Sometimes you just can’t: within-person variation in working memory capacity moderates negative affect reactivity to stressor exposure

ABSTRACT The executive hypothesis of self-regulation places cognitive information processing at the center of self-regulatory success/failure. While the hypothesis is well supported by cross-sectional studies, no study has tested its primary prediction, that temporary lapses in executive control underlie moments of self-regulatory failure. Here, we conducted a naturalistic experiment investigating whether short-term variation in executive control is associated with momentary self-regulatory outcomes, indicated by negative affect reactivity to everyday stressors. We assessed working memory capacity (WMC) through ultra-brief, ambulatory assessments on smart phones five times per day in a 7-day ecological momentary assessment (EMA) study involving college-aged adults. We found that participants exhibited more negative affect reactivity to stressor exposures during moments when they exhibited lower than usual WMC. Contrary to previous findings, we found no between-person association between WMC and average stress reactivity. We interpret these findings as reflecting the role of executive control in determining one’s effective capacity to self-regulate.

Individuals with higher fluid cognitive ability (e.g.general fluid intelligence, executive control) tend to achieve superior health and well-being outcomes (Batty & Deary, 2004;Gottfredson & Deary, 2004).These positive long-term outcomes are thought to emerge from a combination of engagement in health and well-being promoting patterns of thought and behaviour, as well as avoidance of negative risk factors (Gottfredson & Deary, 2004).Health and well-being promotion and risk avoidance are often the products of successful self-regulation (e.g.choosing a nutritious meal, managing negative perseverative thought patterns), and recent evidence suggests that self-regulatory success may depend, at least in part, upon one's executive cognitive abilities (Garrison & Schmeichel, 2020;Hofmann, Gschwendner, et al., 2008;Hofmann et al., 2012;Schmeichel et al., 2008).More effective self-regulation may help explain the epidemiological link between cognitive ability and long-term health and well-being outcomes.However, few studies have been conducted outside of laboratory settings where dynamic interactions between cognition, exposures, and regulatory responses observed under naturalistic circumstances might help shed light on this issue.In the present study, we examined whether short-term, intraindividual variation in executive control is associated with self-regulatory outcomes in the form of negative affect reactivity to everyday stressors.
Self-regulation refers to the process of aligning one's thoughts, emotions, and behaviours with one's standards and longer-term goals (Baumeister & Heatherton, 1996;Carver & Scheier, 1998;Cole et al., 2019;Hall & Fong, 2007;Nigg, 2017).Emotion regulation can be described as the management or influence over one's experienced emotions (whether they be up-or down-regulated or maintained; Cole et al., 2004;Gross, 1998;Thompson, 1994).In practice, emotion regulation is often investigated through response to lab-based, experimental manipulations such as affect and stress induction experiments (Mauss et al., 2007;Seeley et al., 2015).More recently, ecological momentary assessment (EMA) methods have been leveraged to study the factors surrounding emotion regulation in everyday life, inspiring new directions for theory development (Benson et al., 2019;Troy et al., 2019).Stressor exposures are common over the course of everyday life and the stochastic manner in which they are encountered can be leveraged as a naturalistic experiment through which emotion regulation processes can be studied (Gross, 2015;Sliwinski et al., 2009;Smyth et al., 2018).Consistent with previous studies, we chose stressor reactivity as a proxy for emotion regulation and operationally defined reactivity as the change in negative affect experienced following a stressor exposure compared to moments lacking an exposure (Almeida, 2005;Garrison & Schmeichel, 2020;Stawski et al., 2019).
Recent evidence suggests that individuals with greater executive control ability (e.g. higher working memory capacity, WMC) exhibit more effective emotion regulation in lab-based, experimental contexts and in response to everyday stressor exposures (Kobayashi et al., 2021;Schmeichel & Demaree, 2010).These findings have advanced an "executive" hypothesis of self-regulation that places control over information processing at the center of self-regulatory success/failure (e.g.active goal representation in working memory; Hofmann et al., 2012).The executive hypothesis has received significant cross-sectional support via lab-based studies (Hofmann, Friese, et al., 2008), and more recently, EMA and experience sampling studies (Garrison & Schmeichel, 2020;Kane et al., 2007).However, no study, to date, has tested the core prediction of the hypothesis, that individuals are more likely to experience selfregulatory failure during moments when they are experiencing a temporary lapse/reduction in executive control.This is likely due to a number of factors including theoretical tradition (stationarity of cognition) and statistical assumptions (ergodicity).However, probably the most significant barrier to date has been that computerised cognitive assessments are often lengthy and not amenable to intensive, repeated sampling study designs.To overcome this barrier, we recently developed methods for adapting performance-based cognitive assessments to EMA study designs and demonstrated that WMC can be reliably measured via repeated, ultra-brief administrations on smart phones (Hakun et al., 2023;Sliwinski et al., 2018).Utilising ultra-brief, ambulatory assessments, we and others have shown that WMC varies substantially, within-individuals over short periods of time in daily life settings (e.g.within and between days; Brose et al., 2012;Hakun et al., 2023;Sliwinski et al., 2018).
The aim of the present study was to investigate whether short-term variation in WMC is associated with the extent of within-person negative affect reactivity to everyday stressors.We hypothesised that if temporary lapses in executive control underlie selfregulatory failure, then greater negative affect reactivity to stressor exposures should be observed when scores on ambulatory assessments of WMC are lower than usual.To address this aim, participants were recruited to complete up to 6 EMA surveys per day for 7 days, including momentary assessments of stressor exposure, negative affect, and an ultra-brief Rotation Span task (administered at the end of 5 of the 6 daily EMA surveys).We administered the EMA protocol via the Mobile Monitoring of Cognitive Change ("M2C2") platform.
Open Practices Statement: The data and scripts for replicating the analyses presented are publicly accessible at https://osf.io/cq6up/.The study was not publicly pre-registered, but the hypotheses, predictions, and approach were documented in proposal K99/R00AG056670, which provided funding for the research.

