OUP user menu

A model of successful ageing in British populations

Yvonne G. Doyle, Martin Mc Kee, Martyn Sherriff
DOI: http://dx.doi.org/10.1093/eurpub/ckq132 71-76 First published online: 29 September 2010


Background: European populations are ageing rapidly yet, although, it is widely recognized that some individuals age more successfully than others, an agreed concept of successful ageing remains elusive. We sought to develop a model of successful ageing in a British population, which combines the subjective and objective elements of successful ageing that have previously been proposed. Methods: Structural equation modelling was used to provide a model of successful ageing, defined in subjective and objective terms. The data comprised 15 000 subjects aged ≥50 years in England, recruited to the Health and Lifestyle Surveys (HALS) (followed up >7 years) and the English Longitudinal Survey of Ageing (ELSA) (followed up >2 years). A model was developed using a 50% random sample from HALS and tested in the other 50% and in ELSA. We examined the association of risk of disease, functioning and engagement with society at baseline with confidence and continued engagement at follow-up. Results: The model developed in the 50% sample of HALS was strong and reproducible in the other data sets. Low risk and good function at baseline are associated with confidence and continued engagement at follow-up, with engagement at baseline independently associated with engagement at follow-up. Conclusions: This model provides reproducible insights on the nature of ageing well, confirming the importance of personal resilience and continued involvement in physical and social activities. Far from retiring, engagement with life and society should be the norm for ageing populations.


In the developed world, the population aged between 60 and 74 years will more than double between 1950 and 2050, whereas the numbers aged >75 years will increase by over five times in this period.1 An important question is whether projected gains in longevity in later life will be accompanied by increases in disability, vulnerability and service use2 and what a good ageing process might entail. The term ‘successful ageing’, first used >40 years ago, mainly described how to reduce illness in old age.3 Subsequent definitions of success varied and an agreed definition of the concept remains elusive.4,5 Criteria have included physical and cognitive performance, dying without disability, life satisfaction and societal benefits such as low demand on services.6,7

There have been attempts to reconcile these criteria with older people’s own aspirations.8 When ageing is defined in terms of social interaction and feelings of resilience, a much higher proportion of older people express satisfaction with the process than when it is defined purely as physical functioning.9 People place a high value on the achievement of social goals,10 frequently related to family and leisure or learning activities. Thus, 45–50% of older people express satisfaction with their ageing process even though, on objective measures, they function poorly. Here, the concept of ‘possible selves’, defined as an individual’s ideas of what they might become, what they would like to become11 and what they are afraid of becoming, may be relevant in building resilience.

We examined two sentinel definitions of successful ageing. One encompasses a combination of avoidance of disease and related risk factors, maintenance of high function and sustained engagement with life.12 The other defines a process of selection, optimization and compensation, by which older people adapt to old age.4 The available evidence suggests that both are highly variable in the ageing process.13

We consider that there is a need to reconcile these different definitions. We commence by illustrating two main themes from the literature in figure 1. Figure 1(a) shows the risk, function and social engagement elements, and figure 1(b) seeks to capture how the person undergoing ageing feels in themselves. We sought to examine whether these subjective and objective components could be combined in a single model, reproducible in different British populations? We hypothesized that low risk (incorporating the absence of disease and a subjective feeling of contentment), good functioning and active engagement with society at a baseline period would lead to future outcomes that could be reasonably described as desirable. Therefore the model would be longitudinal and, cognizant of older people’s views,8–10,14 we conceived optimal well-being as a desirable outcome and a proxy for ‘success’ [figure 1(b)]. We defined well-being as feeling resilient, enabling continued engagement with life and able to express the forward-looking goals considered critical for growing old well15 as well as potentially conveying a survival advantage.16 This model combines a baseline of physical and social measurements with an outcome related to adaptation and social engagement.

