Background: Resilience is having good outcomes despite adversity and risk and could be described in terms of preserving the same level of the outcome or rebounding back to that level after an initial set back. Using the latter definition, resilience as “bouncing back”, this paper aims (1) to identify those members of a panel survey who demonstrated resilience, and (2) to identify the characteristics of the resilient individuals and the predictors of their resilience.
Methods: The study subjects were the 3581 participants in the British Household Panel Survey, selected from waves 1–14, who satisfied three requirements: exposure to an adversity; availability of consecutive General Health Questionnaire (GHQ)-12 scores; aged 50 or more years. The primary outcome variable was resilience, operationalised as a GHQ-12 score that increased after exposure to adversity and returned to its pre-exposure level in the next (after 1 year) wave of the survey. The adversities were: functional limitation; bereavement or marital separation; poverty.
Results: The prevalence of resilience, as defined, was 14.5%. After adjusting for regression to the mean, the GHQ-12 score of the resilient dropped by a mean of 3.6 points in the post-adversity period. Women predominated among the resilient, with this gender difference stronger among older women than younger women. The resilient were more likely to have high social support than the non-resilient, but otherwise were not different socioeconomically. High social support pre-adversity and during adversity increased the likelihood of resilience by 40–60% compared with those with low social support.
Conclusions: Resilience is relatively rare and favours older women. It is fostered by high levels of social support existing before exposure to adversity.
Statistics from Altmetric.com
Health at older ages has gained prominence due to the increase in life expectancy at middle age. At older ages mental health, unlike physical health, often is reported as being better than earlier in life. For example, the life-time risk of depression decreases with age almost linearly, with the greatest risk in young adult cohorts and the smallest risk in those aged 65 years or more.1 However, others have found that mental health deteriorated with age.2 The apparently better mental health of those at older ages has been explained in terms of response bias, differential morbidity, inappropriateness of measures used and insufficient power of diagnostic criteria.3 Also, occasionally, older people are described as having greater resilience,4 which is the focus of the present paper.
Resilience was recognised first in studies of children who, despite exposure to adverse conditions, sometimes developed into successful adults. Resilience could involve either rebounding after adversity5 or preserving competence in the face of adversity.6 More generally, resilience is a good outcome despite adversity and risk.7 Although these definitions refer to the same phenomenon, the definition chosen has knock-on effects in terms of the data and methods that can be used. The present paper defines resilience as bouncing back after adversity, which requires longitudinal data and methods suitable for them.
Studies of resilience at older ages are rare relative to those of younger age groups. Nevertheless, they have shown that old age is not necessarily a period of deterioration8–12 and that resilience-promoting factors can be identified at the psychological and social as well as at the material and demographic level.13–15 Building on this literature, the objectives of the present study are (1) to identify those members of a panel survey who demonstrated bouncing back after an adversity; (2) to describe their characteristics; and to discover the predictors of their resilience.
The British Household Panel Survey (BHPS), which started in 1991, is an annual interview survey of a representative sample of more than 5000 UK households.16 The present analyses use BHPS waves 1–14 to construct a working dataset of consecutive triplets of waves. Using waves1 (1991), 2 (1992) and 3 (1993), for example, we selected all participants aged 50 years or more who reported one or more adversity (see below) in the second wave, but not in the first wave, and had General Health Questionnaire (GHQ)-12 scores for all three waves (once thus selected, the participants were excluded from all subsequent selection). Moving on to waves 2, 3 and 4, we repeated the process; and we continued to do so till the last possible triplet (waves 12, 13, and 14). By this method we produced a dataset in which each participant was seen at three time points and exposed to one or more adversities between the first and second time points. We termed these time points as pre-adversity, adversity and post-adversity.
Primary outcome variable
We derived a variable called resilience as our primary outcome. Resilience for this purpose was defined as bouncing back after adversity. The impact of adversity was measured using the 12 item GHQ-12 with a range of 0 to 36.17 Higher scores on GHQ-12 represent poorer mental health. We used GHQ-12 as a continuous variable. The adversities were transitions between the first and second time points into functional limitation due to illness; change in marital status to single; transition into poverty. All three adversities have been shown to be associated strongly with poor mental health.18
Functional limitation was assessed using the question, “does your health in any way limit your daily activities compared to most people of your age?” (This question was not asked in BHPS waves 9 and 14; we kept the variable as missing, rather than substitute from somewhat similar questions in SF-36.) Moving into single status was assessed using reported marital status: defined as anyone who was married or living with a partner at time point 1 and reported their marital status as “widowed”, “divorced” or “separated” at time point 2. Poverty in any wave was defined as having a household equivalent income of less than 50% of the median income for the whole sample for that wave. Household equivalent income was calculated as the total income of all members of the household during the previous 12 months, with missing information imputed, divided by the equivalence scale before housing costs.
