Participants were recruited from existing databases of participants from prior studies, and advertisements in university listservs, newsletters, and community locations. To meet inclusion criteria, participants had to be English-speaking adults aged 40 years or older, not have a clinical diagnosis of dementia, or suffer from any significant visual or hearing impairment. Participants were also required to wake regularly between the hours of 4-10 a.m. (i.e., to adhere to tablet prompts in momentary component of study).

Ninety adults completed the first testing session, but only 81 individuals completed the second testing session that involved cognitive testing. Reasons for discontinuing the study included time constraints, occupational interferences, and personal reasons. There were 59 females and 22 males, and the mean age was 61.10 years (SD = 12.24; 41-94 years). Nearly all participants, were Caucasian (92.6%), 54.3% were married, and 71.6% lived with others. Average educational attainment was 17.35 years (SD = 2.79). Most participants (64.2%) were working, and the rest either completely or partially retired. Approximately 75% reported household incomes higher than $50,000, and 95.1% rated their health as good or better.

Participants were entered in a drawing for 10 tablets (~$150 value), and 20 $50 gift cards. Each participant who completed the study received one entry, and if they achieved at least 90% compliance in the momentary daily portion of the study, they received an additional entry.

The data were collected with approval by the Institutional Review Boards at Colorado State University, Protocol ID 15-5950H, approved August, 2015.

Data for the present analysis were collected over two testing sessions. All testing sessions occurred on campus and were conducted by a trained lab member. The first session was conducted in a small group setting and took approximately two hours to complete. Following informed consent, participants completed a series of questionnaires and surveys that collected basic demographic and health information, a general activity questionnaire, and other questionnaires about personality and meaningful activity. Two to three days later the participant returned to the lab for the second testing session where a series of cognitive tests were administered. This session took approximately an hour to complete.

Statistical analyses were performed using SPSS Statistics version 26. Skewness and kurtosis were examined to ensure each variable was normally distributed. Little’s Missing Completely at Random Test [32] for both activity (χ²(200) = 224.19, p = 0.12) and meaningfulness questionnaires (χ²(8) = 11.76, p = 0.16) indicated the data was missing at random. Less than 5% of the data was missing from the VLS-AQ and EMAS, and expectation maximization procedures were used to input missing data. There was no missing data for the cognitive tasks.

A series of factor analyses were then performed for each of the main variables: activity, meaningfulness, and cognition. This was necessary to confirm prior factor solutions for the VLS-AQ [24], and the EMAS [33]. Confirmation of these prior factor solutions was first conducted. If needed, exploratory analyses were conducted following the pattern of exploratory factor analysis (EFA) followed by confirmatory factor analysis (CFA) evaluating model fit. These analyses use a reflective measurement model based on the assumption that the individual items measured the underlying constructs of engagement and meaningfulness. The same process of EFA and CFA was performed for cognition to classify tasks with similar functional properties. Once the ideal factor structure underlying each latent variable was confirmed, mediation analysis proceeded. Although mediation analyses cannot speak to causal pathways in a cross-sectional sample, the analyses provide a starting point to examining the associations between these constructs.

Two primary SEM models were tested. Model 1a evaluated the direct effect of activity on cognition and Model 1b tested the indirect effect of activity on cognition mediated by meaningfulness. Model 2a evaluated the direct effect of meaningfulness on cognition and Model 2b tested the indirect effect of meaningfulness on cognition as mediated by activity. Model fit was evaluated using several indicators: chi-square (χ²) statistic, where lower values typically indicate better fit; root-mean-square error approximation (RMSEA), where values <0.05 indicate good model fit; incremental fit index (IFI) and Tucker-Lewis index (TLI), where higher values indicate better fit (e.g., IFI > 0.90, TLI > 0.80); and comparative fit index (CFI), where values ≥0.9 indicate good model fit [34]. The significance of the indirect effect was tested using bootstrapping procedures. Unstandardized indirect effects were computed for each of the 1,000 bootstrapped samples, and the confidence interval was computed by determining the indirect effects at the 5th and 95th percentiles. In all models, age and education were included as covariates.

A calculator for SEM was used to estimate power [35]. Both direct models with two latent variables with N = 81 had >99% power to detect large effects and 76% power to detect medium effects. The indirect models with three latent variables were estimated to have >99% power to detect large effects, and 60% power to detect medium effects. Both models were estimated to have limited power to detect small effects (10-13%). Given that there was no research thus far directly examining relationships between activity meaningfulness and activity participation, it was unclear what effect sizes to expect. Due to limited power for detecting small-to-medium effects, non-parametric bootstrapping (1000 samples) was also used. This approach is a robust alternative method that provides more reliable estimates.

In Model 1a, activity was regressed onto cognition (see Table 2). Model 1a provided poor fit, χ²(23) = 33.19, p = 0.08, RMSEA = 0.09, IFI = 0.89, TLI = 0.80, CFI = 0.87. The direct effect hypothesis was not supported: β = 0.29, p = 0.77, demonstrating the link between activity and cognition was not significant.

