Introduction

Precarity has emerged as an influential concept in the Social Sciences aiding understandings of the social consequences of neoliberalism for employment conditions and the concomitant reduced state investment in areas such as welfare, infrastructure and local government. Definitions and applications of precarity vary, but it is perhaps most often used to identify and/or describe the experiences of a group of people (the precariat) who share working conditions marked by instability and insecurity, a lack of control, limited worker rights, a lack of opportunity for career progression, a lack of work-based identify with often, but not always, low and uncertain pay (Standing, 2011). Broader definitions have extended precarity beyond the labour market to wider experiences of ‘life worlds that are inflected with uncertainty and instability’ (Waite, 2009) (p416). Accordingly, notions of ‘precarity in life arrangements’ now extend beyond employment including to relationships, care and housing (Colic-Peisker et al., 2014; Motakef, 2019). Recent research has developed theoretical perspectives on later-life precarity as a lens to understand growing exposure to insecure conditions in the context of- a. population ageing and policy responses to it, b. neoliberalism and c. major disruptive events such as the policy responses to the financial crisis of 2008 and most recently the Coronavirus pandemic (Colic-Peisker et al., 2014; Grenier & Phillipson, 2018; Grenier et al., 2017; Lain et al., 2018). Literatures on later-life precarity draw on theories of intersectionality and accumulation of (dis)advantage proposing that the risks and precarities experienced over the life-course deepen disadvantage in later life with the specific nature of the disadvantage varying according to intersections between characteristics such as sex, disability, social class and ethnicity (Grenier et al., 2017).

Following the financial crisis, research on the prevalence of low and unstable income has become particularly salient as growing numbers struggle to meet basic costs of living in the UK and elsewhere (Nau & Sooner, 2019), as marked by growing dependence on food banks, for example (Strong, 2020). Standing (2011), identifies an ‘old-age precariat’ increasingly forced into insecure employment as the statutory retirement age is increased and as a result of the financial crisis which de-valued many pensions significantly requiring returns to the labour market. Less secure employment and job losses in later life have been linked to both involuntary retirement (either due to job loss of poor health) and subsequent vulnerabilities in terms of depressive symptomatology in retirement (Voss et al., 2020). Although the triple lock has protected the value of the UK State Pension against inflation over the past decade, a broader context of neoliberalism and arguments around the affordability of the State Pension have, over the past 40 years, served to reduce the relative value of the State Pension while encouraging the uptake of occupational or private pensions (Foster, 2018). At the same time, changes in the pensions market are serving to reduce the value of many occupational and private pensions with a shift away from defined benefit (most common in Occupational pension schemes) to defined contribution pensions (Banks, 2006). Those relying on benefits to supplement their income are likely at greater risk of low income but also stress around uncertain income since the criteria and eligibility of benefits are subject to changes over time and across the life-course. Most recently, a series of cuts to welfare provision following the financial crisis have drastically reduced the income received by individuals and families and while these were focussed on working age population, they had important impacts on those in the final years of work and beyond (Beatty, 2016). A broader context of policies that seek to extend working lives in the context of reducing value of pensions and benefits, brings the risk of greater penalties for those who retire in an unplanned manner due to poor health or job loss.

In line with broader housing market trends, Colic-Peisker et al. (2014) observe a rise in home renting in later life as older people drop out of home-ownership, or are lifetime renters due to increasingly unaffordable housing, with a consequence of reduced disposable income in later life. Those renting are at a significant financial disadvantage since, unlike home-owners who have cleared their mortgage, their rental housing costs continue into retirement while at the same time they are the least likely to have savings to draw on (Storey & Coombes, 2020). Older tenants report significant stress around the potential to lose their home and in securing adaptions so that housing meets their needs (Colic-Peisker et al., 2014; Storey & Coombes, 2020).

The death of a spouse is an important later life event that is often associated with financial penalties, especially for women, and on average, subsequent elevated rates of various indicators of poor health and mortality (Das, 2013; Seiler et al., 2020). As norms around marriage and divorce shift, the population who experience divorce in later life is increasing as is living alone, and the impact of divorce on women’s income in later life remains, on balance, unfair (Buckley & Price, 2021).

Cuts in local authority budgets following the implementation of austerity policies from 2010 have led to 26% fewer older adults receiving social care in 2016, with the potential for increased levels of unpaid informal caring and unmet care need (Humphries et al., 2016). In a context of uncertainty and reduced access to social care, many older people who require support to perform activities of daily living face stress and/or additional costs to cover care needs. Those older adults who provide significant levels of unpaid care have reduced capacity to engage in the final years of work (Vlachantoni, 2010) with implications for income and pensions, reduced time for engagement in social life (Akgun-Citak et al., 2020) and poorer health outcomes (Vlachantoni et al., 2013).

Although, emerging research has developed concepts of later life precarity and, in some cases, has explored these in qualitative studies, limited attention has been paid to the empirical measurement of later life precarity. Similarly, later-life income is not used as frequently as a marker of social position and financial circumstances compared to wealth or educational position; for example, a recent paper noted that no other research had examined the association between frailty and income (Watts et al., 2019) while the association between frailty and wealth is regularly modelled (Davies et al., 2021; Marshall et al., 2015; Rogers et al., 2017). Where later-life income is used, typically, a snapshot is taken rather than considering the potential for change in income including declines to low income at particular points of later life or phases of income volatility. In this paper we respond to these gaps in the literature by capturing characteristic income trajectories in later life and analysing how membership of particular trajectories (including low and unstable income) associate with broader experiences of precarity in later life using the English Longitudinal Study of Ageing. We examine how pathways into certain income trajectories are structured by gender and social class and consider whether there is evidence for accumulation of disadvantages across the life course and generations.

