hoc comparison: χ² = 12.3, df = 1, p_corr = 0.001). Global community (χ² = 12.41, df = 2, p = 0.002, Monte Carlo simulated p = 0.005, Cramer’s V = 0.198) and support/solidarity (χ² = 8.87, df = 2, p = 0.012, Monte Carlo simulated p = 0.008, Cramer’s V = 0.168) were each significantly associated with orientation. Post hoc pairwise comparisons initially indicated that heterosexual participants were more likely to mention global community (χ² = 4.730, df = 1, p = 0.03, p_adj = 0.089) and support/solidarity (χ² = 4.130, df = 1, p = 0.042, p_adj = 0.126) than LGBQA participants, but these comparisons were not significant following correction for multiple comparisons. Accessibility/convenience was significantly associated with orientation (χ² = 7.68, df = 2, p = 0.022, Monte Carlo simulated p = 0.014, Cramer’s V = 0.156), with LGBQA individuals being more likely than heterosexual individuals (post hoc comparison: χ² = 5.95, df = 1, p_corr = 0.044) to mention accessibility/convenience. Reference to specific likes did not significantly differ between orientation groups (χ² = 4.07, df = 2, p = 0.131, Monte Carlo simulated p = 0.095, Cramer’s V = 0.114).

For dislike theme codes, there were significant associations between the pattern of dislikes and participants’ age (pseudo-R² = 0.01, p = 0.002) and gender (pseudo-R² = 0.01, p = 0.030) (Supplementary Table S7). For age, the top 5 dislike codes with highest δ were (1) lack of community (δ = 3.72), which was negatively correlated with age (⍴ =  − 0.251, p < 0.001); (2) no dislikes (δ = 3.15), which was positively correlated with age (⍴ = 0.152, p = 0.007); (3) technology issues (δ = 2.57), which was positively correlated with age (⍴ = 0.117, p = 0.039); (4) distractions (δ = 1.47), which was negatively correlated with age (⍴ =  − 0.162, p = 0.004); and (5) reference to specific dislike (δ = 0.97), which was not significantly correlated with age (⍴ = 0.034, p = 0.553) (Supplementary Figure S2). For gender, the five dislike codes with highest δ were (1) no dislikes (δ = 3.06), which was not associated with gender (χ² = 4.91, df = 2, p = 0.086, Monte Carlo simulated p = 0.089, Cramer’s V = 0.125); (2) lack of community (δ = 2.59), which was significantly associated with gender (χ² = 13.8, df = 2, p = 0.001, Monte Carlo simulated p = 0.002, Cramer’s V = 0.209), with gender expansive individuals more likely than men (χ² = 7.72, df = 1, p_corr = 0.016) or women (χ² = 6.89, df = 1, p_corr = 0.017) to mention lack of community; (3) technology issues (δ = 1.73), which did not significantly differ between gender groups (χ² = 4.58, df = 2, p = 0.101, Cramer’s V = 0.12); (4) reference to specific dislike (δ = 1.41), which did not significantly differ between gender groups (χ² = 1.53, df = 2, p = 0.464, Cramer’s V = 0.07); and (5) prefer in person (δ = 0.75), which did not significantly differ between gender groups (χ² = 1.04, df = 2, p = 0.594, Cramer’s V = 0.057) (Supplementary Figure S2).

As described in the Supplementary Materials, the optimal number of clusters to describe the meditation variables was two (Supplementary Table S8). Cluster 1 (n = 204) was characterized by individuals who practiced meditation more than individuals in Cluster 2 both before and since the onset of the pandemic, felt more diligent in their practice, indicated that the pandemic improved their practice, and reported an increase in practicing with a group since the onset of the pandemic. Conversely, Cluster 2 (n = 111) was characterized by individuals who practiced less both prior to and since the onset of the pandemic, felt less diligent in their practice, and indicated that the pandemic hindered their practice, and half of the individuals were not practicing with a group either before or since the onset of the pandemic. Mean values of the meditation variables for each cluster are depicted in Fig. 3 and complete descriptive statistics can be found in Supplementary Table S12.

