Introduction
In many European countries, including the Netherlands, older workers are obliged to extend their working lives owing to retirement policy reforms (Hofäcker and Radl Reference Hofäcker, Radl, Hofäcker, Hess and König2016). The extension of working lives occurs in an environment of labour market flexibilisation that has led to the deregulation of the employment relationship, yielding lower wages and fewer fringe benefits for many (Gevaert et al. Reference Gevaert, Van Aerden, De Moortel and Vanroelen2021). In several European countries, employment protection has decreased, layoffs have become easier, while flexibility in the employment contract and working hours has increased (Heyes and Lewis Reference Heyes and Lewis2014). This means fewer older workers are likely to work full-time in their long-term jobs close to retirement. These developments are often assumed to have led to increased variation (Fasang Reference Fasang2010; Riekhoff Reference Riekhoff2019) or ‘de-standardisation’ of pathways to retirement (Turek et al. Reference Turek, Henkens and Kalmijn2024).
Increased variation in retirement pathways results from older workers ‘reinventing’ – out of necessity or choice – their routes to retirement. Some older workers value a lower job intensity over employment security or high earnings (Platts et al. Reference Platts, Ignatowicz, Westerlund and Rasoal2023). Older workers in privileged positions (usually employees with more seniority and high income) have more opportunities to craft their jobs by adjusting their work tasks (Garthe and Hasselhorn Reference Garthe and Hasselhorn2022). This can occur through an employer change or reducing hours. This may also be the case for older workers with health issues and/or in physically demanding jobs, who may be forced to adjust their pathways to retirement out of financial necessity (Lain et al. Reference Lain, Van Der Horst, Vickerstaff, Czaja, Sharit and James2020). Other older workers in less favourable positions may be displaced from their long-lasting jobs and forced into precarious employment or unemployment. This process can be quite unpredictable, characterised by job changes, displacements and unemployment (Scotti Reference Scotti2022). Therefore, the transition to retirement is increasingly becoming a process rather than a single event.
In liberal contexts such as the United States, the above-mentioned processes have produced several ‘atypical’ routes such as phased retirement (gradual reduction in working hours) (Cahill et al. Reference Cahill, Giandrea and Quinn2015) or partial retirement (combining the receipt of pension with employment) (Maestas Reference Maestas2010). Evidently, receipt of social protection and participation in the labour force can occur as distinct or overlapping processes, making them an integral part of the retirement process. These routes establish a deviation from ‘cliff-edge retirement’ (Loretto and Vickerstaff Reference Loretto and Vickerstaff2015), that is, one-time exit from full-time paid employment to full-time leisure at age 65 (Calvo et al. Reference Calvo, Madero-Cabib and Staudinger2018). These developments have fuelled the literature on the emergence of a new life stage, coined as ‘encore adulthood’, in which older workers engage in a multitude of paid and unpaid activities at the end of their ‘career’ and child rearing and before ‘old age’ (Moen Reference Moen2016).
Despite establishing the trend towards de-standardisation of the retirement process, hardly any research has assessed this process considering several overlapping aspects. Previous empirical research focuses on a single or at most a few dimensions of employment quality such as work-time arrangements, employment security, financial security and social protection (Andrea et al. Reference Andrea, Eisenberg-Guyot, Oddo, Peckham, Jacoby and Hajat2022; Tang and Burr Reference Tang and Burr2015). This has been even less the case for regimes such as the Netherlands, where the pension system was previously orientated towards ‘early employment exit’ (Schils Reference Schils2008) but shifted later towards ‘employment maintenance’ (Hofäcker Reference Hofäcker2015), and the labour market is very flexible (Van Heuvelen et al. Reference Van Heuvelen, Bettendorf and Meijerink2021). Current research on features of employment quality has been explored for mostly early to mid-career individuals in the Netherlands (Eberlein et al. Reference Eberlein, Pavlopoulos and Garnier-Villarreal2024; Mattijssen and Pavlopoulos Reference Mattijssen and Pavlopoulos2019; Peckham et al. Reference Peckham, Flaherty, Hajat, Fujishiro, Jacoby and Seixas2022), rarely focusing on older workers. Existing research on retirement for the Netherlands is based on older cohorts that were not entirely subjected to recent retirement reforms (Riekhoff Reference Riekhoff2018, Reference Riekhoff2019; Visser et al. Reference Visser, Gesthuizen, Kraaykamp and Wolbers2016).
The aim of this study is to explore multi-dimensional retirement trajectories among post-war baby boomers in the Netherlands. For this purpose, we draw on recent data (2008–2019) on all contemporary older men and women in the Netherlands, regardless of whether they were employed or not at baseline. Focusing on labour market exit, we use a trajectory approach to concentrate on at least four years before and two years after state pension age (SPA). We apply a mixture hidden Markov model (MHMM) and explore retirement trajectories using ‘late employment quality’ (LEQ) as a latent variable measured by employment security, work-time arrangements, financial security and social protection. This enables us to determine the degree to which atypical retirement routes are becoming common in the Netherlands by including the relevant labour market factors that can vary as well as overlap for older workers in later life. Moreover, this allows capturing more of the variation in pathways to retirement of women, which is otherwise underestimated (Madero-Cabib et al. Reference Madero-Cabib, Le Feuvre and König2023).
