Positive Emotion Correlates of Meditation 1

POSITIVE EMOTION CORRELATES OF MEDITATION 1

Online Supplemental Material (OSM) for

Positive Emotion Correlates of Meditation Practice:

A Comparison of Mindfulness Meditation and Loving-kindness Meditation

Recruitment Details

Study 1. Participants in Study 1 were recruited from the counties surrounding the cities of Chapel Hill and Durham, North Carolina, via flyer, listserv, and email advertisements. The advertisements stated: “Science has shown that meditation improves people’s health and well-being. Help UNC researchers learn how.” A total of 270 people responded, learned the time commitment required to join the study, and were initially screened for eligibility by phone. To be eligible, participants were required to be between 35 and 64 years old, fluent in English, new to meditation, absent any chronic illnesses or disabilities, and to have home internet access for completing daily surveys. The procedure for assessing whether participants met the inclusion criterion of “new to meditation” involved asking the following two questions: “How much experience with meditation do you have?” and “Currently, how often do you meditate?” Each question had a range of response options and volunteers were included in the study if they responded with “None” and “Never,” respectively. If the response chosen for the first question was “Read about it or participated a few times,” then the study coordinator would inquire further about degree of exposure and volunteers who had never taken a class or undertaken formal self-study were also invited to participate.

Among the 270 who expressed interest, 198 met the eligibility criteria, of which 176 provided informed consent and were scheduled for baseline assessment (see Figure S1 for Study 1 CONSORT Diagram). Participants received compensation after completing various portions of a larger study on sustained behavior change, up to a maximum of $205 per participant.

Study 2. Participants in Study 2 were recruited from the same region of North Carolina using the same advertising media. Advertisements were similar to Study 1 and in addition requested individuals who were “interested in making lifestyle changes.” A total of 640 individuals responded, learned the time commitment involved, and were screened using the same eligibility requirements (with one additional exclusion criterion being participation in Study 1). Among the 640 who expressed interest, 312 met eligibility criteria, of which 231 provided informed consent and were scheduled for baseline assessment (see Figure S2 for Study 2 CONSORT Diagram). Participants received compensation after completing various portions of a considerably larger study on health behavior change, up to a maximum of $640 per participant.

CONSORT Diagrams

Figure S2

Consort Diagram for Study 2

Meditation Workshops

To develop the interventions, a subset of the authors with expertise in psychological science (BLF, SBA, and AMF) or meditation practices (MMB, SLK, JB, & SS) held two, day-long, face-to-face meetings. Each of the meditation experts had decades of meditation experience, including with both MM and LKM (J. Brantley, 2007; M. Brantley & Hanauer, 2008; Loundon, 2001; Salzberg, 1995, 2011). The purpose of these meetings was to articulate the training objectives and curricula for the two meditation workshops, with aims to both adhere to traditional teachings and coincide with previous research. Although all authors were aware of the goal to compare MM and LKM, the specific research questions to be tested were not discussed with meditation instructors.

As a framework to delineate and describe the similarities and differences between MM and LKM, this group adopted Shapiro and colleague’s three axioms of mindfulness (Shapiro, Carlson, Astin, & Freedman, 2006). These three axioms are intention, attention, and attitude, which in combination produce a particular hallmark “meta mechanism” or fundamental shift in perspective characteristic of each training type. Whereas the attitude of both MM and LKM were identical, namely, open and nonjudgmental, their respective intention, attention, and hallmark meta mechanism differed, as described in the accompanying article. The two meditation instructors (SLK and MMB) ultimately designed the meditation workshops (MM and LKM, respectively) with iterative feedback from the entire group to ensure intended similarities and differences were achieved.