Participants
A total of 108 college students participated in the study.From the total sample, 7 participants were excluded for lack of compliance with the study protocol (see statistical analysis section for more details).Participants in the analysis sample (n = 101) were on average 21 years of age (SD = 3.41; Range = 18-44 years), 28% male, 74% identified as White (4% Asian or Pacific Islander, 5% Black or African American, 3% Latinx, and 14% more than one race).Participants were recruited from Pennsylvania State University's College of Education via e-mail listserv communications.All participants provided written informed consent and received compensation for their participation.All experimental procedures were approved by Pennsylvania State University's Institutional Review Board for the ethical treatment of human participants.

Materials and procedure
The study protocol involved two phases: (1) an inperson study onboarding visit where participants provided consent and received an overview of the study, study materials (smart phones, M2C2 app, EMA questions), and instructions for the rotation span task; (2) a subsequent 7-day EMA measurement burst.

Ecological momentary assessment (EMA) protocol
Data analysed in this report were generated from study of academic engagement and self-regulation.Recruitment was parameterised in a manner to generate generalisable inferences regarding an average week of an average semester.To accomplish this goal, participants were recruited only during 1 Fall and 1 Spring semester.Recruitment was only scheduled to occur during weeks 3 through 13 of a 15week semester.This was done to avoid sampling during weeks early in the semester where no assignments or exams were yet scheduled and during weeks late in the semester where participants may have already completed their activities (e.g.courses with no final exam or projects) or were solely focused on final exams.In addition, coordinators aimed to recruit approximately 5 participants per week to avoid an over-concentration of observations during any given week in the semester.
Participants were asked to carry a lab-provided smart phone for a period of 7 days (Xiaomi Mi A1 model phones configured with Android One OS, v8.0; Xiaomi Corporation, Beijing, China).A custom Android EMA mobile application currently under development for the National Institutes of Health (NIH), the "Mobile Monitoring of Cognitive Change" or "M2C2" app was loaded onto the investigator-provided phones and used to administer the EMA protocol, including the self-report survey items and the ultra-brief Rotation Span task.Phones were configured into a "kiosk" mode where only the study application was available to the participant.

