Theoretical underpinnings and outcomes of micro-breaks

We can rely on multiple explanatory theories to understand how micro-breaks act on individual outcomes. From a fundamental standpoint, the cognitive load theory states that the mental capacity in working memory is limited, and if a task requires too much ability, learning will be hindered [31–33]. Individuals have limited cognitive resources; when allocating resources to one task, their availability becomes limited for other jobs [34]. In this explanatory context, micro-breaks can be seen as natural reactions of the cognitive system to a possible cognitive overload that could affect performance.

Narrowing down towards the applied workplace context, the Conservation of Resources theory (COR) [35] and the Effort Recovery Model (ERM) [8] provide the theoretical foundation for the role of recovery in the relationship between work demands, resources, and stress. The general assumption is that employees have a particular supply of personal resources, such as directed attention or mental resilience instrumental for achieving work goals [36]. Recovery is thus necessary, achievable when no demands similar to those related to the task at hand are put on the person [8], or when new resources are built up (e.g., energy, feelings control) [35]. Individuals lacking recovery experiences tend to endure fatigue and to feel negative affect [37], whereas recovered individuals feel more vigorous and engage in helping behaviors [38]. Moreover, the restoration theories such as Attention Restoration Theory (ART) [39] and The Stress Recovery Theory (SRT) [40] are traditionally used to explain the mechanisms through which exposure to nature can improve mental well-being and performance by reducing the impact of stress, mental fatigue, and negative affect [41–43].

In terms of specific outcomes, there are (at least) two individual-level components of well-being relevant for recovery: vigor (a pleasant activation) and fatigue (unpleasant deactivation) [28, 44]. Moreover, in COR theory, energy is an intrinsic resource (i.e., vigor) that must be replenished when exhausted [45]. Vigor contributes to the willingness to invest effort into the tasks at hand and persist when difficulties arise [46]. For instance, in one study many of employees reported break activities were negatively associated with increased energy (i.e., vitality) but positively related to fatigue [5]. Such results suggest that "employees seek out these strategies when they are already fatigued" [5, p. 34]. The results of a diary approach suggest that break activities positively predict vigor but negatively predict fatigue [21].

Performance represents another key outcome on which micro-breaks are considered to have an impact. It is well known that cognitive (i.e., declarative knowledge, procedural knowledge, and skills) and motivational factors (i.e., effort investment and persistence) are the main determinants of work performance [47]. Breaks from work can improve task performance through beneficial resource-strain, cognitive, affective, and motivational mechanisms [16]. Breaks are essential for performance on sustained attention tasks, suggesting that the vigilance sensitivity decrement is influenced by the frequent use of cognitive resources [48].

However, the current knowledge on micro-breaks relies considerably on experimental data testing their effects, such as improved self-control capacity, mood, or work engagement [49–51]. Specific answers are still needed to understand the effects of micro-breaks on well-being and performance, to explore whether they have an optimal duration and how vital the contextual factors are. To our knowledge, these important questions have no clear answers to date.

Based on the theoretical and empirical rationale described above, we explored two main questions in this meta-analysis. The first refers to the efficacy of micro-breaks (as defined earlier) on participants’ well-being. In other words, we would like to see if we could collect enough meta-analytical evidence to support the assumption that micro-breaks increase individuals’ vigor and decrease their fatigue levels. The second research question refers to the efficacy of micro-breaks in enhancing participants’ performance. In this respect, we would like to see whether we could find sufficient meta-analytic support for the idea that, despite slightly reducing the time allocated to the task, could micro-breaks actually augment one’s performance? In addition to these central questions, we wanted to refine our finding by exploring the impact of several moderator variables.

Potential moderators for the effectiveness of micro-breaks

Eligibility criteria

Based on the PICOS approach, the included studies had to meet the following criteria: the studied population to be represented by healthy individuals, either employees or students involved in depleting work or comparable work-related tasks (to be able to generalize the findings to the workplace context) (P); to be embedded in micro-breaks literature or focused on short break activities during tasks of 10 minutes maximum (I); a control group to be included as a comparison (C); among the monitored outcomes to be at least one of interest, such as vigor, fatigue, or performance (O); and to have a between-groups experimental or quasi-experimental design (S). Also, the studies had to be published in the last 30 years, written in English (language), and available in full text (availability). The 30 years for this search was used according to Scholz et al.’s [22] argument on the changes in the workplace due to the inclusion of computers and subsequent changes following the reliance on technology.

