The Mediating Role of Affective States in Short-Term Effects of Activity Engagement on Working Memory in Older Age

Active activity engagement has been shown to protect against cognitive decline in older age [1, 2]. The research gap to understand mechanisms linking everyday activities and cognitive abilities has long been acknowledged but has rarely been examined [3, 4]. Recent ambulatory assessment studies have shown that everyday activities are associated with cognitive performance over the course of hours and days in older adults [5, 6]. For example, more frequent social interactions and more diverse activity engagement are related to better performance in ambulatory cognitive tasks (e.g., processing speed, working memory) on a daily level [6, 7]. Engagement in cognitive activities is associated with older adults’ better cognitive performance (e.g., semantic memory, executive functioning) at the same or the next measurement point over hours within a day [5, 8].

A recent ambulatory assessment study from our lab explicitly examined the duration of the time-lagged association between a bout of activity engagement and working memory performance in older adults [9]. More specifically, with seven assessments per day over 2 weeks, the study examined if activity engagement related to working memory concurrently, over one measurement point, or longer. Results showed that engagement in sociocognitive activities was associated with better working memory performance 6 h later, and engagement in passive leisure activities was associated with worse working memory performance 6 h later – and the effects completely faded out within 8 h [9]. In explaining the observed time-lagged activity effects on working memory, we hypothesized an emotional pathway where affective states triggered from activity engagement may have needed several hours to unfold their influence on cognitive performance. Yet, the proposed emotional pathway has not been empirically tested. As a follow-up to this recent study, the current paper examined whether affective states played a mediating role in the relationship between activity engagement and working memory using the same dataset.

In summarizing mechanisms that possibly link activity engagement and cognitive aging, a recent review identified an emotional pathway where activity engagement may enhance well-being and thus cognitive ability [4]. Positive affect, as the hallmark of well-being, has been proposed to engender success, such as having higher chances of being employed, having higher income, and maintaining good social relationships [10]. In other words, positive affect could lead to successful characteristics that arguably require higher cognitive functioning. In the same vein, the broaden and build theory proposes that positive emotions broaden people’s momentary thought-action repertoires (e.g., promoting discovery of novel and creative actions) and that such a broad repertoire in turn helps build enduring personal resources, such as in the intellectual domain [11]. In other words, positive affect enables the building of cognitive ability. Additionally, the resource allocation model proposes that processing negative affect occupies mental capacity and thus could impede ongoing cognitive task performance [12]. That is, negative affect might reduce cognitive ability. Taken together, the above theories suggest that higher emotional well-being, i.e., higher positive affect and lower negative affect, could enhance cognitive ability. In support of this proposition, ambulatory assessment studies of young adults showed that daily working memory performance was associated with higher daily positive affect [13] and lower daily negative affect [14].

In fact, both valence and arousal dimensions of affective states could be associated with cognitive performance. According to the circumplex model of affect, affective states could be characterized by the dimensions of valence and arousal [15]. For example, the cue-utilization hypothesis posits that an increase in emotional arousal reduces the amount of information (i.e., “cues”) that could be attended to (i.e., “utilized”) when performing tasks [16]. In support of this hypothesis, an ambulatory assessment study showed that the feeling of nervousness (i.e., high-arousal negative affect) was associated with worse working memory that was assessed at the same time in middle-aged and older adults [17]. Thus far, most theories have not specified the timeframe that is needed for affective states to take effect on cognitive performance. Moreover, existing evidence has focused on concurrent associations between affective states and working memory performance [13, 14, 17] but sheds little light on potential time-lagged associations. Thus, it is unclear how the positive and negative valence of high and low arousal are related to cognitive performance, particularly over the course of several hours.

On the other hand, older adults’ activity engagement has been shown to have concurrent associations with affective states [18, 19]. For example, social activities have been shown to be associated with higher positive affect and lower negative affect in older adults [20, 21]. Compared to engagement in social and cognitive activities, TV watching (a passive leisure activity) was associated with lower high-arousal positive affect, lower low-arousal positive affect, lower high-arousal negative affect, and higher low-arousal negative affect in young and middle-aged adults [22]. However, similar to research on affective states and cognitive performance, existing research has focused only on concurrent associations between activity engagement and affective states.