The research questions that we tackle are:

  1. 1.

    What are the characteristic income trajectories in later life and how do these relate to income instability and low income?

  2. 2.

    Are people who experience particular income trajectories, including income instability and low income, more or less likely to be exposed to precarity in other areas including employment, retirement, relationships, pensions and benefits, housing and care?

  3. 3.

    Is membership of later-life income trajectory groups structured by sex and life-course markers of social class?

Methods

Data and Variables

This paper analyses nine waves of data from the English Longitudinal Study of Ageing (ELSA), spanning the period 2002 to 2019 (Banks et al., 2023). ELSA is a nationally representative sample of the population, aged over 50 and living in private households, originally drawn from another nationally representative survey, the Health Survey for England, in 2002. Since its inception, ELSA data has been collected from its original sample bi-annually, with refreshment samples added in waves 3, 4, 6, 7 and 9. A major strength of ELSA is that its respondents provide a rich set of sociologically-informed data on experiences and circumstances and events in later life through computer-aided questionnaire and a self-completion module. A life history module in wave 3 provides information on circumstances across the life-course extending back to childhood. In this paper we utilise the full ELSA sample, including its refreshment samples, giving 13,005 respondents who participated in at least two waves of ELSA and, to minimise the impact of outliers with very high income, had an income between £0 and £1000 per week. Full details on the methodology of the English Longitudinal Study of Ageing are provided elsewhere (Pacchiotti et al., 2021).

The income variable in this analysis is the equivalised total income at ‘benefit unit’ level. A benefit unit comprises a couple or a single person plus any dependent children they may have. So, where a respondent is married or cohabiting, the survey figures relate to a combined income for this couple. Where couples share finances, the person best able to answer questions on income is asked to complete the Income and Assets (IA) module and these figures are used, although the survey methodology does require that both members of the couple complete the module. Where couples have separate finances, they each complete the IA module and incomes are summed. The income figures are equivalised to allow for comparison according to household size with the income measures indexed to a single person household. Full details of the ELSA data methodology on equivalising income are available in the survey metadata that is provided with the data extract (see Financially Derived variables user guide; an OECD equivalence scale is used that assigns a weight of 0.5 to second adults and dependent children aged 14 and over and a weight of 0.3 to children under 14 years of age). Finally, since we are examining income trajectories over time, we adjusted all incomes for inflation between 2002 and 2019 using 2018/9 as our index.

We used two education variables and a variable on occupational status as a proxy measures for social class across the life course. Our models used respondent’s highest qualification, the age at which a respondent’s father left education and the National Statistics Socio-Economic Classification (NS-SEC) applied to respondent’s occupation. Where there was a change in social position, we used the highest social position observed and for respondents who retired we used the NSSEC as recorded in earlier waves. Our interest here is to evaluate whether the influence of social position on membership of later life income trajectory groups accumulates over the life course (childhood, entering adulthood and at the end of working life) and additionally whether it persists across generations. We included sex in our models to determine whether there were differences in membership of income trajectories for men and women, particularly in light of research that has pointed to disadvantages for women in the labour market and in terms of access to pension wealth, especially following divorce (Buckley & Price, 2021).

Informed by our review of the literature, a set of variables on various forms of later life precarity were included in our models of membership of income trajectory groups. The indicators of precarity as listed below were selected on the basis of being, on average, linked to uncertainty and/or vulnerability in later life including in relation to experience of volatile/low income.

Relationship Dissolution

whether a respondent had ever experienced divorce/separation or the death of a spouse either prior to joining ELSA, if not remarried at wave 1, or during the full period of ELSA data collection.

Unplanned Retirement

whether a respondent retired due to poor health or due to job loss. Each of these variables captures an unplanned exit from the labour market and a lack of autonomy in the nature of a key later life transition.

Housing

whether a respondent always rents their home in later life as opposed to owning a property for at least one wave of ELSA data collection (either with a mortgage or owning outright).

Pensions and Benefits

we capture whether a respondent had ever held an occupational pension or not and whether a respondent had ever held a private pension or not. ELSA includes questions across waves 1 to 9 on whether a respondents’ job included a pension and whether they are receiving income through a private pension. In 2018/19, the UK full rate State pension was £164.35 per week, a level generally regarded as not sufficient to cover essential costs of living, although in such cases access to benefits bridges shortfalls (Davis et al., 2021). Holding an occupational and/or private pension then, is increasingly viewed as essential to cover non-essential costs such as participation in social and leisure activities, running a car or going on holiday. We included in our models a variable on whether respondents had ever claimed any benefits, such as Income Support, Job Seekers Allowance and Widows pension, Pension credit, Universal Credit, Child Benefit or other State benefits.

Caring

whether a respondent ever received care to perform activities of daily living (e.g. washing, dressing or getting in and out of bed) and whether a respondent ever provided care to another. Our measure of giving informal care distinguishes time spent in this caring role with divisions of < 20 h, 20–35 h and 35 + hours to try to get at levels of care provision that might carry more significant impacts on the lives of carers.

