Why are there differences in retirement age preferences between job categories? : the explanatory role of wealth, health, late-career work disengagement, and job satisfaction

(1)

Why are there differences in retirement age preferences between job

categories?

The explanatory role of wealth, health, late-career work disengagement,

and job satisfaction.

10

th

_{of July 2017}

Master Algemene Sociologie

Written by

Jip Teegelbeckers (10829954)

First readers

Kène Henkens & Marleen Damman

Second reader

Levi van den Bogaard

University of Amsterdam

Netherlands Interdisciplinary Demographic Institute

(2)

Preface

To find a research subject is for me always a hard tasks. But, after many contemplations about the reading material I have read during my master courses, I started to gain interest in the Dutch pension system. And, although within those courses the pension system was mainly linked to the formation of the welfare state, I wanted to give my own twist on the subject. Also, since I did my bachelor thesis in Applied Psychology, I thought I could combine my bachelor insights with the sociological knowledge I have gained during my master and pre-master period. But, when starting this research, I had only general knowledge about the pension system and it’s psychological, sociological and economic fundament. So, after brainstorming with my former supervisor, Levi van den Bogaard, my knowledge and interest in the subject started growing. Without Levi’s insights and the interesting articles he gave me I would not have been able to gain a better understanding of the academic pension literature.

However, after a few weeks in the master thesis process, Levi got a new job. Thereafter, Marleen Damman was assigned to me, as my new supervisor. Quickly it became possible to meet with her, and my first reader, Kène Henkens. They helped me to steer my research in new directions and gave it more focus. Also, they made it possible for me to work with a new dataset, recently collected by the Netherlands Interdisciplinary Demographic Institute (NIDI). I am very grateful for that opportunity, since it made my research more contemporary and meaningful. Also, since the dataset was only for internal use within the NIDI, I had to travel several times to the Hague to work there at the research institute. Not only did this increase my motivation to complete a purposeful study, but it also showed me what a pleasant working environment such a research institute is.

I especially want to thank Marleen for her helpful and beneficial supervision. I really enjoyed our conversations and discussions about the subject. Thanks to her I gained more knowledge and a better understanding about the work to retirement process. Also, her feedback gave me insights on how to write an academic study, and how to structure my research reports. So, briefly said, I have learned a lot, and I could not have imagined beforehand that I was able to participate in such and interesting master thesis project.

Jip Teegelbeckers Utrecht, July 2017

(3)

WHY ARE THERE DIFFERENCES IN RETIREMENT AGE PREFERENCES BETWEEN JOB CATEGORIES?: THE EXPLANATORY ROLE OF WEALTH, HEALTH, LATE-CAREER WORK DISENGAGEMENT AND JOB SATISFACTION.

Amount of words: 16.274 (excluding preface, abstract, table of contents, literature and appendix) Student number: 10829954

Abstract

Objective: Since the aging of the Dutch population, the pension system came under pressure. Already policy reforms tried to release this pressure in several ways with the intention to increase the labor force participation of older employees and postponing their retirement age. Already academic studies analyzed the differences of this timing of the retirement age between occupations. However, within these studies it is still under-researched what the underlying factors are that facilitate the relation between the job category wherein an employee’s occupation is classified, and his or her preferred retirement age. Therefore, the aim of this study is to analyze both the influence of economic factors (wealth and health) and psychological factors (late-career work disengagement and job satisfaction) on that relation.

Method: Using data collected in 2015 among Dutch older employees (N = 4999) a mediation analysis was performed to measure if wealth, health, late-career work disengagement and job satisfaction explain the underlying mechanism of the relation between one’s job category and the preferred retirement age.

Results: In line with the hypotheses, a positive relation between one’s job category and the preferred retirement age was found. Herein, mainly employees who practice non-manual labor prefer to retire relatively later than employees who practice manual labor. These differences are inter alia mediated by wealth, health and job satisfaction. Thus, when employees are in good health and are satisfied with their job, they also tend to postpone their retirement. And, employees with more wealth are inclined to quit working at an earlier age. Nevertheless, the differences in preferred retirement age between job categories are, in this study, mainly influenced by late-career work disengagement. Hence, when people are more disengaged with their job in proximity of their retirement, they also prefer to retire relatively early. This mainly occurs at semi-skilled or unskilled occupations, which are ranged lowest on the axis of job categories.

Discussion: To gain a broader understanding of the retirement age preferences between job categories, it is best to not only reckon economic arguments about wealth and health, but to predominantly take into account work-place related psychological factors, especially late-career work disengagement.

Key words: Retirement age preference – Job categories - Late-career work disengagement – Job satisfaction – Economic factors

(4)

Retirement is always in revision (Ekerdt, 2010) and the Dutch pension system (Box 1.1.) is currently under pressure as a result of different economic trends, like low rent and unsure investment income (Been et al., 2014). But, this pressure it is mainly caused by demographic changes, namely the decline of the amount of children per family since the anti-conception pill (Gjaltema & Broekman, 2001), the baby boom generation (Clark, 2003) and the rise of the life expectancy. A consequence of the ageing of the population is that it becomes increasingly difficult for younger employees to satisfy the basic old pension demand of the elderly. Henkens (2011) argues that this aging is one of the important social issues of the 21st_{century. Whereas}

critics warn against the possibility of old-age poverty because pension benefits can get cut (Taylor-Goodby, 2004).

To lower the pension costs and to enlarge the pension funds, policy reforms tried to increase the labor force participation of older workers. First, the early retirement arrangements (Vervroegde uittreding / VUT) and pre-pension were abolished in 2005, with the idea that people would retire later than the age of 60, which was previously the average retirement age (Heemskerk, 2013). This early retirement arrangements still existed for people born before 1950. However, people who were born later had to pay a premium for these arrangements, whereby it became less attractive for them to retire early. Second, another known policy reform to create a foundation for prolonged labor force participation, is the gradual rise of the basic old age pension age (AOW-leeftijd) (De Beer

et al., 2017; Goudswaard, 2010). This was done in small steps with one or two months each year since 2013, but shall increase with four months each year between 2018 and 2021. Also, there is further upward development of the retirement age planned which is linked to the average life-expectancy from off 2022. Third, there was a switch from a final wage scheme (eindloonregeling), whereby the final pension income depends on the employees final wage in his career, to a middle wage scheme (middelloonregeling), whereby the final pension depends on the amount of years in wage labor and the salary rate during that period

Box 1.1. The Dutch pension system

The Dutch pension system has three tiers. First, there is the public basic old age pension (AOW). All Dutch residents are qualified to receive such a benefit. In this system the working populations pays the expenses of the basic old age pensions of the current retirees. This is called a pay-as-you-go system (i.e. omslagstelsel). Second, there are collective earnings-related occupational pension plans. These plans are accommodated by a pension fund. Although these occupational pensions are not mandatory, 90% of the Dutch employees have such a pension scheme at their employer. Third, there are individual private pension savings. Mostly self-employed workers and employees in industries without a pension scheme use this third tier, for instance, to supplement their personal pension savings (Damman, 2014).