EMA protocol
EMAs were administered to each participant over 7 consecutive days in a measurement burst design.Each day involved one self-initiated morning survey, followed by four pseudorandomly signalled "beeped" surveys, followed by one self-initiated evening survey.At the start of the study, participants provided their typical waking time.From this information, the EMA protocol was set-up for each participant so that their first beeped survey would be delivered at a random time within 2 h of their self-reported typical morning waking time.Subsequent notifications were separated by an average of 3.75 h and pseudorandomly jittered around this interval by up to 30 min.Stressor exposures were assessed during each beeped and evening survey.The Rotation Span (RSpan) task was administered at the end of each morning and beeped survey, but not during the evening survey in order to limit participant burden as several additional daily diary items were included in the evening survey and participants were asked to complete this survey before going to bed.The primary analysis, which is focused on contemporaneous self-report of stressor exposure, negative affect, and WMC makes use of the four beeped surveys each day where all three measures were collected.For the main analysis, the morning surveys were not used because stressor exposure data were not collected, and the evening surveys were not used because Rotation Span task data were not collected.However, for the follow-up analyses, we used the Rotation Span task data from the morning surveys to facilitate examination of associations between both lag-1 WMC, concurrent WMC, and concurrent negative affect.

Negative affect
Participants answered questions about the extent to which they were currently experiencing several discrete negative valence emotions ("Right now, … what is your level of worry?… are you feeling sad? … are you feeling fatigued?").Responses to each of these questions were made using a touch visual analog slider ranging from Not at all (0) to Extremely (100).The average within-person correlations among the discrete emotion items were 0.39 for worry and sad, 0.22 for worry and fatigue, and 0.27 for sad and fatigue.A negative affect composite was computed as the average response to the three discrete negative emotion items at each measurement occasion.Multilevel reliability analysis for the 3-item negative affect composite (mlr function in psych pkg; Revelle, 2020) revealed between-person reliability (average of all ratings across all items and times) of RkF = 0.98.Reliability of within-person change for fixed time points and fixed items was Rc = 0.56.

Stressor exposure
During each beeped survey and evening survey participants were asked whether "Since the last survey, has anything stressful happened to you?".Response options included yes (1) and no (0).

Working memory capacity
An ultra-brief RSpan task was selected to assess WMC based on the results of our previous work where we found an ambulatory adaptation of the RSpan task to be the strongest indicator of latent WMC among a battery of ultra-brief complex span tasks (Hakun et al., 2023).The RSpan task involved memorising oriented arrows, one at a time, while performing interleaved mental rotation judgments on letter stimuli that were either forward or backward facing if oriented back to standard typeface orientation.Study arrows were oriented at 45°increments around an invisible clock-face and were either short or long in total length, allowing for 16 possible orientation-length combinations.After memorising a set of arrows and completing the interleaved mental rotation judgments, participants were asked to recall all of the oriented arrows memorised throughout the current trial by touching the respective arrow heads on a recall screen depicting an array of all possible studied arrows.Three trials were administered at each occasion.Set size, defined as the number of total arrows memorised during each trial, was 5. Partialcredit scores were given for each correctly recalled arrow regardless of whether all arrows for a given trial were correctly recalled (range 0-15).