Models

In this paper we capture characteristic income trajectories in later life by fitting a mixture of nonlinear trajectories using the lcmm package in R. The non-linear trajectories were modelled using second order B-splines with equidistant knots placed at ages 50, 60, 70, 80 and 90. Latent class analysis was used to identify clusters of income trajectories to which respondents were matched. We explored different numbers of clusters and settled on 10 as optimal to robustly capture the complexity of income trajectories in later life. In coming to this decision, we first took into consideration an elbow plot (see Supplementary material and Fig. S1 and Table S1) which suggests a levelling off in the improvement in model fit (AIC) by 10 clusters with clear declines in AIC exceeding a threshold of 2 (Burnham & Anderson, 2004) for each additional cluster up to 10. We visually examined the raw income trajectories for all 10 cluster (see Supplementary material and Fig. S2 in Supplementary material) confirming the underlying shape of specific income trajectories within clusters despite the heterogeneity. Finally, we are further reassured on the robustness of our 10-cluster solution by the concordance between our substantive findings on drivers of volatility in income trajectories in later life across clusters and other literature/theory as described in the “Discussion” section. In terms of sample sizes across clusters we have 3 larger clusters with over 10% of the sample and exceeding 1,900 respondents. There are 7 smaller clusters with between 3% and 6% of the sample equating to 398 and 844 respondents exceeding sample size thresholds for cluster analysis (Dalmaijer et al., 2022). Figure 1 provides the full details of sample sizes for each cluster.

Fig. 1
figure 1

Income trajectories distinguishing the 10 clusters and 4 groups based on levels of income during retirement

Full size image

A range of approaches exist to capture low income and fluctuations in income including income volatility characterised by sharp increases or decreases in income; economic insecurity marked by periods of time where income falls below poverty thresholds and income precarity which focuses on sharp declines in income (Nau & Sooner, 2019). We examine the characteristic later-life income trajectories that emerge from our analysis and consider the evidence for volatility, precarity and insecurity from age 50 upwards. When assessing whether income is ‘low’ a range of measures exist including self-reported ability to meet essential costs of living, evaluating whether income falls below a level necessary to meet essential living costs or according to a threshold based on a particular point in the wider national income distribution. In this paper we draw on recent research that has determined the following thresholds for later life income which connect to differing standards of living:

Luxury

Recent research on expenditure in later life identified that an income of around £600 would allow a single older person to cover costs of items such as a new/2 year old car, long-haul and European holidays, leisure club membership, employing a gardener or cleaner, a Samsung Galaxy smart phone (unlimited calls/texts and 6GB data) and TV steaming services (Davies, 2022; Padley & Shepherd, 2021).

Comfortable

Approximately £360-£380 per week has been identified as the income threshold for a single older person that would allow for expenditure including a European holiday, running a car (3 years or older, a Samsung Galaxy smart phone (unlimited calls/texts and 3GB data) and TV streaming services (Davies, 2022; Padley & Shepherd, 2021).

Low Income

Around £200-£230 per week is thought to be needed for a single older person to cover essential costs in retirement (e.g. groceries, utilities), limited rail travel, an entry-level smart phone, the most basic TV, broadband and streaming services and not the other items listed in the Luxury and Comfortable groups above (Davies, 2022; Padley & Shepherd, 2021). This level of income for essential costs is supported by other research, such as that commissioned by the Joseph Rowntree Foundation that found its Minimum Income Standard for a single pensioner to be £201 per week in 2019 (Hirsch, 2019).

We use three sets of models to explore how membership of later-life income trajectories is associated with life-course social position and later-life precarity. For the first two sets of models, we simplify our analysis and reporting by combining clusters to four later-life income trajectory groups of ‘Luxury’, ‘Comfortable’, ‘Always Poor’ and ‘Boom to Bust’ (See Fig. 1). Clusters are combined to groups on the basis of sharing broadly stable and similar income levels after the statutory retirement age, with the exception of cluster 9 which forms its own group of Boom to Bust. The labels for the groups are informed by the thresholds on living standards referred to earlier. The first set of models report a set of bivariate analyses (cross-tabulations) that explore whether the composition of the four income trajectory groups differ according to our independent variables on life-course social position and experiences of later-life precarity (see Table 1). The second set of models utilise multinomial logistic regression to model correlates of membership of income trajectory groups controlling and investigating confounding effects across our independent variables (see Fig. 2 and Supplementary material Table S2). We take the Comfortable income trajectory group as the reference category in these models, as it facilitates exploration of the impact of our independent variables on the relative risk ratios of experiencing higher or lower income in the ‘Luxury’ or ‘Always Poor’ groups. The third set of models consider correlates in the membership of all ten income trajectory clusters, again, using a multinomial logistic regression (see Fig. 3 with the full model output available on request) and focussing on differences within income trajectory groups. Since income trajectory clusters within groups differ most prior to the Statutory Retirement age (65), where they are also most volatile (see Fig. 1 and the Comfortable and Luxury clusters), to ease presentation of results, our models take an income trajectory cluster within each group (comprising more than one cluster) as the reference category, facilitating comparison to other clusters within the same group. For the analysis that focuses on the Luxury group, we take the most stable cluster 3 as our reference, which also has the highest sample size of the Luxury income trajectory clusters (see Fig. 3b). The analysis focussing on the Comfortable group takes the most stable cluster 7 as the reference category (see Fig. 3c and d) and for the Always Poor group we take the higher income trajectory cluster (cluster 8) as the reference (see Fig. 3a).

Table 1 Cross-tabulations of income trajectory groups with sex, social position and markers of later-life precarity
Full size table
Fig. 2
figure 2

Multinomial relative risk ratios (RRR) of income trajectory group membership

Full size image
Fig. 3
figure 3figure 3

Multinomial relative risk ratios (RRR) of income trajectory cluster membership focussing on within group differences

Full size image

Since the life history information on the age at which father left education applies to a sub-sample of the full ELSA data (collected in wave 3), we conducted sensitivity analysis that excluded this variable confirming that all substantive results still hold.