Moreover, the studies were eligible if they included all the necessary data in computing effect sizes. If these details were not reported, the corresponding author was contacted to provide them.

Information sources and search strategy

The systematic search strategy was based on enquiring exclusively online databases (SCOPUS, MEDLINE, and PsycINFO) for research published between January 1990 and April 2021. Searches were performed using Boolean operators. Details about how the search was conducted for each database can be found in S1 Table.

Selection process and data management

We used a two-step approach in study selection: (1) titles and abstracts screening and (2) full-text screening of the remaining records. In the first step, we enquired if the studies investigated internal rest breaks and the outcomes of interest, such as well-being and performance, the design of the studies, and details about the population tested. The next step in the study selection process consisted of screening the full texts of records selected in the previous step and applying the eligibility mentioned above criteria. Cited references were also checked, and eligibility criteria were also applied to these records.

Two independent reviewers (PA and AB) did the screening of both titles and abstracts, and full-text records was done by two independent reviewers (PA and AB). Any disagreement between them over the eligibility of particular studies was resolved through discussions with a third reviewer (AR).

Data items and collection process

For each eligible study, we proceeded to extract the following information: general bibliographical information (i.e., authors and year of publication); sample characteristics such as country, professional category (i.e., employees or students), age, and gender (P); recovery activity duration (expressed in minutes), type (i.e., work vs. non-work-related; and cognitive, physical, relational, relaxing, or nature-related), and setting (i.e., organization vs. laboratory) (I); type of control group (i.e., also break vs. no break) (C); type of outcome (i.e., vigor, fatigue, performance) and the operationalization (O); research design (i.e., randomized vs. non-randomized) (S). Moreover, we included antecedents of recovery as the type of work task (i.e., clerical, cognitive, creative, or emotional work), the relevance of the task for the workplace (relevant or irrelevant), as well as task duration (i.e., time on the task before the break, expressed in minutes).

We recorded effect sizes for measures of vigor, fatigue, and performance (objective and subjective). Objective performance is expressed as a measure of accuracy representing the amount of work performed in a specific time (e.g., mean correct responses, number of words remembered, errors reported during the completion of tasks, etc.), response speed (reaction time), as well as ideas generated (e.g., where the task require generating new ideas on a topic). Subjective performance variables were measures of self-reported levels of work productivity. Vigor and fatigue represent scores on subjective, self-report instruments.

The data collection and coding process were performed by two reviewers working independently (PA and AB). Differences or inconsistencies were re-examined with a third reviewer (AR) until an agreement was reached. To evaluate the degree of agreement between the reviewers, we computed Cohen’s kappa coefficient. The results suggest a moderate average agreement between the assessors (72.56%; average k = 0.45).

Risk of bias

The internal validity of the studies was performed based on the Cochrane risk of bias tool [72] by taking into consideration six domains: sequence generation and allocation concealment (selection bias), blinding of outcome assessor (detection bias), incomplete outcome data (attrition bias), selective outcome reporting (reporting bias), and other potential threats to internal validity.

Each criterion was evaluated for each study by assigning it a "low risk," "high risk," or "unclear risk" of bias rating. At the low risk of bias were labelled the studies which reported a clear description of how that specific internal validity criterion was handled. The more criteria with a low risk of bias a study meets, the higher the study’s internal validity can be concluded.

The same two assessors performed the risk of bias assessment. The same third expert discussed any incongruence between them and consensually settled. Inter-rater agreement was consistent across domains, ranging from fair (k = 0.27 for blinding of outcome assessor and participants) to substantial (k = 0.67 for incomplete outcome data).

Summary measures and synthesis of results

The main research questions were addressed using random-effects meta-analyses based on Borenstein et al.’s [73] framework. We used effect size estimates based on the between-groups standardized mean difference (Cohen’s d) [74], with 95% confidence intervals and two-sided p-values. Positive estimates represent effects in the hypothesized direction (i.e., increased vigor and performance; decreased fatigue), while negative values indicate opposed ones.

