The study complied with the ethical standards of the Declaration of Helsinki39and was approved by two German ethics committees (Technische Universität Dresden: SR-EK-270052021; Friedrich-Schiller Universität Jena: 2021-2121-BO).
During the lecture-free period between the winter semester of 2020/2021 and the summer semester of 2021, lecturers willing to participate in the study were recruited via mailing lists and personal contacts of the Mindful Universities network. Inclusion criteria for lecturers were the participation in an online workshop led by a certified MBSR senior teacher (up to three appointments of 2 h each), the willingness to implement mindfulness exercises and a control condition in at least two teaching sessions each. Mindfulness was not allowed to be the main topic of the course curriculum itself. Additionally, only participants (lecturers and students) older than 18 years, and capable to give informed consent were included. Lecturers recruited students during their first teaching session of the semester via verbal and written information about the study (e.g., study flyer). Participation was voluntary for both lecturers and students. At the end of the study, n = 30 students were randomly drawn from the post-sample and received a monetary reimbursement. Psychology and biomedical engineering students received study credits for participation.
Procedure and intervention
After informing about the study, lecturers provided the links for a baseline survey (lasting approximately 15 min for students and 5–10 min for the lecturers). Completion took place until the next teaching session. In the following weeks, lecturers presented two conditions (course start with vs. without brief mindfulness exercise), in alternating order. The study design followed an ABAB design. Thus, all student participants received both conditions. We chose this design to ensure an easy implementation for lecturers despite varying durations between courses. The intervention consisted of a brief (3- to 4-min) mindful breathing space, as introduced in the MBCT curriculum38. The aim of the exercise is to first mindfully observe their own state of mind (all present thoughts, feelings, body sensations), second, observe their own breath and follow its movement through the body, and third, turn back mindfully to one’s own state of mind and potential changes in comparison to the beginning of the exercise.
Across conditions, students and lecturers completed an interim survey I (1 min.) before the regular course started (for the treatment condition: after the breathing space). At the end of each course, the students had the option to complete an interim survey II (1 min.) to re-assess their current mental state.
After the last teaching session of the semester, all participants were asked to complete a post-survey (again 15 min for students, 5–10 min for lecturers). All surveys were implemented in LimeSurvey and administered online.
Our primary outcomes were the current mental states of students and lecturers after the implementation of a brief mindfulness exercise at the beginning of the course, compared to no such exercise (interim survey I). Students received the same items again after each teaching session (interim survey II). At each interim survey, students and lecturers were asked to assess their current mental state with visual analogue scales ranging from not at all (0) to completely (100) with the following items: „How do you feel at this moment?” (a) “concentrated”, (b) “alert”, (c) “present”, (d) “distracted by thoughts”, (e) “energized”, (f) “stressed”. One extra item assessed current mood with a slightly different scale (g) “My mood is” negative (0) to positive (100). Only prior to the teaching session (interim survey I), students were also asked to rate their (h) “motivation for the upcoming teaching session”.
In the post-survey, all participants were asked to evaluate the intervention throughout the semester regarding its effect on their mental state with separate items asking for (a) “concentration”, (b) “alertness”, (c) “presence”, (d) “distraction by thoughts”, (e) “energy”, (f) “stress perception”, (g) “mood”, (h) “learning success”, (i) “contentment”, and (j) “closeness between participants within the course”. Lecturers were asked to answer these items for themselves and for the perceived effects on students’ mental states.
At baseline- and post-survey, students received multiple questionnaires: To assess trait mindfulness, we used the Freiburg Mindfulness Inventory (FMI-1440; Cronbach’s α at baseline = 0.77), which includes 14 items, each presented with a 4-point response scale ranging from rarely (1) to almost always (4). A mean score was calculated after excluding the negatively phrased item (item 13) as recommended41, 42.
To assess personality dimensions, the brief 25-item version of the Personality Inventory for DSM-5 (PID-5 BF43) was implemented. Responses are rated with a Likert scale ranging from very false (0) to very true (3). The PID-5 has five subscales (Negative Affectivity, Antagonism, Disinhibition, Psychoticism, and Detachment), all of which are calculated by building a mean score. In this study, we only focused on the Negative Affectivity subscale (Cronbach’s α at baseline = 0.64).
Additionally, we assessed psychological stress using the Perceived Stress Scale (PSS-1044; Cronbach’s α at baseline = 0.85). It includes 10 items, each rated on a 5-point response scale ranging from (never) 0 to very often (4). A sum score was calculated.
