Results

Because all children lack self-control on occasion, we defined a child’s sustained self-control style using an omnibus measure of self-control that comprised reports collected at ages 3, 5, 7, 9, and 11 y. These reports by researcher-observers, parents, teachers, and the children themselves assessed capacities including lack of control, impulsive aggression, hyperactivity, lack of persistence, inattention, and impulsivity (SI Appendix, SI Methods 1). They were combined into a highly reliable composite measure for each study child [α = 0.86 (2)]. Children with better self-control tended to come from more socioeconomically advantaged families (r = 0.27, P < 0.0001) and had higher tested IQs (r = 0.45, P < 0.0001).

Does Better Self-Control in Childhood Forecast Slower Aging of the Body and Fewer Signs of Brain Aging?

From ages 26 to 45 y, we measured the pace of study members’ physiological decline across multiple organ systems. At age 45 y, we also collected structural MRI measures to derive estimates of brain aging as well as the volume of white matter hyperintensities, a clinical index of microlesions that accrue across the life span and predict accelerated cognitive decline and dementia risk. In addition, we conducted assessments of study members’ functional capacity and asked several independent raters to judge each study member’s apparent age from facial photographs (Fig. 1 and SI Appendix, SI Methods 2). These outcomes were correlated with each other (Table 1) and were therefore combined to form a composite measure of accelerated aging using principal components analysis. Children with better self-control displayed slower aging in later life, as assessed by this composite (β = −0.35, 95% CI [−0.42, −0.29], P < 0.0001; Table 2 and Fig. 2). This remained the case even after controlling for their social class origins and IQ (β = −0.20 [−0.26, −0.13], P < 0.0001; Table 2). Self-control was also associated with each constituent measure individually. As adults, children with better self-control aged more slowly across different organ systems, had lower brain age scores, had a smaller volume of white matter hyperintensities, walked more quickly, and appeared younger in facial photographs shown to independent raters (Table 2). Associations with brain age and white matter hyperintensities became nonsignificant when we controlled for childhood social class and childhood IQ, but the remainder of associations were independent of these fundamental causes (Table 2).

Table 1.

Correlations among childhood predictors and aging outcomes assessed in the Dunedin Study

	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22
1. Childhood self-control
2. Childhood IQ	0.45
3. Childhood social class	0.27	0.41
4. Accelerated aging PC	−0.34	−0.40	−0.31
5. Pace of aging*	−0.29	−0.29	−0.24	0.75
6. BrainAGE†	−0.12	−0.17	−0.11	0.48	0.20
7. White matter hyperintensities (mm3)‡	−0.10	−0.09	−0.10	0.42	0.18	0.14
8. Gait speed (m/s)	0.25	0.34	0.21	−0.64	−0.34	−0.11	−0.09
9. Facial age	−0.20	−0.25	−0.24	0.66	0.33	0.16	0.11	−0.25
10. Health preparedness PC	0.30	0.33	0.28	−0.44	−0.37	−0.14	−0.15	0.29	−0.32
11. Practical health knowledge	0.34	0.50	0.33	−0.38	−0.28	−0.16	−0.11	0.24	−0.32	0.55
12. Pessimism toward aging	−0.16	−0.13	−0.12	0.25	0.22	0.07	0.13	−0.16	0.16	−0.72	−0.12
13. Self-predicted life expectancy	0.15	0.12	0.15	−0.30	−0.27	−0.07	−0.06	0.21	−0.21	0.75	0.16	−0.28
14. Financial preparedness PC	0.31	0.37	0.26	−0.45	−0.40	−0.20	−0.08	0.25	−0.34	0.40	0.35	−0.28	0.21
15. Practical financial knowledge	0.39	0.60	0.36	−0.44	−0.37	−0.19	−0.08	0.29	−0.31	0.38	0.53	−0.17	0.15	0.66
16. Financial planfulness	0.20	0.20	0.17	−0.31	−0.28	−0.18	−0.04	0.15	−0.23	0.29	0.22	−0.22	0.17	0.75	0.39
17. Credit scores	0.15	0.13	0.17	−0.28	−0.25	−0.11	−0.09	0.13	−0.21	0.22	0.15	−0.17	0.12	0.67	0.23	0.31
18. Informant-reported financial problems	−0.15	−0.13	−0.04	0.24	0.21	0.08	0.02	−0.13	0.19	−0.23	−0.09	0.22	−0.14	−0.70	−0.24	−0.34	−0.35
19. Social preparedness PC	0.18	0.13	0.10	−0.21	−0.21	−0.07	−0.05	0.13	−0.14	0.48	0.12	−0.54	0.27	0.34	0.17	0.26	0.24	−0.29
20. Social support	0.15	0.09	0.07	−0.16	−0.16	−0.03	−0.04	0.11	−0.09	0.36	0.10	−0.37	0.23	0.22	0.10	0.16	0.17	−0.19	0.80
21. Loneliness	−0.11	−0.09	−0.06	0.15	0.13	0.08	0.02	−0.10	0.10	−0.35	−0.08	0.41	−0.17	−0.24	−0.13	−0.20	−0.15	0.21	−0.82	−0.48
22. Life satisfaction	0.17	0.13	0.10	−0.22	−0.22	−0.08	−0.05	0.11	−0.15	0.48	0.12	−0.54	0.27	0.37	0.18	0.29	0.27	−0.30	0.83	0.49	−0.54

