Unconditional quantile regression analysis of volunteering and cognitive functioning: Implications for cognitive impairment and dementia

Abstract

Introduction:

Volunteering is widely recognized for its health benefits, including improved cognitive functioning and delayed onset of cognitive impairment and dementia. However, the extent to which these cognitive benefits vary across individuals with different levels of cognitive functioning, particularly among those with cognitive impairment, remains unclear.

Methods:

This study used 195,633 person-wave observations from 31,371 adults in the Health and Retirement Study (HRS; 1998–2020). Unconditional quantile regression (UQR) models with fixed effects were employed to examine how the cognitive benefits of volunteering—measured by participation status and annual time commitment—differ across the population distribution of cognitive functioning.

Results:

Volunteering was associated with better cognitive functioning on average, but the benefits were most pronounced at the lower end of the distribution, corresponding to those experiencing cognitive impairment or probable dementia. Moderate engagement (approximately 2–4 hours per week) yielded the largest cognitive benefits for these groups, while benefits plateaued at higher levels of engagement.

Conclusions:

Volunteering may serve as a scalable, lifestyle-based strategy for maintaining or improving cognitive health, particularly among those at risk for or already experiencing cognitive decline. These findings highlight the importance of tailoring volunteer opportunities to match individual capacities and suggest that even modest engagement can produce meaningful cognitive benefits.

Introduction

In the context of population aging and growing concerns about cognitive impairment and dementia, promoting social engagement and reducing isolation have emerged as promising lifestyle-based approaches to promote healthy cognitive aging.¹ A recent UN report estimated that over 860 million people engage in volunteering each month worldwide, including more than 11 million individuals in the U.S., over 11% of whom are aged 55 and older.^2,3 Given its broad reach and relatively low barriers to participation, volunteering is increasingly viewed as a viable and scalable public health intervention to prevent or delay cognitive decline and dementia.⁴

The evidence supporting this view continues to grow, as indicated by a set of reviews linking volunteering to improved cognition, slower cognitive decline, and reduced dementia risk.^5–7 Yet a critical gap remains: it is unclear whether these benefits vary across levels of cognitive functioning. Most prior research has emphasized average effects while addressing reverse causality and selection into volunteering, given that suboptimal cognition can be a barrier to engagement.⁸ However, far less is known about whether and how volunteering is linked to cognitive functioning once decline has begun. This is consequential because many older adults with cognitive impairment or dementia continue to volunteer.⁹ Understanding whether—and to what extent—volunteering benefits those with suboptimal cognitive functioning presents essential knowledge for developing interventions that promote cognitive health across all levels of functioning.

Selective entry into volunteering based on cognitive functioning—and better health in general—often leads to the presumption that benefits of volunteering may be greater for those who are already cognitively fit. However, there are both theoretical and empirical reasons to expect the opposite.⁴ Volunteering is widely known to promote psychosocial resources, including social connection and support, a sense of meaning, and self-efficacy, while also helping to buffer the detrimental effects of daily and chronic stress that can impair physiological functioning and accelerate cognitive decline.^10,11 These mechanisms are especially relevant for individuals with compromised cognitive functioning, who are more likely to lack such psychosocial resources.

Further, although the authors are not aware of any work to date that has directly examined whether the cognitive benefits of volunteering vary across the population distribution of cognitive functioning, a growing body of research shows distributional effects on other health outcomes. For example, analyses of longitudinal data from European countries reported stronger physical health benefits of volunteering among those in poorer health.¹² A similar pattern of distributional effects was found among U.S older adults, where volunteering was more strongly associated with improvements in frailty among subpopulations in worse health compared to their healthy counterparts.¹³ These findings bolster the possibility that the health-promoting mechanisms of volunteering may be especially potent for individuals at the lower end of the health distribution. Whether a similar pattern exists for cognitive functioning, however, remains to be tested empirically.

To address this gap, this study examines whether and how the cognitive outcomes associated with volunteering vary across the population distribution of cognitive functioning, using nationally representative data and analytic methods—unconditional quantile regression models—that allow for the investigation of distributional heterogeneity. Specifically, the study evaluates whether individuals with lower cognitive functioning, including those with potential cognitive impairment or dementia, experience greater cognitive benefits from volunteering compared to their higher-functioning counterparts.