Statistical analyses
Momentary occasions were included in the analysis if the participant provided complete data for negative affect, stressor exposure, and WMC.Seven participants were excluded from the analysis sample because they did not complete any of the beeped surveys (n = 6), or because they did not complete any of RSpan task administrations that occurred at the end of each beeped survey (n = 1).Participants included in the analyses (n = 101) provided, on average, 22 beeped surveys (SD = 5.41, Min = 8, Max = 32), and 90% of the sample completed at least 50% of the expected beeped surveys.We did not observe any significant differences in compliance rates based on demographic or other central variables under study (e.g.gender, race, class year, stressor exposure rate, etc.).Beeped surveys were uniformly completed across days in the measurement burst such that, on average, participants responded to at least one beeped survey during 6.64 of the 7 study days (SD = 0.78, Min = 3, Max = 8*).*Note, that while participants were instructed to respond to the EMA protocol for a period of 7 days, the application did not prohibit notifications/responding outside the scheduled window and one participant provided one survey response on an 8th day.
Data preparation steps included separating timevarying predictor variables into the between-person and within-person components (Bolger & Laurenceau, 2013).For model parsimony, we chose to use a 2level modelling approach (occasions nested withinpersons), taking into consideration the relatively low levels of day-level variance observed for each factor measured repeatedly within-persons (ICC range = 0.04-0.16from 3-level, intercept-only, random effects models).The time-invariant between-person component, WMC_BP i , was calculated as the intraindividual mean across the repeated measures, yielding one score per person.Similarly, the momentary within-person component, WMC_WP it , was calculated for each observation for each person as the deviation from their intraindividual mean (WMC_BP i ).We also created a lag-1 WMC variable, WMC_WP i,t-1 , by aligning the WMC_WP it scores with the EMA data collected on the subsequent occasion (i.e.shifting each score "down" one row).For example, the score for the first beeped survey was aligned with the EMA data from the second beeped survey data for analysis.The last score for each day was not shifted forward to the next day.The between-person component, StressorExposure_BP i was calculated as the percentage of surveys that the participant reported a stressor experience across their available occasions.The momentary within-person component, StressorExposure_WP it was left in its original units of 0 (no stressor exposure since the last survey) and 1 (stressor exposure since the last survey).Prior to analysis, person-level variables (WMC_BP i , StressorExposure_BP i ) were sample-mean centred to facilitate interpretation with respect to the prototypical person in the sample.

Primary analysis
To examine whether between-person differences and within-person variation in WMC moderated negative affect reactivity to stressor exposure, we fit a multilevel linear regression model, where the repeated measures of negative affect for participant i during moment t are modelled as a function of a person-specific intercept, β 0i , changes driven by stressor exposure since the last EMA prompt, β 1i , changes driven by concurrent WMC, β 2i , the interplay between stressor exposure and WMC, β 3i , and residual differences, e it .Person-specific coefficients were simultaneously modelled as a function of person-level predictors, b 0i = g 00 + g 01 (StressorExposure BP i ) (2) where γ 00 , γ 10 , γ 20 , and γ 30 , and are sample-level parameters describing the prototypical person, and γ 01 , γ 02 , and γ 11 describe how individual differences in stressor exposure and WMC are associated with participants' negative affect and/or moderate the within-person associations between stressor exposure, WMC, and negative affect.Random effects (u 0i , u 1i , u 2i , u 3i ) were allowed to covary with one another, but not with e it .

Secondary analysis
To examine whether WMC assessed at the prior occasion moderated the within-person association between stressor exposure and negative affect, we fit a separate multilevel linear regression model, where the repeated measures of negative affect for participant i during moment t are modelled as a function of a person-specific intercept, β 0i , changes driven by stressor exposure since the last EMA prompt, β 1i , changes driven by concurrent WMC, β 2i , changes driven by lag-1 WMC assessed at the prior EMA prompt, β 3i , interplay between stressor exposure and lag-1 WMC, β 3i , and residual differences, e it .Person-specific coefficients were simultaneously modelled as a function of person-level predictors, in a similar way as was done in Equations ( 2)-( 5).

Sensitivity analyses
A series of sensitivity analyses were conducted following the primary analysis provided in Equations ( 1)-( 5), with each momentary discrete emotion (sad, worry, fatigue) swapped in as the outcome variable (Equation ( 1)) in place of the momentary negative affect composite.See Supplemental Materials for results and supplemental discussion of these analyses.
In the Bayesian framework, the mean of the posterior probability distribution is obtained for each parameter of interest, rather than directly estimating a point estimate for each parameter as is done in the frequentist framework.The posterior mean for each parameter is then evaluated with respect to its 95% Credible Interval (CI), which is similar to a Confidence Interval in the frequentist framework.The interpretation differs slightly though from the frequentist Confidence Interval.A Bayesian CI is interpreted as a 95% probability that the true parameter falls within the 95% CI.A posterior mean with a CI that does not span the value of the null hypothesis is considered to reflect a credible association.For the models in this paper, the null hypothesis is that the value of the posterior mean is zero.Thus 95% CIs that do not span zero indicate that the posterior mean estimate for that parameter is credibly different from zero.