We operationalised bouncing back as a sequence of GHQ-12 scores in which the GHQ-12 score at the post-adversity time point was within the 95% CI of the GHQ-12 score at the pre-adversity time point, after being elevated at the adversity time point to above the pre-adversity 95% CI. Resilience was coded 1 as resilient and 0 as non-resilient. It should be noted that adversities were still persisting, and in some cases new adversities started, in the post-adversity period. However, this did not vitiate our operationalisation of resilience
The choice of these variables are informed by our own and other research demonstrating their influence on mental health and well-being.18–20 Age was categorised as 50–64, 65–74, 75+ years. Education as up to age <16 years, 16–18 years, >18 years. Social class as the Registrar General’s classification. Being male, owning house outright and having a car available for use were noted as binary variables. Similarly, binary variables were used for those who like their neighbourhood, were not planning to move, were members of any organisation, took an active part in them, and expected an improvement in their finances next year. In all binary variables, 1 denoted ‘yes’ and 0 denoted ‘no’.
Social support was calculated on the basis of five questions about having someone who will listen to the participant; who will help in a crisis; with whom they can relax; who appreciates them; and on whom they can count to offer comfort. The response to each question was coded 0 for no such person, 1 for one such person and 2 for more than one person. A summary score for each wave was calculated. As the questions were asked only during odd-numbered waves, we extrapolated the score for even-numbered waves by taking the arithmetic mean of two adjacent odd numbered waves (except for wave 14, where the value for wave 13 was carried forward). The support score for each wave was then dichotomised at the median to yield a variable representing high social support.
We use descriptive statistics to describe the resilient and non-resilient at the three time points. In describing the sample and the distribution of resilience within different groups we use the explanatory variables at the adversity time point. We used logistic regression to identify the determinants of resilience. We needed to estimate the reduction in GHQ-12 scores at the post-adversity time point due to regression to the mean effect. We did this using the following equation:21 reduction due to regression to the mean = (1−ρ)(χ−μ), where χ is the mean GHQ-12 score at the adversity time point, μ is the common mean for GHQ-12 scores for the adversity and post-adversity time points, and ρ is the correlation of GHQ-12 scores between adversity and post-adversity time points.
All analyses were done using STATA Version 9.2 on participants with complete information.
We included 3581 participants in the British Household Panel Survey (BHPS) who satisfied three requirements: exposure to an adversity; GHQ-12 scores available at the three relevant waves; aged 50 or more years. This study sample was 9% of the total panel members present in BHPS waves 1–14. Table 1 show that almost half of this sample was in the 50–64 years age band and women were in the majority (57.2%). Nearly two-thirds were educated to less than 16 years of age; 3% were in the highest social class; 47% owned their home outright, and 58% had a car available for their use.
Table 2 shows that 14.5% of this study sample was resilient, in the sense of having bounced back after exposure to an adversity; and that 60% of these resilient individuals were women. The prevalence of resilience increased with age, as did the female–male ratio; in women the age gradient was significant (p = 0.033).
Table 3 shows that at the pre-adversity time point there was no difference in mean GHQ-12 score between the resilient and non-resilient; both groups having the same mean (11.3) with similar spread. At the adversity time point, however, the resilient had a significantly higher mean GHQ-12 score (15.8; 15.4, 16.1) than the non-resilient group (12.0; 11.8, 12.2). At the post-adversity time point the mean GHQ-12 score of the resilient (9.2; 9.0, 9.4) had returned to below their pre-adversity level, while that of the non-resilient (12.4; 12.2,12.6) remained little changed. A proportion of the drop of 6.6 GHQ-12 units among the resilient at the post-adversity time point may have been due to regression towards the mean. We calculated the average reduction due to regression towards the mean as 3.0 units (the corresponding value for the non-resilient was −0.05). After accounting for the regression to the mean, therefore, the resilient showed a reduction in mean GHQ-12 score at the post-adversity time point of 3.6 units.
Table 4 shows that the resilient did not differ from the non-resilient in terms of education, social class, housing tenure or car access. Of the other six factors examined, only high social support showed a significantly higher prevalence among the resilient than the non-resilient. More than one-third of the resilient had high social support at all three time points compared with around one-quarter of the non-resilient (the differences between the groups were significant at all time points). There were no significant differences between sexes in the association of these factors with resilience except for social support and expected improvement in financial situation, both of which were significantly associated with resilience in women.
These results were supported by the logistic regressions carried out to identify which of the factors can predict resilience at what time points (table 5). The only significant predictor was high social support at the pre-adversity and adversity time points, which increased the chance of being resilient by 40–60% compared with those with low social support. In the pre-adversity period, liking the neighbourhood in which one lives and not planning to move from there were also on the borderline of significance in improving chances of being resilient. As adversities accumulate, the probability of resilience decreased. None of the factors was significant in the post-adversity period, suggesting that it was the circumstances before and during adversity which could render resilience.