Our contribution to the literature is threefold. First, we contribute to the knowledge regarding the pathways to retirement of cohorts that have been theoretically more exposed to a wide range of institutional changes targeted towards the extension of working lives. Second, we combine the growing literature on de-standardisation of retirement with the literature on employment quality. We do so by highlighting that the multi-dimensional nature of retirement is closely related to several aspects of employment. Third, by analysing LEQ in the pathways to retirement, we advance the knowledge on retirement by including more of the previously identified important dimensions than considered before in Andrea et al. (Reference Andrea, Eisenberg-Guyot, Oddo, Peckham, Jacoby and Hajat2022) or Tang and Burr (Reference Tang and Burr2015) to better capture the retirement process.
The remainder of the article is organised as follows: we first draw on previous research to explain the link between de-standardisation of retirement and employment quality in later life. Then, we elaborate on data and methods used to explore this variation in pathways to retirement. In the results section, we describe the variations in pathways, as well as how common these are for men and women. Finally, in the discussion and conclusion, we reflect on the finding that not all pathways easily fall in the ‘cliff-edge’ or ‘atypical’ categories and quite a large group of older people, mainly women, in the Netherlands seem to be doing neither.
De-standardisation of retirement pathways
Existing research conceptualises retirement as a process and asks for a more dynamic analysis of the lifecourse. Several researchers have aimed to do this by assessing retirement as a sequence of labour market statuses (Calvo et al. Reference Calvo, Madero-Cabib and Staudinger2018; Fasang Reference Fasang2010; Madero-Cabib et al. Reference Madero-Cabib, Le Feuvre and König2023; Riekhoff Reference Riekhoff2019). In this stream of literature, de-standardisation of retirement typically refers to heterogeneity in institutionalised pathways (i.e. labour market exit using early retirement, unemployment or disability benefit) as well as non-standard retirement pathways – via self-employment or part-time employment or non-employment – that are not shaped by explicit institutional pathways (Fasang Reference Fasang2010; Riekhoff Reference Riekhoff2019).
Retirement pathways can differ in their combination of timing, order and duration of employment and non-employment spells (Fasang Reference Fasang2010; Riekhoff Reference Riekhoff2019). More transitions in and out of these spells is understood as a higher degree of complexity or differentiation in the pathways to retirement (Riekhoff Reference Riekhoff2019). Typical pathways that emerge from these studies are associated with the timing of transition to retirement (e.g. ‘early retirement’, ‘standard retirement’ or ‘late retirement’) and the types of labour market status such as unemployed, employed, inactivity or disability (see e.g. Calvo et al. Reference Calvo, Madero-Cabib and Staudinger2018; Riekhoff Reference Riekhoff2018).
In this study, we are interested in the multi-dimensional nature of these de-standardised retirement pathways. Diversity in retirement conceptualisation is well documented in existing literature. As Denton and Spencer (Reference Denton and Spencer2009) have summarised in their literature review, retirement can be understood as (1) non-participation in the labour force, (2) reduction in work hours and/or earnings, (3) work hours or earnings below a specified threshold, (4) receipt of retirement income, (5) leaving the main career employer, (6) change of career or employment later in life, (7) self-assessed retirement and (8) combinations of the above.
Subsequently, Ekerdt (Reference Ekerdt2010, p. 70) asserted that ‘the designation of the retirement status is famously ambiguous because there are multiple overlapping criteria by which someone might be called retired, including career cessation, reduced work effort, pension receipt or self-report’. Similarly, research has highlighted the ‘fragmented nature of the retirement process that may include bridge employment, second career, (part time-) disability retirement, unemployment, unsalaried periods, part time leave, part time pension, early retirement, work while drawing pension, salaried while not working, “regular” retirement, work past retirement age, unretirement, etc’ (Hasselhorn and Apt Reference Hasselhorn and Apt2015, p. 21). Therefore, de-standardisation of retirement should be conceptualised as a combination of several processes over time.
Conceptual model of the work-to-retirement process
De-standardisation of retirement can be better understood by focusing on the de-standardisation of employment relationship in recent times. The standard employment relationship (SER) – commonly defined as permanent full-time waged employment – has been on the decline owing to economic restructuring of the labour market (Gevaert et al. Reference Gevaert, Van Aerden, De Moortel and Vanroelen2021). De-regulation of the labour market has led to the decline of organisational careers and job security vis-à-vis the emergence of new forms of non-standard employment, such as part-time work, short-term temporary work and contract work (Lewchuk Reference Lewchuk2017).
Subsequently, awareness regarding the quality of jobs has grown in recent years in Europe. Changes in labour market conditions may lead to new ways in which older workers exit the labour market (Andrea et al. Reference Andrea, Eisenberg-Guyot, Oddo, Peckham, Jacoby and Hajat2022), just as they have led to variation in employment quality for labour market entrants (Eberlein et al. Reference Eberlein, Pavlopoulos and Garnier-Villarreal2024). Previous research suggests that older workers are more likely to be affected by deteriorating labour market conditions (Madero-Cabib Reference Madero-Cabib2015; Visser et al. Reference Visser, Gesthuizen, Kraaykamp and Wolbers2018), with downward mobility in income or working hours becoming more prevalent for those at a disadvantage in the labour market.