Methods

Study Sample

This study used data from the Health and Retirement Study (HRS), an ongoing biennial panel study of adults aged 51 and older in the U.S.¹⁴ The HRS began in 1992 with a nationally representative sample of adults aged 51–61 and, through the addition of subsequent birth cohorts, became nationally representative of the population aged 51 and older starting in 1998. The HRS was approved by the Institutional Review Board (IRB) at the University of Michigan; because this study used publicly available, deidentified data, it was exempt from a separate IRB review.

The current study employed 12 waves of data from 1998 to 2020. Of the 36,466 total respondents interviewed during this period, data from 35,082 respondents who met the age criteria and completed at least one non-proxy interview were included. Missing data on key variables were minimal, accounting for 0.35% of person-wave observations (from 74 individuals; see Appendix Table 1), which were excluded. Following earlier recommendations,¹⁵ respondents with only one observation who do not contribute to the key estimates were excluded (n=3,637). The final analytic sample consisted of 195,633 person-wave observations from 31,371 participants (6.2 observations/participant; see Appendix Figure 1 for a flow diagram).

Measures

At each wave, cognition was measured using the modified Telephone Interview for Cognitive Status (mTICS)¹⁶. The measure included episodic memory tests of immediate and delayed word recall (range=0–20), a working memory test of serial 7s subtraction (range 0–5), and a processing speed test of backward counting (range=0–2). The composite score was calculated as the sum of the cognitive tests (range=0–27) with higher scores indicating better functioning. The composite score was also used to classify probable dementia (0–6), cognitive impairment without dementia (CIND; 7–11), and normal cognitive functioning (12–27), based on concordance with neuropsychological assessments from the Aging, Demographics, and Memory Study ¹⁷. The m-TICS battery has demonstrated good criterion validity and classification accuracy relative to clinical assessments from ADAMS, with approximately 74% of respondents correctly classified across all three cognitive categories.¹⁷

Volunteering was measured in terms of both status and time commitment. Volunteering status was assessed with the question, “Have you spent any time in the past 12 months doing volunteer work for religious, educational, health-related or other charitable organizations?” Responses were coded dichotomously (1=yes; 0=no). Participants who answered yes were then asked how many hours they spent doing volunteer work. Based on the responses, time commitment was coded as: 0=did not volunteer, 1=1–99 hours (low commitment), 2=100–199 hours (moderate commitment), and 3=200+ hours (high commitment); this variable was treated as categorical in analyses to capture potential threshold effects.⁸

The analyses controlled for a set of time-varying covariates (TVC) identified in the literature, measured at each wave, that could confound the key estimates.¹⁰ These included age (linear, quadratic), marital status (1=married, 0=not married), household wealth inverse hyperbolic sine-transformed), work status (1=working for pay, 0=not working), activities of daily living limitations (range: 0–6), and chronic health conditions (range: 0–8). Because person fixed effects account for all time-invariant characteristics, birth cohort, gender, race/ethnicity, and education are reported descriptively only.

Statistical Analysis

To address the key study objective—examining the associations between volunteering and cognitive functioning across the population distribution of cognitive functioning—UQR methods are employed, which allow for investigation beyond the average effect. Conventional approaches based on general linear models, including fixed effects (FE) models, uncover how predictors are associated with the conditional mean of the outcome. For example, a FE model is expressed as:

$$y_{it}={\beta }_v{v}_{it}+{\gamma }_x{x}_{it}+{\alpha }_i+{\epsilon }_{it}$$

where $y_{it}$ represents cognitive functioning for individual $i$ at time $t,v_{it}$ denotes to the time-varying indicator of volunteering, and $x_{it}$ represents a vector of TVCs. Importantly, the TVCs include interactions between age (linear and quadratic) and birth cohort to account for cohort differences in cognitive decline.¹⁸ The person fixed effects, ${\alpha }_i$, absorb all time-invariant individual characteristics (e.g., cohort, gender, race/ethnicity, education), effectively using each individual as their own control; and ${\epsilon }_{it}$ is the error term. The association between volunteering and cognitive functioning is estimated by the coefficient ${\beta }_v$, which represents the average within-person change in cognitive functioning associated with volunteering, adjusted for time-varying covariates and net of all time-invariant person-level characteristics.