Figure 1

Representations of successful ageing influenced by functionality and activities, or by how the person feels. Part a depicts a functional, objective representation of successful ageing. Part b is a subjective, personalized view


The model commenced with the baseline elements: low risk, high function and active engagement.12 At follow-up, optimal well-being had two elements. The first element, as shown in figure 1(a), encompassed continued engagement over time. The second element encompassed how the person felt—particularly in relation to peers, as shown in figure 1(b). We called this ‘confidence’.

Physical, social and psychological function, resilience and feelings of well-being as set out in figures 1(a) and (b) are latent variables. Latent concepts are used in everyday life—for instance satisfaction, intelligent quotient (IQ) or beauty. They are measured through manifest variables or indicators, proxies for the hidden world.17 We used structural equation modelling (SEM) to gain access to the latency inherent in figure 1. SEM measures the co-variances of indicators, grouped into factors and presented in a matrix.18 We, therefore, proposed five factors for our model: low risk, good function, active engagement at baseline and confidence and continued engagement at follow-up. All data were re-coded to produce interval data with a consistent expected direction of effect. Interactions among the factors can also be measured, addressing directly one of the knowledge gaps cited in relation to successful ageing—the relative importance of its various elements.4 SEM also deals with influences not explained by the model itself, which are quantified as ‘errors’.

The Health and Lifestyle Surveys (HALS 1 and 2) were used to develop the model and the English Longitudinal Survey of Ageing (ELSA) was used to confirm it.19–21 Both are longitudinal surveys with good follow-up data and have been described elsewhere.22 HALS comprises 9003 individuals aged ≥18 years. The baseline was in 1984 with a full follow-up 7 years later and mortality measured at 20 years. We used a sub-set of HALS comprising 3005 subjects aged >50 years at baseline for our model; ∼80% were aged <75 years and ∼25% had died within the 7 years before follow-up. ELSA comprises over 12 000 people aged ≥50years commencing in 2002 with follow-up at 2-yearly intervals, and we used the first round of follow-up in 2004.

Variables from the HALS and ELSA data sets related to risk, function active engagement were identified and examined for their validity and missing data. Table 1 shows the list of these indicators where the data were of good quality in both HALS and ELSA. We defined low risk of disease as low prevalence of major risk factors. We had intended to include smoking, respiratory function, blood pressure, body mass index (BMI) and alcohol history but there was a high frequency of missing data for the last two in both data sets and so we were limited to using the first three, with the addition of the presence of cardiovascular disease and diabetes in ELSA. We used the 12-item General Health Questionnaire (GHQ12), the presence of confirmed disability or long-term illness and self-reported contentment as measures of function. Active engagement at baseline included employment or voluntary work, as either is cited as adding social value.12

View this table:
Table 1

Indicators chosen from HALS and ELSA for the model

Factors drawn from US and European literature4,12Indicator in HALSIndicator in ELSA
Low disease, low riskSmoking statusSmoking status
Hypertensive statusHypertensive status
Measured respiratory function (nurse exam)Presence of respiratory disease
Presence of heart disease, diabetes
FunctionLong-term illness or disabilityLong-term illness or disability
Feeling content—and health benefiting from thisFeeling content with life
Actively engagement with lifeSocial activities, hobbies and leisureSocial activities, hobbies and leisure
In touch with friendsHaving and seeing friends
Being in current employment or undertaking voluntary work or trainingUndertaking voluntary work
Being employed
Success at 7 years follow-up (confidence)Self-reported health at follow-upSelf-reported health at follow-up
Has a healthy life been led over the 7 years?Quality of life module comprising questions on goals and abilities
How active the person feels now in comparison to their peers?
Success at 7 years follow-up (engagement)Physical activities undertaken in the previous fortnightMild, moderate and vigorous physical activities undertaken in the previous fortnight
Leisure and hobbiesLeisure and hobbies
Getting out and about frequentlyGetting out and about frequently
[No] Recent hospitalization or deterioration in healthVisiting relatives at follow-up

Table 1 also shows the chosen indicators of well-being from the follow-up data in each survey. The confidence indicators in HALS include self-rated health, health experience since baseline and comparison with peers. The ELSA confidence indicators also included self-reported health, but had a set of other measures from a quality of life module. In each survey, we sought measures of resilience for the future and positivity about self. Indicators on the second well-being factor, continued engagement, involved being out and doing things with no recent health problems. We proposed that confidence would lead to engagement, although we tested the reverse possibility.