There is no consensus in the literature on the definition of the quality of a job (Findlay et al. Reference Findlay, Kalleberg and Warhurst2013; Sengupta et al. Reference Sengupta, Edwards and Tsai2009). However, an analytical distinction is made between work quality and employment quality. The former represents the subjective aspects of a job (e.g. job satisfaction, work intensity), whereas the latter is related to the objective characteristics of the employment relationship (e.g. work-time arrangements, income, employment security) (Gallie Reference Gallie2007; Van Aerden et al. Reference Gevaert, Van Aerden, De Moortel and Vanroelen2014). There is growing recognition that employment quality is a multifaceted construct characterised by various combinations of employer–employee relations that are deviations from SER (Peckham et al. Reference Peckham, Fujishiro, Hajat, Flaherty and Seixas2019, Reference Peckham, Flaherty, Hajat, Fujishiro, Jacoby and Seixas2022). Recent studies suggest that employment quality can be effectively assessed using the seven employment characteristics: (1) employment stability, (2) work-time arrangements, (3) material reward, (4) social protection and worker’s rights, (5) training and employment opportunities, (6) collective organisation and (7) interpersonal power relations (Gevaert et al. Reference Gevaert, Van Aerden, De Moortel and Vanroelen2021; Peckham et al. Reference Peckham, Fujishiro, Hajat, Flaherty and Seixas2019, Reference Peckham, Flaherty, Hajat, Fujishiro, Jacoby and Seixas2022).
Applying the concept of de-standardisation to employment quality in later life, it can be expected that fewer older workers (than previously) transition from SER-like jobs to ‘on time’ retirement. Using the SER as a reference point, different combinations of these dimensions can occur as distinct or overlapping processes which yield variation in employment quality. For instance, some jobs may entail high working hours, high pay and good benefits, while others may include a short-term contract, low pay and few working hours (Gevaert et al. Reference Gevaert, Van Aerden, De Moortel and Vanroelen2021). Many older workers may be occupying these non-SER types of job at the end of their working lives. Therefore, the study of de-standardised retirement pathways should assess job security, financial security, work-time arrangements and receipt of social protection simultaneously as they cannot be seen independently from one another.
The Dutch institutional context
Early-exit culture
Historically, early retirement was common in the Netherlands, using early-exit schemes (VUT) or other social security benefits (Kohli et al. Reference Kohli, Rein, Guillemard and Gunsteren1991; Van Oorschot and Jensen Reference Van Oorschot and Jensen2009). However, owing to concerns regarding financial sustainability, the VUT schemes were replaced by capital-funded pre-pension plans in the 1990s (Fleischmann and van den Broek Reference Fleischmann, van den Broek, In Léime, Ogg, Rašticová, Street, Krekula, Bédiová and Madero-Cabib2020). Unlike VUT schemes, pre-pension schemes were based heavily on the older workers’ own contributions, with lower replacement rates for those who exited earlier. Lastly, incentive for early exit was further reduced by taxing both the premiums and the benefits received through this plan.
Disability and unemployment pathways
An alternative route to early exit was through the disability (WAO) benefits and to a lesser extent via unemployment benefits (WW), which became prevalent among older workers, known as the ‘Dutch disease’ (Ebbinghaus and Hofäcker Reference Ebbinghaus and Hofäcker2013). In the late 1980s, access to these benefits was curtailed, shifting the policy emphasis towards work rehabilitation. For WAO, individuals were expected to re-enter the labour market and work according to ability. For WW, stricter job search effort became a requirement for entitlement to unemployment assistance (Riekhoff Reference Riekhoff2018). Such measures may have forced older workers to rely more on the labour market than previously, leading to more employment pathways to retirement.
Changes to mandatory retirement age
Most collective labour agreements at the sector level rule that employment relationships automatically terminate at SPA (van Solinge et al. Reference Van Solinge, Damman and Hershey2021). This mandatory retirement age has gradually increased from 65 in 2013 to 67 in 2021, with further increases made dependent on life expectancy (OECD 2021). The system of mandatory retirement could be consequential for the degree to which atypical pathways may be observed in the Dutch context.
Mandatory retirement may, on the one hand, act as a barrier in the degree to which atypical pathways may be observed before SPA. More older workers may continue to work in the same job until SPA instead of transitioning to alternative arrangements. This happens because, in the Netherlands, older workers experience seniority-based wage growth, tenure and employment protection (Oude Mulders et al. Reference Oude Mulders, Van Dalen, Henkens and Schippers2014). Such a system combined with mandatory retirement may diminish mobility before SPA and therefore atypical pathways may not be observed to a very large extent. On the other hand, beyond SPA, the possibility for de-standardisation of the retirement process is greater as pensioners who remain in paid work generally need to negotiate a new labour contract, which, most likely, is a fixed-term contract (van Solinge et al. Reference Van Solinge, Damman and Hershey2021) or switch to self-employment.
Provisions for flexible employment and retirement
The incidence of part-time employment in the Netherlands is the highest among the Organisation for Economic Co-operation and Development (OECD) countries (Bloemen et al. Reference Bloemen, Hochguertel and Zweerink2016), where those who work part-time have the same pro rata rights as individuals working full-time (Yerkes and Hewitt Reference Yerkes, Hewitt, Nicolaisen, Kavli and Jensen2019). Most occupational pension funds offer opportunities for older workers to draw part of their pension and at the same time continue paid employment with a reduction in working hours (van Solinge et al. Reference Van Solinge, Vanajan and Henkens2023). This makes the Netherlands a relevant case for studying gradual withdrawal from the labour force.
