Meditation—particularly mindfulness meditation—has dramatically risen in popularity in western cultures over the last several decades (see Van Dam et al. 2018). Despite empirical support that meditation may be beneficial for a variety of populations and may both improve quality of life (e.g., Reibel et al. 2001) and reduce negative mental health outcomes (e.g., Tickell et al. 2019), understanding the mechanisms through which meditative practices work is crucial for further refinement. Exploring mechanisms will help to highlight the potentially crucial components of meditation, which in turn can help to maximize the benefits of practice. One of the key components of various meditations is a trained ability to focus on the present moment (Hölzel et al. 2011; Wallace 2007). As such, improved attentional ability may be one mechanism that drives the outcomes of meditative practices (Kok et al. 2013).

Understanding the mechanisms behind meditation are only possible after answering the question, “which meditation?” as meditative traditions and techniques can be drastically different from one another. Moreover, Western versions of mindfulness and other meditative practices tend to prioritize quantifiable outcomes that can be readily studied (e.g., symptom relief or productive gains) rather than embrace the holistic Eastern perspectives from which they came (see Monteiro et al. 2015, for a review). Dahl et al. (2015) described three families of meditative practices into which many can be classified. The attentional family includes focused attention (FA) and open monitoring (OM; alternatively, open presence) practices, which are centered around attention regulation (see also Lutz et al. 2008; Lutz et al. 2007). FA involves narrow, sustained attention on an object of attention, which may be internal (e.g., breath) or external (e.g., sound), whereas OM involves taking a broad and open approach to noticing experiences as they appear. The constructive family, which includes compassion and values-based practices, for example, centers around cultivating empathy or cognitive reappraisal. The deconstructive family, including Vipassana and Zen practices, centers around fostering self-inquiry and gaining insight into one’s own thoughts and behaviors.

Popular clinical mindfulness meditations, such as mindfulness-based stress reduction (MBSR) and mindfulness-based cognitive therapy (MBCT), typically integrate both FA and OM in their practices. Often, FA techniques are taught prior to OM techniques. Once foundations of focusing on the present moment have been established, practitioners are encouraged to pursue a broader focus. Researchers have begun to dismantle these mindfulness-based interventions into discrete FA and OM components (Britton et al. 2018) and have shown that both lead to improvements in emotional regulation (Lohani et al. 2020). However, mental states, including those which could ostensibly be described as FA or OM, should best be understood as continuous among many dimensions (Lutz et al. 2015), and there is likely functional overlap between FA and OM techniques. Nevertheless, to date, most studies examining the attentional effects of mindfulness meditation have not distinguished FA and OM (see Ainsworth et al. 2013 for an exception), much less plotted their dimensionality.

To connect literature on meditation to that of attention, Hölzel et al. (2011) presented the mechanisms of mindfulness as being comprised of four parts: (1) increased attention regulation, or ability to maintain focus during mediation; (2) stronger body awareness, or ability to recognize and distinguish physical sensations; (3) acute emotion regulation, or meta-awareness and non-reactivity; and (4) a change in perspective on the self, or utilizing a growth mindset (see Dweck 2000). On the other hand, Bishop et al. (2004) posited two primary mechanisms: trained self-regulation of attention (sustained attention) and a growing orientation to experience, which can be described as presentism and meta-awareness. Common to these two theories is a focus on attentional mechanisms: mindfulness training improves practitioners’ ability to sustain attention and inhibit distractions. If meditation’s benefits are due at least partially to improvements in attention, it is important to understand which aspects of attention are being trained and which types of meditations are training attention.

Posner and Petersen (1990) developed a seminal model of attention consisting of three networks that are distinct from other cognitive functions: alerting (or vigilance), orienting, and executive control (sometimes referred to as target detection, conflict monitoring, cognitive control, attentional control, executive function, etc.). The alerting network is responsible for initial attentional capture as well as sustained attention. In other words, the alerting network represents attentive temporal readiness; an alerting network that is running well will quickly attend to a novel stimulus and be prepared for the occurrence of other new stimuli. The orienting network, on the other hand, is responsible for using spatial information to anticipate stimuli in a specific direction. For example, a parent reading a book may have their attention oriented to where their child is playing, despite not looking in that direction: they are ready for something to happen. Quickly reacting to the child falling would be indicative of a high-functioning orienting network. Finally, the executive control network distinguishes between competing stimuli. Attentional functions that relate to prioritizing stimuli, such as moving attention from one stimulus to another, relies on the executive control network. The existence and relative independence of these three attentional networks have been well-researched using behavioral methods (e.g., Fan et al. 2002) and neuroimaging (e.g., Fan et al. 2005).

