Derived originally from ancient Buddhist practice, mindfulness (“[…] to pay attention in a particular way – on purpose, to the present moment, nonjudgmentally,” Kabat-Zinn 1994
, p. 4) has received considerable attention and developed enormously over the past 30 years. A large amount of studies showed the beneficial effects of mindfulness and mindfulness-based interventions in different domains of life (e.g., Eberth and Sedlmeier 2012
; Grossman et al. 2004
). The development of the research in mindfulness was facilitated through the recent advancement of valid and reliable measures of mindfulness (for reviews, see Baer 2011
; Sauer et al. 2013
). One of the most comprehensive instruments in terms of dimensional coverage is the Five Facet Mindfulness Questionnaire (FFMQ; Baer et al. 2006
), which was developed on the basis of the combined pool of items from five other mindfulness scales. Exploratory factor analysis of the combined 112 items yielded five clear factors labeled as observing, describing, acting with awareness, non-judging of experience, and non-reactivity to inner experience (Baer et al. 2006
). The FFMQ was therefore considered to measure mindfulness through these five facets. The scale demonstrated satisfactory psychometric properties and has been validated across cultures (e.g., Aguado et al. 2015
; de Bruin et al. 2012
; Michalak et al. 2016
Typically, the five facets were positively inter-correlated and also loaded on a single overall mindfulness construct (Baer et al. 2008
). A second pattern was found with observing being different from the others. More specifically, often a non-significant correlation between observing and non-judging (e.g., Baer et al. 2006
; Lilja et al. 2011
; Michalak et al. 2016
) was found, and only four of the facets (all but observing) constituted to the overall mindfulness construct. The former pattern was found in participants with meditation experience (meditators) or patients after participating in mindfulness-based cognitive therapy (MBCT; Segal et al. 2013
), while the latter was found in participants without meditation experience (non-meditators) or patients before participating in MBCT (Baer et al. 2006
; Gu et al. 2016
; Williams et al. 2014
). By the same token, a few studies used confirmatory factor analysis (CFA) to compare the correlated five-factor model (which assumes that the scale measures five distinct, but related, facets of mindfulness) with the hierarchical five-factor model (in which the five factors were indicators of an overall mindfulness construct). They found out that the latter performed worse than the former, especially among the non-meditators (e.g., Hou et al. 2014
; Veehof et al. 2011
). In a nutshell, we encountered convincing evidence suggesting that the inter-correlations of the five facets and the hierarchical five-factor model of the FFMQ cannot be replicated consistently across different samples (meditators vs. non-meditators) and time points (before vs. after mindfulness-based cognitive therapy).
What could account for the discrepancy regarding the observing facet between people with and without meditation experience? Baer (2016
) suggested that attention to the present moment can be either reactive and judgmental (i.e., not mindful for the non-meditators) or open, curious, and accepting (i.e., mindful for the meditators). People with limited meditation experience tend to observe in a judgmental way, while experienced meditators tend to observe mindfully (Baer et al. 2006
). Reviewing previous research on the findings of the FFMQ, two additional interpretations were put forward. The observed effects might be related to (1) the constrained scaling of the FFMQ and (2) the heterogeneous sampling.
First, enhanced scores (because of being a meditator or completing a MBCT training) might yield a ceiling effect when there is a constrained scaling. This hypothesis was preliminary supported by a few observations in the literature. At first, we found that across different samples using different language versions of the FFMQ, the mean scores of certain facets were always higher than the middle value, even among people with very limited or no meditation experience (e.g., Aguado et al. 2015
; Taylor and Millear 2016
). Second, combining the studies that used the English version of the FFMQ (see Table 1
), we also noticed that across the 11 samples from five different studies, a high mean score for a given facet has usually been linked to a low standard deviation.