(7)

(Heemskerk, 2013; Kennis et al., 1997). Hereby, it was the idea that people who formerly did not have to work for the whole forty years, now have to, to earn a full pension.

The effects of these policy reforms become already clear from data from Statistics Netherlands (CBS, 2016). It shows that there is an upward retirement age trend since 2006 (figure 1). Also, the bruto and netto labour force participation among older workers (55+) has seen an obvious rise1_{(CBS, 2017). However, it is still}

the question if the rise of the retirement age, and the rise of labor force participation of elderly, is only the result of the above named policy reforms, or also due to perceptions of work and retirement of near pensioners (Van Dalen & Henkens, 2004). Therefore, this research analyzes the job perceptions and retirement age preferences of different birth cohorts of Dutch near pensioners, born between 1950 and 1955. Notably, the youngest of these cohorts receive their public basic old age pension at an increasingly later age (figure 2), and therefore have to work longer. The question is, however, when and why these younger cohorts can and want to work longer (Heemskerk, 2013; Polat et al., 2012). The prolonged labor force participation of older employees, whereby both the ability and the willingness to work are improved, is called ‘sustainable employability’2_{(Schaufeli, 2011). It is expected that, whether or not these employees can and want to work}

longer, differs per job category and occupation. For instance, people who practice physical demanding manual work (for example, in mechanic and repair) prefer to retire earlier than people with sedentary jobs (for example, administrative jobs), because their health deteriorates faster (Chung, et al., 2009). Therefore, the research questions of this study are:

(i) Are there differences in the preferred retirement age of near pensioners from different job categories? And (ii) can these differences be explained by factors from an economic perspective (health and wealth) and factors from a psychological perspective (late-career work disengagement and job satisfaction)?

1_{This is also due to population growth.}

2_{‘Sustainable employability’ mainly includes the vitality, health, productivity, optimal utilization of competences,} skills, flexibility and mobility of older workers. Hereby, they are able to extended their working lives, without invoking too much on younger generations (Schaufeli, 2011).

(8)

This study has its focus on retirement timing (i.e. the age of retirement). This is affected by different factors, and has already been widely researched within academic pension-literature. For example, research about: the effects of employers’ actions and attitudes (e.g. Billet et al., 2011; Van Dalen et al., 2009), financial incentives (e.g. Laitner & Sonnega, 2013), health (e.g. Van Den Berg et al., 2009), social norms and values about rights and obligations for (older) employees (e.g. Van Dalen & Henkens, 2005), social embeddedness in the organization (e.g. Ekamper, 2006), the ‘social contract’ an employee has with the organization (e.g. Henkens, 2011; Lazear, 1979), the labor productivity of older employees (e.g. Van Dalen et al. 2010), the composition and social embeddedness in the household (e.g. Henkens 1999; Henkens & Van Solinge, 2003), circumstances in the individual lifecycle (e.g. Henkens, 2011) and other work-related factors (e.g. Henkens & Van Solinge, 2014). Also, a social stratification perspective about retirement timing has been applied in a few studies (Blossfeld, et al., 2006; Blossfeld et al., 2011; Radl, 2013; Visser et al. 2016). It is herein argued that self-employed and service workers retire later than employees in the working class. Also (un)skilled manual workers, manual supervisors and lower sales and service workers have a risk of early voluntary and involuntary retirement, when compared to the upper and service class workers.

However, these social stratification studies mainly focused on education and social class. And, although these researches did assess social class differences in retirement timing to varying occupations, it payed little attention to the underlying mechanisms concerning the influence of one’s occupation. Nevertheless, research by Visser et al. (2016) did hint towards the idea that economic factors (like savings) and psychological factors (like attachment), may be of influence to differences in retirement timing. However, 58 59 60 61 62 63 64 65 A ve ra ge r et ir em en t ag e Year Source: CBS 2016

Figure 1. Average retirement age per year _{Figure 2. Retirement age per year of birth}

65 and 3 months 65 and 6 months 65 and 9 months 66 and 4 months 66 and 8 months 67 and 3 months January 1st

1950 January 1st1951 January 1st1952 January 1st1953 January 1st1954 January 1st1955

(9)

Visser et al. (2016) did not further study these assumptions. Therefore, this study extends the expectation that different psychological and economic factors intermediate between the differences in preferred retirement age along the range of different job categories of older workers.

Hence, this study contributes to the work field of pension-studies in three ways. First, job categories (and their corresponding occupations) are used as sociological conditions to analyze work related social stratification. These job categories are influenced by job-related and occupational norms, values, beliefs, assumptions and rules, by means of professionalization, socialization and institutionalization (Vollmer & Mills, 1966; Forsyth & Danisiewicz, 1985; Levine & Moreland, 1994; Johnson, 1995; Barley & Tolbert, 1997). The research of job categories as a measure for social stratification, instead of education, elevates the problem that some people have a different intellectual or executive level at the end of their career, than at the beginning. For instance, many employees make career within one company. Educational-level would only analyze one’s intellectual or executive level at the beginning of the career, whereas job categories analyzes them at the end of the career. Second, retirement timing is in this study not seen as an actual behavioral component, but as a behavioral tendency (i.e. preference), as it is used in psychology. The research of preferences, instead of behavior, has as an advantage that it transcends the problem that some employees retire involuntary or without a choice (Van Solinge & Henkens, 2007). Thus, the research of individual preferences about the retirement timing shows the unforced retirement intention, instead of (possible) forced behavior. For example, some employees are forced to retire because their employer or spouse wants them to (Szinovacz & Davey, 2005; Smith & Moen, 1998). A third way this research contributes to the work field, is the analysis of different underlying mechanisms that influence the relation between job categories and retirement age preferences. These underlying mechanism are: wealth and health from and economic perspective, and late-career work disengagement and job satisfaction from a psychological perspective (Polat, et al. 2012; Schaufeli, 2011; Weeden, 2002).

This study is organized as follows: First, in the theoretical chapter (Chapter 2), retirement and retirement timing are defined. Thereafter, the retirement decision making process is described. Then, the economic factors that influence the retirement timing, namely wealth and health, are outlined. Hereafter, the psychological factors, namely late-career work disengagement and job satisfaction, are conceptualized and expounded. On the basis of these conceptualizations, and the influence of these concepts on the relation between job categories and the retirement age preferences, several hypotheses are composed. Second, in the methodological chapter (Chapter 3), the sample size, data collection, used variables and the statistical analyses are described. Third, the results of the analyses are visualized, written down, deciphered and interpreted (Chapter 4). Finally, in the last chapter (Chapter 5), the conclusions of this study are described. Thereafter, this research’ limitations, and suggestions for future research, are noted.