Data and sample
In this research, we used data from Statistics Netherlands, namely the basic integral registration dataset (system of social statistical datasets [SSD]), which contains micro-level register data on welfare, jobs and other characteristics for individuals registered in the Netherlands (Bakker et al. Reference Bakker, van Rooijen and van Toor2014). At the time of research, data were available for an observation period of 12 years (2008–2019), containing information for older workers spanning a maximum of 144 months. This dataset has information on the types of employment such as employed and/or self-employed; contract type (e.g. permanent or non-standard); number of weekly (contractual) employment hours; income amounts from various sources; and the receipt of various forms of benefits (e.g. pension, disability benefits and unemployment benefits).
As we are interested in observing the retirement process, we selected individuals who were living and registered in the Netherlands for a minimum of four years before and two years after their SPA. We selected individuals who were born between January 1947 and March 1952 as these were the individuals that reached their SPA between January 2012 (to be able to observe individuals for a minimum of four years before SPA) and December 2017 (to be able to observe individuals for a minimum of two years after SPA). In accordance with Statistics Netherlands (2012), individuals born between 1946 and 1955 are referred to as the post-war baby-boom generation in this study. Older workers born in 1947 had 65 years as their SPA whereas those born from 1948 onwards had an SPA higher than 65 years old. Our sample therefore contains a mix of older workers who were both unaffected and affected by the policy of higher SPA in the Netherlands. This yielded a sample consisting of 1.7 million cases. A random sample of 5 per cent (n = 16,027 individuals) was then drawn from this dataset to manage computational time.
We excluded director-major shareholders (DGAs) from the data. This is because the DGAs (<4% of the random sample) are a special case of self-employed individuals that classify themselves as employees in their own company. Additionally, DGAs can directly influence their own income from the company. This income can take the form of either wages or dividends. The latter is hardly ever the case for regular employees of a firm. Thus, DGAs are neither typical self-employed nor workers, and were excluded from the analysis.
In this study, people born in different years reached their SPA at different chronological ages. Therefore, the observations within individuals are organised based on distance from one’s individual SPA (zero is SPA, negative numbers are months before SPA and positive numbers months after SPA). To optimise estimation time, we chose measurement occasions to be three months apart. This selection resulted in a final sample of 3-monthly information for 15,470 individuals.
Indicators of LEQ
The various observed indicators that we use are, theoretically, measures of different aspects of association with the labour market. This way we will measure the latent variable of LEQ as an overall indicator of wellbeing within the labour market. As indicators of employment security and work-time arrangements, we included the employment contract type and the weekly contractual working hours, respectively. To further distinguish types of employment, we included a separate indicator for self-employment. To distinguish types of non-employment, we included receipt of pension and receipt of benefits. Moreover, we included income as an indicator of the overall financial security of older workers. The distinct categories within each indicator are mutually exclusive. For example, a wage-employed individual can only have one type of contract: permanent, fixed-term, on-call or temporary agency work. However, individuals can combine locations from different indicators. For example, individuals can be employed and receive a pension.
Contract type
The employment contract type measures the level of employment security of older workers. The labour market has been argued to have broadly two forms of employment contracts: standard contracts (permanent employment with fixed working hours for the same employer) and non-standard contracts where the duration of the contract, the employer or the working hours are not fixed. We take into consideration the most common forms of non-standard paid employment in the Netherlands: fixed-term contracts, temporary agency contracts and on-call contracts. Even though the latter two contract types can be either permanent or fixed term, they have been usually attributed to higher labour market insecurity through introducing variation in the actual employer where the worker is employed (temporary agency workers) or the stability of working hours (on-call work). These contract types have been associated with precarious careers (Mattijssen and Pavlopoulos Reference Mattijssen and Pavlopoulos2019). Therefore, for contract type, we made a distinction between permanent, fixed-term, other flexible contract and no contract. Temporary agency work (3.6% of the sample) and on-call work (5.3% of the sample) are merged into the category ‘other flexible contract’ regardless of whether they are permanent or fixed-term contracts.
Contractual working hours
Contractual weekly working hours is an indicator of work-time arrangements in paid employment. The categorical variable on weekly working hours of individuals in paid employment consists of seven categories: ‘zero hours’, ‘less than 12 hours’, ‘12–20 hours’, ‘20–25 hours’, ‘25–30 hours’, ‘30–35 hours’ and ‘more than 35 hours’. Individuals who were not in paid employment (i.e. the self-employed and the non-employed) were coded as having zero contractual working hours.
Self-employment
We include self-employment as an additional indicator of employment type. This is a dichotomous variable which identifies a self-employed individual. If an individual is reported to have either business profit or income from other work excluding salaried employment, they have been coded as being self-employed. As self-employment is a separate indicator from contract type, individuals can simultaneously be self-employed and have a different type of contract. A word of caution is needed here. In our register data, self-employment status is determined from yearly information on business profit or income from other work. Therefore, if an individual is registered as self-employed in a given month, they are registered to have been self-employed that entire year. Therefore, in practice, we do not observe any transitions in and out of self-employment within any given calendar year.