Results

Latent Class Analysis - Income Trajectory Clusters and Groups

Figure 1 displays the ten income trajectory clusters that emerged from the latent class analysis, presented in four charts that are organised according to four super groups (groups from here on) of income trajectory: Luxury, Comfortable, Boom-to-Bust and Always Poor. The rationale for these groups is that, with the exception of cluster 9 (Boom-to-Bust), there are broadly stable and comparable within-group income trajectories during retirement (from age 65) that equate approximately the various living standards identified in the literature (Davies, 2022; Davis et al., 2021). Within groups, a key difference separating income trajectory clusters is the volatility in income between ages 50–65. For example, in the Luxury group, those in cluster 1 experience a strong decline in income from around £800 per week at age 50 to a stable income of around £500 after the Statutory retirement age (65), in contract to the other stable income trajectories in this group. Clusters 5 and 6 in the Comfortable group are characterised by a spike in income at age 60 (around £800 and £650 per week respectively) with lower incomes either side of age 60 that settles around an average of around £400 and £350 per week respectively in retirement. Cluster 4 has a decline in income from £600 at age 50 to £300 at age 65 before settling at around £400 per week from age 70. For presentational purposes we initially focus on models that predict membership of the four groups, however, we also go on to refer to models of membership of all ten clusters in order to identify key differences for clusters that sit in the same group.

Membership of Income Trajectory Groups – Bivariate Analysis

Table 1 provides some descriptive statistics on the socio-demographic characteristics of the four income trajectory groups. Women (54.8% of the full sample) are over-represented among the Always Poor (58.6%) and under-represented in the other groups especially in the Luxury and Boom-to-Bust groups (48.6 and 48.7% respectively). Those with the lowest qualifications are over-represented in the Always Poor group and under-represented in the other groups, most notably the Luxury group. 7.6% of those in the Luxury group have no qualifications compared to 19.0% in the Comfortable group, 12.1% in the Boom-to-Bust group, 44.2% in the Always Poor group and 30.7% for the full sample. Similarly marked social gradients across the four groups are found for the variable on the age at which the father of respondent left education and occupational social class. There is a strong gradient of increasing risk of exposure to all of the various forms of precarity moving from Luxury to Comfortable to the Always Poor group. For example, the proportion of individuals always renting a house (16.6% for the full sample) increases from 1.6% (Luxury) to 7.1% (Comfortable) to 26.3% (Always Poor). Similarly, the proportion who retired on the grounds of poor health (15.2% for the full sample) increases from 9.1% (Luxury), 13.1% (Comfortable) to 18.1% (Always Poor). 70.9% of those who are Always Poor did not provide care compared to around 60% in the other groups. However, 13.2% of the Always Poor provided 35 hours of care a week or more, slightly higher than for the Luxury (12.2%) and Comfortable (12.6%) groups. Higher levels of informal care receipt are found in the Always Poor group of 50.0% compared to 27.8% in the Luxury group, 35.0% in the Comfortable group and 32.6% in the Boom-to-Bust group. The Boom-to-Bust group is similar to the Comfortable group in socio-economic composition, but is distinguished by having higher proportions who hold private pensions (78.5% versus 70.1% for the Comfortable group) and are highly educated (29.2% have a degree versus 19.8% for the Comfortable group) and lower proportions who hold occupational pensions (31.9% versus 36.5% for the Comfortable group).

Multinomial Models of Membership of Income Trajectory Groups

Figure 2 shows the adjusted relative risk ratios (RRR) of membership of the Luxury, Boom-to-Bust and Always Poor groups compared to membership of the Comfortable group for each independent variable (for the full regression output see Supplementary material and model 4). We divide the presentation of key results into three sections describing the association between income trajectory group membership and (i) social position across the life-course, (ii) sex and (iii) indicators of precarity.

Figure 2 illustrates that, compared to being in the Comfortable group, there is an increased risk of being in the Always Poor group for those in the lower social classes across all measures (father’s years in education, respondent’s educational qualifications and respondent’s occupational social class). Conversely there is a decreased risk of being in the Luxury group for those in lower social classes across all social class measures with a similar result for the Boom-to-Bust group for educational qualifications (but not measures of social class based on occupation or father’s education). Relative to risks of membership of the Comfortable group, those with no qualifications are 0.34 times (p < 0.001) less likely to be in the Luxury group, 0.44 times (p < 0.001) less likely to be in the Boom-to-Bust group, and over two times more likely (2.36, p < 0.001) to be Always Poor. Those who have fathers with no education beyond age 14 are more likely to be Always Poor (RRR = 1.45, p < 0.001) and less likely to be in the Luxury (RRR = 0.65, p < 0.001) group compared to being in the Comfortable group. Those who work/worked in semi-routine occupations are more likely to be Always Poor (RRR = 2.14, p < 0.001) and less likely to be in the Luxury (RRR = 0.45, p < 0.001) group compared to being in the Comfortable group.