Descriptive data of the five facets of mindfulness across English-speaking samples
Community sample (n = 293)
Students (n = 259)
Highly educated (n = 252)
Meditators (n = 213)
Students (n = 735)
Curtiss and Klemanski 2014
Heterogeneous clinical sample (n = 153)
Pre-MBCT participants (n = 238)
Post-MBCT participants (n = 238)
Community sample (n = 940)
Participants with depressive disorder (n = 424)
Meditators (n = 235)
Correlation between M and SD (Pearson’s r)
As shown in Table 1
, the correlations (Pearson’s r
) between the mean and the standard deviation across the 11 samples were − .53 for the observing facet, − .92 for the describing facet, − .62 for the awareness facet, − .83 for the non-judging facet, and − .53 for the non-reacting fact, respectively. This negative association might accord to the hypothesis that the meditation experience made participants reach a similar level of mindfulness facets and thus made them more homogeneous. However, it might also be that the scales were skewed: certain facets (e.g., describing and non-judging) might have been constrained by the scaling even among people without meditation experience. This is in line with a few studies that reported the multivariate non-normality of the FFMQ scales (e.g., Christopher et al. 2012
; Gu et al. 2016
Second, while one can expect randomly drawn samples to show the same pattern of inter-correlations of the five facets, this will change once a training is involved (as learning curves are different and the training effects would be differently sustained), or when the meditators with different levels of experience are jointly investigated. The reason is that the variation in the scores increases when some yield a stronger training effect than the others (just as when the meditators and the layperson are mixed in one sample), and this subsequently also affects the covariations; i.e., there will be higher inter-correlations. Furthermore, if for some individuals some components of mindfulness are more easily improved than the others or faded out more quickly than the others once training was stopped, then the inter-correlations among the facets will also be affected. Not systematically controlling for these factors might explain the differences found for different studies as mentioned above.
However, no comparison has been made between people who are currently practicing meditation and people who had meditation experience in the past and stopped training at the moment. The criteria to categorize participants as meditators remained somewhat unclear and inconsistent in the literature. A few selected open issues are as follows: how intensive should their meditation experience be until they could be considered as meditators; do meditators have to practice meditation regularly at the moment; or once they practiced meditation in the past, is it already enough to categorize them as meditators. People stopped their meditation practice for various reasons, such as a change of personal interests or not having enough time. Simply framing them as meditators because of their past meditation experience or simply framing them as non-meditators because they gave up meditation was not appropriate. The meditation experience of these people could range from “tried once and never was interested again” to “constantly practiced for a few years and then stopped.”
In sum, we listed three possible explanations (incl. Baer’s and ours) for the observed discrepancies that could have an impact on the inter-correlations and factor structure of the FFMQ in different samples and we aim to test all proposed interpretations empirically. First, we aimed at exploring if there is constrained scaling of FFMQ, if it occurs in our German-speaking samples as well, and if this leads to an alternative explanation of the current issue regarding the observing facet. Second, we aimed at ruling out possible diversities in the sample and for the first time exploring the unique features of an unstudied group in the mindfulness research—the past meditators, who practiced meditation a while ago, but stopped for different reasons. Third, we aimed at replicating the factor structure of the FFMQ in three German-speaking samples, which contain participants with different levels of meditation experience (the current meditators, the past meditators, and the non-meditators). Finally, to highlight these effects (components of mindfulness being differently related to each other in a predictable way in different samples), we aimed at illustrating how the factor structure of the FFMQ changes from level to level across participants with different levels of meditation experience by employing a top-down method. We expect to provide an empirically based answer to the question why the discrepancy regarding the loading of the “observing” facet on an overall mindfulness construct occurs.
In total, 2582 participants registered for the study on the website, of which 2474 participants completed the questionnaire. Seventy-two participants’ data were omitted because they gave unusual/inconsistent responses (e.g., those who rated at least 80% of the items with the same value, or those who claimed to have no meditation experience but reported at the same time practicing meditation regularly/unregularly). Participants who did not specify whether they were practicing meditation or what their meditation type was, as well as those who practiced meditations other than Buddhist-based meditations (such as Christian meditation, Yoga, and Tai Chi), were not included for further analysis (n = 155). The final sample consisted of 2247 German-speaking volunteers (571 men, 1676 women). Three samples were identified according to their meditation experience. The current meditators (sample 1, n = 745) comprised a sample of adults who were currently practicing meditation and most of them had intensive meditation experience. The past meditators (sample 2, n = 791) consisted of participants who had practiced meditation in the past, but have currently stopped practicing. The non-meditators (sample 3, n = 711) comprised people who had no experience with meditation at all.