Taken together, theories on affective states and cognitive performance suggest that affective states could be a correlate and even determinant of cognitive performance. Moreover, prior daily life evidence showing concurrent associations of affective states with activity engagement and with cognitive performance suggests that affective states could be a plausible psychological mechanism linking short-term effects of activity engagement on cognitive performance in older adults’ daily lives. Most existing theories and studies have focused on concurrent associations between activity engagement, affective states, and cognitive performance but have rarely examined time-lagged associations. Our recent study indicating a short-term time-lagged effect of activity engagement on working memory over hours [9] hinted that affective states, if they indeed played a role, could have taken effect somewhen in the course over hours. As the first exploratory step toward establishing knowledge about a potential underlying mechanism behind the short-term effect of activity engagement on working memory, research is needed to examine if affective states are involved in the time-lagged process over hours.

This study examined the role of affective states (both valence and arousal) in linking activity engagement and working memory over hours. Following previous work [9], we examined data from the “Mobility, Activity, and Social Interactions Study” (MOASIS) [23]. For seven times per day over 2 weeks, participants reported their present activity engagement, affective states, and completed working memory assessments on a smartphone. The data were collected every 2 h with a random interval of plus or minus 0–15 min. Thus, the study had approximately equally spaced assessment points which is important for time-lagged association analyses. We examined both valence and arousal by categorizing affective states into four dimensions: high-arousal positive, low-arousal positive, high-arousal negative, and low-arousal negative. As an exploratory study on time-lagged associations between activity engagement, affective states, and cognitive performance over hours, we did not have directed hypotheses on the proposed effects. We pre-registered our study on OSF (https://osf.io/3bhk5).

Previous results from this dataset showed that sociocognitive activities had a positive effect on working memory, and passive leisure activities had a negative effect on working memory, and the effects only showed 6 h later [9]. Because of this time lag, it is possible that activity engagement may influence other determinants of working memory at a point in time before working memory itself. The goal of our analysis is to determine if activity engagement influenced affective states prior to influencing working memory and to then determine if affective states subsequently influenced working memory levels. That is, we examined if affective states mediated the time-lagged associations between activity engagement and working memory. We focused on two activities, as in the prior work: sociocognitive activities and passive leisure activities.

Figure 1 illustrates all possible ways in which activity engagement may directly influence working memory 6 h in the future while also influencing affective states, given the short-term effects of activity engagement on working memory over 6 h. The previously discovered time-lagged effect (lag 3 or 6 h) of activity engagement on working memory is shown as the c path. Further, there are different possible paths how, through affective states, activity engagement could exert effects on later working memory over 6 h at lag 3. For example, activity engagement could be associated with concurrent affective states (a1), which subsequently influence working memory 6 h later at lag 3 (b1). Alternatively, activity engagement could be associated with affective states 2 h later at lag 1 (a2), which subsequently influence working memory 4 h later at lag 3 (b2). Taken together, we expected that there would be four different possible paths linking activity engagement to affective states within a 6-h period (concurrent, lag 1, lag 2, lag 3 [a1, a2, a3, a4]) and that there would also be four different possible corresponding paths linking affective states to working memory within a 6-h period (lag 3, lag 2, lag 1, and concurrent [b1, b2, b3, b4]).

We recruited participants through the participant database of the institute’s survey center, advertisements in local newspapers, and snowballing among already registered participants. Participants’ inclusion criteria included being 65 years and older, having sufficient eyesight to operate the smartphone, having computer and internet access at home, and having a score higher than 26 in the Mini-Mental State Exam (MMSE) [24]. Participants were compensated with up to 200 Swiss Francs.

A total of 150 participants met the criteria. Participants had an average age of 73.48 years (SD = 5.60 years, range = 64.59–91.36 years, 47% men, 100% white) and an average of 14.02 (SD = 3.34) years of education. About 86% of participants were retired, and 14% were working part time or full time (n = 1). Participants provided 15,178 (90%) data points out of 16,800 (150 participants × 16 days × 7 assessments) possible data points.

As documented in the study protocol [23], the study had multiple phases. In particular, after signing informed consent, participants took part in a baseline assessment via telephone and in the lab during which they reported demographic information and completed a wide range of questionnaire and psychometric test assessments. Immediately afterward, participants carried a smartphone with them for a 2-week ambulatory assessment phase. Through the smartphone, participants were alerted to report their current activity and affective states and to complete a working memory task. The alerts were scheduled every day around 8:30, 10:35, 12:40, 14:45, 16:50, 18:55, and 21:00, i.e., every 120 min with a random interval of plus or minus 0–15 min.