(10)

2. Theory

2.1. Retirement and retirement timing

To understand the relation between retirement age preferences and different job categories, first of all, the concept of retirement itself has to be conceptualized. Denton & Spencer (2009) reviewed and assessed the many concepts and measures that reflected the concept of retirement. They argue that no definition dominates and that the existing definitions are constantly refined and shifted as social conditions and individual expectations change. Its different definitions include: exit from the labor force, bridge employment (e.g. reduced work hours and earnings), pension reception or other retirement income, working hours below a minimum amount, exiting the main career, changing late career employment and self-assessed retirement. In the pension literature, psychologists tend to define retirement more in terms of job or career, whereas economists define it as exit from the labor force and the reception of pensions and retirement income (Beehr & Bowling, 2013). In this research, retirement is defined as the retreat of paid labor and exit from the labor force, whereby an employee receives a pension or other retirement income. Herein, retirement is both the physical and psychological retreat of one’s current occupation.

Next to different definitions about the conceptualization of retirement itself, the concept of retirement timing has also a variety of conceptions, which are used differently in various academic disciplines. Beehr (1986) defined it in the 1980’s in a relative way, as retiring “early”, “on-time” or “late”. Those notions where mostly seen as relative to the eligible pension age or the mandatory age for retirement in relation to when one expected, planned, or preferred his retirement age, or in relation to co-workers, spouse and others in the social surroundings. Thus, this timing depends on economic, sociodemographic, cultural and organizational circumstances, and employer -and country-specific pension and retirement rules and policy (Fisher et al. 2016).

A more specific conceptualization of retirement timing is wielded in the psychological work-field. Herein, psychologists have their focus on the psychological and behavioral processual antecedents, results and reactions of and towards retirement on a micro-level. Hereby psychologists analyze both inter -and intra-individual differences within the social environment (Shultz & Wang, 2011). This includes, among other things, the willingness, confidence, behavioral tendencies and preferences about retirement and the work-to-retirement transition. On the other hand, pension-sociologists and economists do not necessarily study this subjective micro/individual level, but mainly focus on analyzing retirement-related behavior and conditions (Davis, 1998; Kingston, 2000; Lazear, 1986). In this study, however, a psychological retirement

(11)

timing perspective is utilized. Or, more specifically, retirement timing is defined in terms of retirement age preferences.

2.2. The retirement decision making process

This retirement timing (e.g. retirement age preference) is mainly being referred to as the cause of a three-layered decision-making “process” (Feldman & Beehr, 2011). During the first stage retirees are mainly oriented on the future, and employees start thinking about the replacement of their activities and social relations that are currently afforded by their work (Appold, 2004; Dunegan, 1993). In the second stage employees are oriented on the past, whereby they estimate their willingness or reluctance to leave their long-time career, job or occupation behind (Bailey & Hansson, 1995; Moon, 2001; Shultz, 2003). Finally, in the last stage, employees are oriented on the present. Hereby they look at specific actions they need to take, and the resources they need, to make a successful work-to-retirement transition. (Allen et al., 2003; Fronstin, 1999; Herrbach et al., 2009). These stages are not necessarily distinct, but can occur simultaneously (Feldman, 1994). However, they do vary across the time period of the retirement decision process (Beehr et al., 2000; Shultz et al., 1998) and across different job categories (Feldman & Beehr, 2011). Moreover, this three-dimensional understanding of the retirement decision corresponds with what Radl (2013) describes as a mechanism of push and pull factors. This mechanism is primarily based on structural constrains and rational choice theory (Shultz et al., 1998). These theories presume that individuals retire at a certain moment when benefits of retirement and the abandonment of work are maximized, and the costs of these actions are minimized. Herein, push factors are negative considerations about the current situations of older works, whereas pull factors are seen as positive considerations about the future situation.

The different studied push and pull factors of the retirement transition vary between different academic disciplines. Pension-sociologists and economists mainly used health and wealth as push and pull factors that influence the work-to-retirement transition. Hence, several studies indicated wealth and health as the most strongest predictors of the retirement preference (Beehr et al., 2000; Talaga & Beehr, 1989). Concerning these factors it is argued that people can work longer if they have a stable healthy situation and people want to work shorter if they have a wealthy financial situation (Polat, et al., 2012). On the other hand, psychologists research the psychological and behavioral push and pull factors of and towards retirement on a micro-level at the work floor. This includes mainly the relation of individuals towards their work and the work environment. Recent studies included job satisfaction as a psychological push factor for the retirement decision (Mein et al., 2000; Topa et al., 2009). Whereas other research by Damman et al. (2013) included late-career work disengagement as a psychological factor that steers the retirement timing. In the next paragraphs the concepts of wealth, health, late-career work disengagement and job satisfaction are described in relation to different retirement age preferences and differences in job category.

(12)

2.3. Wealth and economic status

Wealth operates as a pull factor for the retirement decision. This is an argument that is mainly consistent with rational choice theory and economic life-cycle models (Wang & Shultz, 2010). Herein, retirement timing is affected by the financial resources that an employee potentially has for his retirement (Beehr, et al., 2000; Kim & Feldman, 1998; Laitner & Sonnega, 2013; Madero-Cabib et al., 2016; Mein et al., 2000; Pienta & Hayward, 2002). The main trend herein is that people with more wealth retire relatively early, and, on the other hand, those who have more to gain by postponing retirement, tend to retire relatively late (Fields & Mitchell, 1984). Behind this trend there are three arguments: (1) people with more wealth would have more to spend during their post-pension leisure-time; (2) people with a higher economic status would appreciate extra income of extended work less than people who stand lower on the economic ladder; (3) people with more wealth have more goods to complement during their retirement years (like sailboats or vacations homes).

Wealth, and it’s influence on the preferred retirement age, differs per job category. For instance, the average economic status of the lower service classes is lower than the upper service classes, which makes early retirement less performable. It is thus expected that people with jobs with the most wealth, like high intellectual headwork and intermediate intellectual headwork, retire relatively early. Whereas people with a lower economic status, like employees within manual labor, retire relatively late (Radl, 2013).