The definition of executive control in Posner and Petersen (1990) has been expanded upon. Multiple functions are involved in attentional juggling; Miyake et al. (2000) presented three interrelated subsystems of executive control. The shifting (or, switching) aspect of executive control refers to the ability to move quickly and accurately from one stimuli or task to another. A poorly operating shifting function might be illustrated by difficulty disengaging from a stimulus. The updating (or, updating working memory) aspect of executive control involves the constant refreshing of information in the attentional area. For example, during a busy shift, a server may be holding the orders and statuses of tables in their mind, updating their working memory as customers come, go, order food, and ask for their individual bills. The inhibition function of executive control refers to the cognitive ability to refrain from updating or shifting attention to another stimulus. Simply, it is the ability to not be distracted, and requires volition.

Multiple meta-analyses and systematic reviews have already been published on the effects of mindfulness and meditation on attention and related cognitive variables. Eberth and Sedlmeier (2012) published a brief meta-analysis, examining 39 experimental studies with inactive control groups. Across eight studies, they found a mean effect size of \( \overline{r} \) = 0.301 (equivalent to d = 0.630) where meditation predicted attentional improvements. In a second meta-analysis across 163 studies, Sedlmeier et al. (2012) distinguished different types of meditation: transcendental meditation, mindfulness meditation, and other meditative practices. Furthermore, they investigated a host of different cognitive variables. Across the studies, the relations between meditation and attention/cognition produced mean weighted effect sizes of \( \overline{r} \) = 0.28 (equivalent to d = 0.58). However, these effects were not explicated to subsystems of attention.

On the other hand, Chiesa et al. (2011) conducted a systematic review examining the neuropsychological consequences of mindfulness meditation based on the Posner and Petersen (1990) attentional networks. Based on the results across studies, the authors concluded that alerting, orienting, and executive control networks all benefited from mindfulness meditation. Further, mindfulness meditators seemed to perform better on working memory tasks that could also be considered measures of updating, relative to controls. Therefore, the review’s conclusions emphasize attentional improvements for meditators on both the general and subsystem-specific levels.

In a comprehensive systematic review, Gallant (2016) examined the effects of mindfulness meditation on executive control. Relying on the Miyake et al. (2000) model of executive control, outcomes within studies were classified as measuring shifting, updating, or inhibition. The author reviewed 12 experimental or quasi-experimental studies that each used at least one of 10 tasks. Gallant concluded that there was sufficient evidence for an effect of mindfulness on inhibition across the studies: mindfulness meditators were better at inhibition of distractors than controls. Results for shifting and updating were mixed, however.

Considering the findings of previous reviews, the purpose of the present meta-analysis is to re-examine the attentional benefits of attentional meditation techniques (FA and/or OM) in terms of the subsystems of attention and executive control. We hypothesize that (1) there will be generally positive effects of meditation on attention and executive control; (2) meditation will specifically predict improvements in the alerting attentional network and the inhibition component of executive function; (3) studies that utilized randomization will show smaller effect sizes than those that did not; and (4) differences between meditation experimental groups and active control groups will be smaller than with inactive control groups. Examination of differences in findings between FA and OM techniques is exploratory.

Method

Search Strategy

Published studies and dissertations were identified for inclusion first via database search on September 2, 2019. The PsycINFO database, which includes articles from the ProQuest database that relate to psychology, the Scopus database, and the PubMed database, was queried using the following search string within study abstracts: (mindfulness OR meditation) AND (“cognitive control” OR cognition OR executive OR (attention* NOT “mindful attention awareness scale”)). In simpler terms, the search string identified abstracts which contained either the words mindfulness or meditation, and mentions of cognitive control, cognition, executive, or attention, the latter of which could be suffixed in any way (e.g., attentional). If the only mention of attentional constructs was the Mindful Attention Awareness Scale, the study was excluded. The search was not limited by date. After studies were classified, a lateral search was conducted to find other studies that may have been missed by our original search terms.