Because only two studies collected and analyzed follow-up data [75, 76], we decided to include only the post-manipulation data in our analysis.

To ensure that all effect sizes are computed on independent groups, in cases where studies assessed the effect of different micro-break interventions against the same control group, we combined those intervention groups into one by using standardized or pooled data [17, 49, 59, 75, 77–80]. For the study conducted by Rees et al. [81] several steps were taken, namely: (1) for each of the five post-interruption blocks for each outcome variable tested (i.e., correct responses, reaction times, response bias, sensitivity), the average was calculated; (2) from every five experimental groups (i.e., free break, music, music with video, choosing between listening to music or watching a music video, no activity break), because they were tested against the same control group, we created a single intervention group by calculating their weighted mean and pooled standard deviation; and (3) because three outcome measures fell into our accuracy operationalization of performance, we aggregated them into a single accuracy indicator.

In three other studies, we estimated the effect size based on the result of the analysis of variance with one degree of freedom at between-group level [82, 83] or based on the p-value of a χ2 test and sample size [57].

Besides the weighted average effect size for each outcome, we computed the between-studies heterogeneity. Following the recommendations made by Borenstein et al. [84], we reported the Q, I², and τ² indices. A statistically significant Q reflects if the true effect sizes vary across studies, whereas τ² estimates the between-studies variance. Based on the standard deviation of true effects (τ), a prediction interval for the true effects can be estimated, informing how widely the effects vary. Moreover, the I² statistic quantifies the dispersion observed due to true variations in effect sizes.

Additional analyses

Selection and inclusion of studies

The systematic search yielded 4868 unique records (after removing the duplicates) (see Fig 1). These were first reviewed based on title and abstract, removing another 4825 entries. In the next step, we analyzed the full text for 43 articles, two of which were included after screening the reference lists for potentially relevant studies. Out of these records, 25 were excluded for the following reasons: (1) 6 for not using a control group, (2) 9 employed a within-subjects design, and (3) 10 had incomplete data for computing effect sizes. The corresponding authors were contacted for this latter category, making available data from only one more study [81]. Hence, the final sample included 22 independent study samples that resulted from 19 publications (see Fig 1 for the complete flow of the literature search).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. The PRISMA flow diagram of the literature search and selection process.

https://doi.org/10.1371/journal.pone.0272460.g001

Description of the included studies

A systematic overview of the studies’ characteristics (e.g., participants, design) is displayed in Table 1. A summary of attributes of experimental manipulations (e.g., breaks, tasks before breaks) and outcomes are shown in Table 2. Moreover, a detailed narrative overview is provided in S1 File.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Summary of studies included in the meta-analysis and their description.

https://doi.org/10.1371/journal.pone.0272460.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Characteristics of the interventions and outcomes.

https://doi.org/10.1371/journal.pone.0272460.t002

Eleven studies used samples of students, whereas the other ten studies tested their hypotheses on samples of employees. One study used "normal volunteers" to characterize its sample without further clarifications. The total number of participants in these studies was 2335, with a mean age of 31.2 years old.

Most of the studies employed an experimental design, where participants were randomly allocated to different interventions (n = 15), whereas fewer used non-random allocation but with equivalent groups (n = 7). Most of the studies took place in a laboratory (n = 13); a smaller number was in an organizational/workplace setting (n = 9). In most studies, participants in the control group were engaged in some break or free time between work tasks (n = 12). In contrast, in a slightly smaller number of studies, these participants continued working without respite (n = 10).

To study the resource replenishing effects of breaks, participants had to complete a series of tasks before taking a respite. These tasks were either relevant to organizational life, such as work simulations and actual work-related tasks (n = 13), or irrelevant, such as various cognitive tests (n = 9). Participants were exposed to different type of demands, such as cognitive (n = 7), emotional (n = 3), or clerical (n = 8). Studies in which participants had to generate new ideas were considered as having creative demands (n = 3). Time on task is one of the most widely studied contributors to the depletion of resources. In our sample, this information was specified only for 13 studies, with participants spending between a minimum of 2 minutes and a maximum of 4 hours on resource-demanding tasks before getting a break.