2.4. Health

Health, as opposite to wealth, operates as a push factor for the retirement timing. Older workers can be the most experienced workers within a company, but their health is also the most vulnerable of all employees (Van Den Berg et al., 2010). Becker (1967) describes the concept of health capital (which is linked to his concept of human capital). This influences someone’s productivity and time that someone can invest in their work (Becker, 1967; Ben-Porath, 1967; Grossman, 1999), and thus the level of someone’s employability near their pension age. Poor health, however, makes quitting the labor market rather a necessity than a voluntary choice (Feldman, 1994) and thus pushes employees into retirement. Research by Henkens & Leenders (2010) show that high workload, a lack of challenge, and a lack of social support by managers and colleagues are positively related to health complaints and burn-outs, which are on their turn positively related to a higher intention to retire. Moreover, part of the older workers see retirement as a strategy to invest in their health (Henkens, 1999). Also, longitudinal research by Van Den Berg et al. (2009) has shown that poor health and, somewhat also, physical and psychosocial work load are important parameters for early retirement. Furthermore, Hoeymans (2011) argues that, on average, people who are chronically sick or

(13)

disabled tend to leave the labor market earlier than healthy employees, because they are either being discriminated, or transport and work place facilities do not fit their needs.

Several studies already showed occupational effects on ill-health and consequently the retirement timing. For instance, Vermeer et al. (2014) found that policy debate emphasizes that low-skilled workers cannot be expected to work longer than average, because their heavy work have worsened their health. Also, Ybema & Geuskens (2011) found in their research that employees with occupations that negatively influence their health, for instance in construction and industry, have a tendency to retire early. Additionally, Radl (2013) emphasizes that employees who practice manual labor (both skilled and semi-skilled) prefer to retire earliest, because of their high risk of deteriorated health or to become disabled.

2.5. Late-career work disengagement

Adjoining wealth and health, there are psychological work-place related variables that play a role in whether or not an older employee has a preference for an early or a late retirement age. Research by Callan & Lawrence (2008) concluded that job engagement is positively related to job retention and negative to turn-over. The engagement and investment people do in their job would eventually withhold employees to withdrawal from their work, because the on-the-floor working conditions and tasks support them to keep working (Kahn, 1990). However, it is mainly the opposite concept of job engagement, namely late-career work disengagement, that is focused on in pension literature about retirement timing and retirement age preferences (Damman et al., 2013).

To understand that concept, it is necessary to conceptualize job engagement first. It is described by Kahn (1990) as the “harnessing of organizational members’ selves to their work roles; in engagement, people employ and express themselves physically, cognitively, and emotionally during role performance” (p. 694). Thus, the concept of job engagement has a physical, cognitive and an emotional dimension. Cognitive job engagement is the awareness of someone’s role in the occupation, whereby employees also have what they need to do their work and can develop themselves (Harter et al., 2002b; Luthans & Peterson, 2003). Emotional job engagement is the meaningful connection with colleges, managers, supervisors, clients, costumers and others who are involved in the work sphere. Finally, physical job engagement is the behavior of employees that shows extra effort in their work.

Kahn (1990) argues that people vary in the way they engage in their working role. In these roles people are to an extent involved in their work, committed to their organization, or self-estranged and alienated from their work. The extent of job engagement influences the centrality of the job in a person’s life, the eager to put effort in one’s work and the organization, and the degree of the alienating effects of the social system of the job. Also, when employees maximize their involvement in their working role, they increase their

(14)

vigor, absorption and dedication3_{in their work (Schaufeli & Bakker, 2004; Schaufeli et al., 2002). The}

concept of the working role, and its influence on work-place based behavior, can be traced back to different micro-sociological theories, as described in Box 2.1.

Kahn (1990) describes further that there is a varying degree of attachment and detachment of the specific ‘working roles’ and the ‘self’ as an autonomous person. When there is a separation between an employee’s expected ‘working role’ and his ‘self’, a distance from work and its corresponding role is shown. But when the ‘self’ and the ‘working role’ are connected, the employee is embracing his work. This separation, or connection, with the working role and the ‘self’, is found back in human’s intrinsically contradiction about being a member of a group. On the one hand, people move towards a group to overcome isolation, and on the other hand they move away from it to keep privacy and prevent social engulfment.

People who are in proximity of their retirement tend to detach and disengage themselves from their working role (Damman et al., 2013) because the transition phase from work to retirement is accompanied by its own retirement socialization procedure (Evans et al., 1985). In this retirement socialization procedure a ‘self’ is formed that is connected to retirement and that abandons the ‘working role’. Hence, during this late-career work disengagement people become less likely to engage in different facets of work, like training, new colleges, developments and responsibilities (Henkens & Van Solinge, 2003). So, in sum, when employees approach their retirement, they estrange or alienate themselves from their working activities. Hereby,

3_{This includes inter alia: an abundance of energy, persistence, a sense of significance, pride, challenge, inspiration} and intense concentration by the employee (Schaufeli & Bakker, 2004; Schaufeli et al., 2002).

Box 2.1. Role theory and working roles

Role theory has it’s basis in the dramaturgical theory of Erving Goffman (1978). He describes that social behavior is aimed toward impression management. Hereby people reveal or conceal certain aspects from themselves when interacting with others. In his theory Goffman uses the acting stage as an analogy for the ‘self’ within social interaction. During social interaction, people engage in front-stage behavior, which is behavior aimed at others (i.e. this is the self as if playing out a stage-character). Thus, when engaging in social interaction with others, people act out a certain role. Someone’s social role as a worker becomes more clear in Ralph Turner’s role theory (J. Turner, 1988). Turner draws on the same idea as Goffman of the ‘self’ as “a force guiding social interaction” (ibid., p. 86 ). He argues, however, that behavior is organized into identifiable roles. In this view, people operate on expectations on how to act (Turner calls these “folk norms of consistency”) that “predisposes an assessment of behaviors in terms of their internal consistency and their capacity to signal what role others are playing” (ibid., p. 86). Moreover, this “processes of role-making and role-taking constitute a cycle that reinforces (or fails to do so) self-conceptions, inventories of role-conceptions, cultural frameworks, and folk norms” (ibid., p. 89). Thus, in short: roles are cultural frameworks, ordered by expectations, and lead people to develop, by means of socialization and experience, an inventory of role-conceptions, which represent clusters or behaviors depicting both general classes and more specific types of roles. Concerning work related roles these are, for example: colleague, manager, employee, and so on.

(15)

they do not correspond to the cultural framework and role-conceptions of their job anymore, because their socialization as a retiree elevates their role as an employee. In this way, the ‘working role’ becomes detached from the ‘self’.

Hence, when people reduce their work investments and activities, when they approach the retirement age, they prefer to retire relatively early. It is however expected that the timing of this disengagement, and consequently the preference for the retirement age, differs between job categories (and their corresponding occupations). For instance, employees who gained promotion or whom have improved working conditions, like more autonomy, authority and self-direction, are less disengaged, because these employees have a more meaningful working role and more job opportunities during the preretirement period (West & Nicholson, 1989). Also, cross-sectional analysis by Damman et al. (2013) shows that employees who perform low intellectual manual labor have higher levels of late-career work disengagement. Moreover, supervisors seem to be less disengaged in proximity of their retirement than workers with a non-supervisory role. Consequently, low skilled manual workers are expected to have a preference for a relatively early retirement. And, employees with a higher intellectual or executive position prefer to retire relatively late, because of their lower late-career work disengagement.