Figure 2 shows that after controlling for other variables, there is no significant association between sex and income trajectory group membership. Since there are strong sex differentials across groups in bivariate analysis (see Table 1 and Supplementary material Table S1-3 (model 1)) it appears that the independent variables in the model explain the different risks of group membership for males and females. Exploration of the specific independent variables that attenuate the sex differentials in risks of group membership suggests that pension access and experience of partnership dissolution are responsible. The relative risks of being Always Poor, compared to Comfortable, are higher for women compared to men in a model with just sex (RRR = 1.34, p < 0.001). After including independent variables on whether individuals held a private pension or an occupational pension the elevated relative risk ratio for women being Always Poor is attenuated (RRR = 1.13, p < 0.01), and further attenuated and are no longer statistically significant (RRR = 0.97, p = 0.453) after accounting for partnership dissolution, either through divorce/separation or bereavement (for the full results see models 1–3 in Table S1-3 in Supplementary material). No other independent variables attenuate the elevated relative risk of being Always Poor for women to the same extent. A model with all independent variables, except those capturing pensions and partnership dissolution, still reports a significantly elevated relative risk ratio of 1.31 (p < 0.001) for women being Always Poor compared to the Comfortable group (model result not shown in output but available on request).

Figure 2 shows the relative risk ratios of membership of each of the income trajectory groups, compared to the Comfortable group, across other indicators of precarity including housing (tenure), partnership dissolution (through divorce, separation or death of a spouse), pension and benefit access, unplanned retirement (due to poor health or job loss) and caring (giving and receiving). The broad finding is that exposure to precarity across indicators is generally associated with elevated risks of being in the Always Poor group and lower relative risks of being in the Luxury group (all compared to the Comfortable group) with the exception of receiving care which is not significant in the full model.

Those who always rent their home are over two times more likely to be Always Poor than Comfortable (RRR = 2.16, p < 0.001) and 0.32 (p < 0.001) times less likely to be in the Luxury compared than the Comfortable group. Experience of retirement due to job loss elevates the relative risks of being in the Always Poor group (RRR 1.43, p < 0.001) compared to the Comfortable group. Respondents with no access to occupational or private pensions are, not surprisingly, more likely to be Always Poor than Comfortable with relative risk ratios of 1.49 (p < 0.001) and 1.69 (p < 0.001) respectively. Those with no private pension are 0.69 times less likely to be in the Luxury group compared to the Comfortable group with the equivalent relative risk ratio not statistically significant for those with no occupational pension. Receipt of benefits in later life is associated with greater risks of membership of the Always Poor group (RRR = 2.14, p < 0.001) compared to the Comfortable group with the opposite true for membership of the Luxury group (RRR = 0.66, p < 0.001). Turning to partnership dissolution, those who experience divorce/separation or bereavement are more likely to be Always Poor than Comfortable (for divorce/separation, RRR = 2.18, p < 0.001 and for bereavement, RRR = 1.34, p < 0.001). Those who experience divorce/separation are less likely to be in the Luxury compared to the Comfortable group (RRR 0.64. p < 0.001) with the corresponding relative risk ratio not statistically significant for spousal bereavement. Older people who provide the highest levels of care (more than 35 hours per week) are 1.19 times (p < 0.05) more likely to be Always Poor rather than Comfortable with no significant association observed for membership of the Luxury compared to the Comfortable group. Receiving care was not associated with membership of any groups on controlling for the other independent variables and so it would seem that the association between income trajectory groups and receiving care observed in bivariate analysis (see Table 1) is attenuated by the inclusion of other independent variables in the model.

Experience of precarity in terms of care (giving or receiving), unplanned retirement and housing do not appear to impact on the relative risk ratio of being in the Boom-to-Bust group as compared to the Comfortable group. However, those with no private pension and those who have experienced the death of a spouse are less likely to be in the Boom-to-Bust than the Comfortable group both with a relative risk ratio of 0.71 (p < 0.001) and 0.68 (p < 0.001) respectively. Conversely, those who do not hold an occupational pension are 1.43 (p < 0.001) times more likely to be in the Boom-to-Bust group than the Comfortable group. Thus, these results are in line with the bivariate analysis in Table 1 which, for example, highlighted that people with private pensions are over-represented in the Boom-to-Bust group, with 79% holding private pensions compared to the Comfortable group population, with 70% holding private pensions.

Multinomial Logistic Regression Models of Membership of Income Trajectory Clusters

Figure 3 displays relative risk ratios (RRRs) for membership of income trajectory clusters focussing on differences in RRRs between clusters that are in the same group. In the results presented below, we alter the reference category, selecting a stable income trajectory within a particular group to best understand the predictors of membership of more volatile income trajectories with the same group. Where RRRs are statistically significant, results are often in line with the group level analysis in terms of similar social gradients and experiences of precarity according to the broad levels of income across the within-group clusters. Additionally, a new finding emerges around the correlates of income volatility (or not) in the years leading up to the Statutory retirement age (65). Three key results are described below and are shown in Fig. 3a-d.

First, focussing on the Always Poor group (see Fig. 3a): compared to cluster 8 (the higher income trajectory within the group), those in cluster 10 (the lower income trajectory in the group) are more likely to have worked in semi-routine occupations (RRR = 1.74, p < 0.05) and more likely to have a father who did not attend school or left by age 14 (RRR = 1.58, p < 0.001). Compared to cluster 8, those in cluster 10 are more likely to have no private pension (RRR = 1.55, p < 0.001), no occupational pension (RRR = 2.73, p < 0.001), to experience divorce/separation (RRR = 1.75, p < 0.001), to always rent their home (RRR = 1.74, p < 0.001) and to have received benefits (RRR = 2.27, p < 0.001). Thus, a similar set of results observed across groups, holds when comparing clusters within groups. Here, the poorest income cluster (10) comprising 20.8% of the population is more strongly associated with life-course measures of lower social class and experience of precarity than the other Always Poor cluster (8) which has slightly higher income levels.