The indirect effect Model 1b was tested with meaningfulness added as the mediating variable (see Figure 1). Table 2 reports the standardized regression coefficients (β) for each pathway in this model. The indirect effect Model 1b provided good fit, χ²(46) = 56.40, p = 0.14, RMSEA = 0.05, IFI = 0.94, TLI = 0.90, CFI = 0.93. In comparison to the direct effect Model 1a, the fit indices for the indirect effect were significantly better (i.e., higher IFI, TLI, CFI) and acceptable for a “good fitting” model. However, the regression coefficients linking activity to meaningfulness and meaningfulness to cognition were not significant. These results indicated that activity did not predict meaningfulness, and neither activity nor meaningfulness predicted cognition. Further, the association between activity and cognition was not mediated by meaningfulness. The bootstrapped unstandardized indirect effect of activity on cognition as mediated by meaningfulness was not significant (-0.10, SE = 0.64, 95% CI = -2.39, 0.53, p = 0.51).

The standardized regression coefficient (β) for the direct effect of activity on meaningfulness was quite large (β = 2.67) as was the effect of age and education on meaningfulness (β = -1.84). Standardized values outside the range of -1 to 1 can sometimes be valid [36], indicating a high correlation between predictor variables. It can also indicate that unreasonable model constraints have been imposed, however no error messages were incurred nor was the maximum iteration reached; therefore, these values were deemed acceptable. Rather, the range for standardized coefficients was likely exceeded due to the small correlations between the EMAS factors and the social activity factor. However, because the multicollinearity was considered quite small (>0.4) by research standards [37], it did not pose a significant threat to this analysis. The correlation matrix for all observed variables is reported in Table 3.

In Model 2, the roles of the latent predictor variables were reversed where meaningfulness served as the independent variable predicting cognition. Model 2a provided good model fit, χ²(10) = 10.09, p = 0.43, RMSEA = 0.01, IFI = 0.99, TLI = 0.99, CFI = 0.99. However, the direct effect of meaningfulness on cognition was not significant, demonstrating that greater meaningfulness was not associated with higher cognition (see Table 4 for model parameters).

Next, activity was added to the model as the mediator variable. Figure 2 displays the simplified path diagram for the indirect effect Model 2b. Model 2b provided adequate model fit, χ²(46) = 56.35, p = 0.14, RMSEA = 0.05, IFI = 0.94, TLI = 0.90, CFI = 0.93. Meaningfulness significantly predicted activity; as meaningfulness scores increased by 1 unit, the frequency of activity engagement increased by 0.82. However, higher activity levels did not predict cognition, nor was the relationship between meaningfulness and cognition mediated by activity engagement frequency. The bootstrapped unstandardized indirect effect of meaningfulness on cognition as mediated by activity was not significant (0.20, SE = 0.84, 95% CI = -0.47, 4.57, p = 0.40).

Our hypotheses of mediation of cognition relied on the expectation of a positive relationship between activity and meaningfulness. Meaningfulness positively predicted activity, where higher levels of meaningfulness was associated with higher levels of activity participation. This is in line with the selective engagement hypothesis [21], which proposed that activities with greater meaning are selected more often and engaged in more frequently than less meaningful activities. This also aligns with occupational therapy research, reporting that more meaningful and frequent activity engagement in older adults explains higher life satisfaction and other favorable psychological health and well-being outcomes [33]. The positive relationship may particularly apply to physical, social, and developmental activities (see Table 3). All three EMAS factors were positively correlated with the social activity factor, suggesting that social health, a key part of successful aging, is linked to the perceived value of an activity. The physical and developmental activity factors were also positively correlated with the social value and competence EMAS factor.

Meaningful activity experiences may also foster intrinsically motivated behavior. Eakman [38] has argued meaningful activities as assessed by the EMAS are commonly imbued with experiences of autonomy, competence, and pro-social relatedness, which serve to motivate engagement and promote well-being through fulfilling psychological needs [39]. Additionally, activity engagement as assessed by the VLS-AQ may be sensitive to the EMAS meanings of pleasure and satisfaction which are known to motivate persistence in goal-directed action [40]. Given these findings, encouraging meaningful activity may represent a promising way to promote engagement, a key component in a successful aging trajectory.

In contrast, the path coefficient of activity predicting meaningfulness was not significant. Therefore, while directionality cannot be determined with these cross-sectional data, this pattern of association aligns with the idea that the meaning attached to engagement may be more closely tied to driving activity participation rather than vice versa. These findings can be explained by theories of leisure and recreation, which state that meaningfulness motivates self-creation and promotes participation in novel experiences [41]. According to Atchley [14], individuals select and develop ideas, relationships, environments, and activities based on personal dispositions about meaning in their lives. Personal meaning influences both individual and social expectations regarding activity engagement and suggests that when an activity is perceived to be more meaningful (i.e., feelings of self-determination and competence), it is more likely to be expressed through behavior [41].