In almost all of the studies, the break was a non-work one (n = 19), whereas only one study included a work-related break. Two studies were incorporated into a category with work and non-work micro-breaks because they shared the same control group, and their results were pooled in single indicators. Thus, the low variability on this factor also impedes treating it as a moderator (despite our initial intention). Regarding what participants did during the micro-breaks, five studies used a combination of two activities. For three of these studies, the combination of multiple types of activities was a result of using the same control group to compare different interventions. Five studies included a cognitive micro-break, where participants were involved in activities such as watching movie clips. Six studies offered a physical break between work bouts, whereas the other six activities during breaks were relaxing. Hence, the heterogeneous and mixed nature of the micro-breaks employed in the included studies impedes us from exploring this aspect as a potential moderator. Break duration varied significantly between studies, ranging between 8 seconds and 10 minutes.

Considering the outcomes’ operationalizations, vigor and fatigue were exclusively assessed with self-report scales. The measures varied from multi-factor well-established instruments (e.g., Activation-Deactivation-Checklist [94]; Utrecht Work Engagement Scale [95]; Profile of Mood States, [96]), to one-item ones [e.g., 92] The reporting of reliability estimates was inconsistent across studies (and not possible for single-item measures), but when mentioned, generally exceeded α = .80 or even α = .90 [e.g., 17, 79, 92]. As for performance, there were studies using both objective and subjective measures of performance, as well as measures of creativity. We grouped accuracy (e.g., mean correct responses, number of words remembered, errors reported during the completion of tasks, etc.) and reaction speed in cognitive tasks into a category of objective measures (n = 8), self-perceived performance was considered as a subjective measure (n = 4), and idea generation was considered as creative performance (n = 3). A more fine-grained categorization (especially for the objective measures) was hard to impose because of the limited number of studies. In terms of reliability, with two exceptions (Studies 2 and 3 from Paulus et al. [80]), there were no such estimates reported. While the mentioned cases focused on creative performance with interrater reliabilities of > .90, for all the other operationalizations it was not possible to make any accurate inference from this perspective. However, as also previously discussed [70], the objective measures typically imply the employment of arbitrary tasks and/or unaudited performance measures or paradigms with reduced numbers of trials. Hence, it is less likely to expect such approaches as being psychometrically precise. Unfortunately, in the absence of reliability estimates, such measurement artefacts cannot be taken into account meta-analytically.

Quality of the included studies

The risk of bias assessment results for each threat to the internal validity are shown in Table 1 (for each study) and Fig 2 (as a visual summary).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Risk of internal bias summary.

https://doi.org/10.1371/journal.pone.0272460.g002

The results following the selection bias assessment show an unclear risk for both domains (i.e., sequence generation and allocation concealment), as thirteen of the twenty-two studies did not make explicit the randomization method. Detection bias follows the trend, with most studies in the unclear risk of bias category (72% of included studies). Most studies were at low risk for attrition bias (n = 20). Twenty-one studies were assessed with unclear risk regarding reporting bias. For other potential threats to validity, fifteen studies out of twenty-two have been evaluated as having low risk.

Overall, only four out of twenty-two studies were assessed with a low risk of bias, whereas one presented a high risk for at least half of the criteria. Thus, we are inclined to consider the risk of bias in our overall sample as being somewhat unclear.

Preliminary analyses

To detect potential outliers, we analyzed the studies for which the confidence intervals failed to overlap with the confidence interval of the pooled effects. We found only two potential outliers. One of them was in the sample of effects regarding the role of micro-breaks on fatigue [49], with an effect size of d = 13.43, and a 95%CI which stretched between 11.11, and 15.75, being completely non-overlapped with the overall one [0.29, 1.52]. The other one was in the sample of effects of micro-breaks on performance outcomes [77], with an effect size of d = 3.85, 95%CI of [3.38, 4.33], as compared to the meta-analytical one laying between [-0.14, 0.86]. The decision to exclude them was proven legitimate, as the results differed greatly, with improvements in heterogeneity. Specifically, for the effect on fatigue; before exclusion the heterogeneity was statistically significant (Q₍₉₎ = 127.55, p < .001), while after excluding the outlier became non-significant (Q₍₈₎ = 6.24, p = .619), result also reflected into the I² value (before exclusion: I² = 92.94%; and after: I² = 0.00%). In the case of performance, heterogeneity with the outlier was also higher (Q₍₁₅₎ = 242.57, p < .001; I² = 93.82%), as without it (Q₍₁₄₎ = 34.52, p = .002; I² = 59.45%). The detailed analyses including the outliers can be found in S2 Table.