Descriptive statistics
The average intensity of experienced negative affect was M = 37.63 (SD = 15.25), and the average performance on the RSpan task was M = 10.73 (SD = 2.14).Prior to the main analyses, we calculated intraclass correlation coefficients to examine the extent to which each variable differed across moments and between persons (Bolger & Laurenceau, 2013).For negative affect and WMC, approximately half of the variance (53%, 51%, respectively) reflected withinperson variation, and the other half between-person variation (47%, 49%, respectively).In contrast, for stressor exposure, the majority of the variation was at the within-person level (90%), and very little variance existed at the between-person level (10%).Between-and within-person correlations between all study variables can be found in Table S1.Given that there was substantial variation at both the withinand between-person levels of analysis for negative affect (the outcome variable), we proceeded to fit multilevel models.Results are described in the following sections and reported in Table 1.

Between-person negative affect reactivity to stressor exposure
Participants reported being exposed to a stressor during an average of 16.60% (Min = 0%, Max = 56%) of beeped surveys, which equates to approximately five stressor exposures over the 7-day measurement burst (consistent with previous coordinated analysis; Zawadzki et al., 2019).75.5% of the sample (77/102) reported two or more stressor exposures during the study period.The range of non-stress exposed occasions observed across participants was Min = 7 to Max = 34.We observed evidence that betweenperson differences in stressor exposure were associated with between-person differences in negative affect.In particular, participants who experienced a higher percentage of stressor exposures across the study period also tended to have higher negative affect (γ 01 = 0.32).

Within-person negative affect reactivity to stress exposure
We also examined whether within-person variation in momentary stressor exposure was associated with within-person variation in negative affect.The results indicated that for a participant with prototypical stressor exposure rate and WMC, their negative affect was expected to be 35.94(γ 00 ) on occasions when they did not report stressor exposure.In contrast, on occasions with a reported stressor exposure, their negative affect was expected to be 10.86 units higher (γ 10 ).

Between-person cognition
Between-person differences in WMC were not credibly associated with negative affect (γ 02 = 0.06).In contrast to previous evidence (Garrison & Schmeichel, 2020), between-person differences in WMC over the study period did not credibly moderate negative affect reactivity to stressor exposure (γ 11 = 0.84).

Within-person concurrent WMC
Although between-person differences in WMC were not systematically related to average negative affect experienced over the study period (γ 02 = 0.06), when WMC was higher than a given participant's usual level, negative affect also tended to be higher (γ 20 = 0.59).Further, consistent with our hypothesis, intraindividual variation in WMC moderated negative affect reactivity to stressor exposure.That is, the difference in change in negative affect between occasions with and without a stressor exposure (i.e.reactivity) was lower during moments when an individual's WMC was higher than usual (γ 30 = −1.28; Figure 1).

Time-lagged associations between WMC and subsequent reactivity
The above analyses suggest systematic associations among stressor exposure, subsequent negative affect, and WMC.To better understand the temporal sequence of associations, we also examined whether WMC at the prior EMA occasion (lag-1) was associated with individuals' negative affect reactivity to stressor exposure.Results from this analysis showed that prior WMC (estimated before a stressor exposure) did not moderate negative affect reactivity to the stressor exposure (γ 30 = 0.88; see Table S2 for all model parameters).