While such average effects are informative, they may obscure heterogeneity across the cognitive functioning distribution, particularly at levels most relevant for cognitive impairment and dementia. To investigate distributional heterogeneity, UQR models that estimate associations at specific points of the unconditional outcome distribution are employed.¹⁹ Here, the dependent variable is transformed using the recentered influence function (RIF):

$$RIF\left(y_i;q_{\tau }\right)=q_{\tau }+\frac{\tau -I\left(y_i\le q_{\tau }\right)}{f_Y\left(q_{\tau }\right)}$$

where $\tau$ denotes a specific quantile (e.g., 25^th), $q_{\tau }$ is the value of the outcome at that specific quantile, $I\left\{y\le q_{\tau }\right\}$ is an indicator function, and $f_Y\left(q_{\tau }\right)$ is the density of the outcome distribution at $q_{\tau }$.²⁰ The RIF-transformed outcome variable then replaces the outcome in the model structurally similar to the FE model:

$$RIF\left(y_{it};q_{\tau }\right)={\beta }_{v\tau }{v}_{it}+{\gamma }_{x\tau }{x}_{it}+{\alpha }_{i\tau }+{\epsilon }_{it\tau }$$

where ${\beta }_{v\tau }$ captures the marginal effect on the unconditional ${\tau }^{th}$ quantile of outcome resulting from changes in the distribution of volunteering ($v_{it}$). That is, ${\beta }_{v\tau }$ reflects how the observed distribution of cognitive functioning would change in the ${\tau }^{th}$ quantile due to changes in the share of volunteers in the population, which allows us to detect potential heterogeneity in the strength of the association at different levels of cognitive functioning.¹⁹ To address potential overcontrol from adjusting for TVCs measured concurrently with volunteering, supplementary analyses included TVCs lagged by one wave.²¹ Additional supplementary models further incorporated lagged cognitive functioning to account for potential feedback effects, whereby prior cognitive status may influence subsequent volunteering.²² Finally, an asymmetric fixed-effects approach was implemented to distinguish transitions into volunteering from transitions out of volunteering, allowing these changes to be temporally ordered with respect to subsequent cognitive functioning.²³

UQR models were estimated at key quantiles of the cognitive functioning distribution that correspond to established thresholds for cognitive status based on the mTICS scale. Specifically, the 2^nd percentile is within the score range indicative of probable dementia, the 5^th and 10^th percentiles correspond to CIND, and the 25^th, 50^th, 75^th, 90^th, and 95^th percentiles fall within the normal cognition range. All UQR models with individual fixed effects were estimated using the xtrifreg command in Stata (version 18).²⁴ For comparison, conventional fixed effects models were also estimated to describe the average effects. All models incorporated the sampling weights provided by the HRS, and robust standard errors were reported.

Results

The descriptive characteristics of the analytic sample (195,633 person-wave observations from 31,371 participants) are summarized in Table 1. On average, participants scored 15.87 (SD=4.43) out of a maximum of 27, indicating normal cognition. The full distribution of cognitive functioning, along with the key quantiles, is presented in Figure 1. While the majority of observations (84%) fell within the normal cognition range, approximately 2.9% were classified as probable dementia and 13.1% as CIND. Volunteering was reported in 35.8% of observations, with most participation occurring at a low level (1–99 annual hours; 20.8%), followed by moderate (100–199 hours; 8.1%) and high levels (200+ hours; 6.9%). To explore how volunteering varies across cognitive functioning, the prevalence of volunteering across of the cognitive functioning deciles were also examined (see Appendix Table 2 and Appendix Figure 2). Although rates of volunteering declined toward the lower end of the distribution, approximately 16% in the lowest decile still reported volunteering.

Table 1.

Weighted Descriptive Characteristics of the HRS Sample

Variable	M	(SD)	%
Time-varying characteristics (N = 195,633 person-wave observations)
Cognition: mTICS (0–27)	15.87	(4.43)
Volunteer status
Did not volunteer			64.2%
Volunteered			35.8%
Volunteer hours
1–99 hours			20.8%
100–199 hours			8.1%
200+ hours			6.9%
Age (in years; 51–109)	65.45	(9.78)
Married			61.6%
Working for pay			44.8%
Household wealth (IHS-transformed; −8.6–14.7)	5.05	(2.76)
ADL limitations (0–6)	0.30	(0.90)
Chronic health conditions (0–8)	1.92	(1.46)
Background characteristics (N = 31,371 persons)
Birth cohort (1892–1965)	1945.82	(13.60)
Female			53.8%
Race-ethnicity
White, non-Hispanic			75.0%
Black, non-Hispanic			11.0%
Other, non-Hispanic			4.5%
Hispanic			9.5%
Education (in years; 0–17)	12.96	(3.11)

Notes: ADL = activities of daily living. IHS = inverse hyperbolic sine. HRS = Health and Retirement Study. mTICS = modified Telephone Interview for Cognitive Status.