The meditation experience of the current meditators and the past meditators was measured following a procedure adapted from Baer et al. (2008
). First, participants were asked if they had any meditation experience before (yes; yes, but a while ago; no). If they answered “yes” or “yes, but a while ago,” they were instructed to provide the following information: (1) duration of regular practice (less than 1 year; 1–5 years; 6–10 years; more than 10 years); (2) frequency of meditation sessions (less than once a week; 1–2 per week; 3–4 per week; 5–6 per week; 7 or more per week); and (3) length of a typical meditation session (less than 10 min; 10–20 min; 21–30 min; 31–45 min; 46–60 min; more than 60 min). If they answered “no,” they were instructed not to provide the information. In addition, participants were also asked about the type of meditation they were practicing or have been practicing. A list of options was provided. Participants could mark all that apply from the following: Mindfulness, Breathing, Zen, Focused-awareness, Vipassana, Tibetan, Samatha, and others. They were required to specify which applied. The characteristics of the meditation experience for the current meditators and the past meditators are shown in Table 2
Meditation experiences of the current and past meditators
Duration of regular practice
Less than 1 year
More than 10 years
Frequency of sessions
Less than once a week
1–2 per week
3–4 per week
5–6 per week
7 or more per week
Length of typical session
Less than 10 min
More than 60 min
Type of Buddhism meditation a
Others (please specify) b
As shown in Table 2
, compared to the past meditators, the current meditators had extensive meditation experience. Around one third of them had meditated regularly for more than 6 years and around another one third had done so for 1 to 5 years, and the remaining third had meditated less than a year. Most of them (85.1%) reported practicing meditation quite frequently (more than once a week). A typical practice session for most of them (68.5%) lasted for 10 to 30 min. In contrast, slightly less than two thirds of the past meditators (65.1%) stopped practicing meditation regularly within a year. The majority of them (67.9%) practiced less than once a week, while a typical practice session for them usually lasted less than 20 min (79.0%).
Participants’ age ranged from 18 to 80 years (M = 42.7, SD = 11.9) and more than half of them (n = 1418, 63.1%) had a university degree or were studying at the time they filled in the questionnaire. Around half of the participants classified themselves as Christians (n = 1189, 52.9%), while only a few participants adhered to other religions (n = 95, 4.2%), such as Buddhism (n = 55, 2.4%). The rest of the participants either reported that they did not have a religion (n = 674, 30.0%) or chose not to provide an answer (n = 259, 11.5%). Around a quarter of all participants (n = 539, 24.0%) were practicing their religion and slightly more than one third (n = 851, 37.9%) were not, while the rest of the participants either did not report religious affiliation or chose not to answer this question (n = 857, 38.1%).
Participants were requested to complete the FFMQ on a well-established website (www.charakterstaerken.org
; hosted by the Section of Personality and Assessment at the Department of Psychology at the University of Zurich) for research purposes between May 2015 and June 2017. The study was promoted by different means through the Internet (e.g., online forum, social media, and the university mailing list). To reach a larger audience of meditation experts, the contact details of German-speaking meditation practitioners were sought on the Internet, after which an invitation letter/email, as well as the instruction of how to participate in the study, was sent to the meditation experts. The volunteers registered on the website with their personal computers and completed the questionnaires online. Respondents were not paid for participating but were provided an automatically generated feedback of their individual results. The procedure was in line with the guidelines of the Ethics Committee of the Department of Psychology at the University of Zurich.
The Five Facet Mindfulness Questionnaire-German
was adapted (FFMQ; Baer et al. 2006
; German translation from Michalak et al. 2016
The FFMQ (Baer et al. 2006
) is a self-report questionnaire. It consists of 39 items, which measure mindfulness as a trait with five facets: observing, describing, acting with awareness, non-judging of experience, and non-reactivity to inner experience. Answers are given on a 5-point frequency scale ranging from 1 = “never or very rarely true” to 5 = “very often or always true”. The instrument showed adequate psychometric properties across different samples. For instance, Cronbach’s α
ranged from .75 (non-reacting) to α =
.91 (describing) in the original publication (Baer et al. 2006
) and .74 (observing) to .90 (non-judging of experience) in the German version (Michalak et al. 2016
Descriptive statistics included internal reliability of instrument (using Cronbach’s alpha), mean, standard deviation, and correlations with demographics, as well as distribution characteristics (using skewness, skewness divided by the respective standard errors, and kurtosis).
One-Way Analyses of Covariance
Differences in mindfulness levels (in the form of five facets) across three samples were assessed using analysis of covariance (ANCOVA) with assumption of homogeneity of slopes was met. Demographics (e.g., age and education) were controlled as covariances as they were shown to be related to mindfulness in previous studies (Baer et al. 2008
). Subsequently standardized effect sizes were calculated using Cohen’s d
family of effect sizes (Cohen 1988
). According to Cohen’s logic, an effect size of .80 or larger was considered as large, .50–.79 as medium, and .20–.49 as small, and an effect size smaller than .20 as negligible.