Second, considering the Luxury group, cluster 1 has a different shape of income trajectory exhibiting a sharp decline in income in the final years of work that is not observed in clusters 2 and 3. Figure 3b. shows that compared to cluster 3 (stable trajectory), those in cluster 1 (pre-retirement decline in income) are more likely to have experienced divorce/separation or the death of a spouse and to have retired for health reasons. It appears then that the sharp decline in income in the final years of work may be connected to unplanned retirement (for health reasons) or the consequences of relationship dissolution due to bereavement or divorce/separation.

Third, Fig. 3c and d display RRRs comparing membership of clusters within the ‘Comfortable’ group. Here, again, there are interesting differences in the shapes of income trajectories around Statutory retirement age; cluster 7 has an income trajectory that is relatively stable, cluster 4 has a trajectory that declines sharply between the ages of 50 and 60, whilst clusters 5 and 6 each have a spike in income at age 60 with lower incomes at younger/older ages. Compared to cluster 7 (stable income), those in cluster 4 (pre-retirement decline in income) are more likely to have no occupational pension and less likely to have no private pension (see Fig. 3c.). It appears then that holding a private pension may be associated with decline in income prior to retirement while possessing an occupational pension may have a protective effect against experiencing such income decline. Compared to those in cluster 7, those in cluster 5 are less likely to be in the lower social classes according to all our measures (Fig. 3d). Thus, higher social position appears to be associated with a pre-retirement spike in income in the final years of work with a similar result for RRR comparing clusters 7 and 6 (results not shown but available on request).

Discussion

This paper makes six key contributions. First, we implement an innovative technique to capture meaningful and distinct later-life income trajectories that exhibit marked differences in income volatility around retirement ages and, for all clusters except one, broadly stable income after retirement that corresponds to stark inequalities in standards of living. Second, the risks of membership of income trajectory groups (and clusters) follows a strong social gradient, that appears to accumulate over the life-course, as well as according to experiences of later-life precarity in terms of unplanned retirement, partnership dissolution, housing, care provision, pension access and benefits. Third, women are overrepresented in the Always Poor income trajectory group and it appears that this is as a consequence of inequalities in access to pensions and in the financial consequences of partnership dissolution due to divorce/separation or the death of spouse. Fourth, there are sharp declines in two income trajectories in the Luxury and Comfortable groups in the years prior to Statutory retirement age that appears to be associated with a range of factors including type of pension (private or occupational) unplanned retirement and experience of partnership dissolution. Fifth, two income trajectory clusters in the Comfortable group exhibit a spike in income at age 60 and this seems to be associated with high social position in terms of occupational prestige and education. Finally, a small proportion of the population (4%) in the Boom-to-Bust group experience a sharp increase to very high income at age 70 with a similarly sharp decline to a very low income at the very oldest ages that is sufficient only to cover essential costs of living. We discuss each of these findings in turn before considering the policy implications and limitations of the research.

Identifying Income Trajectories

Identification of different shapes of income trajectory brings a valuable insight for research on inequalities in later life with the potential to better understand inequalities in the experience of ageing and adverse outcomes such as frailty, falls, depression and mortality. Most research on inequalities in ageing tends to use wealth, a more stable indicator than income, or a snapshot of income. Our research suggests that there are meaningful patterns in later life income trajectory that can be captured and that might be accommodated within models of later life outcomes for greater insights on inequalities. The greatest volatility in income trajectories, as might be expected, is around retirement ages reflecting the diverse routes and often phased nature of retirement that usually occurs ahead of the Statutory retirement age (Banks, 2006). We consider the drivers of different forms of income volatility at retirement later in the discussion. After retirement, the broad stability in incomes is perhaps a reflection of measures taken to protect older people’s financial position such as the triple lock maintaining the value of the State Pension and other benefits that top up low incomes in retirement (Davis et al., 2021). Significantly, over half of older people in England (54%) are assigned to an Always Poor income trajectory with persistently low-income during retirement that is only sufficient to cover only essential costs of living. This finding tallies with critiques of Laslett’s Third Age (Laslett, 1989) that points to inequality and low income as a barrier for many older people to take up the various opportunities for personal fulfilment that Laslett envisaged in retirement (Rowland, 2012). The low income of a significant proportion of the ELSA sample is confirmed in the wave 9 report which gives the 25th percentile of weekly equivalised income to be equal to £265 and £235 for men and women respectively (Banks et al., 2020).

Income Trajectory Membership and Relation to Social Position and Later-Life Precarity

A central finding of this paper is that both social class and experience of precarity are associated with membership of income trajectories. Those who are in the lower social classes or who experience precarity in later life in terms of housing, pensions, relationships, care (giving/receiving) and unplanned retirement are at greater risks of being Always Poor and less likely to be in the Luxury compared to the Comfortable group. Since, each of father’s education, respondent’s own education and final occupational status are independently associated with income trajectory group membership, the result is consistent with the notion that the impact of social position on later life income trajectories accumulates across the life course and across generations. Those who have attained a high occupational status despite the challenges of coming from a background where their father had little schooling or having not securing high qualifications themselves, are, nonetheless more likely to be Always Poor than an individual with the same occupational status but who holds high educational qualifications and/or has a highly educated father. That social position across the life course, and across generations, is associated with income trajectories in later life is in line with theories of accumulation of dis(advantage) which observe a “systemic tendency for interindividual divergence in a given characteristic (e.g., money, health, or status) with the passage of time” (Dannefer, 2003). For example, adverse experiences in childhood, such as poverty, poor housing and nutrition, stress or low access to education, are socially patterned and influence subsequent outcomes such as occupation and career trajectories, wellbeing and health, capacity to manage stress and form stable relationships and the ability to own a house (Bartley, 2012). All these factors interconnect and combine across the life-course to drive stark inequalities in income trajectories in later-life in various ways. This finding is in line with previous research that has demonstrated empirically the accumulation of (dis)advantage over the life-course and across generations for outcomes such as wellbeing and allostatic load in later life in the UK (van Deurzen & Vanhoutte, 2019; Vanhoutte & Nazroo, 2016).