Discussion
In the current study, we investigated a key, previously untested, prediction of the executive hypothesis of self-regulation, that temporary lapses in executive control may underlie self-regulatory failure (Baumeister et al., 2007;Hofmann et al., 2012;Nigg, 2017).We assessed WMC in close temporal proximity to stressor exposure using ambulatory cognitive assessments to determine whether negative affect reactivity to everyday stressors is associated with the state of one's executive control ability.We found that negative affect reactivity was higher than usual during moments when individuals' WMC was lower than usual.This finding provides initial evidence linking short-term variation in executive control and self-regulatory outcomes under naturalistic circumstances.
Our ambulatory approach of assessing WMC provides confirmatory evidence for a primary assumption behind the executive hypothesis, that executive control ability varies over short timescales.Approximately 50% of the total variance in ultra-brief RSpan performance was determined to be within-person, over time, and greater negative affect reactivity to stressor exposure was observed during moments with poorer RSpan performance.The executive hypothesis has, to date, been extrapolated from cross-sectional studies examining individual differences in performance-based measures; specifically, showing that individuals with lower WMC tend to exhibit less well-regulated behaviour when compared   to individuals with higher WMC (Garrison & Schmeichel, 2020;Hofmann, Friese, et al., 2008;Schmeichel et al., 2008).Our results extend these findings and provide the first documentation of this phenomenon occurring within-individuals, over the course of their daily lives.WMC describes the control processes involved in what gains access to and is maintained in shortterm mental representation/focus of attention (Engle, 2002;Kane & Engle, 2003).Momentary impairments in WMC may constrain one's ability to selfregulate in several ways.Expressive suppression and cognitive reappraisal are two of the most wellstudied emotion regulation strategies.Recent evidence from EMA studies suggests that both methods can be effective, depending upon the context (Aldao, 2013;English et al., 2017;Haines et al., 2016;Troy et al., 2013).The availability of greater executive control resources may contribute to the success of each strategy through distinct mechanisms.Hofmann et al. (2012) highlight at least two distinct pathways, proposing that greater control over working memory can lead to better suppression of repetitive thought or mind-wandering through passive inhibition, and that reappraisal may be supported by the availability of the operational/representational canvas in WM for information processing.We cannot adjudicate between these potential mechanisms with the available data but propose that future studies be designed to investigate these hypotheses further.
We operationalised reactivity as the change in negative affect observed following exposure to a stressor in the context of participants' daily lives.This ecological approach is advantageous in that it reflects the impact of naturalistic forces, thereby increasing generalisability of our findings to everyday life.At the same time, because we did not experimentally manipulate our design with respect to regulation/reactivity (e.g.randomisation to conditions with/without regulatory strategies) or administer assessments at a finer timescale than once every ∼3-4 h, a few caveats need to be taken into consideration.
We examined WMC as a modifier of stress reactivity on either side of the stressor exposure event, conducting concurrent and lagged analyses.Results of timelagged analysis indicated that WMC assessed in the window before a reported exposure did not modify reactivity to the subsequent stressor to the same degree as concurrent WMC.While this observation might be taken to suggest that the state of WMC entering a stressor exposure is less important than WMC in the time period following an exposure, we hesitate to draw this conclusion without knowledge of how close, in real-time, the assessment of WMC occurred in proximity to the stressful event (i.e.prior, lag-1, WMC, assessed in the moments just prior to an exposure, might be more strongly associated with reactivity).What we know based on our current design is that "concurrent" WMC was measured in context of having been exposed to a stressor, which may reflect the action of processes invoked by the stressful event (e.g.reactive processing).
This issue raises a second important possibility that negative affect may be acting as a potential driver of short-term changes in WMC.In fact, previous work has shown that negative affect and working memory may be negatively coupled within-persons (suggesting a potential resource allocation tradeoff; Brose et al., 2012;Sliwinski et al., 2006).Results of our primary analysis revealed the opposite pattern showing that, in the absence of a stressor, moments with higher negative affect were associated with higher WMC.Taken together with the pattern observed during moments following stressor exposures this view would suggest that negative affect in the absence of a stressor improves WMC while negative affect in the presence a stressor impairs WMC.While we did not measure other qualitative aspects of experienced negative affect (e.g. did the participants attribute their experienced affect to the stressful event), it remains possible that the experience of negative affect brought about by exposure to a stressful event is more salient than that experienced during unexposed periods, which in turn may have performance-modifying impacts on cognition.Because this interpretation requires the assumption of a non-linear, or at least contextually constrained, association between negative affect and WMC that we did not measure or manipulate, we prefer to interpret these results through the lens of the executive hypothesis.In any case, we hope to highlight that the current study represents a significant initial step forward beyond the cross-sectional approach to study these processes that unfold over relatively fast timescales in daily life.We urge that researchers consider these important timing and contextual factors when designing future studies.While our findings represent a conceptual replication/extension of the cross-sectional work described above, our results differ from the findings of Garrison and Schmeichel (2020) who showed that between-person differences in WMC (assessed on a single occasion in a laboratory setting) moderated the extent of negative affect reactivity to everyday stressors observed during an EMA measurement burst.It is possible this may be due to differences in the complex span task selected for each study (e.g.verbal Operation Span vs visuospatial Rotation Span).In their study, the observed interaction between in-person complex span task performance and EMA stress reactivity was specific to a verbal Operation Span task (an emotional complex span task was not associated with reactivity differences; Garrison & Schmeichel, 2020).WMC is often investigated as a domain general construct, indicated by performance across multiple complex span tasks covering different stimulus/memory domains (Kane et al., 2004).Utilisation of a single indicator might risk less reliable estimation of latent WMC, which could attenuate observed associations.Future research should consider a multi-indicator approach to estimating WMC, for which we recently developed ultra-brief adaptations for ambulatory administration (Hakun et al., 2023).A second possibility relates to the potential influence of the assessment context (full-length lab vs ultra-brief ambulatory) on span task scores, which we elaborate further below.