Participants reported their current activity by choosing only one of 13 different types of activities (0 = absence, 1 = presence). Sociocognitive activities included the items of education/mental stimulation, cultural/religious activity, hobbies, social interactions, and work/volunteer. Passive leisure activities included the items of watching TV/listening to music and rest. Please refer to the 13 different activity items (the binary variables) and a confirmatory factor analysis of the activity categorization in previous work [9]. About 32% of observations reported engagement in sociocognitive activities, and about 18% of observations reported engagement in passive leisure activities.

On a 7-point scale (0 = not at all to 6 = very much), participants reported their current affective state using a set of items from the Positive and Negative Affect Schedule (PANAS) [25] and the experience-sampling version of the Multidimensional Affect Balance Scale [26]. High-arousal positive affect was specified as the average score of the following two items: happy and awake (M = 4.61, SD = 1). Its reliability estimates are not reported here because these estimates are not valid for constructs with two items. Low-arousal positive affect was specified as the average score of the following three items: content, relaxed, and balanced (M = 4.44, SD = 1.07, omega_within = 0.50, omega_between = 0.98). High-arousal negative affect was computed as the average score of the following four items: angry, annoyed, nervous, and worried (M = 0.49, SD = 0.81, omega_within = 0.67, omega_between = 0.98). Low-arousal negative affect was computed as the average score of the following four items: unwell, sad, restless, and without energy (M = 0.63, SD = 0.76, omega_within = 0.67, omega_between = 0.98).

Working memory was assessed on smartphones with two versions of a numerical memory updating task. The task involved a 2 × 2 grid with a total of four digits between 0 and 9. After starting the task via button press, the initial digits were randomly updated by an addition or subtraction between −8 and +8, which appeared consecutively in a random order in each of the four cells. Participants continuously updated and remembered the new resulting digit in each cell. There were, in total, two trials: an easier task version [27] and a more difficult version [28, 29]. In the easier (vs. more difficult) version, presentation time was initially 6,000 ms (vs. 4,000 ms) and 3,500 ms (vs.1,250 ms) at each of the following operations; the time between operations was 500 ms (vs. 250 ms), and there were 5 (vs. 8) operations in total. After the last operation, participants filled in the final result in each of the 4 cells. Working memory performance was indicated as proportionally correct across both trials, with a possible range of values between 0 and 1 (M = 0.59, SD = 0.24). Higher scores indicated a greater number of correct answers and thus better working memory performance.

Age was assessed as years since birth. Sex was assessed through a binary variable (0 = women, 1 = men). Education was assessed as years of education received. Time of day was the number of beeps on each day (range = 1–7). Time since start of the study period was the occasion number since the study began (range = 1–112 [7 assessments × 16 days]).

Similar to previous work [9], we used a multilevel vector autoregression (mlVAR) model to study the interplay of activity engagement, affective states, and working memory. A mlVAR model is a network-based approach to modeling time series data nested within multiple participants [30, 31]. For the purposes of this study, we simultaneously estimated within-person concurrent and time-lagged associations between activity engagement, affective states, and working memory across time to determine how activity engagement influences affective states, which then in turn influence working memory.

In the mlVAR model, variables are represented as nodes and the relationships between these variables are represented as directed or undirected edges. The mlVAR models examine the relationships between multiple variables at multiple time lags. It estimates separately within-person temporal networks, concurrent networks, and between-person networks. Specifically, the mlVAR algorithm first estimates the relationships between all variables across time, centered within individuals, to determine within-person temporal dynamics, and then assesses the association between the residuals of this model to determine within-person concurrent effects that occur above and beyond within-person temporal dynamics. With mlVAR, we are able to test multiple time lags to determine the optimal time lag observed between variables. According to the previous related work [9], the models were estimated with up to 4 lags, which was the highest possible number of lags to be included in the models without convergence issues. We estimated separate mlVAR models for the two different activity types and the four different affective states. This led to eight different models as reported in Tables 1 and 2.