Reliability Analysis

To estimate the reliability of the emotion composite scores, we followed the approach suggested by Bolger and Laurenceau (2013, p. 138) and detailed in the Supplemental Material document accompanying Isgett, Algoe, Boulton, Way, & Fredrickson (2016). Specifically, we used the daily diary data to evaluate the reliability of the between- and within-person components corresponding to each individual’s daily data time series. The between-group component corresponds to the reliability of each individual’s aggregate level of positive (negative) emotion across the duration of the study, and the within-person component represents the day-to-day fluctuations in positive (negative) emotion, which was of primary interest in the current article. Reliability was calculated using a multilevel structural equation model (MSEM) estimated in Mplus (version 7.2; Muthén & Muthén, 1998-2015). McDonald’s omega was calculated for a single between-person and single within-person factor in the MSEM model; furthermore, each item was regressed onto time (scaled in weeks, as discussed below) in the within-group model to control for linear trends in the data. Full-information maximum likelihood estimation with robust standard errors and test statistics (MLR) was used. Using the entire 11-week reporting period (i.e., two weeks of baseline data plus the targeted nine weeks after the first workshop session), reliability coefficients for the positive emotion composite (within: .87; between: .96) and negative emotion composite (within: 79; between: 96) were considered sufficient. Additionally, both models appeared to fit the data adequately according to conventional guidelines at both levels for positive emotions, overall χ2 = 3496.587, p < .001, overall CFI = 0.877, overall RMSEA = 0.049, SRMRWithin = .050, SRMRBetween = .032, and negative emotions, overall χ2 = 3366.640, overall CFI = 0.774, overall RMSEA = 0.048, SRMRWithin = .060, SRMRBetween = .067, although the Comparative Fit Index appeared to suggest potential misfit, particularly in the negative emotions model.

Meditation Practice Measurement

Because the response prompts for meditation practice duration variable was open-ended in Study 1 – “How much time (in minutes) did you spend on meditation since the last time you answered this question” – as opposed to constrained, as in Study 2 – “How much time (in minutes) did you spend on meditation in the last 24 hours – we were interested in whether any systematic differences existed between studies in the reporting of meditation practice duration. The evidence does not appear to suggest any systematic differences. The lags between observations in the two studies (i.e., between adjacent meditation practice reports) were distributed similarly, M = 1.24 (SD = 0.79) and M = 1.25 (SD = 0.79) days for studies 1 and 2, respectively, and were not significantly different, t (11831) = -.60, p = .55. Thus, for the majority of daily reports in Study 1, the most recent report would be within approximately 24 hours of the current report. Furthermore, mean levels of meditation practice duration were similar (Study 1: M = 14.54, SD = 11.62, Study 2: M = 14.46, SD = 13.62) and did not differ significantly between studies, t (13561) = .42, p = .68. Finally, none of the interactions tested that involved study membership and meditation practice – either duration or frequency – were significantly different from zero. As a result, we believe that it was reasonable to pool the meditation practice variables across studies.

Missing Data, Model Specification, and Model Testing

Missing data. Response rates for the daily reporting protocol were high. Participants were allowed to skip items and/or portions of the online daily diary report; thus, response rates differ by measure. A total of 26,103 daily emotion reports were possible for the 339 participants (339 × 77 days) in the pooled dataset; participants completed positive emotion reports on 20,684 days (79.24%), negative emotion reports on 20,673 days (79.20%), duration reports on 20,088 days (76.96%), and frequency reports on 20,619 days (78.99%). Although participants were allowed to skip items on the daily emotion report, for each item, observations were missing on less than 1% of the days that positive or negative composite scores were calculated (i.e., days in which at least 1 item was completed); therefore, the composite scores were calculated by averaging over all available item responses. Regarding missingness at the level of the emotion composites and the meditation practice variables, the restricted maximum likelihood (REML) estimation procedure uses all available data points on the dependent variable for a given participant (i.e., a participant was not listwise deleted if they did not fill out all 77 daily reports). Thus, all 339 participants were included in model estimation, with individual rows excluded if observations were missing on the emotion composite or mediation practice variables for a given day. REML is predicated on the missing at random (MAR) assumption, and thus we assume that missingness was not due to the unobserved values of the missing data points (i.e., missing not at random, or MNAR).