Receipt of pension
Receipt of pension is an indicator of retirement status. An indicator variable was created depicting whether an individual receives pension income in a given month. We have pension receipt as a separate category instead of including it in the benefits variable as there is substantial overlap in receiving both pension and various other benefits in the data.
Receipt of benefits
Regarding benefits, a distinction is made between sickness and disability benefits, unemployment benefits, other benefits and no benefits. The category ‘other benefits’ includes those on welfare and other social benefits. There was little overlap between these benefits, and therefore we classified individuals based on the highest amount of benefit received per month. For example, if an individual received both sickness and unemployment benefits but received higher unemployment benefits, then this individual was categorised under unemployment benefits.
Income
We include income as an indicator of financial security of older workers. Income was constructed as the sum of income (in euros) from wage employment, self-employment and all forms of benefits (including pension, sickness and disability benefits, unemployment benefits and other benefits). Income from wage employment includes only monthly income from contractual working hours and does not include holiday pay or overtime pay. Income from self-employment was available only at the yearly level. A monthly average is calculated based on the yearly amount. Furthermore, the total income was adjusted for the yearly inflation rate to be able to observe real changes in income. Lastly, income was categorised based on percentiles with 10 per cent increments. Therefore, income consists of ten categories: ‘zero income’, ‘0–744’, ‘745–1030’, ‘1031–1279’, ‘1280–1623’, ‘1624–2038’, ‘2039–2490’, ‘2491–3070’, ‘3071–4069’ and ‘>4070’ euros per month.
Demographic characteristics
We consider variation by demographic features to be an important descriptive measure of who may belong to which trajectory. Specifically, we consider gender, country of origin and SPA to be social characteristics of interest. Gender is constructed as a distinction between men and women. Country of origin was constructed as a dummy variable indicating whether the individual was born in the Netherlands. Lastly, a distinction in SPA being 65 versus higher than 65 is made using the year of birth as 1947 as opposed to higher than 1947.
Methods
We are assessing the latent structure of LEQ using an MHMM, which is a probabilistic model (Vermunt Reference Vermunt, In Van Montfort, Oud and Satorra2010) that is a longitudinal extension of latent class analysis (Di Mari et al. Reference Di Mari, Oberski and Vermunt2016; Vermunt and Magidson Reference Vermunt and Magidson2016). It is used to perform an exploratory analysis to identify an underlying grouping variable (i.e. a categorical latent variable) that is not directly observed but inferred from a set of indicators and to see developments of this grouping variable over time. In our analysis, the MHMM has three components that correspond to different substantial processes: measuring LEQ (measurement part), transitions between different classes of LEQ over time (structural part) and of groupings of LEQ longitudinal patterns (trajectories, mixture part).
Measuring LEQ
In the measurement part of the model, we conceptualise LEQ as a categorical latent variable. This measurement model is depicted in Figure 1(a). Here, LEQ is measured by six observed indicators: contract type, contractual working hours, self-employment status, receipt of pension, receipt of benefits, and income. Formally, by including a measurement component, MHMM corrects for random measurement error by defining the latent variable as a function of the commonalities between the (imperfect) observed variables (Vermunt Reference Vermunt, In Van Montfort, Oud and Satorra2010). The categorical latent variable consists of multiple latent classes estimated based on distinct response patterns of the observed indicators. A latent class is considered to be highly homogeneous when individuals in that latent class are likely to have the same observed response pattern, implying that one response pattern is highly characteristic of that latent class. For example, in a latent class where older workers are only pension receivers, they will have a very high probability of receiving a pension given that they belong to the pension receivers category and a low probability of being (self-)employed. The item-response probabilities for the categorical indicators describe the relationship between the indicators and the latent variable. Testing MHMMs up to ten latent classes and using statistical criteria and theoretical interpretations, we identified the categories of LEQ.
Figure 1. Visualising pathways of LEQ. (a) The measurement model of LEQ as a categorical latent variable at each time point. (b) The arrows represent the transition probabilities from one latent state of LEQ at a certain time point to another latent state of LEQ at the next time point.
Transitions between different classes of LEQ
Transitions between the various classes of LEQ can be understood from the structural part of MHMMs. The structural part refers to longitudinal change in the latent variable over time. This part of the analysis deals with LEQ as a dynamic categorical latent variable (Vermunt Reference Vermunt, In Van Montfort, Oud and Satorra2010), meaning that individuals can transition between different categories of LEQ over time. We expect to observe different classes of LEQ before and after SPA (see also Table 1 with many differences four years before and two years after SPA). Therefore, we centred time in our analysis at SPA and allowed for a quadratic effect of time in our model, making it a mixture time-varying hidden Markov model.
Table 1. Descriptive statistics before and after state pension age
Source: Own estimations, SSD, 2009–2019.
An illustration of the structural part of the model, that is, transitions between latent classes of LEQ, is given in Figure 1(b). In this figure, the bubbles represent the latent classes that individuals belong to at each time point. A key feature of MHMMs is that transitions over time follow the first-order Markov assumption (Collins and Lanza Reference Collins and Lanza2010). This means that the value of LEQ at any given time point (t) is dependent only on the previous time point (t − 1), and independent of other previous time points.