The following inclusion criteria were used for the meta-analysis: (1) Studies were published in an academic journal or as a dissertation. Conference presentations or book chapters are not held to the same standard of peer review as academic articles or dissertations. (2) Studies were unique, empirical, and relied on statistical modeling. (3) Studies targeted to the general, healthy, and adult population. A recent meta-analysis examined the effects of mindfulness on executive control in adolescents (Dunning et al. 2019). Another examined attentional and executive function results in older adults who used mind-body exercises (Chan et al. 2019). These findings may not extend to the broader adult population. Therefore, studies that exclusively targeted healthy older adults or adolescents were excluded. (4) Studies used objective, behavioral measures relevant to attention. People have limited capacity to gauge their attentional ability (Buchanan 2016). Though studies often included self-report or physiological measures, only behavioral data were extracted. (5) Studies used experimental or quasi-experimental designs. These designs are better suited to test the causal effects of practice than correlational designs. (6) Studies had a control group. Papers that only compared meditation styles (e.g., comparing FA and OM) or exclusively recruited experienced meditators were excluded because they did not provide comparable effect sizes. (7) Meditative practices used either FA or OM techniques. Transcendental meditation (TM) and lovingkindness meditation (LKM) were excluded, as they rely on techniques different to FA or OM. Consequently, the attentional mechanisms affected by these meditations may also be dissimilar. (8) Physical activity was not a primary component of the meditation. Exercise provides benefits both physically and psychologically (e.g., Guiney and Machado 2013). To avoid misattributing the effects of physical activity to meditation, we excluded studies involving techniques that merged meditation and physical activity as the primary meditative practice (e.g., hatha yoga). Techniques or interventions that utilized physical activity to a lesser extent and in the context of a larger program (e.g., MBSR) were not excluded. (9) Studies were written in English or had readily accessible translations.

Coding of Study Variables

Behavioral data were extracted from the studies included in the analyses. Tasks were classified by the authors into seven different outcome types (Online Resource 1), reflecting distinctions outlined in literature on attentional networks and executive control (see Fig. 1). Using the Posner and Petersen (1990) model of attention, task outcomes were categorized as either pertaining to the alerting, orienting, or executive control networks. Tasks classified as measuring executive control were further sorted into inhibition, shifting, and updating categories, as outlined by Miyake et al. (2000). Several outcomes were not easily classifiable into a single category or could be perceived as measuring attention more broadly (e.g., anagram performance) or non-specific executive control (e.g., the defined intensity stressor simulation). These outcomes were thus classified only by the appropriate superordinate category (i.e., general attention, executive control).

Fig. 1
figure 1

Models of attention and executive control

Full size image

Importantly, behavioral measures often have multiple ways of measuring a theoretical construct. For example, the most commonly reported effect in the Stroop color–word task is the “interference effect,” calculated as the difference in reaction time (RT) between the incongruent and congruent conditions. Yet, the Stroop task also provides a measure of accuracy (or error) across trials. It is possible that a participant may shorten their RT after an intervention but increase their error rate. To test whether meditation leads to genuine attentional improvements (vs. a speed-accuracy trade-off) effect sizes for RTs and accuracy were compared. Further, to ensure parsimony, data extraction was limited to one RT and one accuracy measurement per outcome type in each task.

Some studies included multiple experimental or control groups. In these cases, data were extracted from the experimental group that had the strongest meditation component (e.g., a meditation retreat vs. a one-time lab intervention) and the control group that was most active (most likely to lead to attentional improvements).

Each measure for which an effect size was calculated was coded according to study characteristics and moderator variables. The following nine moderator variables were coded: (1) task type: whether the behavioral measure was reliant on accuracy, reaction time, a combination of both, or neither; (2) meditative practice: whether participants were taught (for experimental designs) or practiced (for designs which recruited experienced meditators) FA techniques, OM techniques, or both; (3) meditative experience: experienced meditators’ experience in years; (4) intervention length: the hours of meditative practice assigned; (5) random assignment: if the study conducted analyses comparing randomly assigned treatment and control groups; (6) control group: whether the study had an active control group that would likely lead to attentional improvements, an active control group that would not lead to attentional improvements, or a no-contact (NC) control group (e.g., waitlist); (7) risk of bias: studies coded having low/some/high risk of bias; (8) publication status: an unpublished dissertation or published journal article; and (9) publication year.