We estimated mlVAR models using the R package mlVAR version 0.5 [32]. The package yields two p values for within-person concurrent relationships because these networks are based on a function of two parameters that are standardized and averaged within each participant [32]. To determine significance of the within-person concurrent associations, we used the AND rule, which requires both of these p values to be significant [32]. p values <0.05 were considered to be statistically significant.

We controlled for demographic variables and time trends related to time of day and time since start of the study period before estimating the mlVAR models. That is, we estimated separate multilevel models on each focal variable (activity engagement, working memory, affective states) as a function of age, sex, education, and linear and quadratic effects of time of day and time since start of the study period. We then obtained the residual scores from these models and used those scores for the mlVAR analyses. This procedure requires all included variables to have no missingness. Multilevel models were estimated in R version 4.2.1 [33] using the packages “lme4” version 1.1–29 [34] and “lmerTest” version 3.1–3 [35]. A total of 3,161 observations out of the 15,178 (21%) data points had missing data in some of the focal variables. Thus, the final analytical sample included 12,017 (79%) observations from 149 participants.

We used augmented Dickey-Fuller tests to assess if our detrending approach was successful at creating stationary time series [36, 37]. At the highest lag studied in our paper, lag-4, 95% of participants' time series were significant at the 0.05 significance level, indicating our detrending process was successful. An additional analysis with only the participants who had stationary time series showed no differences in statistical significance. We thus consider our results to be robust. The augmented Dickey-Fuller tests were conducted with the R package “aTSA” version 3.1.2 [38].

We tested the significance of possible mediation effects using a Sobel test [39]. The Sobel test is conducted by comparing the strength of the indirect effect to the point null hypothesis that it equals zero. The test calculates the standard errors of the indirect effect to determine a test statistic that can be used to determine the significance of mediation. This test statistic can then be compared against a z distribution to determine a p value. We applied a Sobel test to any cases in which we found significant paths which may indicate mediation between variables and across time.

Consistent with previous analyses using the same data [9], sociocognitive activities had a positive effect on working memory performance at lag 3 (Table 1, c paths in models 1–4). Moreover, sociocognitive activities were associated with concurrent higher high-arousal positive affect (Table 1, model 1, a1 = 0.06, ps = 0.007 and 0.009) and higher low-arousal positive affect (Table 1, Model 2, a1 = 0.05, ps = 0.018 and 0.017). We did not find any concurrent or time-lagged associations between affective states and working memory (Table 1).

Likewise, passive leisure activities had a negative effect on working memory performance at lag 3 (Table 2, c paths in models 1–4). Passive leisure activities were associated with concurrent lower high-arousal positive affect (Table 2, model 1, a1 = −0.12, ps = <0.001) and lower high-arousal negative affect (Table 2, model 3, a1 = −0.07, ps = 0.002 and 0.001). Moreover, high-arousal positive affect was associated with lower working memory performance at lag 3 (Table 2, model 1, b1 = −0.05, p = 0.020). Additionally, passive leisure activities were associated with higher high-arousal positive affect at lag 2 (Table 2, model 1, a3 = 0.07, p = 0.001) and lower low-arousal negative affect at lag 2 (Table 2, model 4, a3 = −0.05, p = 0.003).

Notably, the concurrent association between passive leisure activities and high-arousal positive affect (Table 2, model 1, a1 = −0.12, ps = <0.001) and the lagged association between high-arousal positive affect and subsequent working memory performance at lag 3 (Table 2, model 1, b1 = −0.05, p = 0.020) seemed to indicate a possible mediation. We thus conducted a Sobel test on these findings and found a significant mediation effect of high-arousal positive affect on the relationship between passive leisure activities and working memory (Sobel test statistic = 5.64, p < 0.001).

This study examined the mediating role of affective states in the process of activity engagement, exerting its effects on working memory performance over hours. A prior study using this dataset showed that engagement in sociocognitive activities was associated with better working memory performance 6 h later and engagement in passive leisure activities was associated with worse working memory performance 6 h later in the daily lives of older adults [9]. Building upon these findings, we hypothesized different possibilities for how affective states may link the short-term effects of activity engagement on working memory performance over hours (Fig. 1).

As shown in Figure 1, we hypothesized four different possible paths linking activity engagement to affective states within a 6-h period (concurrent, lag 1, lag 2, lag 3 [a1–a4]). In models of sociocognitive activities, activity engagement was associated with concurrent higher high- and low-arousal positive affect. These findings are in line with prior evidence that has shown social activities are associated with higher positive affect [20, 21]. In models of passive leisure activities, activity engagement was associated with lower high-arousal positive and negative affect. These findings are also in line with the prior study showing TV watching being related to lower high-arousal positive and negative affect in young and middle-aged adults [22].