Figure 1.

Distribution of cognitive functioning scores based on the modified Telephone Interview for Cognitive Status (mTICS) from the Health and Retirement Study (HRS; 1998–2020). The histogram is based on weighted data from 195,633 person-wave observations. The mTICS scores corresponding to the 2^nd, 5^th, 10^th, 25^th, 50^th (median), 75^th, 90^th, and 95^th percentiles were 6.0, 8.3, 10.3, 13.6, 16.7, 19.4, 21.6, and 23.0, respectively. The sample mean was 15.8 (dashed vertical line). Shaded regions represent cognitive status categories based on mTICS cutpoints from Crimmins et al. (2011): probable dementia (0–6), cognitive impairment without dementia (CIND; 7–11), and normal cognition (12–27).

Results from models examining the linkages between volunteer status and cognitive functioning are presented in Table 2 (see also Panel 1, Figure 2). The first column presents estimates from FE models, where volunteering was associated with a higher level of cognitive functioning, adjusted for the full set of covariates (b=0.17; p<.001). However, UQR models revealed substantial heterogeneity in the strength of the association across the cognitive functioning distribution. Specifically, the estimated effect was largest at the 2^nd percentile (b=0.32), which is within the range suggestive of probable dementia, followed by the 5^th (b=0.24) and 10^th (b=0.22) percentiles, which fall within the range suggestive of CIND. These effects were all statistically significant (p<.001). Associations attenuated steadily across higher quantiles and became small and statistically nonsignificant at the upper end of the distribution. This pattern indicates heterogeneous distributional effects, with the strongest associations observed among individuals with CIND or probable dementia.

Table 2.

Weighted Estimates Linking Volunteering Status and Cognitive Functioning Across the Cognitive Functioning Distribution

	FE				UQR-FE
Variable	M	2nd	5th	10th	25th	50th	75th	90th	95th
Volunteered	0.17*** [0.12,0.22]	0.32*** [0.18,0.46]	0.24*** [0.13,0.35]	0.22*** [0.13,0.32]	0.22*** [0.14,0.30]	0.16*** [0.09,0.22]	0.08* [0.00,0.16]	0.08 [−0.03,0.18]	0.04 [−0.12,0.20]
Linear age	−0.08*** [−0.09,−0.08]	−0.05*** [−0.07,−0.03]	−0.07*** [−0.08,−0.06]	−0.07*** [−0.08,−0.06]	−0.08*** [0.09,−0.07]	−0.08*** [−0.09,−0.08]	−0.08*** [−0.09,−0.07]	−0.07*** [−0.08,−0.06]	−0.06*** [−0.07,−0.04]
× Birth cohort a	1.08*** [0.96,1.20]	0.34** [0.09,0.59]	0.51*** [0.32,0.70]	0.66*** [0.51,0.82]	0.76*** [0.63,0.88]	0.81*** [0.70,0.91]	1.12*** [1.00,1.24]	1.59*** [1.42,1.75]	2.10*** [1.85,2.35]
Quadratic age a	0.03 [−0.04,0.09]	−0.40*** [−0.57,−0.23]	−0.33*** [−0.46,−0.21]	−0.27*** [−0.38,−0.17]	−0.16*** [0.24,−0.08]	−0.03 [−0.10,0.03]	0.17*** [0.10,0.25]	0.33*** [0.23,0.43]	0.50*** [0.36,0.65]
× Birth cohort a	0.17 [−0.04,0.39]	5.95*** [5.08,6.83]	4.42*** [3.83,5.01]	3.35*** [2.93,3.78]	1.81*** [1.51,2.11]	−0.14 [−0.37,0.09]	−1.50*** [−1.75,−1.25]	−2.51*** [−2.85,−2.18]	−3.26*** [−3.75,−2.77]
Married	0.03 [−0.05,0.10]	−0.01 [−0.25,0.25]	0.11 [−0.09,0.32]	0.09 [−0.07,0.24]	0.13* [0.01,0.26]	0.06 [−0.04,0.16]	−0.01 [−0.12,0.11]	−0.06 [−0.21,0.08]	−0.2 [−0.42,0.03]
Working for pay	0.10*** [0.04,0.15]	−0.08 [−0.22,0.07]	0.00 [−0.12,0.12]	0.09 [−0.01,0.19]	0.09* [0.01,0.18]	0.12** [0.04,0.19]	0.08 [−0.00,0.17]	0.02 [−0.10,0.13]	0.09 [−0.09,0.26]
Household wealth b	0.02*** [0.01,0.04]	0.08*** [0.04,0.12]	0.06*** [0.03,0.09]	0.07*** [0.04,0.09]	0.03** [0.01,0.05]	0.01 [−0.01,0.02]	0.00 [−0.01,0.02]	0.00 [−0.02,0.03]	0.04* [0.01,0.07]
ADL limitations	−0.21*** [−0.25,−0.18]	−0.74*** [−0.90,−0.59]	−0.61*** [−0.71,−0.50]	−0.47*** [−0.54,−0.39]	−0.25*** [0.30,−0.20]	−0.15*** [−0.18,−0.11]	−0.08*** [−0.11,−0.04]	−0.03 [−0.06,0.01]	0.01 [−0.05,0.06]
Chronic health conditions	−0.10*** [−0.13,−0.06]	−0.13* [−0.24,−0.03]	−0.11** [−0.19,−0.04]	−0.12*** [−0.19,−0.06]	−0.11*** [0.16,−0.06]	−0.09*** [−0.13,−0.05]	−0.05* [−0.09,−0.01]	−0.06 [−0.11,0.00]	−0.09* [−0.17,−0.00]