Further, to assess trait abilities in attentional control, we used the Effortful Control subscale (19 items) of the Adult Temperament Questionnaire, which contains a total of 77 items (ATQ45, 46). Internal consistency at baseline for the Effortful Control subscale was Cronbach’s α = 0.74. The items are assessed with a Likert scale ranging from not at all applicable (1) to completely applicable (7). We calculated a mean score for further analyses.
Finally, we assessed meditation experience at baseline as a control variable (“meditation frequency”, ranging from 0 = not at all to 100 = a lot and “experience” in years). Only lecturers were asked to rate their “experience in instructing meditation exercises” on a visual analogue scale ranging from not at all (0) to a lot (100).
To control for whether students actually participated in the entire 3- to 4-min exercises, we included four options for students to rate the offered “condition” in the interim survey I: (a) “The exercise was offered today, and I fully participated”, (b) “The exercise was offered, but I only participated partly”, (c) “The exercise was offered, but I did not participate” and (d) “There was no exercise offered today” (control condition). To further verify the students’ answers regarding the condition, we also asked lecturers whether they had offered the exercise or not.
Students also rated the following items: “teaching session held in person” vs. “online teaching session” vs. “the teaching session was recorded”, as well as camera “off” vs. “on” during the exercise, and quality of internet connection during exercise (“stable” vs. “not stable but did not bother me” and “not stable and it bothered me”).
The lecturers rated their experience during the exercise and their adherence to the instructions (“During the instruction, I …” (a) “felt well”, (b) “felt authentic”, (c) “was following the standardized instructions”) with a visual analogue scale, ranging from not at all (0) to completely (100).
Power analysis and descriptive statistics
All analyses were done using R version 4.2.147 and MPlus version 8.148.
We used the summary-statistics-based power analysis to calculate the minimum sample size (https://koumurayama.shinyapps.io/summary_statistics_based_power/;49). As we are not aware of another study with a similar within-group design, we based the power analysis on the meta-analytic results of Dawson et al.50, who reported an effect size of d = − 0.47 for the post-comparison of distress scores between mindfulness-based interventions and passive controls. The summary-statistics-based power analysis tool requires t-values, but because the t-distribution is almost identical to the z-distribution in large samples51, we used the reported Z-value (see Supplementary Fig. 1,49). Thus, based on Z = 7.06 and a level-2 sample size of N = 2201, the minimum sample size to achieve 80% power is N = 324. With an expected drop-out of approximately 30%, N = 200 * (1 ÷ 0.7) ≈ 462 student observations are needed. The data analysis of potential effects on the lecturers was exploratory, that is why no power estimation was done.
Next to our main outcome, we wanted to determine differences between pre- and post-measurements of students. To conduct an Intent-to-Treat analysis and thus account for missing data at post-measurement, we used regression analyses with full information likelihood estimation. This estimation method is implemented in the sem() function of the lavaan package52. The respective trait variables were used as outcomes and time as a dummy-coded predictor (pre = 0, post = 1). Effect sizes indicate a very small effect for all d < 0.1, a small effect for all 0.1 ≤ d < 0.3, and a moderate effect for all 0.3 ≤ d < 0.553.
Spearman’s correlation coefficients were calculated to determine construct validity between primary and secondary variables as well as test–retest reliability (see Supplementary Information Figs. S1 and S2).
Our analysis differed slightly from our preregistered analysis plan regarding three aspects. First, instead of using eight different outcome variables, we aimed to use composite scores of the primary outcome items to reduce dimensions. Second, to probe our second hypothesis, we did not use change scores, but the mental state scores reported after the lecture (interim survey II). Third, instead of including every control variable step by step, we modelled a base model with all control variables, and added the predictor of interest (condition) in a second step.
To reduce dimensions, we applied exploratory factor analyses. As our data included multiple measurements per subject and condition, we followed the guidelines for multilevel exploratory factor analysis (for more details see explanation and Tables S1–S4 in the Supplementary Information54). Briefly, we computed an exploratory maximum likelihood factor analysis across all repeated measurements, conditions, and individuals, and two separate factor analyses for within-subject and between-subject variance. Finally, we performed a multi-level exploratory factor analysis using MPlus by computing the factor structure at each within-subject and between-subject level simultaneously. After examining all factor solutions carefully, we averaged the items “concentration”, “energy”, “presence”, and “alertness” to build factor 1 (presence composite score, Cronbach’s α = 0.90), and “stress” and “distraction by thoughts” to build factor 2 (stress composite score, Cronbach’s α = 0.66). Thus, the presence composite score refers to a state of being engaged in or focused on the present moment. The stress composite score reflects a state of being distracted or mentally absent and stressed in the current situation.