The model was developed on the first 50% of randomly split HALS data and was confirmed in the other 50%. Then the model was replicated in ELSA. The data sets were analysed using Stata 9 SE23 and modelling was conducted in linear structural relations (LISREL).24 Significance was pre-determined at α = 0.05.


Figure 2 shows the HALS model with the best fit of data. The measurement model is the relationship between the factors and their indicators. The function and risk factors were merged into one called ‘funcrisk’ and the indicators were contentment, smoking status, measured respiratory function, hypertension history, long-term illness/disability and GHQ12. The strongest indicator relationships with the function/risk factor were long-term illness/disability, hypertension history, GHQ12 in that order, respectively. There were also strong relationships between the active engagement factor and two of its indicators: employed/voluntary work and social activities, respectively.

Figure 2

The model with new latent variables for success, in a randomly selected 50% sample of data, and modified to give a non-significant chi-square. Content, My health benefits from my feeling of contentment; Smoking, Regular smoker; FEV1, Forced expiratory volume in the first second; Disease, Long-term illness or disability; BP/Cardio, Past history of hypertension /cardiac with or without medication; GHQ Scr, GHQ 12 Score; Employed, Currently in work, undertaking voluntary work or training; Activities, Hobbies and social; Friends, Visiting friends regularly; Lifestyle, Opinions as to whether a healthy life was led in the 7-year follow-up period; Own health, Self-assessed health in 1991; Comparison, Comparison of physical activities with those of peers; Outabout, Getting out and about; Hospital, Recent deterioration in long-term illness or hospitalization; Physical, Recent physical activities; Leisure, Hobbies, leisure and/or social activities

At 7-year follow-up, the measurement model shows that self-reported health, and how active the person felt in comparison to their peers had the strongest relationships to their subjective factor ‘confidence’. These indicators, in addition to views about leading a healthy lifestyle, also clustered together strongly—that is their joint relationship strengthened the model. Engagement in social activities, physical activity, getting out and about and recent health experience (such as hospitalization) also clustered and demonstrated the strongest overall relationships with their objective factor: ‘engaged’.

The structural model in figure 2 shows the relationships between the four factors—the ‘middle section’ of the model. It shows that function and risk (‘funcrisk’) at baseline influenced confidence at 7 years which in turn enabled continued engagement. Active engagement at baseline only directly influenced continued engagement with life. The outcomes are not the same measures as those at baseline, because we are measuring something different at follow-up—objective and subjective well-being. Furthermore, it would be entirely predictable but unremarkable to have a strong SEM that simply related the same indicators to each other at different times.

The structure in figure 2 emerged as strong and stable. Although it was derived from 50% of the HALS data, it reproduced immediately on the other 50% of the data, as well as in ELSA with good model statistics. All t-values of the paths between the four factors were strongly significant at the 0.05 level (as are all relationships between factors and indicators). Table 2 shows the overall goodness of fit of this model in two random samples of HALS, and in the ELSA data. Both random samples in HALS produced models with relevance to a wider population, as shown by the non-significant chi-square. Counterintuitively, in SEM a non-significant chi-square is good because it shows no difference with the larger population from which it is drawn.25 The fit of the model to the ELSA data was good. It was not possible to achieve non-significance in this very large data set; however, the model addressed many of the requirements of a robust model. These include a large number of observations, validated survey data, a minimum and maximum number of variables per factor, inclusion of major contributory variables within the model and no known prior strong correlations between variables.26

View this table:
Table 2

The goodness-of-fit statistics for models in 50% random samples of the HALS data, and in the ELSA model