Correlations were estimated controlling for sex. Estimates with absolute values of 0.09 or greater are statistically significant at P < 0.01 (with the exception of the relations between white matter hyperintensities and gait speed [P = 0.013] and between white matter hyperintensities and credit scores [P = 0.011]). Cells shaded in grey indicate correlations with our three predictors (childhood self-control, childhood IQ, and childhood social class). Cells shaded in blue, green, yellow, and red indicate correlations among the variables in the four aging domains. Bolded estimates in columns 1 to 3 indicate correlations between our three predictors and the principal components. Bolded estimates in columns 4 to 21 indicate the correlations among the variables within each aging domain. PC = first principal component. This table shows the correlations among the childhood predictors and aging outcomes of primary interest in the current study. Correlations with additional study variables (participant sex and adult self-control) are provided in SI Appendix, Table S4.

* Years of physiological change per chronological year.

^† BrainAGE is the difference between participants’ predicted age from MRI data and their exact chronological age.

^‡ Measure was natural log-transformed.

Table 2.

Do children with better self-control age more slowly and exhibit better preparedness for later-life health, financial, and social demands?

	Baseline associations		Adjusted for childhood social class and IQ*
Outcome	β (95% CI)	P value	β (95% CI)	P value
Accelerated aging
Accelerated aging PC (z-score)	−0.35 (−0.42, −0.29)	<0.0001	−0.20 (−0.26, −0.13)	<0.0001
Pace of aging†	−0.30 (−0.36, −0.24)	<0.0001	−0.20 (−0.26, −0.13)	<0.0001
BrainAGE‡	−0.13 (−0.20, −0.06)	<0.001	−0.06 (−0.13, 0.02)	0.134
White matter hyperintensities (mm3)§	−0.11 (−0.17, −0.04)	0.003	−0.07 (−0.15, 0.01)	0.073
Gait speed (m/s)	0.26 (0.20, 0.33)	<0.0001	0.13 (0.06, 0.20)	<0.001
Facial age	−0.21 (−0.28, −0.14)	<0.0001	−0.10 (−0.18, −0.03)	0.005
Health preparedness
Health preparedness PC (z-score)	0.30 (0.24, 0.37)	<0.0001	0.18 (0.11, 0.25)	<0.0001
Practical health knowledge	0.34 (0.28, 0.41)	<0.0001	0.13 (0.07, 0.20)	<0.0001
Pessimism toward aging	−0.16 (−0.23, −0.10)	<0.0001	−0.12 (−0.20, −0.05)	0.001
Self-predicted life expectancy	0.15 (0.09, 0.22)	<0.0001	0.11 (0.04, 0.19)	0.003
Financial preparedness
Financial preparedness PC (z-score)	0.32 (0.26, 0.39)	<0.0001	0.18 (0.11, 0.25)	<0.0001
Practical financial knowledge	0.40 (0.34, 0.46)	<0.0001	0.15 (0.09, 0.21)	<0.0001
Financial planfulness	0.21 (0.14, 0.27)	<0.0001	0.14 (0.06, 0.21)	<0.001
Credit scores	0.15 (0.09, 0.22)	<0.0001	0.11 (0.03, 0.18)	0.005
Informant-reported financial problems	−0.15 (−0.22, −0.09)	<0.0001	−0.12 (−0.19, −0.04)	0.002
Social preparedness
Social preparedness PC (z-score)	0.18 (0.12, 0.25)	<0.0001	0.15 (0.08, 0.23)	<0.0001
Social support	0.16 (0.09, 0.23)	<0.0001	0.14 (0.07, 0.22)	<0.001
Loneliness	−0.12 (−0.18, −0.05)	0.001	−0.09 (−0.17, −0.02)	0.016
Life satisfaction	0.18 (0.11, 0.24)	<0.0001	0.14 (0.07, 0.21)	<0.001

All models combined men and women and controlled for sex. Supplementary analyses stratified by sex showed that self-control forecast aging outcomes in both men and women (SI Appendix, Table S5). β = standardized linear regression coefficient, CI = confidence interval, PC = first principal component.