Model Specification and Model Testing. As noted in the accompanying article, multilevel models were used to answer the research questions, estimated via the PROC MIXED procedure in the SAS 9.2 software. The model building strategy we employed mirrors that reported in Isgett et al. (2016) and can be considered a forward selection or “build-up” approach (Snijders & Bosker, 1999, p. 94), such that relevant predictors and random effect terms were added in a sequential and exploratory fashion. Positive and negative emotions were analyzed in separate models; additionally, the frequency and duration of meditation practice variables were entered in separate models, thus resulting in four primary analysis models (see Table 2 in the accompanying article). Because restricted maximum likelihood estimation was used, additions to the random effects structure (e.g., variances and covariances) were evaluated using deviance tests, and fixed effect parameters were tested via each coefficient’s associated t-test statistic. Given the exploratory nature of the study, all hypotheses were tested using an α level of .05; the p-values were halved, however, for deviance tests of the random effect parameters (Snijders & Bosker, 1999, p. 90), as variances are subject to boundary constraints (i.e., must be greater than zero). For tests of fixed-effects in the positive emotions models, we employed the Kenward-Rodger method (Littell, Stroup, Milliken, Wolfinger & Schabenberger, 2006) to obtain adjusted standard errors and degrees of freedom. For the negative emotions models, non-normality of model residuals were observed in preliminary model runs; therefore, robust standard errors (e.g., the Huber-White sandwich estimator) were used. Multilevel models with these specifications were also used for calculating the intraclass correlation coefficients and conducting inferential group comparisons as described at the beginning of the Results section in the accompanying article.

The approach was designed to test the following questions in sequence: (a) do MM and LKM each increase people’s day-to-day experiences of positive emotions over time, and is there a difference in growth rate between the two conditions? (b) is there a dose-response relation between time spent meditating and daily experiences of positive emotions? And (c) if a dose-response relation exists, does it emerge within individuals, between individuals, or both, and does it differ between MM and LKM? To answer (a), we first fit a linear growth curve model to evaluate the average growth rate in the emotion reports for the entire sample. Time was scaled in weeks and we omitted the two-week non-meditation baseline period to simplify the model. A linear model was considered adequate after inspection of individual participants’ time series. Using this model, we also tested for individual differences in participants’ emotion reports on the day of the first workshop attended (intercept) as well as the average growth rate (slope). We also at this point tested an alternative covariance structure for the level-1 residuals to account for potential serial correlation (a spatial power structure; Bolger & Laurenceau, 2013, p. 93); for both positive and negative emotions, this alternative structure provided a significant improvement in model fit and was retained in subsequent models. Next, we added experimental condition as a person-level predictor of the random intercept and slope coefficients in subsequent models to test for condition differences. To test questions (b) and (c), we entered the meditation practice duration and frequency variables as predictors in separate models; within each model, a person-mean centered and average practice variable were entered simultaneously. The within-person effect was permitted to vary randomly across individuals in a subsequent model and this random effect was retained based on improvement in model fit. The final step involved entering two interaction terms, crossing experimental condition with the within- and between-person practice variables, respectively. Simple intercept and slope analyses are presented belowfor the interaction terms that were found to be significant in the multilevel models (see Table S1).

To control for potential study heterogeneity, study membership was included as a covariate in all models, thus adjusting for any study-specific differences in the emotion composite variables. In addition, every time a new predictor was introduced into the model – including interaction terms (e.g., experimental condition × time) – we tested to see whether there was a study membership × predictor interaction; if significant, the interaction was retained in subsequent models. This occurred once in the positive emotions models, with a significant interaction between study membership and experimental condition (discussed below). Although we attempted to account for between-study heterogeneity during data analysis using this fixed-effects approach, we acknowledge that it remains unclear whether differences in measurement (e.g., differential item functioning) existed between studies. Advanced psychometric procedures have been proposed to account for measurement differences across studies within the integrative data analysis framework (Curran et al., 2014); however, it is unclear whether and how these procedures can be applied appropriately to intensive longitudinal data as collected in these two studies.