Cluster analysis was conducted to group the past meditators based on their different profiles of scores on the five FFMQ scales and their meditation experience (duration, frequency, and length) using the clustering algorithm PAM (partitioning around medoids; Reynolds et al. 2006
). The “cluster” package in R was used and we chose “Gower distance function” as the distance measure because the data consisted of both continuous variables (i.e., the FFMQ) as well as ordinal variables (i.e., duration, frequency, and length). The silhouette width (Rousseeuw 1987
) was used to identify the optimal cluster solution, which is an aggregated measure of how similar an observation is to its own cluster compared to its closest neighboring cluster. The metric can range from −1 to 1, where higher values indicate a better fit.
CFA was performed to examine the factor structure of the FFMQ using robust maximum likelihood (RML) estimation with the R package lavaan (Rosseel 2012
). RML was used due to the non-normality nature of the scales. Three models were specified, namely (1) a correlative five-factor model, which was identified via EFA and allowed the five factors to inter-correlate, (2) a hierarchical five-factor model, in which the five factors were themselves indicators of an overall mindfulness factor, and (3) a hierarchical four-factor model, which defined describing, awareness, non-judging, and non-reacting as facets of an overall mindfulness construct but excluded observing.
A top-down method namely the hierarchical factor analysis (HFA; Goldberg 2006
) was employed to highlight the components of mindfulness being differently related to each other in a predictable way in different samples. Iteratively the number of factors extracted by the algorithm was increased (e.g., one, two, and three) until one reached a point where a component would have been extracted on which no variable has its highest factor loading. Factors were represented as rectangles, whose width corresponds to the factor’s size, i.e., to the amount of variance accounted by that factor. The factor scores of adjacent factor solutions were correlated with each other, and the salient relations (r
> .35) were represented using arrows. By this means, we could examine how the factors unfold and how they split up or stayed stable from solution to solution. This method elucidates the hierarchical structure of a set of variables top-down, as opposed to the bottom-up tradition that first identifies lower order trait structures and then defines higher order traits based on the patterns of covariance among those. In the present case, we expect that the observing items will not be represented well in the first un-rotated principle component and gain independence (i.e., form a separate factor) at earlier stages in the unfolding in the non-meditators compared to the current meditators and the past meditators.
Data Availability Statement
The primary goals of the study were to (1) examine the constrained scaling of the FFMQ; (2) explore the unique features of the past meditators; (3) replicate the factor structure of FFMQ using CFA in German-speaking samples; and (4) using HFA to illustrate how the factor structure of FFMQ changes from level to level across participants with different levels of meditation experience.
We indeed found evidence that three scales of the FFMQ were constrained (i.e., observing, describing, and non-judging). An average of 15% of the respondents of these scales reaching the highest value was too high and we would not expect it from a normal distribution. In an ideal situation of the normal distribution whose mean equals three and which ranges from one to five, the respondents reaching the highest value should be less than 3% (simulated with a sample size of 10,000, and SD of .50 to .75). This meant that some items of the FFMQ were too simple, and those who could have had a higher score were more prone to reach the highest level possible, in particular the meditators or participants after the MBCT. Therefore, as reported in previous studies (Baer et al. 2006
; Gu et al. 2016
; Williams et al. 2014
), in those samples whose scores were skewed in a similar way, the facets correlated higher and all load on the overall construct, whereas for the non-meditators or participants before the MBCT, not as many people reached the highest value possible and thus the correlation and the loadings were also lower. These results could also be linked to our CFA results. Because our non-meditators scored high in the five facets and reached the constrained scaling in observing, describing, and non-judging as what usually happened in the meditators, the CFA model which assumes all five facets loaded on the overall mindfulness also gained a great fit despite a slight lower loading of the observing facet across all three samples. Unlike what has been found in the literature, observing loaded significantly on the overall construct of mindfulness even in our non-meditator sample. This should be link to the fact that our non-meditator sample obtained higher mean scores on almost all facets compared to previous samples. The reason behind it could be that our non-meditators were probably very open-minded and curious because they came and filled in the questionnaires on our website voluntarily without any incentives and were curious to get a feedback on their own scores. On the other hand, our sample was also rather well educated: More than half of the participants had a university degree or were studying at the time they filled in the questionnaire.