* In secondary analyses suggested through peer review, we tested whether there were interactions between childhood self-control and childhood social class and between childhood self-control and childhood IQ in models predicting the four aging principal components. No interactions survived correction for multiple testing.

^† Years of physiological change per chronological year.

^‡ BrainAGE is the difference between participants’ predicted age from MRI data and their exact chronological age.

^§ Measure was natural log-transformed.

Fig. 2.

Childhood predictors and midlife aging. Children with better self-control, who came from more socioeconomically advantaged backgrounds, and with higher IQs aged more slowly (A) and were more prepared to manage later-life health (B), financial (C), and social (D) demands. The means are adjusted for sex.

Do Individuals Change in Their Self-Control across Age?

Children’s self-control forecast their pace of aging as well as their aging preparedness in adulthood. Is self-control a stable predictor of midlife outcomes, or do some children change in their self-control over time? We created a composite measure of study members’ self-control in adulthood using measures of self-control that were administered to multiple raters when the study members were 38 and 45 y of age, including close informants and Dunedin Study personnel [to parallel the multi-occasion, multi-informant strategy of our childhood self-control measure (Methods)]. Self-control in childhood was moderately correlated with self-control in midlife (r = 0.33, P < 0.0001; Fig. 3), indicating that some study members shifted in their rank in level of self-control across age. To contextualize this degree of child-to-adult stability, we calculated two benchmarks in our sample: height, for which individuals’ rank order is highly stable across age (r = 0.77, P < 0.0001; Fig. 3), and accidental injuries, which are, by definition, highly unstable (r = 0.07, P = 0.03; Fig. 3). The degree of child-to-adult stability in self-control fell in between these two benchmarks. This was also the case for children’s social class (r = 0.36, P < 0.0001; Fig. 3). By contrast, IQ was as stable across age as height (r = 0.77, P < 0.0001; Fig. 3). Although our findings are observational, the natural history change in self-control that we observed suggests that self-control might also be subject to intervention-induced change. Furthermore, whereas children’s IQ may be highly difficult to change, there is the possibility that their self-control may offer a more malleable target. A caveat is that our adult measure of self-control was not identical to our childhood measure. However, both measures had good reliability and validity and it is unlikely that measurement factors alone can explain the lower stability in this estimate relative to IQ. Supporting the potential benefits of self-control change, individuals who became more self-controlled from childhood to midlife also aged slower and had better aging preparedness at age 45 y (SI Appendix, Fig. S1). Some individuals became less self-controlled across age; they aged faster and had poorer aging preparedness (SI Appendix, Fig. S1).

Fig. 3.

Child-to-adult stability in self-control, social class, and IQ. The figure shows that some Dunedin Study members shifted naturally in their self-control from childhood to adulthood. The degree of stability in self-control was similar to that for social class and substantially lower than that for IQ. Also shown are benchmarks for low (accidental injuries) and high (height) developmental stability, computed within this analytic sample. Child and adult accidental injuries were measured as annual accidental injury rates from birth to age 11 y and from 38 to 45 y, respectively. Child and adult height in millimeters were measured as a composite (standardized within sex) from ages 5 to 11 y and at age 45 y, respectively. Correlations are adjusted for sex. The bars are 95% CIs.

Given that individuals change in their self-control across age, midlife might represent another period during which to build aging preparedness. However, are both adult and childhood self-control important for midlife aging, independent of each other? Participants’ self-control in adulthood was associated with their pace of aging and their health, financial, and social preparedness, even after accounting for their self-control in childhood (βs [absolute values] = 0.33 to 0.54, Ps < 0.0001; SI Appendix, Table S2). Participants’ self-control in childhood also predicted their pace of aging, health preparedness, and financial preparedness after accounting for their self-control in adulthood (βs [absolute values] = 0.14 to 0.22, Ps < 0.0001). It did not independently predict their social preparedness (β = 0.06, P = 0.091; SI Appendix, Table S2), suggesting that associations between childhood self-control and midlife social integration are largely accounted for by individuals’ contemporaneous behavior.

Discussion

Higher self-control is known to forecast longevity (567–8). Our findings suggest that youth with better self-control do not just live longer, but also age more slowly as adults and are better prepared by midlife to manage the aging process. In this five decade prospective study of a population-representative birth cohort, children who exercised better control over their emotions, thoughts, and behaviors later experienced less age-related decline in their bodies, showed fewer signs of brain aging, were more attentive to practical health and financial information, were more consistent in implementing positive health and financial behaviors, exhibited more positive attitudes toward and expectancies about aging, were more socially integrated, and were more satisfied with life. For nearly all associations, the effects of children’s self-control could be disentangled from their socioeconomic origins and intelligence, suggesting the possibility that self-control itself might be an active ingredient in healthy aging.