In addition, our results showed that passive leisure activities were associated with higher high-arousal positive affect and lower low-arousal negative affect at lag 2. These findings suggest that engagement in passive leisure activities could also influence affective states that were captured approximately 4 h later. Constrained by the little research on time-lagged effects of activity engagement on affective states, we do not have an explanation on the observations.

As seen in Figure 1 again, we also hypothesized four different possible corresponding paths linking affective states to working memory within a 6-h period (lag 3, lag 2, lag 1, and concurrent [b1–b4]). When considering our findings, only some of these proposed paths were indeed found: In models of sociocognitive activities, we found no significant relations between affective states and working memory performance, be it concurrently or with time lags. We reason that the relatively broad operational definition of sociocognitive activities in this study (including visiting church and reading) might have led to weaker associations with affective states (e.g., Table 1, model 1, a1 = 0.06, SE = 0.03, ps = 0.007 and 0.009), whose time-lagged effects on working memory might then be too weak to be captured in our study. In contrast, parameters in the model of passive leisure activities with high-arousal positive affect seemed stronger (Table 2, model 1, a1 = −0.12, SE = 0.07, ps < 0.001).

Further, in models of passive leisure activities, higher high-arousal positive affect was associated with lower working memory performance at lag 3. We will return to this finding in the next section. Additionally, we did not find concurrent or other time-lagged associations between affective states and working memory performance.

Notably, we did not find any concurrent associations between affective states and working memory performance in models including sociocognitive activities or passive leisure activities as predictors. These findings seem to contradict those from previous ambulatory assessment studies that showed concurrent associations between affective states and working memory performance when assessed on the same day [13, 14] or at the same time [17]. However, the concurrent associations in the mlVAR models are correlations between variables after accounting for time-lagged associations (i.e., a correlation of residuals after accounting for time-lagged associations). When there is meaningful time dependence between variables, the concurrent associations in our models can yield substantially different values compared to lag-0 concurrent correlations reported in the previous studies [13, 14, 17], which did not take into account time-lagged associations. In turn, our findings highlight the importance of further examining concurrent and time-lagged associations between affective states and cognitive performance.

As seen in Figure 2, passive leisure activities were negatively associated with concurrent high-arousal positive affect, which was negatively associated with working memory performance 6 h later. Results of the Sobel test showed that high-arousal positive affect mediated the time-lagged relation of passive leisure activities on subsequent working memory performance. Specifically, high-arousal positive affect may have a suppression effect on the statistical relationship between activity engagement and working memory. Within a mediation model, a suppression effect is present when the direct and indirect effects of an independent variable on a dependent variable have opposite signs [40]. In Figure 2, the direct effect is the c path and the indirect effect is the product of the a1 path and the b1 path. When a suppression effect exists, the magnitude of the relationship between an independent variable and a dependent variable becomes larger when a third variable is included [40]. To put our findings in context, if high-arousal positive affect was not accounted for, the estimated relationship between passive leisure activity and working memory (Table 2, model 1, c = −0.08, SE = 0.02, p = 0.001) may be null or weaker than what we observed. This was confirmed when we conducted a mlVAR model to estimate the relationship between the two variables of passive leisure activities and working memory: There was a significant lag-3 association over 6 h (estimate = −0.06, SE = 0.02, p = 0.006). Taken together, our findings suggest that passive leisure activity related to reduced working memory approximately 6 h later and this relation was altered when including high-arousal positive affect in the model. Passive leisure activity related to decreased concurrent high-arousal positive affect. Although lower high-arousal positive affect related to increased working memory 6 h later, the total effect of the activity on working memory was negative. It is plausible that high-arousal positive affect is an essential but not the only relevant element in this process. However, without testing for additional mediator or moderator variables on this mediation effect, we cannot be sure of this.