2.6. Job satisfaction

The second topic within the scope of psychological job attitudes, next to late-career work disengagement, that influences the retirement age preference, is job satisfaction. And, although it seems that the concepts of job engagement and job satisfaction mostly overlap, these two most be separated. Job engagement is the organizational commitment, involvement and role-taking (concerning the content of the work), whereas satisfaction is the appraisal and affective reaction to one’s job (i.e. the affective psychological perception of work) (Alarcon & Lyons, 2011).

Beehr (1986) argued that negative views and low satisfaction of the job pushes employees towards retirement. Already Fillenbaum (1971) made a similar argument in the early 1970’s, when he concluded in his research that positive satisfying perceptions towards work reflects negative perceptions towards retirement. However, his research mainly included professionals and self-employed. Subsequently, this argument was strengthened in the mid-1970’s by Goudy et al. (1975). They wrote that people who see themselves become incompetent because of their age have a negative satisfactory perception of their work and have a higher intention to retire. Also, Hanisch & Hulin (1990) found in their research that differences in satisfaction of someone’s current job and expected satisfaction of the retirement predicts the age someone prefers to retire. More recent research found similar relations. Mein et al. (2000), found low job satisfaction as an indicator for early retirement for men and women, as well as Topa et al. (2009) who found that job

(16)

satisfaction was one of the strongest predictors of retirement planning activities within a meta-analysis in Europe.

Job satisfaction is an ambiguous concept that has many different approaches within academic literature. It is however one of the most widely investigated subjects about well-being on the work floor and it involves a central role in many theories (Judge & Klinger, 2008). Therefore, it entails also many definitions, conceptualizations, contextualization’s and measurements. In this research the concept of Locke (1976) is used. He outlines job satisfaction as “a pleasurable or positive emotional state resulting from the appraisal of one’s job or job experiences” (ibid. p. 1304). Additional Hulin & Judge (2003) and Eagley & Chaiken (1993) describe a tripartite psychological response towards one’s job: cognitive, affective and behavioral. Cognitive satisfaction has been seen as the cognitive appraisal of the characteristics of one’s job (Judge & Kammeyer-Mueller, 2012). Affective

satisfaction, however, is the experience of positive or negative emotions towards one’s job. Finally, the behavioral component includes someone’s manner of conduct during the job. Also, satisfaction has been related to other psychological factors and mental conditions, like anxiety and self-esteem, which is explained in Box 2.2.

It is expected that employees with an occupation that generates higher satisfaction are inclined to retire relatively late. This was found, for instance, in the study of Fillenbaum (1971). His research shows that satisfaction, and consequently the preferred retirement age, was proximately late among

upper-echelon, self-employed and professional occupations. Moreover, research by Statistics Netherlands (CBS) (Moonen, 2011) found that managers and employees with high intellectual occupations are the most satisfied with their work. Hence, it is expected that they prefer to retire latest. Whereas manual workers, and

Box 2.2. Satisfaction, anxiety, self-esteem and the retirement decision Steger & Dik (2009) describe that employees are motivated to work because they get a sense of satisfaction, purpose and meaning out of it, when their psychological needs are met. Consequently, when psychological needs are met, people also report less anxiety on the workplace (Deci et al., 2001). Hence, in the view that work alienates and retirement liberates, retirees experience less anxiety than unsatisfied workers (Drentea, 2002). These pre -and intra-retirement anxiety differences are also found in an experimental study by Thériault (1994). Thus, work dissatisfaction can lead to anxiety (Faragher et al., 2005), and the perspective of a less anxious well-being during retirement may lead to earlier withdrawal from work.

Moreover, workers who experienced higher satisfaction also experienced higher self-esteem (Ryan et al., 1993), a concept that is seen as the opposite of anxiety (Deci et al., 2001). Thus, when psychological needs are met, employees become more satisfied, and thus have more self-esteem to accomplish what they want to do in their work. It is therefore assumed that employees who are more satisfied, are afraid to lack self-esteem during retirement, wherefore they are disposed to retire later (Rosenberg & Pearlin, 1978). Hence, it is also found that some retirees enter bridge employment to regain a sense of self-esteem and satisfaction which may have been lost since retirement (Mor-Barak, 1995; Dendinger, 2005).

(17)

employees with elementary jobs, are the least satisfied with their work. It is therefore expected that they prefer to retire earliest.

2.7. Hypotheses

As argued above, the differences in retirement age preference between job categories are expected to be influenced by economic factors (wealth and health) and psychological work-place related factors (late-career work disengagement and job satisfaction). These hypotheses are stated below.

Hypotheses 1: Economic factors

Hypothesis 1a: Employees with a higher intellectual or managerial non-manual job have more wealth than employees with an occupation that is ranged lower on the scale of intellect and level of management, and employees who practice manual work, and consequently prefer to retire relatively early.

Hypothesis 1b: Employees with a higher intellectual or managerial non-manual job have better health than employees with an occupation that is ranged lower on the scale of intellect and level of management, and employees who practice manual work, and consequently prefer to retire relatively late.

Hypotheses 2: Psychological factors

Hypothesis 2a: Employees with a higher intellectual or managerial non-manual job have lower late-career work disengagement than employees with an occupation that is ranged lower on the scale of intellect and level of management, and employees who practice manual work, and consequently prefer to retire relatively late.

Hypothesis 2b: Employees with a higher intellectual or managerial non-manual job are more satisfied with their job than employees with an occupation that is ranged lower on the scale of intellect and level of management, and employees who practice manual work, and consequently prefer to retire relatively late.

In short, a higher occupation on the axis of job categories leads to more wealth, better health, lower (late-career work) disengagement and a higher (job) satisfaction. Consequently, more wealth leads to a preference for an early retirement. Better health leads to a later retirement age preference. High disengagement leads to a preference for an early retirement age. And, finally, high satisfaction leads to a preference for a later retirement age. This is summed in figure 3.

(18)

From the above hypotheses and flowchart it becomes clear that most factors (except wealth) point in the direction that, the lower one’s occupation is ranked on the axis of job categories, the earlier that employee prefers to retire. This assumption is stated in hypothesis 3.

Hypothesis 3: Employees within non-manual labor categories prefer to retire later than employees within manual labor categories. Also, employees within a job category classified by a higher intellect or a higher level of management prefer to retire later than employees who score lower on these parameters.

Additionally, the job categories that are used in this research are based on that of an ISSP87 analysis by De Vries & Ganzeboom (2008), and were manually adjusted to make it more feasible for the analysis (described in paragraph 3.2.). To give a visual overview of these job categories and their corresponding expected retirement age preference, the third hypothesis 3 is summed in table 1.