As indicated previously, we use three-monthly data between January 2008 and December 2019. Therefore, the arrows in the path diagram in Figure 1(b) demonstrate that individuals can transition from one latent state of LEQ to another every three months in the observation window. The output of the structural part of the model consists of two types of estimated probabilities: (1) the initial state probabilities, which refer to the probability of belonging to each class of LEQ in the first time point, and (2) the latent transition probabilities which are the three-monthly transition probabilities between the categories of LEQ. Since we are estimating time-varying transitions, we have a different set of transition probabilities between each time point instead of assuming the same probabilities across all time points. These transition probabilities are estimated using a multinomial logistic regression as the probabilities to move (or stay) between all possible categories of LEQ (Biemer Reference Biemer2011).
Pathways to retirement: trajectories of LEQ
The final component of the MHMM refers to the estimation of different trajectories (mixtures) of older workers. This part seeks to identify the most homogeneous classes of the longitudinal process, based on differences in initial states and transition probabilities. In this part, we accounted for the possibility that there is not one general set of transition probabilities that applies to all older workers. Testing models from one to seven trajectories and using statistical criteria as well as theoretical interpretations, we identified the most likely classes of longitudinal processes. To ascertain who belongs to which trajectory, after estimating the MHMM, we produced descriptive statistics of trajectory affiliation with gender, country of origin and increased SPA.
Advantages of MHMM
The bulk of prior research on retirement trajectories has been done using sequence analysis (SA) (Calvo et al. Reference Calvo, Madero-Cabib and Staudinger2017; Fasang Reference Fasang2010, Reference Fasang2012; Madero-Cabib Reference Madero-Cabib2015; Riekhoff Reference Riekhoff2018, Reference Riekhoff2019). However, we argue that, in our case, using MHMMs presents several methodological advantages. The measurement part of the model ‘combines’ multiple variables into a single latent categorical variable. In contrast, describing, visualising and comparing large sequence data using multiple channels in SA can be very complex. Thus, in comparison to multichannel SA, MHMMs study transitions over time for a single latent variable instead of summarising the pathways of several observed variables. Additionally, MHMMs present interpretable parameters that describe the transitions over time. Thus, MHMMs can be used to compress and visualise information (Helske and Helske Reference Helske and Helske2019), allowing for a more parsimonious representation of results.
Moreover, MHMMs are flexible enough to include predictors in each part of the model (measurement, structural and trajectory). This way we can test more complex theoretical models and the possible causes in one cohesive model. These predictor effects can help interpret and explain variability in the latent profiles and trajectories. This way we can test specific hypotheses in each part of the model. In SA, causal analysis is done stepwise. Trajectory membership is calculated deterministically, and afterwards this membership is used as a dependent variable in further analysis.
Results
Latent classes of labour-economic status
The optimal number of latent classes was identified using two key considerations. First, we used model fit statistics, including Bayesian Information Criteria (BIC) and Akaike Information Criteria (AIC) (see Figure 4 in Appendix A1). Usually, lower values of AIC and BIC are better. Second, we examined the interpretability of the latent classes (Masyn Reference Masyn and Little2013). Interpretability refers to the extent to which the categories of LEQ are meaningfully different from each other as well as the extent to which individuals within the sample are classified to one of these categories with the least possible uncertainty (i.e. with a high probability). Formally, interpretability refers to latent class separation and heterogeneity and is summarised in the measure of entropy-R 2. As such, the solution with eight latent classes was deemed to be optimal as it substantially improved model fit and had an entropy-R 2 of 99 per cent.
In Table 2, we present the measurement model estimates of the eight-class solution. The table represents the different cross-sectional categories of LEQ that individuals can transition between every three months. Overall, the results indicate that older workers occupy one of the following three class types: (1) employment, (2) out of employment with a pension or (3) out of employment without a pension. The latent classes representing employment are (1a) mostly full-timers (13% of observations), (1b) mainly part-timers (6%), (1c) self-employed (9%) and (1d) wage-employed pensioners (6%). The percentages in parenthesis refer to the size of the classes identified (first row of Table 2). Another two latent classes represent groups where receiving a pension is the only dominant characteristic, namely (2a) ‘Low-income pensioners’ (23%) and (2b) ‘Medium/high-income pensioners’ (21%). The remaining three latent classes represent mostly non-employed older workers. These are (3a) ‘benefits’ (13%) and (3b) ‘inactive’ (10%). Detailed descriptions of these latent classes are available in Appendix A2.
Table 2. Latent classes of LEQ
Source: Own estimations, SSD, 2009–2019. Note: item-response probabilities may not exactly sum to one within each indicator owing to rounding.
Describing the transitions between latent classes of LEQ
We arrived at a five-trajectory type as the optimal solution by combining statistical criteria (see Figure 5 in Appendix A3) and theoretical interpretability. The five-trajectory solution presented an acceptable level of latent trajectory separation and heterogeneity as it had an entropy-R 2 of 78 per cent.
For the five-trajectory solution, we present index plots for the five different types of retirement trajectory of LEQ that represent how older workers transition in and out of the labour market around their SPA. An overall distinction can be made between these retirement trajectories: (1) trajectories that are relatively stable over time and (2) trajectories that tend to include frequent changes over time.