Our analysis is strengthened by four design features. First, the Dunedin Study has a high retention rate (94%), reducing the potential for attrition bias [and the small amount of study attrition was not related to childhood self-control (SI Appendix, SI Methods 3)]. Second, we employed a rigorous measurement strategy for children’s self-control (reports from multiple informants across multiple occasions), social class (standardized assessments across multiple occasions), and intelligence (full-scale IQ assessments across multiple occasions). Third, we assessed participants’ aging at a period when individuals are still relatively healthy, at midlife, increasing the potential for our findings to inform preventive interventions. Fourth, associations with self-control were present across a range of aging outcomes assessed using different methods.

We acknowledge limitations. First, our data were right-censored at midlife. We therefore could not confirm that cohort members with better health, financial, and social preparedness at midlife will also exhibit better aging in later life [although the indicators of preparedness that we selected have been previously implicated in life span and/or health span (Fig. 1 and SI Appendix, Table S1)]. We also could not test whether building aging preparedness during midlife impacts longevity. Second, our findings are limited to a cohort of individuals born in New Zealand in the 1970s who are primarily white. However, associations between self-control and longevity have been observed across samples from different countries, born at different times, and of different ethnicities (567–8) (although importantly, these samples are from Western countries, and more research is needed in non-Western populations). Third, we utilized an omnibus measure of self-control rather than focusing on specific self-control capacities. Intelligence comprises different verbal, visuoperceptual, working memory, and processing speed abilities, and social class summarizes income, education, and occupational prestige. Similarly, researchers have argued that self-control comprises multiple abilities with diverse neural underpinnings, and different disciplines conceptualize and measure self-control in different ways using different methods [e.g., as conscientiousness, impulsivity, effortful control, ego strength, willpower, self-discipline, self-regulation, delay of gratification, executive function, and temporal discounting (34353637383940–41)]. Future research should aim to determine whether different self-control abilities and concepts relate to aging outcomes. Fourth, our measure of adult self-control was based on brief personality scales rather than comprehensive inventories (but it was strengthened by the use of reports from multiple raters across multiple occasions). Fifth, although our measure of physiological aging was based on longitudinal data, our measure of brain age was derived from imaging data collected at one time point. Prospective imaging data are needed to measure within-subject changes in brain age. Sixth, our observational study can only identify better childhood self-control as an indicator of better adult aging, not necessarily an indicator of causation.

With these limitations in mind, several implications can be noted. First, our results suggest the hypothesis that raising self-control in childhood could promote both life span and health span. Although our observational analysis did not incorporate an intervention component, we found that participants shifted in their rank order of self-control across age (and that participants who became more self-controlled also had better outcomes), suggesting that self-control might be subject to change and a malleable target for intervention. This conclusion is bolstered by data supporting the effectiveness of self-control– and self-regulation–focused interventions (42434445–46) [although some studies have obtained smaller effect sizes or found that the magnitude of effects varies depending on intervention type and design features (46, 47)].

Second, our findings have implications for self-control measurement and theory. Renewed debate over the importance of early-life self-control has emerged from efforts to replicate associations between children’s delay of gratification [as measured by the marshmallow test (48)] and their academic and behavioral outcomes, with researchers reporting smaller associations when employing more representative samples and after controlling for social class and cognitive ability (49). With respect to measurement, this raises the question of whether alternative approaches to assessing self-control (like the multi-occasion/multi-informant approach we used) might yield stronger associations with life outcomes than a single behavioral task. Laboratory tasks provide important information about self-control processes, but findings are mixed concerning how well they predict behavior in the real world (3, 5051–52). Measuring multiple self-control behaviors to ascertain a child’s style of self-control across situations, reporters, and years may capture the broader range of capacities that comprise this umbrella construct and that are important for real-world outcomes. Our multi-occasion/multi-informant approach to assessing children’s self-control yielded robust associations with aging outcomes measured over three decades later.