Table 1. Expected retirement age preference per job category

Job categorya _{Expected retirement age preference}

Higher intellectual or executive Latest

Intermediate intellectual or executive Late

Other non-manual work Average

Skilled executive manual work Early

Semi-skilled or unskilled manual work Earliest

a _{job categories are based on ISSP87 analysis by De Vries & Ganzeboom (2008) and manually adjusted}

(19)

3. Research method

3. 1. Sample

The sample of the survey for this research is taken from the NIDI pension panel study 2015 (wave 1) by the Netherlands Interdisciplinary Demographic Institute. The sample was drawn from the administration of the three biggest occupational pension funds in the Netherlands: ABP (civil servants), PfZW (social work and care) and BpfBouw (construction). The method was a random sample with a stratified design. During the first step a sample was drawn from the organizations that were covered by the three occupational pension funds. In order to uphold variation concerning organizational size, the organizational sample of each pension fund included N = 300 small organizations, N = 200 medium sized organizations, and N = 50 large organizations. During the second step a random sample was drawn from the target population (older workers aged 60 years or older; years of birth between 1950-1955) who worked at least 12 hours a week (which include them in the labour force4_{). This finale sample supported variation in manual and non-manual labor,}

gender, age, education and job category. The final response of this sample was 44,2% (ABP 47,8%, PfZW 41,6%, 40,2% BbfBouw).

This first wave of this sample included both respondents who filled in a short survey (N = 499) and respondents who filled in a long survey (N = 6280). The short survey was created with the intention that it would increase the response. The allocation of the short survey among the respondents in the sample was done at random. However, this short survey lacked crucial information that is required for the analyses of this research. The respondents who filled in that survey are therefore eliminated from the final sample. Also, cases with missing values about the questionnaire version are removed (N = 14). The sample was further shortened by excluding cases wherein information was lacking about the independent, dependent and mediating variables, or respondents who answered ‘I don’t know’ at one of the questions of these variables. This resulted in a final analytical sample size of N = 4999.

4_{This is the definition that Statistics Netherlands (CBS) used for inclusion into the labour force before 2015 (when} the survey was put out). The definition is: “persons who have paid work for twelve hours or more (employed labour force (12-hours threshold)), or - don’t have paid work or for less than twelve hours per week, recently looked for work for twelve hours or more per week and are directly available for it (unemployed labour force (12-hours-threshold))” (CBS, 2017).

(20)

3.2. Measures

3.2.1. Dependent variable

The dependent variable, retirement age preference, was examined by asking the respondents during the data collection at what age they prefer to retire, both by year and month (Question 5: If you had a complete say in matters, what would be (or would have been) your age of retirement?). The NIDI recoded this into a variable that indicates the preferred retirement age by year, and the given month as a number with two decimals after the comma. Some respondents answered that they prefer to retire below the age of 60 (N = 144; 2,2%). However, since the minimum age of the respondents is 60, it is impossible to retire below that age. Also, it is not feasible for employees to retire earlier than the age of 605_{. Those respondents are therefore}

merged with the precise category of 60 years. Moreover, N = 3 (0,1%) respondents answered that they prefer to retire at an age above 75. Since those are outliers, whereby it is arbitrary if they can work that long, they are merged with the precise category of 75. As shown in the descriptive table (table 2) the mean retirement age preference is 63,5 (SD = 2,12). Most respondents want to retire between the age of 60 and 64 (N = 2214; 44,3%). A relatively large amount of the respondents has a preference to retire at the exact age of 65 (N = 1653; 33,1%), which was the common retirement age for employees born before April 1952, whereby they would receive their full basic old age pension (AOW). Furthermore, N = 650 (13%) want to retire at the age of 60 (or lower) and N = 472 (9,4%) of the respondents prefer to retire between the age of 66 and 70. Finally, a relatively small amount of N = 10 (0,2%) prefers to retire between the age of 70 and 75 (or later). 3.2.2. Independent variable

The independent variable, job category, was measured by asking the respondents in what category they think their occupation is grouped (Question 38: In which category could your job or profession be grouped?). This closed question is based on an ISSP87 analysis by De Vries & Ganzeboom (2008) and the answers were provided with example professions. These answers were: (1) higher intellectual or free profession6_(N

= 431; 8,6%), (2) higher executive profession7_{(N = 651; 13%), (3) intermediate intellectual or free}

profession8_{(N = 1884; 37,7%), (4) intermediate executive or commercial profession}9_{(N = 356; 7,1%), (5)}

other non-manual work10_{(N = 809; 16,2%), (6) skilled and executive manual work}11_{(N = 241; 4,8%), (7)}

5_{This is inter alia due to the abandonment of the early retirement arrangements (VUT) and the rise of the basic} public pension age (AOW-leeftijd).

6_{Examples were: architect, doctor, scientist, teacher higher education, engineer} 7_{Examples were: manager, director, executive in big business, high government official} 8_{Examples were: teacher, artist, nurse, social worker, civil servant}

9_{Examples were: chief representative, department manager, shop owner} 10_{Examples were: administrative assistant, book keeper, salesmen, home help} 11_{Examples were: motor mechanic, foreman, electrician}

(21)

semi-skilled manual work12_{(N = 525; 10,5%), (8) unskilled and experience manual work}13_{(N = 102; 2%)}

or (9) agricultural manual work14_{(N = 8)} 15_.

Thereafter, job categories were merged. This was done, first of all, for practical reasons, to make the analysis more feasible and orderly. Second, it was done to overcome small N problems. For instance, the category ‘unskilled and experienced manual work’ only had an N = 102 (2,0%). Third, a large distinction between types of job categories would distract this research from its main interest, namely the differences in retirement age preference between manual and non-manual labor, and higher intellectual and executive and lower intellectual and executive labor. Subsequently, merging job categories leads to more focus on that subject.

12_{Examples were: driver, factory worker, carpenter, baker} 13_{Examples were: cleaner, packer}

14_{Examples were: farm laborer, farmer}

15_{No percentage of the category ‘agricultural manual work’ is given, since these ten respondents were deleted} from the final sample.