Our results estimate the prevalence of resilience at older ages at less than 15% of those exposed to misfortune. This estimate is similar to our as-yet-unpublished results from the Boyd Orr cohort and English Longitudinal Study of Ageing. Resilience at older ages appears to be a relatively scarce phenomenon. The range of prevalence reported in the literature is higher,22 perhaps because most studies were of younger age groups, with the implication that resilience became less common with age, or because most studies had defined and measured both adversity and resilience less carefully. The resilient, in our present operationalisation, showed a difference of more than six points in GHQ-12 score between the adversity and the post-adversity period, and more than half of this difference persisted after allowing for regression to the mean. Thus, the bouncing back in this group was real.
The proportion showing resilience in the present study increased with age, with the oldest showing the largest proportion. In this respect, bouncing back after adversity might be different from successful ageing, which in most definitions was equated with freedom from limiting illness and no reduction in function as age increases.23 The proportion showing resilience also was higher in women than men; and the age gradient was steeper, and significant statistically, in women. Since our definition of resilience depended on exposure to adversity, these results might be explained by an increased probability of women being identified as resilient because GHQ12 scores tend to be higher in women24 25 and the proportion of women was higher in older age groups. However, since our results were stratified by age and sex, these differences cannot explain the higher prevalence of resilience among women. Other studies have shown similar results, sometimes referring to it as the “gender paradox”.26
None of the other sociodemographic variables (education, social class, housing tenure and motor car access as markers of wealth and income) had an effect on the probability of being resilient. Scores on GHQ-12 do not vary with social class24 25 and their variation with education is complex,27 but housing tenure is one of its strongest predictors.28 Therefore one may conclude that the influences on resilience differ from those on mental health, as measured by GHQ-12.
The only variable that was consistently related to resilience was social support, measured in terms of having people who can be trusted and who will offer help, comfort and appreciation, especially in a crisis. The influence of social support on all aspects of health, morbidity and mortality is well documented.29–34 It is possible that the relationship between social support and resilience results from social selection, in which those in better mental health attract more social support.35 Our results tend to discount this possibility. Pre-adversity GHQ-12 scores were virtually identical in the resilient and non-resilient groups; and it was social support before and during adversity, rather than after it and presumably in response to it, that rendered a person resilient. It is the strength of the longitudinal nature of our study that we could disentangle this question.
There is a debate in the literature on resilience about whether it is a trait or a process.36 In the present study we did not find a resilient profile of individual characteristics but strong indications that the social context before and during adversity influences resilience. In the tradition of Bourdieu,37 we can describe resilience as the process which converts social goods into good outcomes.
One important conclusion for social policy is that social support, if it is to be useful in terms of resilience, should be a condition of life before and during adversity; and not something that is provided only after adversity has been experienced. Brown and colleagues,38 who studied depression in women longitudinally, found that appropriate support received at the time of crisis would ameliorate the effects of the adverse event. If social policy wishes to enhance resilience it should adopt a preventative approach that fosters social support at the population level.39
All three of our adversities limit functioning: physical limitation of functioning due to ill health; social limitation due to loss of spouse or partner; and limitation in resources due to poverty. Adaptationist theories of ageing, like Carstensen’s socioemotional selectivity40 or selection, organisation and compensation in the Berlin Ageing studies,41 try to explain well-being in the face of declining function with age. Adaptation requires time, but our study was characterised by a very narrow time band (1 year between the impact of adversity and bouncing back). Thus adaptation might not be the explanation here. The emphasis should be on the value of existing social conditions in promoting resilience. Generalising further, we can say that resilience is to be found in the warp and woof of family and civic society. State policies that wish to increase resilience against population-level adversities should try to avoid damaging this social fabric.
The resilient in our study were ordinary people, without superpowers, as indicated by the fact that as adversities add up the probability of resilience decreases; resilience does not imply invulnerability.42 On the other hand, we did not find resilience to be the “ordinary magic” described by Masten.7 Resilience was scarce in this population, even when social support was present. The potential usefulness of this phenomenon for those delivering medical and social care and for the lived experience of older people justifies further investigation. Future research usefully could examine a wider timeframe than that of the present study and investigate more closely the effect of repeat adversities on those exhibiting resilience.
What this study adds
Resilience, bouncing back after exposure to an adversity, is a relatively rare phenomenon in older age groups.
The resilient are ordinary people not different from rest of the population sociodemographically. Social support, existing before and persisting during after the exposure to adversity fosters resilience.
Resilience can be nurtured through social policies that foster social support at the population level.
However, to be useful policy-makers should adopt a preventative approach implementing policies before adversity has been experienced.
Funding: This research was funded by Economic and Social Research Council grant number: L326253061. GN and ZH are supported by Economic and Social Research Council grant number: L326253061.
Competing interests: None.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.