Causal pathways linking income trajectories and specific forms of precarity in other areas are likely complex, often bi-directional, and differ according to the indicator of precarity. For example, income is a key factor in determining capacity to purchase a property, or a private pension, with those on low income more likely to be restricted to rental accommodation and the State Pension in later life as a result. But a similar argument is less obviously true when it comes to divorce/separation, for example, where low income does not appear clearly associated with greater risks of divorce/separation (Brown & Lin, 2012). More generally, low income among younger adults is known to be associated with poor health and wellbeing, although causal directions and pathways are also contested (Mackenbach, 2020), which then plausibly elevates factors such as the risk of involuntary retirement due to poor health/disability (Banks, 2006) and greater need for care in later life (Humphries et al., 2016). On the other hand, experience of partnership dissolution, involuntary retirement, providing or requiring high-levels of care all might also plausibly drive lower income in retirement through mechanisms such as a loss of shared income from a partner (especially for women) (Buckley & Price, 2021; Lin & Brown, 2021; Purdam & Prattley, 2020) and loss of income in the final years of work with implications for subsequent pension wealth (Banks, 2006; Vlachantoni, 2010). Experience of precarity and income are likely often self-reinforcing; for example, those who always rent likely lack sufficient income to buy a house but continue to incur housing costs through later-life whilst those in more affluent groups have no such housing costs having paid off mortgages (Colic-Peisker et al., 2014).

Sex Inequalities in Income Trajectory Membership

Our findings on sex and income trajectory membership are directly in line with a body of research that shows older women have long been at risk of low income and economic hardship in later life due to structural inequalities in employment and pension access, particularly following partnership dissolution through bereavement and divorce/separation (Buckley & Price, 2021; Ginn, 1991; Lin & Brown, 2021; Purdam & Prattley, 2020; Vlachantoni, 2012). The inequalities in pension access in the UK are stark with the median pension wealth six-times higher for men compared to women (£212,000 versus £35,000) (Buckley & Price, 2021). Research shows that after divorce, separation or widowhood, women tend to experience a sharper decline in income linked in large part to inequalities in the way pension wealth is divided, and they consequently suffer a rise in risk of financial hardship (Buckley & Price, 2021; Gillen, 2009; Purdam & Prattley, 2020).

Income Volatility Around Retirement

Two income trajectories show a marked decline in income from the age of 50 to 60 before levelling off through the retirement years. Since the trajectories lie in the Luxury and Comfortable groups, neither approach the levels of income in the Always Poor group, but nevertheless, these declines represent a potentially important change in circumstances during a key phase of later life. The pre-retirement decline in income for the cluster within the Luxury group appears to be associated with partnership dissolution including divorce/separation and the death of a spouse. As noted above, research suggests that divorce/separation and death of spouse impact negatively on income particularly for women (Buckley & Price, 2021; Gillen, 2009; Lin & Brown, 2021; Purdam & Prattley, 2020). Similarly, there is evidence to suggest that the stresses of partnership dissolution in later life may well impact on aspects such as wellbeing (Tosi & van den Broek, 2020) with potential implications for work and income prior to retirement. The pre-retirement decline in income for the cluster within the Comfortable group is associated with holding a private pension and not an occupational pension. One explanation for this result may be linked to the distinction that private pensions are less likely than occupational pensions to be ‘Defined Benefit’ and more likely to be a ‘Defined Contribution’ pension (Banks et al., 2020). In a Defined Benefit pension, the employer guarantees a given level of pension benefits to an employee in retirement based usually on final salary and length of service meaning that there are strong incentives not to retire early or to switch roles with less responsibility or part-time work with lower salaries. In a Defined Contribution pension there fewer incentives for continuing to work up to Retirement age or moving to part-time work in the final years of work (Banks, 2006). Thus, it may be that holding a private pension is associated with declines in income in the final years of work because early-retirement, moving to a role with less responsibility or to part-time work all reduce income and are more common among those with private pensions compared to occupational pensions. Similarly, the spike in income trajectory observed at the age 60 in two clusters may also be linked to the incentives for Defined Benefit pension holders to strive end their careers on the highest salaries. Our analysis suggests that this income spike is associated with being in a managerial/professional occupation or holding degree-level qualifications and such groups are the most likely to have access to a gold-standard Defined Benefit pension (Banks, 2006).

Boom-to-Bust in Later-Life Income Trajectory

The Boom-to-Bust group, who experience very high levels of income around age 70 that drop to the lowest incomes at the oldest ages (80+) are a group of particular policy concern given their financial situation at end-of-life. Those in the Boom-to-Bust group are characterised by having an increased likelihood of holding a private pension and higher qualifications and a lower likelihood of holding an occupational pension compared to the Comfortable group. Again, the greater tendency for private pensions to be defined contribution rather than defined benefit may underpin the results here. Those who hold a defined contribution pension have greater flexibility to ‘cash-in’ their pension leading to a spike in income during retirement and subsequent decline without regular pension payments to rely on.