Effectiveness of micro-breaks

The main meta-analytical results are presented in Table 3 and displayed in Fig 3. These revealed a statistically significant but small effect of micro-breaks on vigor, d = 0.36, p < .001, 95% CI [.16, .55], and fatigue d = 0.35, p < .001, 95% CI [.19, .50], while the effect on performance was not statistically significant, d = 0.16, p = .17, 95% CI [-0.04, .37].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3.

Standardized effect sizes and forest plot for the sample of studies regarding (a) Vigor, (b) Fatigue, and (c) Performance.

https://doi.org/10.1371/journal.pone.0272460.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Effectiveness of micro-breaks on vigor, fatigue, and performance.

https://doi.org/10.1371/journal.pone.0272460.t003

Heterogeneity analyses suggested that the effects on vigor (Q₍₈₎ = 13.61, p = .093; I² = 41.21%; τ² = .04; 95% prediction interval: [-.15, .86]) and especially on fatigue (Q₍₈₎ = 6.25, p = .619; I² = 0.00%; τ² = .00; 95% prediction interval: [.16, .53]) were quite homogenous. However, there was significant unexplained variance in the true effect sizes for performance (Q₍₁₄₎ = 34.52, p = .002; 95% prediction interval: [-.53, .86]). A big proportion of the observed variance being due to real variations of the effects (I² = 59.45%, τ² = 0.09).

Moderator analyses

Even though the effects on vigor and fatigue were relatively homogeneous, and only those on performance revealed significant between-studies variations, we continued with the moderator analyses on all three outcomes (especially since we still had rational/theoretical arguments in this regard). The results of the moderator analyses for all outcomes of interest can be seen in Table 4. Because of the modest number of studies on each category and the low variability in many cases, we only had the methodological possibility to test a low number of moderators (i.e., break duration, the task before break, type of participants, study setting, or control group activity).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Test of significance for each presumed moderator.

https://doi.org/10.1371/journal.pone.0272460.t004

For vigor and fatigue, the results show that none of the considered moderators impacted the efficacy of micro-breaks (all ps > .050). For overall performance, however, two moderators were found. Break duration was revealed to be one of the significant results (b = .07, p = .006, R² = .34). Indicating that the longer the break, the more micro-break leads to a performance increase. The second significant effect was the type of task performed before the break (Q₍₂₎ = 6.53, p = .011). More specifically, when the task was cognitive, the micro-breaks had a very small effect on performance (d = -.09, 95%CI [-.39, .30], p = .541), which was non-significant and still heterogeneous. When the tasks were creative, micro-breaks had a small, significant effect (d = .38, 95%CI [.11, .64], p = .006), and for clerical tasks the effect of micro-breaks was medium and significant (d = .56, 95%CI [0.01, 1.12], p = .047). It is important to mention that the latter effect is based only on two studies; hence, it has to be cautiously interpreted. To have a broader picture of the role of task antecedent to the break, in Table 5, we also report the sub-group results for vigor and fatigue (not only for performance). It may be worth noticing that the effect on vigor when the micro-break is taken from a cognitive task is also minimal and non-significant (important to bear in mind that the sub-groups differences are not statistically significant).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Meta-analytical findings at each level of the antecedent task moderator.

https://doi.org/10.1371/journal.pone.0272460.t005

Finally, we tested if study quality (i.e., total criteria with low risk of bias for internal validity–see Table 1) was associated with the efficacy of the interventions. The results of the meta-regression suggest that for neither of the outcomes the amount of criteria with low risk of bias had a significant impact on efficacy (effect on vigor: b = -.05, p = .642, R² = .00; effect on fatigue: b = -.01, p = .905, R² = .00; effect on performance: b = -.13, p = .329, R² = .00).