Statistical Analyses

All extracted data (Online Resource 2) and conversion formulas (Online Resources 3.1 and 3.2) are available as supplementary material. Effect sizes were extracted from studies using data readily available in papers, presented in appendices, or sent by contacted authors. Using suggestions found in the literature, effect sizes were converted to standardized mean differences (Cohen’s d). Effects at a single time point were converted using (1) from Cohen (1988), where the denominator is the pooled standard deviation, and the formula uses means (M), standard deviations (SD), and sample sizes (n) from each condition. This formula was also used for within-subjects designs, as recommended by Westfall (2016).

$$ d=\frac{M_2-{M}_1}{\sqrt{\frac{\left({n}_1-1\right)\times {\mathrm{SD}}_1^2+\left({n}_2-1\right)\times {\mathrm{SD}}_2^2}{n_1+{n}_2-2}}} $$
(1)

For pre-post control (PPC) and pre-post quasi-experimental (PPQ) designs, standardized effect sizes from factorial ANOVAs tend to be reported in terms of eta-squared or omega-squared. However, these effect sizes are restrictive, as they are not easily compared across multiple studies and designs. Standardized mean differences are more amenable to comparisons but interpolating them from ANOVA results is uncommon and crude. Still, quantifying these effects for the sake of meta-analyses and comparability across studies is important for literature synthesis. This meta-analysis used (2) from Morris (2008) to estimate Cohen’s d for PPC and PPQ designs, where the bias correction where the bias correction cj can be calculated via (3).

$$ d=\frac{c_2\left({M}_{2\ \mathrm{post}}-{M}_{2\ \mathrm{pre}}\right)}{\mathrm{S}{\mathrm{D}}_{2\ \mathrm{pre}}}-\frac{c_1\left({M}_{1\ \mathrm{post}}-{M}_{1\ \mathrm{pre}}\right)}{\mathrm{S}{\mathrm{D}}_{1\ \mathrm{pre}}} $$
(2)
$$ {c}_j=1-\frac{3}{4\left({n}_j-1\right)-1} $$
(3)

Effect sizes were then bias-corrected to Hedge’s g. Higher numbers indicate a more positive attentional outcome for the meditators relative to control data. Equations (4) and (5) represent the conversion formulas for Hedge’s g and its standard error respectively, where d is Cohen’s d, df is the degrees of freedom, and SEd is the standard error of Cohen’s d.

$$ g=\left(1-\frac{3}{4(df)-1}\right)\ast d $$
(4)
$$ {\mathrm{SE}}_g=\left(1-\frac{3}{4(df)-1}\right)\ast {\mathrm{SE}}_d $$
(5)

All analyses were conducted using three-level meta-regressions (see Cheung 2019) with the metafor R package (Viechtbauer 2010). Code is available in Online Resource 4. Meta-analyses are conventionally two-level multilevel models, where level 1 constitutes sampling error and level 2 constitutes between-sample error. To account for non-independence due to extracting multiple effect sizes from studies, three-level meta-analyses add an additional level between these two: one that represents within-sample error. Consequently, three-level meta-analyses can estimate heterogeneity proportions, represented by I2, at the sampling (level 1), within-sample (level 2), and between-sample (level 3) levels. Where significant heterogeneity exists as determined by three-level tests or likelihood ratio tests, moderator subgroup analyses were conducted.