Interpreting Interactions: Simple Intercept and Simple Slope Analyses

In the main article, several interaction terms were reported that were significantly different from zero. To aid interpreting these interactions, simple intercepts and slopes were calculated using the online utility available at (Preacher, Curran, & Bauer, 2006). Estimates of simple intercepts and slopes and associated 95% confidence intervals are provided in Table S1. These estimates represent the effects of a predictor (i.e., study membership, person-mean centered practice duration, person-mean centered practice frequency) within the two levels of the experimental manipulation. For the positive emotions reports, an interaction emerged between study membership and meditation type. As shown in Table S1, participants assigned to MM appeared to report higher levels of positive emotions in the second study, whereas no difference was observed between the two studies for those assigned to LKM. As described above, this interaction term was thus retained in all subsequent models for positive emotions. In addition, positive, significant coefficients were observed for the effects of daily practice duration and frequency within-persons in the LKM condition; in the MM condition, the effects were closer to and not significantly different from zero. Finally, for negative emotions, the effect of daily practice duration was significantly different from zero for those in the LKM condition but not the MM condition. As discussed in the accompanying article, the negative coefficient implies that on a given day, lower levels of negative emotions were associated with higher levels of meditation practice duration, and vice versa, for those in the LKM condition but not those in the MM condition.

Table S1

Simple Slope Analyses for Interpreting Interaction Terms

MM / LKM
Positive Emotions
b / 95% C.I. / b / 95% C.I.
Cond × Study / Simple Intercept / 1.449** / [1.272, 1.626] / 1.743** / [1.567, 1.919]
Simple Slope / 0.374** / [0.154, 0.595] / -.064 / [-0.283, 0.154]
Cond × Duration_PC / Simple Intercept / 1.355** / [1.075, 1.617] / 1.427 / – a
Simple Slope / 0.001 / [-0.000, 0.002] / 0.004** / [0.003, 0.005]
Cond × Frequency_PC / Simple Intercept / 1.376** / [1.021, 1.731] / 1.349** / [0.939, 1.759]
Simple Slope / 0.018 / [-0.014, 0.050] / 0.086** / [0.053, 0.119]
Negative Emotions
b / 95% C.I. / b / 95% C.I.
Cond × Duration_PC / Simple Intercept / 0.469** / [0.331, 0.607] / 0.533** / [0.402, 0.664]
Simple Slope / 0.000 / [-0.000, 0.001] / -.001** / [-0.002, -0.000]

Note. Practice_PC = person mean-centered meditation practice, in number of minutes spent meditating. Practice_M = mean meditation practice levels. Cond = experimental condition

a Standard errors approached infinity; thus, neither a confidence interval nor p-value are available.

References for OSM

Bolger, N., & Laurenceau, J.-P. (2013). Intensive Longitudinal Methods: An Introduction to Diary and Experience Sampling Research. New York: Guilford Press.

Brantley, J. (2007).Calming Your Anxious Mind: How Mindfulness and Compassion Can

Free You from Anxiety, Fear, and Panic. Oakland, CA: New Harbinger Publications.

Brantley, M., & Hanauer, T. (2008). The Gift of Loving-kindness. Oakland, CA.: New

Harbinger Publications.

Curran, P. J., Mcginley, J. S., Bauer, D. J., Hussong, A. M., Burns, A., Chassin, L., … Zucker, R. (2014). A moderated nonlinear factor model for the development of commensurate measures in integrative data analysis. Multivariate Behavioral Research, 49, 214-231. doi:10.1080/0027317.2014.889594

Isgett, S. F., Algoe, S. B., Boulton, A. J., Way, B., & Fredrickson, B. L. (2016). Common variant in OXTR predicts growth in positive emotions from loving-kindness training. Psychoneuroendocrinology, 73, 244-251. doi:10.1016/j.psyneuen.2016.08.010

Littell, R.C., Stroup, W.W., Milliken, G.A., Wolfinger, R.D., & Schabenberger, O. (2006).SAS for Mixed Models. SAS Institute.

Loundon, S. (2001). Blue Jean Buddha: Voices of Young Buddhists. Summerville, MA:

Wisdom Publications.

Muthén, L. K., & Muthén, B. O. (1998-2015). Mplus User's Guide. Seventh Edition. Los

Angeles, CA: Muthén & Muthén.

Preacher, K. J., Curran, P. J., & Bauer, D. J. (2006).Computational tools for probing interaction effects in multiple linear regression, multilevel modeling, and latent curve analysis.Journal of Educational and Behavioral Statistics,31, 437-448.doi:10.3102/ 10769986031004437