With respect to theory, results from our study and others (49) raise the question of whether self-control should be conceptualized as a construct that is distinct from socioeconomic origins and intelligence. It is unclear whether these latter constructs should be viewed as “confounds” to be partialled out of self-control associations or, rather, whether they represent important contributing factors to aspects of self-control (53). Self-control measures have been found to predict success in certain life domains as well as or better than measures of socioeconomic origins (2, 54) and intelligence (2, 27, 5455–56), and some early-childhood programs that improve self-control show positive effects on academic, occupational, and health outcomes even without improving intelligence (or socioeconomic status) (17, 57). Self-control continued to predict nearly all aging outcomes in our study after accounting for socioeconomic origins and intelligence, which is rather remarkable given the degree to which it was correlated with these factors (rs = 0.27 and 0.45, Ps < 0.0001; Table 1). (Exceptions were associations with MRI measures of brain age and white matter hyperintensities, which were attenuated to nonsignificance after controls for childhood socioeconomic status and IQ. This suggests that associations between self-control and neurological aging may be largely explained by socioeconomic origins and intelligence, or it might reflect that effects of self-control on neurological aging—as conferred via an accumulation of environmental factors—are not yet apparent in midlife.) Taken together with meta-analytic data supporting the prospective value of self-control for life success (3), our findings raise an additional question for self-control theory: whether self-control should be considered a “fundamental cause” of children’s later-life outcomes alongside their socioeconomic origins and intelligence (3132–33).

Third, our findings suggest the hypothesis that midlife might be a propitious time at which to revisit the opportunity to promote healthy aging (15, 16). Much work has focused on the importance of intervening in childhood and adolescence, but comparably little attention has been paid to the potential benefits of intervening later in the life course. We found that Dunedin Study members differed in their pace of aging as well as their health, financial, and social preparedness at midlife, indicating that meaningful variation in aging can be detected already at this period. Moreover, individuals’ self-control in midlife was associated with their pace of aging and aging preparedness even after accounting for their self-control in childhood, suggesting that building self-control in midlife—while adults are still relatively healthy—might confer unique benefits for life span and health span. We measured participants’ aging when they were in their mid-forties, but age-related diseases do not typically onset until the sixties in developed nations (58, 59), indicating a 15- to 20-y window for intervention. There is evidence to support the potential utility of intervening at midlife. Smokers who quit in midlife live longer (60, 61), and individuals who build better cardiovascular fitness in midlife live longer and are less likely to develop dementia (62, 63). There is also evidence that positive midlife practices [e.g., use of adaptive coping strategies to manage life stressors (64)] can interrupt the association between childhood adversities and negative long-term health outcomes (15). Furthermore, the population tends to improve in self-control across age (65), and even individuals who were at the low end of the self-control distribution as children may thus be better positioned as adults to implement the practices necessary to stem age-related decline. A key area for future research is to develop, test, and measure the returns on midlife interventions to enhance aging preparedness. Such approaches could range from individual-level therapies [e.g., cognitive-behavioral and transdiagnostic approaches for emotion regulation (44)] to organizational nudge tactics [e.g., opt-out retirement savings plans (66, 67)]. Programs already developed to assist adults in preparing for age-related demands in one domain might be expanded to adopt a “whole-person approach”; for instance, retirement planning programs could incorporate health and social planning in addition to financial planning. Failing to capitalize on these potential opportunities for intervention could have implications not only for individual well-being, but also for economic and health policy. Premature retirement places a burden on pension financing (68), and age-related diseases are incurring growing healthcare costs (69).

Lastly, our findings identify both opportunities and challenges for future aging research. With respect to opportunities, an open question is how childhood self-control impacts individuals’ psychological readiness for old age. Here, we have shown that children with better self-control reach midlife better prepared to handle the demands of later life. Do these individuals also feel subjectively more equipped to navigate the aging process? Our participants who had better self-control as children expressed more positive views of aging and felt more satisfied with life. Answering this question in greater depth could inform our understanding of the cognitive and emotional mechanisms that connect early self-control skills to midlife aging preparedness. A second open question is the potential magnitude of returns on investment in self-control interventions (e.g., number of healthy life-years gained). This can only be ascertained with prospective data that extend into old age. A key challenge for such work will be how to retain participants with the poorest self-control for follow-up. Nonresponse to national surveys has been growing [although the problem is worse for cross-sectional than longitudinal surveys (70)]. In order to obtain unbiased estimates of the returns on self-control interventions, researchers will need to attend carefully to the design and implementation strategies necessary to retain their least conscientious and most rapidly aging participants.

Societal changes have amplified the role of self-control in preparing for later-life health, financial, and social demands: more healthcare providers emphasize patient choice, more jobs are sedentary, more high-fat fast foods are available, more online advertising tempts poor money management, more individuals manage their own retirement savings, and more adults live alone. These changes present new opportunities for prevention and intervention scientists. If the associations documented here are causal, programs that increase self-control might improve not just length of life, but also quality of life: the ability to progress through old age in good health, with financial security, and with strong social bonds.

Methods