Assessment of Bias

Assessment of bias was measured in five ways. Studies published in academic journals may have greater effect sizes than unpublished work (Franco et al. 2014). Studies with positive results are published more quickly than those with null or negative results (Hopewell et al. 2007). On the other hand, a discipline-wide increase in demand for methodological rigor coupled with an increase in comprehensive reporting over time may also lead to smaller effect sizes in newer studies (Fritz et al. 2012). Consequently, publication status and publication year were tested as moderators across all outcomes. Rank correlation tests were conducted as well, examining the correlation between effect sizes and sampling variances (Begg and Mazumdar 1994). Next, trim-fill tests were conducted to probe funnel plot asymmetry (Duval and Tweedie 2000). Analyses with asymmetrical funnel plots were corrected to account for publication bias, providing an estimate of the true value of Hedge’s g for the given outcome. Importantly, however, trim-fill tests are intended to be conducted for two-level meta-analyses and thus three-level results should be interpreted cautiously. Finally, authors coded randomized control trials according to Cochrane’s Risk of Bias 2 (RoB 2; Sterne et al. 2019). The RoB 2 provides criteria for labeling studies as low risk of bias, some concerns, or high risk of bias. This variable was then used as a moderator in statistical analyses.

Results

The PRISMA statement (Liberati et al. 2009) is seen on Fig. 2. The literature search produced 1248 results. Of these, 1103 were excluded at the abstract evaluation stage. To enhance inclusion accuracy, studies were not filtered at the title stage. From the 145 remaining articles, one was excluded for not studying a healthy population, three for not using FA- or OM-based meditations, 20 for not including applicable behavioral measures, 32 for not reporting sufficient data to extract effect sizes, five that used correlational designs, two that reported data from an already-included study, three that did not include a relevant control group, one that used a meditative practice paired with high physical activity, and one that could not readily be translated to English. In cases where data reporting was not sufficient, authors were contacted to provide missing data. A lateral search provided 12 additional studies to include in the analysis. The full list of studies found in the database searches, along with codes and reasons for exclusion at the abstract and full-text phases are available in an Excel document as Online Resource 5.

Fig. 2
figure 2

PRISMA statement outlining study exclusion at each stage

Full size image

Outcome measures within each study were classified by which attentional network(s) or facets of executive control were measured. General attention was measured in 12 outcomes. The alerting aspect of attention was measured 77 times, the orienting aspect 21 times, and executive control 26 times. For specific facets of executive control, inhibition was measured 71 times, shifting 18 times, and updating 40 times. Altogether, data from 265 outcomes were extracted from 87 studies (Table 1).

Table 1 Studies and codes
Full size table

Data were coded in terms of all applicable moderators. Outcomes were classified as measuring participant accuracy (κ = 107), reaction times (κ = 127), both (κ = 29), or neither (κ = 2). Details of these classifications are in the supplementary materials. Examining study designs, 41 used random assignment but 47 did not. Among studies with control groups, 26 used no-contact (NC) controls, 17 used active control groups with interventions that might improve attention (Active+), and 11 used active control groups with interventions that should not improve attention (Active−). Most studies taught or recruited participants that used both FA and OM techniques (K = 67). FA meditation was exclusively used by participants in 21 studies, whereas OM meditation was used in two studies. One study did not report FA vs. OM practitioner information.

Risk of Bias Assessment

All randomized control trials (K = 44) were classified by the authors according to five categories from the Cochrane RoB 2 (Sterne et al. 2019): randomization process (the extent to which studies were randomized blindly), deviations from intended interventions (adherence to conditions), missing outcome data (dropout or unreported data), measurement of the outcome (bias due to selective measurement), and selection of the reported result (analyzed according to preregistration). As per the algorithm, any studies that received a classification of “high risk” across any of the five categories was classified as high risk overall. Similarly, those classified as “some concerns” in a category were classified (at least) as having some concerns of bias overall. Only if a study were coded as having low risk of bias across all five categories would it be classified as having a low overall risk of bias. Altogether, one study was classified as low risk, 30 as there being some concerns of bias, and 13 as having high risk of bias (Fig. 3). Interrater agreement was 95.5% with a Cohen’s kappa of .898. Any disagreements were resolved through discussion. The most common reasons for the “some concerns” classification were non-blinded randomization or lack of a prespecified analysis plan.