Notes: Based on 195,633 person-wave observations collected from 31,371 participants (unweighted). Robust standard errors presented in parentheses. ADL = activities of daily living. FE = fixed effects. UQR = unconditional quantile regression.

^aMultiplied by 100 to reduce leading zeros; age and birth cohort centered at 65 and 1945, respectively.

^bIn $1,000, transformed by inverse hyperbolic sine.

Boldface indicates statistical significance (p < 0.05).

^*p < 0.05.

^**p < 0.01.

^***p < 0.001.

Figure 2.

Cognitive benefits of volunteering across the cognitive functioning distribution. Panel 1 displays fully adjusted estimates from unconditional quantile regression models (Table 2), showing changes in cognitive functioning associated with volunteering status. The dashed horizontal line indicates the mean fixed effects estimate (b=0.17) for reference. Panel 2 presents analogous estimates from models examining volunteering time commitment (Appendix Table 3). Estimates for the 95th percentile were not statistically significant and are therefore not shown.

The subsequent models focused on volunteering hours (Panel 2, Figure 2; see also Appendix Table 4). Results from the FE models indicated a clear dose-response relationship: greater volunteering time was associated with better cognitive functioning, with the strongest effect observed among those who volunteered 200 + annual hours (approximately 4+ weekly hours). However, UQR models revealed substantial heterogeneity in these associations across cognitive functioning distribution. In the lower percentiles reflecting probable dementia (2^nd) and CIND (5^th and 10^th), all levels of volunteering were linked to significantly higher cognitive functioning, but moderate engagement (100–199 annual hours, approximately 2–4 weekly hours) was associated with the largest cognitive gains. In contrast, this pattern did not hold in the normal cognition range. At the 25th and 50th percentiles, estimates generally increased in magnitude with greater volunteering hours, but in the higher percentiles (75^th and 90^th), only high-level volunteering (200+ hours) was significantly associated with better cognitive functioning. Volunteering hours categories were unrelated to cognitive functioning in the 95^th percentile.

Finally, supplementary analyses using lagged TVCs (Appendix Tables 4, 5), and models additionally including lagged cognitive functioning (Appendix Tables 6, 7), showed patterns consistent with the main findings. Notably, the estimated coefficients were consistently larger, suggesting that the cognitive benefits reported here are conservative. Supplementary models based on an asymmetric fixed effects approach (Appendix Table 8) showed that transitions into volunteering were associated with higher subsequent cognitive functioning in a distributionally heterogeneous manner, reinforcing the main findings.

Discussion

Using large, nationally representative panel data and UQR models, this study investigated how the association between volunteering and cognitive functioning varies across the population cognitive functioning distribution. The findings provide strong evidence that the cognitive benefits of volunteering are heterogeneous across the population. Specifically, UQR models revealed that the cognitive benefits of volunteering were most pronounced at the lower end of the cognitive functioning distribution among individuals likely experiencing CIND or probable dementia and diminished steadily across higher quantiles.