Publication bias

For the effect size estimates on vigor, Egger’s test was not statistically significant (intercept = 0.21, p = .915), while the trim and fill procedure imputed two studies to the right of the mean. The adjusted effect was d = 0.45, being similar in magnitude to the observed value (d = 0.35). In the case of fatigue, the statistically non-significant results on Egger’s test (intercept = -0.46, p = .769) indicated a lack of publication bias, but the trim and fill procedure imputed two studies to the right of the mean. The adjusted effect was d = 0.40, being also like the observed one d = 0.34. Finally, for performance, Egger’s test was statistically significant (intercept = -2.63, p = .048), indicating possible publication bias. The trim and fill procedure imputed two studies to the right of the mean, where the adjusted value suggests a significant small effect (d = 0.22, 95% CI [0.02, 0.43]) as compared to the observed one (d = 0.14, 95% CI [-0.07, 0.36]). Even though statistically significant, it still is similar in magnitude. Overall, whereas some evidence for the presence of publication bias exists, it does not seem to be an impactful threat to the observed effects.

Limitations, future research directions, and conclusions

Although this meta-analysis sheds some light on the effect of micro-breaks on individual outcomes, it is not without any limitations, thus prompting the reader to interpret these results with caution.

Firstly, the number of studies was modest for each outcome, which limited the moderator analyses. Specifically, essential moderators that could have helped us better understand the heterogeneity presented for the performance effect were impossible to analyze. Some examples are break type (work or non-work), specific activity performed during the break (physical, relational, cognitive, relaxation, nature-related, etc.), or time working on the task before taking a break. In the recent literature review on the effect of breaks on performance, the aspect of testing moderator variables was also discussed, concluding that underdeveloped methodological approaches for tasks and measures, and extremely heterogeneous perspectives concerning contexts, and designs, make it hard to conduct such differential analyses in the field of rest-break research [70]. Thus, we could not address one of the most important questions for practice about which specific activity is most efficient for recovering lost resources during work: "What to do in these breaks to feel and perform better?". However, while we still need a clear explanation for the performance outcome, at least for well-being, the answer seems to be "any type of decoupling activity".

Secondly, besides the increased heterogeneity in measurement approaches (especially when considering performance), another important cautionary aspect may be their reliability. The fact that we have inconsistent reporting for the well-being self-reports and no information for the other measures makes it impossible to scrutinize the potential bias induced by measurement error. When possible, future studies should not overlook reporting reliability estimates. This way updated meta-analyses on this topic may address the artefactual variance accountable by measurement error [126, 127].

Furthermore, because all the studies considered in the present meta-analysis used self-reports of well-being outcomes, our results could be exposed to response biases [128]. Although this is a well-known risk when using self-reports, the energetic activation (e.g., vigor) and deactivation (e.g., fatigue) of the well-being components considered in this meta-analysis represent a subjective component of a "bio-behavioral system of activation" [129, 130 p. 827], experienced as feelings, emotions, or dispositions [7]. Thus, one’s evaluation of their subjective experience can be more suitable than other individuals’ reports or assessments. However, future studies could also benefit from a combination of self-reports of well-being with objective measures of the bio-behavioral systems involved (e.g., endocrinological indicators) [130].

Lastly, we considered only two aspects of well-being, a pleasant activation (e.g., vigor) and unpleasant deactivation (e.g., fatigue), leaving aside other concepts studied in recovery research such as anxiety or tension [131]. Moreover, the present meta-analysis is based on studies from the pre-pandemic literature. The new Covid-19 pandemic brought forward new challenges at home, at school, and at work, changing the way people work in fundamental ways [132]. Thus, future studies could benefit from including this new challenge related to the pandemic in their designs.

The current meta-analysis was performed on twenty-two experimental studies published in the past thirty years that tested the effects of micro-breaks on vigor, fatigue, and performance. Specifically, it showed that micro-breaks positively impact well-being by enhancing vigor and lowering fatigue, regardless of the contextual factors. Importantly, micro-breaks do not seem to influence performance generally. However, when the break is more extended, the performance tends to improve, especially when individuals are engaged in creative or clerical tasks, and less when performing activities of a cognitively demanding nature.