Table 2. Descriptives

Minimum Maximum Mean Standard deviation

Dependent variable

Preferred retirement age 60 75 63,5 2,12

Independent variable

Higher intellectual or executive 0 1 0,22 0,41

Intermediate intellectual or executive 0 1 0,45 0,50

Other non-manual work 0 1 0,16 0,37

Skilled executive manual work 0 1 0,05 0,21

Semi-skilled or unskilled manual work 0 1 0,13 0,33

Mediators

Wealth in € 0,05 5 1,65 1,50

Health 1 5 3,22 0,87

Late-career work disengagement 1 5 2,78 0,71

Job satisfaction 1 7 5,33 1,03

Control variables

Age 60 65 62,02 1,60

Gender (female = 1, male = 0) 0 1 0,43 0,50

Partner status (partner = 1, single = 0) 0 1 0,82 0,38

Working hours 12 80 32,09 8,49

(22)

The merges are as follows: ‘higher intellecual or free profession’ and ‘higher executive profession’ were merged as ‘higher intellectual or executive’ (N = 1082; 21,6%); ‘intermediate executive or free profession’ and ‘intermediate executive or commercial profession’ were merged as ‘intermediate intellectual or executive’ (N = 2240; 44,8%); ‘other non-manual work’ was not merged with another group (N = 809; 16,2%); ‘skilled and executive manual work’ was also not merged (N = 241; 4,8%); finally, ‘semi-skilled manual work’ and ‘unskilled and experience manual work’ were merged as ‘semi-skilled or unskilled manual work’ (N = 627; 12,5%). For the upcoming analysis, when comparing job categories, the ‘higher intellectual or executive’ category is taken as the reference group.

3.2.3. Mediators

The first mediator, wealth, was measured with a closed question by asking: how large do you estimate your total wealth (own house, savings, stocks, etc. minus debt/mortgage) to be? (Question 134). Choices out of: (1) less than 5.000 euros, (2) between 5 and 25 thousand euros, (3) between 25 and 50 thousand euros, (4) between 50 and 100 thousand euros, (5) between 100 and 250 thousand euros, (6) between 250 and 500 thousand euros, (7) more than 500 thousand euros. To give a better interpretability within the analysis the wealth answer categories were recoded to absolute numbers, and again recoded by dividing them through hundred thousand, to make the coefficients in the upcoming regression output more readable. This recoding is shown in table 3. With this measurement, the mean wealth of the respondents is M = 1,65 (SD = 1,50).

Table 3. Recoding wealth

Answer category Absolute numbers Divided by 100.000

Less than 5.000 euros €5.000 0,05

5.000 – 25.000 euros €15.000 0,15

25.000 – 50.000 euros €37.500 0,375

50.000 – 100.000 euros €75.000 0,75

100.000 – 250.000 euros €175.000 1,75

250.000 – 500.000 euros €375.000 3,75

(23)

The second mediator, subjective health, was measured by a 5-point scale with the question: how would you characterize your health in general? (1 = excellent, 5 = very poor) (Question 123). This is a widely used measurement tool within pension-related research (for instance in Damman et al. 2013). The coding of subjective health was reversed so that the highest score represents excellent health, and the lowest score represents very poor health. On average, the respondents scored a M = 3,22 (SD = 0,87).

Third, the concept of late-career work disengagement was computed by merging several statements which were measured on a Likert-scale (1 = completely agree, 5 = completely disagree), as was also done by Damman et al. (2013). This scale was recoded, so that 1 resembled ‘completely disagree’ and 5 resembled ‘completely agree’. The statements contained: the keeping up of developments (Question 46), the participation in new courses (Question 52), the idea whether or not the company should assign new responsibilities to younger employees (Question 55), the effort in new colleagues (Question 63) and the attitude towards reduced working hours (Question 68). A reliability analysis (Cronbach’s alpha) was performed to examine the internal consistency of these statements. Table 4 shows the item-total correlations and the alpha (α) if one of the items were deleted. From those scores it derives that the highest alpha (α) can be reached if all items are used. Therefore, no items were deleted when computing the variable. Since the Cronbach’s alpha of this final variable is α = 0,7 it can be seen as an acceptable measurement tool. Subsequently, this new variable represents the average score of the respondents on all of these statements. The mean of this new variable is M = 2,78 (SD = 0,71).

Table 4. Cronbach’s alpha (α) late-career work disengagement

Itema _{Alpha (α) if item}

deleted

Item-total correlation Q46: I do not keep up as well with the latest developments in my field

as I did five years ago

0,64 0,45

Q52: They should no longer ask me to participate in new courses 0,62 0,51

Q55: I think they should assign new responsibilities to younger persons 0,62 0,51 Q63: I put increasingly less effort in getting to know new colleagues 0,68 0,37 Q68: I use every possibility to reduce the number of hours I work 0,66 0,42

Standardized Cronbach’s alpha (α) 0,70

(24)

The final mediator, job satisfaction, can be experienced at different levels, and thus measured at different levels (Brief, 1998). First of all, facet job satisfaction can be examined by using the Job Descriptive Index (JDI) measurement tool (Smith et al., 1969), which includes job satisfaction about different facets of: work, supervision, coworkers, pay and promotion. Second, job satisfaction can be measured as a global job attitude (Brayfield & Roth, 1951). With this tool the overall job satisfaction is examined, and it captures the encompassing aspect of job satisfaction among an aggregated variety of job characteristics (Judge & Kammeyer-Mueller, 2012). In this research, the latter measurement of job satisfaction is used, because it is the goal to analyze an overarching level of satisfaction about one’s job. In the questionnaire this question falls within a 7-points Likert-scale measured section with the question: “How satisfied are you with the following aspects of life?” (Question 43) (1 = extremely satisfied to 7 = extremely dissatisfied). The life-aspect ‘your work’ was then taken as a measurement for global job satisfaction (statement G). The scale was then recoded so that a maximum score of 7 represented extremely satisfied and the minimum score of 1 represented extremely dissatisfied. The descriptive table (table 2) shows that, on average, people are M = 5,33 (SD = 1,03) satisfied.

Finally, it is analyzed if there is a cohesion between the mediators. For instance, it has been argued that job engagement has been correlated to health because it limits stress, burnouts and other health complaints (Demerouti et al., 2001; Schaufeli & Baker, 2004; Maslach & Leiter, 1997). Also, other research showed that job satisfaction was positively correlated with psychological health (Faragher et al., 2005). Bad health, on its turn, has been associated with low economic status and income inequality (Rehkopf et al., 2011; Kennedy et al. 1998). Also Alarcon & Lyons (2011) argued that engagement and satisfaction are interrelated, because a positive work environment and work content is positively related to a beneficial emotional perception of one’s work (see also Harter et al., 2002a). Table 5 shows the correlations between the mediators. From this table it can be read that there is a significant correlation between all mediators. However, these are relatively low. For instance, although the concepts of late-career work disengagement and job satisfaction significantly correlate, they do not completely coincide. Therefore, it is assumed that the mediators operate autonomously, and will thus be used so in the analyses.

(25)

3.2.4. Control variables

The analyses are first controlled for gender (Question 2: Are you a man or a woman?) (coding: male = 0, female = 1). Because gender has an influence on one’s life course and thus influences retirement timing (Moen, 1996). For instance, women are inclined to take on caregiving roles, and thereby their propensities for paid work varies during their life. Another example is a research by Danø et al. (2005) who found that healthy women are predisposed to retire relatively early. Also, other researchers found that women are more likely to retire early, which was inter alia related to gender differences in marital status, income, wealth, education and employment status (Chan & Stevens, 2004; Raymo et al. 2011). Of the respondents N = 2847 (57%) were male, and N = 2152 (43%) respondents were female.