Fig. 3
figure 3

Risk of bias assessment

Full size image

Global Effects of Meditation

General results, including breakdowns by outcome, heterogeneity tests, and funnel plot tests are displayed in Table 2. See Online Resource 6 for a forest plot containing all effect sizes and confidence intervals; additional plots are found in Online Resource 7. Overall, meditation predicted greater attention and executive control than controls (κ = 265, g = 0.171, SE = 0.027, z = 6.41, p < .001). Meditation led to greater alerting (κ = 77, g = 0.158, SE = 0.050, z = 3.14, p = .002), executive control (κ = 155, g = 0.203, SE = 0.031, z = 6.54, p < .001), inhibition (κ = 71, g = 0.159, SE = 0.048, z = 3.30, p = .001), shifting (κ = 18, g = 0.185, SE = 0.074, z = 2.52, p = .012), and updating (κ = 40, g = 0.256, SE = 0.041, z = 6.24, p < .001), but not orienting (κ = 21, g = − 0.086, SE = 0.074, z = − 1.16, p = .248). There was significant heterogeneity for generalized attention, both within (I2within = 18.71%, LRT = 6.82, p = .009) and between studies (I2between = 18.41%, LRT = 9.55, p = .002). Further, there was significant between-study heterogeneity for measures of executive control (I2between = 27.40%, LRT = 8.88, p = .003) and, specifically, the inhibition component (I2between = 28.51%, LRT = 5.54, p = .019). Though a three-level test indicated there was heterogeneity among alerting outcomes, likelihood ratio tests did not reach the threshold for statistical significance of heterogeneity within or between studies.

Table 2 Tests of heterogeneity, funnel plot asymmetry, and trim-fill corrected estimates for outcomes
Full size table

Moderator Analyses

For outcomes with significant heterogeneity of effects (general attention, alerting, executive control, inhibition), moderator analyses were conducted (Table 3). Across general attention, task type emerged as a significant moderator. Reaction time tasks had smaller effect sizes (g = 0.12, SE = 0.03, p < .001) than tasks measuring accuracy (g = 0.21, SE = 0.04, p < .001; B = 0.09, SE = 0.04, p = .040) and tasks measuring both (g = 0.27, SE = 0.07, p < .001; B = 0.15, SE = 0.07, p = .039). Studies published in academic journals had larger effect sizes (g = 0.19, SE = 0.03, p < .001) than dissertations (g = − 0.01, SE = 0.09, p = .907; B = 0.19, SE = 0.10, p = .39).

Table 3 Tests for moderation of outcomes
Full size table

The alerting outcome was significantly moderated by two variables: intervention length and publication date. At the average intervention length, there was an effect size of g = 0.17 (SE = 0.05, p < .001). For every 10-h increase in intervention length, the attentional improvement attributable to the alerting network decreased by g = 0.04 (SE = 0.02, p = .044). Similarly, newer studies showed smaller effects than older studies (B = − 0.03, SE = 0.01, p = .004). These findings should be interpreted cautiously, as the distributions for intervention length and publication year were non-normal.

Effects of meditation on executive control were moderated by coded random assignment and publication status. Studies that did not use random assignment (g = 0.28, SE = 0.04, p < .001) provided stronger effect sizes than those that did (g = 0.14, SE = 0.04, p < .001; B = − 0.14, SE = 0.06, p = .022). Studies that were published in an academic journal (g = 0.22, SE = 0.03, p < .001) provided stronger evidence for meditation’s improvements to executive control than those which were not published (B = − 0.26, SE = 0.12, p = .034). Moreover, unpublished studies provided null results (g = − 0.04, SE = 0.12, p = .742). Though the three-level test indicated significant heterogeneity in inhibition-related outcomes, the effect was not moderated by any of the tested variables (ps > .05).

Discussion

This meta-analysis examined the effects of meditation on behavioral measures of attention and executive control. Results from 87 papers were aggregated, coded, and analyzed. Outcomes for each task were classified according to Posner and Petersen’s (1990) model of attentional networks—specifically, the alerting, executive control, and orienting networks, and Miyake et al.’s (2000) model of executive control—focusing on shifting, inhibition, and updating. Overall, data indicated that meditation—whether practitioners used FA or OM techniques—increased attentional capabilities across the studies. This effect was small, however, and was contingent on which aspect of attention was being examined. These results are largely in line with previous reviews on the effects of mindfulness meditation that point to the roles of inhibition and updating in mindfulness (Chan et al. 2019; Gallant 2016).