Model data setχ2RMSEAaCFIbSRMRcAICd
HALS (first 50%) (figure 2)104.68, P = 0.14 (NS)e0.0180.990.072272.00
90% confidence interval for RMSEA = (0.0; 0.031)
P-value for test of close fit (RMSEA < 0.05) = 0.00
HALS (second 50%)85.43, P = 0.24 (NS)0. 0141.000.068272.00
90% confidence interval for RMSEA = (0.027; 0.033)
P-value for test of close fit (RMSEA < 0.05) = 0.00
ELSA full data set511.270.0300.990.054380
90% confidence interval for RMSEA = (0.024; 0.028)
P-value for test of close fit (RMSEA < 0.05) = 1.00
  • a: RMSEA: ≤0.06 as the cut-off for a good model fit

  • b: A value of the Comparative Fit Index of between 0.90 and 0.95 is acceptable, and above 0.95 is good

  • c: Values <0.10 are generally considered favourable

  • d: Smaller values are better

  • e: A non-significant result means the given model’s covariance structure is not significantly different from the (underlying population’s) observed covariance matrix. The model has wider applicability


Successful ageing has so far proven an elusive concept, limiting its utility as a means of describing a good ageing process. However, this research has been able to develop a model incorporating factors that have been associated with ‘success’ in the literature and quantify their relationship to subjective and objective outcomes at follow-up. This links earlier ideas about good ageing that, we consider, conceptually belong together. We hypothesized that factors enabling adaptation and the ability to do what older people say they value most, remain engaged with life, would be important. We were able to demonstrate the paths to engagement in our model, which was powerful and reproducible in surveys from different time periods. We also quantified elements of resilience, where psychological status and contentment at baseline subsequently lead to subjective views of one’s lifestyle and well-being, influencing the activities that older people engage in. The interactions among all factors were quantified, addressing directly one of the knowledge gaps cited in relation to successful ageing: the relative importance of its various elements.

The findings were robust in different data. There are sound epidemiological reasons to why this should be so. For instance, respiratory function is a well-established predictor of good outcomes in later life.7 Physical activities were important at both baseline and follow-up. Being physically active in mid-life protects against long-term illness later in life.27 It is therefore concerning that the proportion of people taking regular, moderate exercise declines steadily in each decade of life and in England <12% are doing so by the age of 65 years.28 Being employed has positive effects on health29 and proved an influential indicator in the model, being associated with strong engagement at follow-up. However, substantial numbers of Europeans leave the workforce prematurely; the majority not by choice.30 The resilience factor contained an indicator about comparison with peers, health and lifestyle. Older people’s perspectives, including their regrets, have been shown to be mediated by their comparison with others.31 Purposeful activities lead to a feeling of being valued, which also appears to predict subsequent mortality.32

These findings signify a need to re-think urgently how people are living in late mid-life if good population ageing is to occur. Personal and societal planning for involvement of older people in a wide range of ways are central to the model. Recent policies on older people in England have begun to address this issue.33 The United Nations has also promoted the importance of full participation of older people in society and ‘ageing in place’.34 The central message is that older people should be encouraged to partake in opportunities for personal development and engagement with society. A good ageing process offers older people the scope to undertake these roles.


Department of Health for England.

Conflicts of interest: None declared.

Key points

  • Successful ageing is an elusive concept but this research has been able to demonstrate a model incorporating factors described in the literature and quantify their relationship to subjective and objective outcomes at follow-up. This links European and North American concepts of success.

  • The interactions among the factors were quantified, addressing directly one of the knowledge gaps cited in relation to successful ageing—the relative importance of its various elements.

  • The model was powerful and reproducible in surveys from different time periods; consequently, it is likely to have relevance to other populations.

  • Success comprises the social and physical activities that people do, mediated by the confidence to do them. It emphasizes particularly the importance of personal resilience and continued involvement in physical and social activities. Far from retiring, engagement with life and society should be the norm for ageing populations.


View Abstract