Markedly, none of the previously discussed theories on affective states and cognitive performance specify a timeframe over hours, such as the broaden and build theory [11], the resource allocation model [12], and the cue-utilization hypothesis [16]. To understand the time-lagged relation between high-arousal positive affect and working memory performance, we refer to the neuroendocrine research on cortisol and cognitive performance. Research has shown that an emotionally arousing experience could trigger the release of hormones, such as norepinephrine and cortisol, which work in concert to influence brain functioning [41]. Specifically, shortly after an arousing event, norepinephrine enhances the function of excitatory synapses – diverting attention to the event – and cortisol promotes this rapid process; later on, cortisol normalizes brain activities that were earlier raised, allowing for restoration of higher cognitive control in the aftermath of the arousing event [42, 43]. The process where cortisol normalizes brain activities once an arousing experience subsides is called slow genomic effects, which have been shown to unfold in a timeframe of several hours [44‒46]. For example, in a randomized controlled trial with young and healthy men, 4 hours after receiving hydrocortisone (i.e., cortisol supplied as a medication) as opposed to a placebo, participants showed better performance in a numerical working memory task and more neuronal activity in the dorsolateral prefrontal cortex (a region for higher-order cognitive processing) [47]. Timeline of exogenous administration of cortisol in experiments of this kind was designed based on prior animal studies which showed the slow genomic cortisol effects appeared within 6 h [48, 49].

The time window of the slow genomic effects of cortisol appears to match our observation of high-arousal positive affect having a negative effect on working memory after around 6 h. Thus, as shown in Figure 3, we consider our results to reflect a neuroendocrine pathway where passive leisure activities influence concurrent high-arousal positive affect – releasing hormones (e.g., norepinephrine, cortisol) – and subsequently influence working memory that was assessed around 6 h later. We also speculate that this hypothesized process might be applicable to general activity engagement, such as other sociocognitive activities, as they also showed time-lagged effects on working memory assessed 6 h later, although we did not observe a significant mediation effect with sociocognitive activities – possibly due to the broad operational definition in this study. We additionally speculate that this process might be applicable to other cognitive abilities because the slow genomic effects of cortisol have been shown in working memory and other cognitive abilities such as inhibition [46]. Nevertheless, future research is needed to test these hypotheses.

Notably, research on the slow genomic effects of cortisol suggests that an arousing experience could evoke a higher cortisol level, leading to higher working memory performance. The described expected pattern, of course, contradicts what we observed: higher higher-arousal positive affect being associated with subsequent lower working memory (Fig. 2, b1 path). However, both arousal and cortisol are theorized to have an inverted-U-shaped relationship with cognitive performance, such that too high or too low levels of arousal or cortisol could impair cognitive performance [16, 50]. The arousal level of positive affect that was captured by our study with the items of happy and awake might be at a high-arousal level, and, thus, the higher level of it impaired working memory performance. There is little research matching arousal level, cortisol, and working memory in daily life. Our speculation should be examined in future research.

Additionally, we only found high-arousal positive affect linking passive leisure activity and working memory performance, but not low-arousal positive affect or high- or low-arousal negative affect. First, positive affect might be specifically relevant to the numerical working memory task in our study. Operations of addition and subtraction (as in numerical updating working memory tasks) dominantly recruit the left hemisphere [51]. Positive affect has been shown to enhance verbal updating tasks that depend on the left frontal lobe, while negative affect has been shown to enhance spatial working memory tasks that rely on the right frontal lobe [52]. Second, compared to low-arousal affect, high-arousal affect might be more relevant to arousal-triggering cortisol and subsequent changes in working memory performance. According to the cue-utilization hypothesis, low-arousal affect may represent circumstances with little resource competition and less intense physiological reactions [16]. This could be particularly true as older adults can be vulnerable to emotional arousal and have greater difficulties returning to homeostasis after experiencing arousal [53]. Taken together, older adults’ working memory performance could be susceptible to the influence from high-arousal positive affect when assessed hours ago.