Policy Implications

There are a number of policy implications that flow from this research. First, we identify that a significant portion of the older population have persistent low income in later life and experience precarity across many other domains. A consequence is that policy responses to significant events such as the financial crisis, the Coronavirus pandemic and its aftermath should consider that for older people in such circumstances, small changes to costs of living or personal income streams and public services might easily tip individuals in a situation where they cannot cover the basic costs of life. Similarly, cuts in UK public spending in areas such as public transport, social care, neighbourhood upkeep and public leisure facilities have most impact on the poorest and elderly, removing opportunities for social participation (Hastings et al., 2017, 2021) which has been connected to adverse health outcomes in later life (Sacker et al., 2017). Cuts in public spending without sufficient attention to the circumstances of a sizeable vulnerable older population may well bring unintended consequences such as rising use of food banks and the levelling off in life expectancy improvement that followed the austerity policies implemented in the UK (Jenkins et al., 2021; Macdonald & Morgan, 2020; McCartney et al., 2022). In the aftermath of the Coronavirus pandemic and an unfolding cost of living crisis in the UK, it is important that policy responses do not neglect vulnerable older populations, particularly around social care provision which is considered at breaking point (Humphries et al., 2016).

A concerning trend for the future is that many of the indicators of precarity that are considered in this paper are increasing in prevalence and evidence-based policies are required to mitigate the impact on a growing later-life precariat. For example, trends point to a growth in older people who live in rented accommodation and who experience later-life divorce/separation (Brown & Lin, 2012; Storey & Coombes, 2020). Older renters are among the poorest and least able to cover rent in later life, they are more likely to experience stress associated with the potential to lose their home or to live in sub-standard or inappropriate housing often lacking the capacity to secure housing adaptions to meet their needs (Colic-Peisker et al., 2014; Storey & Coombes, 2020). As norms around marriage and divorce shift, the population who experience divorce in later life is increasing as is living alone. The impact of divorce on women’s income in later life remains, on balance, unfair (Buckley & Price, 2021) and divorce in later life has been linked with poorer wellbeing among those affected (Tosi & van den Broek, 2020). A growing older population living alone may bring implications for informal care provision in the absence of care that might previously have been covered by a spouse, particularly in the context of sharp reductions in social care budgets (Humphries et al., 2016). There is a need then for policies to ensure pension wealth is fairly split after later life divorce and that social care responds to the needs of a growing population of older people who live alone. Similarly, development of policies to protect the housing security of older renters and to ensure they can secure necessary housing adaptions would seem particularly important in the future particularly in the private rental market.

The Boom-to Bust group are of particular interest in a policy sense given the evidence of spending their final years with very low levels of income. The analysis suggests that membership of this group is particularly associated with holding a private pension and may be linked to older people withdrawing their pension during retirement. Further detailed quantitative and qualitative research would seem valuable to understand the experiences of this group in order to inform appropriate policy responses.

A final policy contribution of this paper is to respond to arguments around intergenerational fairness, that pit the perceived advantageous position of older as responsible for the challenging circumstances faced by younger people (Willett, 2011) and in some cases make a case for removing some the benefit and other protections received by older people. Such arguments sit uneasily with the extent of low income experienced by many older people in England during their retirement. The inequalities within the ‘young’ and ‘old’, and their structural determinants, some shared, would seem more pressing than pitting generations against each other (Bristow, 2019; True, 2019).

Limitations

This paper is subject to some limitations that ought to be taken into account. First, as with all survey data, ELSA is subject to missing data and those who drop out of the survey are more likely to be in poor health or experiencing precarity. It is likely that we underestimate the extent of low income and its intersection with later life precarity as a result. One of the ways we counter issues of non-response is through the inclusion of refreshment samples in our analysis. Other statistical approaches for handling non-response have limitations that led us not to pursue them here given our use of refreshment samples. Multiple imputation is not robust to situations where data is Missing Not At Random (MNAR) which would seem plausible for this research where indicators of precarity themselves (or other unobserved variables) might reasonably be expected to be associated with attrition in the same precarity indicators. While longitudinal weights are available in ELSA, these are not equipped to deal with missingness that is MNAR and weights are only provided for the sample who are present all waves, severely reducing the sample size for our research which draws on data across all of the ELSA waves. Full details of missingness are provided in the English Longitudinal Study of Ageing technical report and we produce our own figure in Supplementary material (Fig. S3) which illustrates the scale of non-response and refreshment samples across the survey.

Second, as noted above, many of the associations likely run in two directions and the nature of the study does not allow us to explore in more details the extent to which associations run in one direction or the other. We do not make specific claims around the direction and relative strengths of any causal connections between indicators of precarity. Rather we would suggest that it is most likely that different forms of precarity intersect with each other in later life in complex ways. Finally, we do not include time in our analysis of income trajectories so major events, such as austerity, are not factored into our latent class analysis of income trajectories. Similarly, we do not distinguish when events such as divorce occur and must acknowledge the potential for the impact of divorce on income trajectory membership to vary according to the stage of life at which it occurs or to be changing over time.

In conclusion, the central contribution of the paper is to empirically capture characteristic groups of income trajectories in retirement and to show that membership of these groups are stratified by gender and social class, but are also associated with and intertwined with precarity in other areas of later life. The Always Poor comprise over half the older population, and are vulnerable not just in terms of low income but also in experiences of other precarity in later life. Those older people who find themselves Always Poor are more likely to be women and in lower social classes, and are particularly vulnerable to adverse outcomes, such as mortality and poor health, in the face of cuts in public spending.