The more stable trajectories are grouped into two large types, namely ‘retirement with medium/high pension’ and ‘from non-employment to low pension’. These trajectory types are presented in Figures 2 and 3, respectively. In these plots, each line represents the pathway to retirement for an individual approximately four years before and two years after SPA, with SPA indicated by the vertical dotted black line in the figure. The plots are constructed using the predicted LEQ for each individual at every time point. Furthermore, after adding descriptive covariates to the model, we found no association between the trajectory type to which an individual is classified and the country of origin or increased SPA. However, using follow-up descriptive analysis, we did find that the distribution of gender was uneven across trajectory types. Therefore, as far as covariates are concerned, we discuss how the trajectories differ only with reference to gender.
Figure 2. Retirement with medium/high pension.
Figure 3. From non-employment to low pension.
Figure 2 presents the trajectory type that is characterised by transitions to retirement with a relatively good pension income (formally to LEQ class ‘medium to high-income pension’). We found that 42 per cent of the sample belongs to this trajectory. The largest group of individuals (about 50%) within this trajectory is transitioning from full-time employment (formally, from LEQ class ‘mostly full-timers’). This transition occurs at various time points until SPA, while some individuals have already exited the labour market at the moment that the shown observation window begins (i.e. four years before SPA). Almost none of these individuals returns to employment after SPA. Such a crisp transition from full-time employment to full retirement represents ‘cliff-edge’ transitions. A distinct group within this trajectory type are the self-employed. These combine retirement with self-employment after SPA, which may partly reflect income necessity as self-employed individuals are less likely to have an occupational pension. This is a male-dominated pathway (69% of this trajectory are men).
Figure 3 illustrates the trajectory type where the most dominant pattern includes the transition from non-employment to retirement with a low pension. We found that 43 per cent of the sample belongs to this trajectory type. Most individuals have been out of employment (either inactive or on benefits) for at least four years before SPA. At SPA, these individuals tend to transition to retirement with a low-income pension and remain retired thereafter. This pathway is typically followed by women (77% of this trajectory are women).
Three smaller trajectory types characterised by more frequent movements are also found. These are the trajectory types ‘eventually partial retirement’ (5%), ‘steps from employment to low pension’ (4%) and ‘alternating work and non-work’ (7%), presented in Figures 6, 7 and 8, respectively, in Appendix A4. Since these trajectories have retirement pathways which are recurrent for many older workers in the smaller trajectory types, we describe an overall summary here in the main results. A detailed description of each trajectory type is available in Appendix A4.
The most prevalent pathways in these smaller trajectories are the moves towards combining retirement with paid employment (formally, the latent class ‘wage-employed pensioners’) or self-employment (formally, the latent class ‘self-employed’). We remind the reader here that in the latent class of ‘self-employed’ there is a 55 per cent probability of receiving a pension. Although these transitions are clearer after SPA, where older workers are seen to combine pension with wage employment or self-employment, they may also occur before SPA. Older workers in these trajectories have an unstable pathway until SPA as they transition between different classes of LEQ on multiple occasions. This makes the trajectories in Figures 6, 7 and 8 more turbulent than those illustrated in Figures 2 and 3, where older workers are typically seen making a one-time transition. When examining gender differences in trajectory types, men are predominantly represented in the ‘eventually partial retirement’ category (75%) and the ‘alternating work and non-work’ category (58%). In contrast, women are more likely to follow the ‘steps from employment to low pension’ trajectory (56%).
Discussion
This study provides a comprehensive picture of pathways to retirement in the Netherlands by conceptualising the multi-dimensional structure of the retirement process. The article reveals three major findings that help us understand the balance between the pathways of cliff-edge retirement and ‘atypical’ retirement within the Dutch context. In other words, these findings can help us determine whether the arguments for the de-standardisation of the retirement process derived from liberal contexts such as the USA are also relevant for more regulated contexts such as the Netherlands.
To what extent do we see cliff-edge retirement?
First, cliff-edge transition, which typically describes the move from full-time employment to complete exit from the labour force, is a considerable but not the most prevalent pathway to retirement in the Netherlands. The cliff-edge transition from SER-like jobs to retirement typically takes place before SPA. This indicates that older workers following this pathway are still using early retirement schemes and/or other supplementary pensions to exit the labour market early. As expected, the cliff-edge transition pathway was more common for working men.
The practice of early exit in the Netherlands, as evidenced in past studies for older cohorts (e.g. Visser et al. Reference Visser, Gesthuizen, Kraaykamp and Wolbers2016), persists for younger cohorts. Pathways of retirement with medium/high pension are consistent with previous research regarding the association of high pension entitlements and early retirement (Kuhn et al. Reference Kuhn, Grabka and Suter2021). The cohorts under study were somewhat more exposed to the full regulations and measures taken to make exit by early retirement, disability and unemployment benefits less rewarding and more difficult. Despite these measures, we found that those who can afford to retire early may still continue to do so. Simultaneously, previous research in the Netherlands found that many retirees want but are unable to find employment while receiving pension. This implies that full retirement may not always be a matter of choice (Dingemans et al. Reference Dingemans, Henkens and van Solinge2016).
To what degree are older workers opting for atypical retirement?
Second, we find only partial support for previous studies that suggest that de-standardisation dominates the retirement process of post-war baby boomers. In more detail, while, for example, Cahill et al. (Reference Cahill, Giandrea and Quinn2015) found a large proportion of early boomers transitioning to intermediate jobs or re-entering the labour market after extended retirements, we document only a group of about 12 per cent of older workers who combine paid work with receiving a pension. This difference with previous literature is possibly attributable to the Dutch retirement context, which includes a stronger social safety net and is therefore less individualised. Within this pathway, however, a higher degree of variation with respect to employment security, work-time arrangements, financial security and social protection was observed relative to the other pathways. For example, some older workers opted for atypical retirement by combining both permanent employment with the receipt of pension while others merged non-standard employment arrangements such as self-employment or fixed-term contracts.