Study limitations
Our findings have several limitations and our interpretations are based on a few key assumptions.The study involved a sample of college-aged adults.While everyday stressors are common in academic environments, reactivity and emotion regulation in adolescence/early adulthood may not reflect the way these constructs operate across other points in the adult lifespan (Blaxton et al., 2020;Sliwinski et al., 2009;Stawski et al., 2019).We reason that the relatively homogenous sample and university setting may contribute to the relatively low between-person variation in exposure rates, despite high variation observed within-persons, over time.Our interpretation of the present study findings is that the within-person moderation results reflect a self-regulatory process (e.g.emotion reappraisal).This assumes that the drive to reduce stressor-induced negative affect is normative (i.e. a hedonic principle/ heuristic).While this is a common assumption, it may not be representative of all individuals or groups.Finally, we argue that the failure to replicate previous between-person findings should not call into question the reliability of previous findings, rather may be attributable to the testing context (e.g.lab versus home/everyday environments).While we assume ambulatory assessments may provide a more ecologically-valid estimate of the state of one's cognitive ability because we are sampling incontext, over a person's daily life, in-lab assessments may provide information about the way a person handles a potentially stressful, longer-form testing session (which may itself relate executive control to real-life self-regulation; see also Kane et al., 2017).The amount of vigilance required to complete laboratory tests that often last between 25 and 30 min is likely reflected in WMC scores obtained in-lab and is unique in some ways from what is captured in the average of many ultra-brief assessments.That is to say that each mode of administration could be taken as a valid assessment of WMC, while still capturing unique information about WMC situated incontext (i.e. a more ecologically situated, low timecommitment state versus a state of sustained attention in a controlled, lab-based setting).

Conclusions
Our findings support and extend the executive hypothesis of self-regulatory capacity to variation in the behaviour of individuals over time.The use of ambulatory assessments increases the ecological validity of the study findings and suggests that shortterm changes in executive control may underlie the success of self-regulated and goal-directed behavior over the course of daily life.Future work should leverage the ambulatory approach to design new experimental and interventional frameworks based on the knowledge that self-regulatory success and failure may, at least in part, depend upon the state of one's executive control ability.

Table 1 .
Results from multilevel models examining associations among negative affect, stressor exposure, and WMC.