The findings suggest that earlier research on the topic, much of which relies on average effects, may underestimate the cognitive benefits of volunteering among individuals with suboptimal cognitive functioning, while overestimating them for those at the higher end of the cognitive distribution (see Figure 2). Although the mechanisms underlying this pattern of heterogeneity remain to be investigated, several plausible explanations warrant consideration. A promising framework for understanding the distributional effects is the caregiving system model,^25,26 which posits that the neurobiological mechanisms that guide prosocial helping behaviors can also yield downstream physiological health benefits. According to this model, engaging in helping behaviors can downregulate stress-response systems, such as the hypothalamic-pituitary-adrenal axis, thereby reducing chronic stress and inflammation, which are well-established contributors to cognitive decline.²⁷ These mechanisms may be especially relevant for individuals with compromised cognitive functioning, who are more likely to experience physiological dysregulation and psychosocial strain.²⁸ Supporting this view, research has shown that volunteering buffers against cortisol dysregulation and systemic inflammation, particularly among older adults facing chronic stress or functional impairment.^29,30 Additionally, volunteering has been linked to improvements in cardiovascular risk factors and outcomes, which contribute to cognitive aging.^31,32 These protective effects may be especially consequential for individuals with poor cognitive health, who may derive greater benefits from volunteering through these physiological pathways. In contrast, individuals with higher cognitive functioning—who are more likely to have better underlying cardiovascular and systemic health—may experience diminishing returns from volunteering.³³

In addition, psychosocial pathways may also help explain the stronger cognitive benefits of volunteering among individuals with compromised cognitive functioning. Volunteering is one of the few socially valued roles available in later life that offers structure, purpose, and a sense of contribution.^34,35 For individuals living with CIND or early-stage dementia, meaningful engagement opportunities are often limited, making volunteering especially impactful.³⁶ It is also possible that despite the lack of immediate benefits, volunteering contributes to cognitive reserve and delay the onset of cognitive impairment or dementia over the longer term.³⁷

The analyses on time commitment offer additional insights that refine the understanding of how volunteering may promote cognitive health. At the lower end of the cognitive functioning distribution—corresponding to probable dementia and CIND—any level of volunteering was associated with better cognitive outcomes, with moderate-level engagement (100–199 annual hours) yielding the greatest benefit. These findings are consistent with prior research showing that structured yet manageable prosocial activity can have protective effects for individuals with cognitive limitations. For instance, a recent review documented that individuals with cognitive impairment often engage in volunteering activities similar in type to those of cognitively healthy peers, particularly service-oriented roles such as intergenerational programs involving reading with children, or helping others through tasks like donating food or clothing.⁹ Importantly, evidence suggests that successful engagement among individuals with cognitive impairment depends less on the specific volunteering domain and more on appropriate task matching, flexibility, and support.⁹ Taken together, the current findings suggest that engaging in such volunteering activities at moderate levels may constitute a sustainable intervention model for older adults with suboptimal cognitive functioning. Thoughtful approaches for recruiting individuals with cognitive limitations—such as designing flexible roles and matching tasks to individual strengths—will be essential for maximizing participation and benefits.

Limitations

First, due to data limitations in the HRS, the different types of volunteering were not considered, which may have distinct effects on cognitive outcomes.¹⁰ Future research should explore whether specific volunteering contexts contribute to heterogeneous outcomes. Second, although the mTICS is a well-validated screening tool for measuring cognitive functioning and identifying CIND and probable dementia, it has limited sensitivity at the upper end of the cognitive functioning distribution,³⁸ which may have constrained the ability to detect nuanced effects among the most cognitively fit individuals. Finally, despite adjusting for a comprehensive set of time-varying confounders and person fixed effects, the estimates remain subject to omitted variable bias, and causal interpretations should be made with caution.

Conclusions

This study contributes to research on volunteering as a viable lifestyle-based strategy to promote cognitive health by employing UQR models to move beyond average effects and uncover distributional effects of volunteering on cognitive functioning. Using two decades of data from a nationally representative sample of older adults, the findings highlight that individuals with suboptimal cognition may stand to benefit the most from volunteering. In this regard, prescribing a moderate level of volunteering may serve as an effective component of lifestyle interventions aimed at improving the lives of those at risk for, or experiencing, cognitive impairment and dementia.

Highlight.

• Volunteering is known to confer cognitive benefits for older adults

• Little is known about whether these benefits vary across levels of cognitive functioning

• Unconditional quantile regression models were used to uncover distributional effects

• Volunteering was most beneficial for those in suboptimal cognitive functioning

• Approximately 2 to 4 weekly volunteering yielded greatest benefits in this population