Compared to the past meditators and the non-meditators, the current meditators had extensive meditation experience. This was why the current meditators scored significantly higher in all five facets of mindfulness. The fact that the past meditators were similar to the non-meditators in acting with awareness, non-judging, and non-reacting, but scored higher in observing and describing as well as different effect sizes among the five facets across all three samples suggested that there might be different cultivation and extension procedures for the five facets of mindfulness. For example, once people learned how to link body sensations to emotions, they would be able to observe and describe feelings more precisely and this is unlikely to wash out fast, maybe not at all. On the other hand, the other three facets might need continuous training or regular training at the current moment to develop them, as they were cultivated by actively applying certain techniques (e.g., using breathing as an anchor). The fact that those past meditators with intensive trainings (cluster 3—more than once a year, more than once a week, and more than 10 min each session) scored higher in all five facets than those past meditators who practiced less (cluster 1 and cluster 2) revealed that giving up training would not be too much of a loss as long as one trained intense enough in the past. These results supported the hypothesis that the past meditators should be separated from the current meditators and the non-meditators and be investigated as an independent sample. Although some of them did have extensive meditation experience, stopping practicing meditation made them notably different in terms of their mindfulness level from both the current meditators and the non-meditators. This can also be revealed in our findings of HFA, how the factors unfolded from solution to solution differed across all three samples.
The HFA illustrated a clear picture on how observing facet loaded on the overall construct of mindfulness differently across the three samples. Observing loaded lower on the overall construct of mindfulness across all three samples, and much lower for those who had less meditation experience (i.e., the past meditators and the non-meditators). Regarding people with no meditation experience, the facet observing emerged at the third iteration and stayed unchanged since then. However, with respect to the current meditators, observing was a more significant part of the general factor from the very beginning and crystallized as a unique factor only at the forth iteration. Furthermore, the results also supported the proposed explanation by Baer et al. (2006
): People with different levels of meditation experience may observe differently with non-meditators tending to observe in a judgmental way and the meditators tending to observe mindfully. This pattern could be discovered within the development of the factor structures from solution to solution. For the non-meditators, instead of being fused with other items into one factor, observing became a separate factor quite early, while for the current meditators and the past meditators, there were levels where observing items were fused with the items from other facets (i.e., awareness and non-reacting). This could provide a statistical support for the statement that meditators and non-meditators observed differently, with the former being more likely to observe mindfully and the latter being more likely to observe without being attentive and non-reactive.
All in all, the study provided empirical answers to the issues of the observing facets of the FFMQ. It could be a combination of different reasons: (1) the non-normality nature of the FFMQ; (2) the heterogeneous sampling, which leads to the different procedures of the cultivation as well extinction of the five facets; (3) differences between laypersons and meditators on how they observe. However, these effects were currently mixed up and future studies might think of investigating each hypothesis specifically.
Limitations and Future Research
The results of the study must be interpreted with caution due to several limitations. First, although by using internet recruitment we could reach large and geographically distributed populations (all over German-speaking countries), it could also cause a selection bias in our data. For instance, we were more likely to recruit participants who were interested in positive psychology in general or who were curious about themselves. We tried to avoid this selection bias by advertising our study as much as we can and by writing invitation letters/e-mails and addressing the importance of the study to the targeted participants. It is different as simply putting the questionnaires on the website and waiting for individuals who happen to have Internet, visit the website, and decide to participate in the survey. However, we must acknowledge that at least some participants were biased, and this could affect the representativeness of the study, and it could also be the reason why we have a rather mindful non-meditator sample. Given that it is hard to reach the relevant sample (in particular the meditators), this sample method is still effective despite its limitations. Future studies could consider it as the first approach and expand to other means such as paper-pencil questionnaire, telephone, or personal interview to obtain a broader sample.
Second, we discussed the impact of meditation experience on mindfulness cultivation (i.e., whether one is currently practicing meditation and how intense one practices) and we found inconsistency across the five facets (i.e., some were easier to cultivate and some fade away faster). However, the practicing experience alone is not the full picture, as the participants in our study as well as reported in other studies all uniformly underwent similar trainings. Other factors should also play a role, which was not controlled in our current study. For instance, why the past meditators stopped their meditation practice could be of importance, and future studies might explore this question in more details. In addition, we could well imagine participants’ education, personal experiences, and their cognitive ability (e.g., memory and learning) could also influence their mastery of mindfulness skills. Taking describing as an example, which encourages practitioners to describe, label, or note the observed phenomena by covertly applying words (Baer et al. 2004
), was related to education as shown in previous studies (Baer et al. 2008
; Van Dam et al. 2009
). Therefore, future studies should consider controlling the possible covariance.
Third, the measure of the meditation experience was not optimal in the current study. The meditation experience of participants was asked by using ordinal scales, which was hard to compare in further analysis. Future studies might consider asking directly the concrete numbers to make linear modeling possible. In addition to the general questions, one should also consider asking participants’ daily behavior in specific context (e.g., event sampling methods).