We used linear mixed-effects models to address the hierarchical structure of our data. All numeric predictors were scaled by two times the predictor’s standard deviation. All outcome variables were scaled by one standard deviation before entering into the models. According to55, this procedure allows to directly compare regression coefficients of continuous predictors and (untransformed) binary predictors. Models were visually checked for relevant model assumptions using the check_model() function56, which also includes the calculation of Cook’s distance. In contrast to the analysis plan in our preregistration, we did not conduct a sensitivity analysis in case of outliers, but decided to use a robust estimation method, which has the advantage that no observations need to be excluded. Therefore, in the case of influential cases, multicollinearity, non-normal distribution of residuals, or heteroscedasticity, we estimated the models again using the robust estimation method offered by the robustlmm R package57, which applies a Huber function aiming for more robust variance components and random effects. Here, we used default settings (computation method = ”DAStau”, k = 1.345, s = 10). Only if the robust and original models showed different results regarding the significance of predictors, we report the robust models rather than the original ones.
To estimate the effect of the mindfulness exercise (vs. control condition), we used the observations from students who participated in the full 3- to 4-min mindful exercise. If not indicated otherwise, “participation” refers to full participation in one exercise (not during the whole semester). Please note that we excluded all student observations for which no or partial participation in one of the exercises was reported. However, these data are reported in the Supplementary Information (Table S6).
To test our first hypothesis (H1), we probed whether the condition (participation in the mindfulness exercise vs. control, fixed effect) predicted students’ mental states at the beginning of the teaching session. The (a) presence composite score, (b) stress composite score, (c) mood item, and (d) motivation for the lecture item served as outcome variables in four distinct models. Some students attended more than one course, which is why subjects and courses were partially crossed and not nested factors58. Using the lme4 package58, it is possible to address partial crossing by modelling two random intercepts. Therefore, we added random intercepts for subject IDs and course IDs in a first model. Additionally, since all participants received both conditions in alternating order, in a second step, we added random slopes for “condition” (varying across subjects and courses). The second model revealed boundary issues indicating that there were too few observations given the model complexity. Consequently, the simple random intercepts model was kept as the final model. This strategy was used for all four outcome variables. All models were controlled for the variables age, gender, meditation experience, meditation frequency, and modality of the lecture (online vs. in-person).
To probe the second hypothesis (H2) stating that the described effects of the mindful exercise condition last throughout the teaching session, we fitted three random intercept models with the (a) presence composite score II, (b) stress composite score II, and (c) mood item II as outcome variables (using the data from interim survey II after the teaching session), condition (participation vs. control) as the independent variable, and subject IDs, as well as course IDs as random intercepts. Also, the same control variables were included as in the analysis of the first hypothesis.
To address our exploratory research questions (secondary outcomes), we added several control variables to our base models (see H1):
First, to probe whether baseline variables moderate the effect of the mindfulness exercise on the mental state of students (E1), we added different interaction terms to our base models (Baseline variable × Condition). More specifically, we modelled (a) the effect of Psychological Stress × Condition on the stress composite score, (b) the effect of trait Attentional Control × Condition on the presence composite score, and (c) the effect of Negative Affectivity × Condition on mood before the course. In each of the models, the effect of trait Mindfulness × Condition was included as well as age, gender, meditation frequency, meditation experience at baseline and modality of the lecture (online vs. in-person).
Second, we aimed to address the potential effects of lecturers’ characteristics on students’ mental states (E2). Here, we dropped all observations of the control condition because lecturers rated these items only after instructing the mindfulness exercise. Thus, all four outcome variables were predicted by the (time-varying) lecturers’ reports of their level of (a) well-being and (b) authenticity during the instruction, as well as the (c) standardization of the instruction.
Third, because our sample of lecturers was relatively small, we refrained from using the same methods as for the student sample, and instead opted to report all lecturers’ data on a descriptive level only (E3).
Finally, next to a descriptive analysis of the evaluation (E4), we investigated which factors influenced students’ retrospective overall evaluation of the intervention at the end of the semester (subjectively rated effects on different aspects of learning and well-being). We calculated a composite score for all ten evaluation items, which we then used as an outcome variable. Participation time per condition (up to six times per control condition or full participation in the mindfulness exercise), setting (“online courses” vs. “courses held in person”), and the individual (time-varying) presence composite score, stress composite score, mood, and motivation for the courses served as predictors in this last model (E5).
Further analyses examining Condition × Time interaction effects, effects of lecturers’ experience in meditation, and effects of setting modalities of the courses on the outcome variables are reported in the Supplementary Information (see Table S6).