To investigate whether features of participants’ recent meditation practice were associated with their reported likes and dislikes, we assessed the relationship between participants’ meditation cluster assignment and patterns of like and dislike theme codes. The means of like and dislike codes for each cluster are depicted in Fig. 4.

Participants’ meditation cluster assignment did not significantly predict the pattern of like codes (pseudo-R² = 0.01, p = 0.153), but did significantly predict the pattern of dislike codes (pseudo-R² = 0.01, p = 0.022). The five dislike theme codes with the highest δ (Supplementary Figure S3) were (1) no dislikes (δ = 3.43), (2) Zoom/screen fatigue (δ = 2.26), (3) less accountability/commitment (δ = 2.09), (4) missing in-person presence (δ = 1.27), and (5) breakout rooms/discussion (δ = 1.00). Individuals in Cluster 2 were significantly more likely to report dislikes of Zoom/screen fatigue (χ² = 5.45, df = 1, p = 0.02, Cramer’s V = 0.14) and less accountability/commitment (χ² = 10.15, df = 1, p = 0.001, Cramer’s V = 0.2) than individuals in Cluster 1 (Supplementary Table S13).

As described in the Supplementary Materials, the optimal number of clusters to describe the living, working, and interaction variables was eight. Individuals in Cluster 1 (n = 72) were characterized as working, staying home and working from home, living in a large city, and having relatively high phone and video contact with others. Cluster 2 (n = 73) was characterized as working, mostly staying home and working from home, living in the suburbs, and having relatively high phone and video contact with others. Cluster 3 (n = 39) was characterized as working, staying home and working from home, living in a small town, and having relatively high phone contact with others. Cluster 4 (n = 37) was characterized as not working, mostly staying home, living in the suburbs, and having relatively low contact with colleagues. Cluster 5 (n = 21) was characterized as working, not staying home—though about half were working from home, living in a large city, and having relatively low contact with family but higher in-person contact with others. Cluster 6 (n = 19) was characterized as mostly working, not staying home—though about half were working from home, living in small towns, and having relatively high phone and in-person contact with others. Cluster 7 (n = 21) was characterized as not working, staying home, living in a large city, not having children under the age of 18, and having relatively low contact with colleagues and family members. Cluster 8 (n = 33) was characterized as not working, staying home, living in small towns, and having relatively low contact with colleagues. Clustering metrics and the descriptive statistics of the living, working, and interaction variables for each cluster are reported in Supplementary Tables S9 and S10.

We investigated whether participants’ living, working, and interaction characteristics were associated with their reported likes and dislikes by assessing participants’ living, working, and interaction cluster assignment and the pattern of like or dislike theme codes. The means of like and dislike codes for each cluster are presented in Supplementary Figures S4 and S5.

Cluster assignment did not significantly predict the pattern of like codes (pseudo-R² = 0.02, p = 0.378), but did significantly predict the pattern of dislike codes (pseudo-R² = 0.03, p = 0.032). The dislike theme codes with the top 5 highest δ were (1) prefer in person (δ = 0.61), (2) missing being physically elsewhere (δ = 0.28), (3) reference to specific dislike (δ = 0.26), (4) distractions of technology (δ = 0.1), and (5) breakout rooms/discussion (δ = 0.1) as shown in Supplementary Figure S6. However, none of these individual theme codes differed significantly across cluster groups (Supplementary Table S11).

Contrary to our predictions, when we assessed whether participants’ personality traits predicted their likes or dislikes, we found no significant associations between participants’ BFI subscale scores and patterns of like (pseudo-R² = 0.02, df = 5, p = 0.380) or dislike (pseudo-R² = 0.02, df = 5, p = 0.537) theme codes (Supplementary Table S14).