This study examined 2-week intensively collected data in real life and had an innovative focus on concurrent and time-lagged association analyses between activity engagement, affective states, and working memory in a short-term time scale, aimed at zooming into processes that link daily life activities and cognitive performance fluctuations. In addition to these strengths, we acknowledge several limitations of the study. First, it has been hypothesized that arousal and cognitive performance have an inverted-U-shaped relationship [16]. We cannot test non-linear associations with the mlVAR model, but future research could consider this possibility. Second, the 6-h time-lagged effects captured in our study might have been due to the study design with measurements arranged approximately every 2 h. Future research could consider collecting data with shorter or varying intervals and using continuous-time models [54] to identify the true duration of time-lagged associations between activity, affect, and cognitive performance. Third, we requested participants to report the types of present activity engagement. Yet, other dimensions to describe activities, such as quality of social interactions or level of cognitive stimulation might be more relevant to affective states [55, 56]. Future studies could consider using sensor-based methods to capture more diverse features of daily activities [57, 58]. Fourth, we could not test the hypothesis of a neuroendocrine pathway that possibly explains the time-lagged effect of activity engagement on working memory in this study. Future studies could consider testing this hypothesis by including hormonal data such as norepinephrine and cortisol in their examination. Finally, our sample of Swiss older adults has a relatively homogeneous background [9]. Future studies could consider examining a sample of older adults with more diverse backgrounds and estimate how the within-person time-lagged associations might differ across different individuals.

To our knowledge, this study is the first to examine associative pathways behind the short-term effects of activity engagement on cognitive performance over hours within a day in the lives of community-dwelling older adults. We found that passive leisure activities reduced concurrent high-arousal positive affect, which was associated with subsequent worse working memory performance. Our findings suggested that high-arousal positive affect could be a possible underlying factor behind the time-lagged effect of passive leisure activities and working memory performance. We speculated that activity engagement could change levels of high-arousal positive affect, which might trigger the release of norepinephrine and cortisol, taking several hours to exert effects on working memory. However, it remains to be examined how norepinephrine and cortisol levels respond to activity engagement and emotional affect in the real world and how they further influence working memory and cognitive performance over hours.

Our results pave the way for a range of further research opportunities. For example, future research could examine biological pathways linking affective states with both activities and cognition, which might eventually contribute to the largely unknown inquiry on understanding mechanisms behind the activity-cognition associations [3, 4]. Furthermore, our focus on short-term processes could encourage inquiries on whether and how the short-term activity effects and the mediating role of affective states could be linked to long-term activity-cognition associations accumulated over years in the aging process. In sum, our findings highlight an emotional, and possibly a neuroendocrine, pathway in activity-cognition associations. More research should be devoted to study underlying mechanisms behind these associations, in short-term or multiple timescales.

The research reported is based on the “Mobility, Activity, and Social Interactions Study” (MOASIS) conducted in cooperation between the URPP “Dynamics of Healthy Aging” and the Department of Geography, University of Zurich. The following persons as members of the MOASIS project team contributed to the initial research idea and the planning and implementation of the project, in addition to one of the co-authors (Christina Röcke, Co-PI): Mike Martin (Co-PI), Robert Weibel (Co-PI), Mathias Allemand, Burcu Demiray, Pia Bereuter, Michelle Fillekes, Marko Katana, George Technitis, and Alexandros Sofios. We thank Brigitte Sonderegger and Corinne Boillat for the recruitment of participants, our student assistants, particularly Victoria Gehriger, for support and orchestration in data collection, and above all, our participants, for their time and willingness to take part in this research. We thank Simona Egli for her support in producing Figure 3 and Laura Breitfelder for supporting literature search in the early stage of the paper. Some of the findings from this paper were presented at the online Annual Scientific Meeting of the Gerontological Society of America 2021.

The study procedures were conducted according to the Declaration of Helsinki and were approved by the Ethics Committee of the Faculty of Arts and Social Sciences of the University of Zurich (permission no. 17.2.4). Written informed consent was obtained from all the participants.

The authors have no conflicts of interest to declare.

The research reported in this paper was supported by the Velux Stiftung (project no. 917) and the University Research Priority Program (URPP) “Dynamics of Healthy Aging” at the University of Zurich. The open access funding was provided by University of Zurich.

Minxia Luo initiated the study, performed data analysis, interpreted results, and drafted the manuscript. Robert Glenn Moulder led the setup of data analysis procedure, supported the interpretation and reporting of the results, and provided critical comments on the manuscript. Elisa Weber supported the setup of data analysis procedure and provided critical comments on the manuscript. Christina Röcke developed the study concept and the study design, oversaw data collection, and provided critical comments on the manuscript. All authors approved the final version of the manuscript for submission.

The data that support the findings of this study are not publicly available because the data belong to an ongoing longitudinal study. The data are available upon reasonable request from the data sharing committee via Dr. Christina Röcke (christina.roecke@uzh.ch).