The first hypothesis tested in this meta-analysis was that there would be positive effects of meditation on attention and executive control. Both global and network-specific measures of attention and executive control showed improvements due to meditation. Consequently, the results support theories positing that one mechanism behind meditation and mindfulness is increased attentional ability. Importantly, however, the magnitude of this effect tempers the one reported by Eberth and Sedlmeier et al. (2012): g = 0.171 (d ≈ 0.191) vs. d = 0.630. This decrease should affect the sample sizes that researchers use to study meditation and attention. Underpowered analyses will result in frequent type II errors. Therefore, we strongly recommend researchers conducting this vein of study adjust their a priori power analyses to be more conservative.

The second hypothesis stated that the alerting attentional network and the inhibition aspect of executive control would show meditation-related benefits. This hypothesis was supported by the meta-analysis. In summary, people who meditate (either assigned to meditate or those who have prior experience meditating) were more skilled at maintaining their focus, juggling multiple objects in their attention, and inhibiting distracting stimuli than were nonmeditators (those who were not assigned to meditate or who did not have meditation experience). These results are in line with theory of how both FA and OM meditations work (see Lutz et al. 2008). FA involves being able to sustain attention on an object for an increasingly long duration. Meditators must inhibit directing their attention to intrusive thoughts, anxieties, or external stimuli and remain focused. Similarly, OM meditation requires developing the ability to stay in the present moment, directing attention to different aspects of experience, but not allowing the mind to become deeply absorbed in thought.

Though the improvements in updating were not hypothesized to emerge in the meta-analysis, this finding mirrors conclusions made by Chiesa et al. (2011). This result is not surprising, particularly for OM meditation, given the meditation style underscores expanding awareness. However, meditators practicing both OM and FA techniques did not show greater improvements than those only practicing FA techniques. It is thus possible that FA meditators have also trained updating through their meditation. Improved updating could be representative of generally improved self-monitoring: keeping track of when attention is been divided and being aware of potential distractors entering the attentional field. Moreover, from a functional, training perspective, listening to and understanding instructions while maintaining focus requires an aspect of divided attention. Further research and theory are necessary to assess these possibilities.

Meta-analyses for orienting and shifting (when corrected for publication bias) indicated null results. Given the small number of outcomes investigating these constructs (orienting κ = 21, shifting κ = 18), it is difficult to make definitive conclusions about effects. Chiesa et al. (2011) suggested that meditation should lead to increases in the orienting network based on results in previous studies. Yet, meditation does not often actively engage the visual system. Further, while OM might encourage re-orienting attention to newly noticed aspects of the environment, it does not necessitate doing so quickly or accurately—qualities key to a strong orienting network. Thus, we believe it is unlikely that meditation improves orienting. Shifting, on the other hand, may have theoretical basis for being trained during meditation. People who practice FA techniques, when failing to maintain attention on an object, must re-focus their attention. Similarly, if meditators practicing OM techniques succumb to mind-wandering or if their thoughts are no longer in the moment, they must realign their attention. Though the effect size estimate for shifting was significantly different from zero, there was evidence of publication bias, with a two-level trim-fill analysis indicating that a corrected estimate would not be different from zero. This finding could be due to the lack of urgency that meditation places in re-focusing. Meditations tend to encourage a forgiving attitude toward mind-wandering: re-orienting attention back to the object of focus does not necessarily need to be quick, in contrast to what measures of shifting often test.

Comparing behavioral measures, reaction time tasks had smaller effect sizes than accuracy-based tasks. This finding suggests that improvements on attentional tasks are not simply a result of a speed-accuracy trade-off, whereby participants become faster but not more accurate. Rather, meditational may improve attentional speed and accuracy, but to a greater degree for accuracy than speed. Speed and urgency are not prioritized during meditation, whereas recognizing mind-wandering and successfully re-focusing attention and inhibiting distractors is.

For the hypotheses investigating methodological factors, results were mixed. Moderator analyses of control group type and study risk of bias did not indicate differences for any outcome type. On the other hand, for executive control, studies utilizing random assignment had smaller effect sizes than those that did not use randomization to condition. This finding is to be expected. Experienced meditators (often with many years of training) should be more strongly distinguished from controls than those who had only recently been introduced to meditation via an intervention. This finding being specific to the executive control network might indicate that regular meditative practice provides increased benefits to executive function over time compared to what can be improved through brief interventions.