Despite part-time employment being a widespread phenomenon in the Netherlands, results did not clearly distinguish partial retirement from gradual retirement in the form of only scaling back by a reduction in the hours worked (i.e. without drawing a pension). This suggests that, in the Netherlands, atypical pathways to retirement are not occurring to a significant degree. It appears that this mostly happens if any loss in income is compensated by receiving pension – and thus financial security may be a prerequisite for atypical pathways – while those who are less dependent on a salary may try alternatives to full-time permanent employment. Alternatively, as mentioned before, part-time work or non-standard employment can be used to compensate for a pension that is considered insufficient.
Moreover, our findings on the existence of de-standardised pathways of retirement in the Netherlands may be seen as partial support for the considerations of Moen (Reference Moen2016) who suggests that older workers may be reinventing retirement within paid work. However, those who follow atypical pathways to retirement may not all be ‘reinventing’ retirement. Some individuals may continue working beyond the ‘traditional’ retirement age out of financial necessity as they cannot afford to retire. Another group concerns individuals who actively seek different forms of work and have more agency in crafting their jobs at the end of their working lives. Arguably, the phrase ‘reinventing’ retirement is less appropriate for those forced to continue working and more appropriate for those who are not. This implies that the extent to which baby boomers in the regulated context of the Netherlands are ‘reinventing’ retirement is probably lower than the 12 per cent that was reported earlier.
To what extent are older workers likely to do neither?
Finally, our third finding reveals that approximately half of post-war baby boomers (mostly women) in the Netherlands follow neither cliff-edge nor atypical retirement pathways. These – predominantly female – older workers experience long spells outside the labour market as either inactive or benefit recipients for several years up to SPA. Most likely, these older individuals were less consistently involved in paid work earlier in their lifecourse. This is consistent with previous literature in which late-life employment and retirement are strongly influenced by gender (Turek et al. Reference Turek, Henkens and Kalmijn2024), with women having a lower pension relative to men owing to the typical female work–family lifecourse of care-giving and weaker attachment to the labour market (Madero-Cabib and Fasang Reference Madero-Cabib and Fasang2016). We therefore add to the the need for a gender-sensitive understanding of the pathways to retirement in the context of extended working lives, especially in traditional breadwinner societies such as the Netherlands.
Limitations
The unique contribution of our study lies in locating the differences in pathways to retirement based on the combination of relevant labour market aspects. Nonetheless, there are some limitations. First, we had only yearly information on the self-employed older workers. This means we may have underestimated the movements of older workers in and out of self-employment. Second, we explored variation in retirement pathways only from a labour market perspective. However, a part of what Moen (Reference Moen2016) defines as reinvention of retirement also includes taking an active role in community service and volunteer work. Owing to the lack of appropriate data, we cannot include non-paid work, and therefore we may be underestimating the degree to which older workers are ‘reinventing retirement’.
Conclusion
Notwithstanding these limitations, in our exploratory approach we were able to discern various types of ‘atypical’ retirement within the labour market, such as partial retirement and moving to non-standard employment. As argued previously, an advantage of viewing de-standardisation of retirement as a process having several overlapping aspects is that it allows for a more nuanced picture of retirement (Andrea et al. Reference Andrea, Eisenberg-Guyot, Oddo, Peckham, Jacoby and Hajat2022; Leinonen et al. Reference Leinonen, Boets, Pletea, Vandenbroeck, Mehlum, Hasselhorn and De Wind2022). We further establish that a large group of baby boomers in the Dutch context were outside the labour market well before SPA, even though it was increased beyond 65 for most older workers in the study. To the extent that previous research mainly focuses on individuals in the labour market, earlier findings may have overestimated the effect of increases in SPA and the degree to which individuals are extending their working lives. However, it is worth noting that the participation of women in the labour market has increased in more recent cohorts (Balleer et al. Reference Balleer, Gomez-Salvador and Turunen2014), which may have implications for their later working lives as well. Within the labour market, the main way in which older workers seem to continue working appears to be through working next to receiving pension. Currently, however, complete withdrawal seems to be more common than partial retirement.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0144686X25000029.
Acknowledgements
Previous versions of this article were presented at the RC28 Spring Meeting at Sciences Po in Paris, 24–26 May 2023. We are grateful to the participants of this conference for helpful comments and suggestions. We also thank Statistics Netherlands (CBS) for providing the data for this research. Furthermore, we wish to express our gratitude to Wendy Smits (Statistics Netherlands) for her constructive feedback on earlier versions of this article, as well as to the anonymous reviewers. All remaining errors are our own.
Author contributions
All four authors contributed substantially to all phases of the study and article and have approved the submission of this article.
Financial support
This article is part of the ‘DYNANSE: Righting the Wrongs. A Life Course Dynamics Approach for Non-Standard Employment’ project, which has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement no. 864471).
Competing interests
All four authors declare no competing interests.
Ethical standards
All four authors have adhered to the data protection standards set by Statistics Netherlands.
Open access