In comparison to the direct effect model of activity on cognition (Model 1a), the model fit statistics for the direct effect of meaningfulness on cognition (Model 2a) were much stronger. However, in the individual parameters, meaningfulness did not directly or indirectly influence cognition. This was unexpected given theoretical predictions stemming from the environmental complexity hypothesis [22] which suggests that activities with greater meaning are selected more frequently due to their level of cognitive stimulation. From this perspective, meaningfulness was thought to enhance cognitive ability by promoting activity selection and motivation [14,17,18]. Although the results showed that meaningfulness predicted activity selection, these selections were not significantly associated with cognitive performance in the current sample. Explanations for this may lie in disentangling the correlations. For example, the reasoning and executive functioning cognitive factor was negatively correlated with EMAS factors 1 and 2 and the social activity factor. Although these associations were small, the mediational results suggest that there may be complex associations between different domains of meaningful engagement and cognition.

In both model variations, the pathways linking activity to cognition were not significant. These findings were unexpected, as associations between activity and cognition are fairly-well established [1,42,43]. Although some cross-sectional investigations have shown that increased participation in certain cognitively demanding activities (e.g., reading the newspaper, Sudoku) is related to higher cognitive performance, others found that higher cognitive performance was associated only with a greater frequency of participation in activities that occurred within a group setting (e.g., bowling, religious activity) [42,43]. However, the absence of a statistically significant overall association between activity and cognition found in the present study is consistent with some prior research [44,45], and there is evidence that associations are more common between certain activity and cognitive domains rather than overall activity [46].

Another point to consider is the activity questionnaire that was used. Some researchers suggest that different time frames over which activity is assessed provide different engagement depictions by tapping into different underlying meaning and memory processes [47]. Although a well-established scale with a wide range of activity was used, the consideration of activity over the past two years may not fully capture fluctuations in activity participation. When activity is assessed over a shorter time frame (e.g., weekly, daily), reported information may be less subject to cognitive biases and memory distortions [1,5,45,47,48]. It is unknown whether assessment at the daily level provides a stronger link to cognition, but research points to the benefit of using multiple types of assessments to measure activity in relation to cognition [46]. Further, there are several challenges in measuring activity engagement (see [1,6,7]), and it remains possible that the scale used to assess activity engagement contributed to the null results.

Given the exploratory nature of this study, these structural models offer a preliminary framework for these relations and highlight several methodological avenues for future research. The sample was not representative of the middle-aged and older adult population, as they were highly educated, primarily White, and reported earning a higher income than U.S. population studies [49]. The lack of sample diversity might have contributed to the non-significant results through lower interindividual variability in cognition and engagement, and limits the generalizability of the findings. Further, it seems probable that the nature of leisure varies greatly depending upon available environmental resources and cultural influences. Future research should examine the associations between activity, meaningfulness of engagement, and cognition cross-culturally, as cultural values towards activity may impact these structural relations.

Several limiting factors may have contributed to the null effects. The sample size complicated interpretations of the factor analyses, especially regarding the model fit for the VLS-AQ, and it is unclear how these complications may have impacted the mediational analysis. The lack of inclusion of various activities due to factor constraints in the structural model may also explain the lack of an association between activity and cognition. Other research reporting significant effects on cognition has generally used larger samples which allowed for more activity items to fit within factor parameters (e.g., n = 153 [20]; n = 267 [24]). Relatedly, there was limited power to detect small effects, implying that links between activity, cognition, and meaningfulness may in fact exist. For example, confidence intervals around the indirect paths of meaningfulness on cognition as mediated by activity, and activity on cognition as mediated by meaningfulness were both quite large, suggesting the effect might exist but the estimates were imprecise in the dataset. The sample size also reduced the range of analytical opportunities that could be explored post-hoc (e.g., moderated mediation). We also note that conducting EFA and mediation testing on the same dataset may cause model overfitting. While our primary goal was examining structural relationships rather than scale validation, future research should use independent samples to confirm these findings.

Further, testing only baseline structural associations between these constructs may not have fully captured the dynamic and contextual influences of the subjective experiences associated with engagement. According to the environmental complexity hypothesis [22], motivation to selectively engage in more meaningful activity occurs when cognitive skills are applied across various settings and situations. Thus, the inability to predict cognition may lie in the static nature of the cross-sectional study design. Relatedly, the EMAS assessed meaningfulness of activity in general versus as they occurred. The relations between activity, cognition, and meaningfulness may be more apparent over micro timeframes such as throughout and across days. Moreover, parsing meaningfulness towards specific activities may also be a fruitful future direction.

Finally, although the present findings adjusted for age and education, other covariates (e.g., gender, subjective health status) were not controlled because of the added complexity that tended to over-saturate the models. Lifestyle and health factors can influence cognition and future work with larger samples should incorporate these relevant influences.