Interestingly, the present meta-analysis found that length of meditation practice increased, there was a decrease in alerting. It is possible that these results may be due to a non-normal distribution of intervention lengths, where heterogeneity of variance in the meta-regression skewed the results. Alternatively, it is possible that longer attentional interventions could overburden the attentional system for a brief period of time; the present meta-analysis did not distinguish length of time between practice and assessment. On the other hand, decreased alerting could represent a benefit in the attentional system. Although greater alerting is typically viewed as desirable, it could also reflect frequent attentional capture, which may occur less frequently with increased meditation practice.

Fortunately, publication bias was only evident to a mild extent in the present meta-analysis. Though studies that were published had higher effect sizes than dissertations, patterns of results did not change by publication year or if the studies were sufficiently powered. Still, the finding that dissertations did not find any attentional benefits to meditation should be concerning. The tests of publication bias in this paper assumes that symmetry and heterogeneity of effect sizes indicates a lack of bias. However, the file drawer problem can still result in a misrepresentation of effects in current literature. We encourage researchers to contact us with relevant unpublished raw data or manuscripts, so we can update our meta-analysis at a later date.

Implications

Given the findings of the present meta-analysis, there are implications for researchers and clinicians who study meditation. First, the steady increase in literature tying meditation and attentional benefits has provided substantial evidence for its benefits. Particularly, these benefits seem to occur for the alerting attentional network and the updating and inhibition executive functions. Meditations—OM meditations in particular—have been theoretically associated with these attentional functions as potential mechanisms for change; however, it appears that FA meditation also may also affect these processes. Future research should investigate which of these mechanisms are most important in bringing about benefits associated with meditation. In developing treatment, it is important to remain parsimonious—that is, to remain as simple as possible, without wasting the time and energy of the clients. Deconstructing mindfulness and meditative practices in this way may help lead to more effective treatment with fewer adverse consequences. However, we also encourage researchers to distinguish between interventions that train meditative mechanisms (e.g., the attention training technique) and those that use meditative practice. It would be remiss to consider these heavily secularized techniques “meditations,” given the cultural, religious, and historical significance of meditative practices.

Limitations and Future Research

There are several prominent limitations to the present meta-analysis that must be considered before drawing conclusions from the results. First, findings cannot be generalized to younger or older populations. Though recent reviews have suggested similar findings for these groups (i.e., Chan et al. 2019; Dunning et al. 2019), excluding papers that exclusively examined these populations prevents age-based comparison. Future meta-analyses should examine how participant age moderates meditation’s effects on attention across the lifespan.

Second, the meta-analysis was broader in scope than many other literature reviews. Rather than examining a specific meditative teaching (e.g., MBSR), demographic, or quantitative methodology, all attentional studies broadly concerning FA and OM meditations in healthy adults were included. Though this may have the benefit of being comprehensive, it has the disadvantage of being too heterogenous in results. Indeed, a more detailed analysis would have been able to test more fine-grained moderators, such as whether meditators (or an intervention) used secular- vs. spirituality-based meditations. Using more refined categories, such as including only experimental/PPC studies, or limiting the attentional factors investigated would have allowed for a broader gray literature search, as well. Because of the larger scope of this meta-analysis, we were unable to contact all authors about additional, unpublished data they might have. Using actual data, rather than relying on statistical indicators of publication bias (e.g., trim-fill), would help elucidate the attentional effects of meditation in greater detail.

Third, outcomes within measures may not measure the construct to which they were classified. Despite thorough research into the behavioral tasks included in the analysis, certain tasks were difficult to sort. Can it be said definitively that the interference effect in the Stroop color–word task captures the same attentional construct as the difference score between the incongruent trials? This subjectivity leaves room for researcher error. Fortunately, many researchers explicitly outlined what they theorized each task would measure (e.g., Josefsson and Broberg 2011). Yet, this limitation extends more broadly into how behavioral measures are interpreted. The n-back task, for example, was classified as measure of updating, but it is difficult to disentangle updating from working memory capacity. Therefore, it is possible that the effects found in the analyses represent advantages that can be attributed to other, highly correlated cognitive processes. This limitation is likely to be present in all research that uses performance on behavioral tasks in order to generalize to attentional processes. Nevertheless, meditation shows clear attentional benefits, irrespective of the neurocognitive path it takes to get there.