Second, the analyses are controlled for age. This is done because the Dutch retirement age is increasingly rising. So, people who are younger receive their basic old age pension later, and this can influence their retirement preferences. The “age variable” was constructed by subtracting the respondents birth year (Question 1: In what year were you born?) from 2015 (the year that the survey was put out). However, this age variable can show a small deviation from reality, because it was not the exact birth date (by year and month) that was asked. Therefore, respondents who’s birthday is later than the emission of the survey are in this study a year older than in reality. On average, the respondents are M = 62,02 (SD = 1,60) years old. Third, partner status was taken in as a control variable, because several studies have shown that retirement is a decision that is made by both partners (Gustman & Steinmeier, 2000; Ho & Raymo, 2009). Thus,

Table 5. Correlation between the mediators

Correlation coefficients

Wealth Health Disengagement

Health 0,16**

Late career-work disengagement -0,10** -0,22**

Job satisfaction 0,10** 0,29** -0,33**

Note: The correlation coefficient that is used is Pearson R ** p < 0,01; * p < 0,05

(26)

whether or not someone has a partner or spouse can influence the retirement decision in different ways. For example, older workers without a partner tend to delay their retirement compared to workers who do have a partner (Van Solinge & Henkens, 2014). The question about someone’s partner status was: “Do you have a partner?” (Question 94). And the answer categories were: (1) Yes, I am married, (2) Yes, I cohabit with a partner, (3) Yes, I do have a partner, but we do not live together, (4) No, I am single. The first three answer categories were merged as the category: “partner” (coding = 1) and the last category was named “single” (coding = 0). From the respondents N = 4111 (82,2%) had a partner, and N = 889 (17,8%) were single. Finally, the analyses are controlled for working hours (Question 32: How many hours do you work on average?), because this can have an influence on an employee’s wealth. For instance, when the respondent has much wealth, but low working hours, it might be that he or she has another revenue source than his or her regular job. Also, working hours can influence retirement behavior (Gielen, 2008). For instance, a lower amount of working hours can discourage employees to postpone their retirement. Furthermore, since the survey only involved employees who are part of the labor force16_{, they have to work at least twelve hours a}

week. Therefore, the N = 44 respondents who answered that they work less than that amount of hours are coded as working twelve hours a week. The respondents worked a weekly average of M = 32,09 (SD = 8,49) hours.

3.3. Analyses

The research of this study involves a mediation analysis, as described by Baron & Kenny (1986), to analyze if the economic and psychological factors operate as an underlying mechanism between the relation of job categories and retirement age preferences. The mediation analysis that Baron & Kenny describe is an expansion on the causal analysis. However, in this study, it cannot be determined that one variable causes another17_{, but only if there is a cohesion between two variables (variable X and variable Y). Herein, the total}

effect of X on Y (or vice versa) is called c. This is shown in figure 2.

However, the expected relation between X and Y may be utilized through one are more other variables. These are called mediating variables (M). Figure 3 shows this mediation analysis. In this figure, path c’ is called the direct effect, and the path that runs from a to b through M is called the intervening or mediation

16_{As defined by Statistics Netherlands in 2015.} 17_{See also the analytical limitations in chapter 5.2.2.}

(27)

process. When variable X does not longer effect Y (or vice versa) directly after M has been controlled, it is called complete mediation. During complete mediation path c’ becomes zero.

In this research, the independent variable ‘job category’ is the X variable, and the dependent variable ‘preferred retirement age’ is the Y variable. The variables wealth, health, (late-career work) disengagement and (job) satisfaction are expected to be the mediating variables (M1 to M4). The mediation analysis for this

research is shown in figure 4.

There are four steps during a mediation analysis. First, there must be relation determined between the X-variable and the Y-X-variable. In this research, the relation between the job categories (X) and the preferred retirement age (Y) will be analyzed via an ANOVA-analysis (table 6). During this ANOVA-analysis, group means between the job categories of the preferred retirement age are compared. An F-test and post hoc analyses are used to analyze if these mean differences are significant. During the second step it is analyzed if the X-variable has a relation with the mediating variables. This is also done via multiple ANOVA-tests (table 6).

Figure 3. Mediation analysis

Figure 4. Mediation analysis for the relation between one’s job category and the preferred retirement age

(28)

The third and fourth steps are analyzed within the same model. The goal of this model is to analyze if the mediators have a relation with the preferred retirement age (Y). However, the analyses must be controlled for the independent variable first. This is done during the third step. The reason for this, is that differences on the dependent variable (i.e. preferred retirement age), as well as on the mediators (i.e. wealth, health, late-career work disengagement and job satisfaction), could already be effected by the differences on the independent variable (i.e. job category). Finally, when there is controlled for the independent variable, it is analyzed, during the fourth step, if there is a relation between the mediators (M1 to M4) and the dependent

variable (Y). This eventually shows which part of the X-Y relation can be explained by the mediators (path a to b), and which part of X-Y is (in this analysis) a direct relation (path c’).

The third and fourth step are preformed via multivariate OLS regression analyses (table 7). First, the relation between X and Y is determined in model 1. Then, it is analyzed if the mediators wealth, health, late-career work disengagement and job satisfaction mediate between the X-Y relationship. This is first done for wealth and health (model 3) and then for late-career work disengagement and job satisfaction (model 4). So, in sum, in the final two models it is analyzed if the M variables completely mediate the X-Y relationship (whereby c’ becomes zero) or partly.

The interpretation of the models in table 7 is done as follows: first, the unstandardized B coefficients (B) of the independent categorical variables (i.e. the job categories) explain the differences in the preferred retirement age between the reference group (i.e. ‘higher intellectual or executive’) and the other four job categories. If the difference of the B coefficient between the reference group and the other job categories becomes smaller between each model, it means that the new variables that are taken in explain (partly) the difference in retirement age preference between that job categories. For example, if the B coefficient of the category ‘other non-manual’ work comes closer to zero in model 4 compared to model 1, 2 and 3, it would mean that the difference in retirement age preference between ‘higher intellectual or executive’ work (i.e. the reference group) and ‘other non-manual work’ is (partly) effected by the difference in late-career work disengagement or job satisfaction. Second, it is shown in the table if these B coefficients are significant (p < 0,01 or p < 0,05). Third, the standardized Beta (β) coefficients are shown in the text. Given that standardized Beta’s can directly be compared, these show if one of the continuous variables has a relatively larger effect on the preferred retirement age than others. Finally, for each model the R2_{is shown in the text}

as a percentage, to display the amount of variance of the preferred retirement age that is explained by the variables in each model.

(29)

4. Results

4.1. Differences between job categories

The ANOVA-analyses in table 6 show the differences of the dependent variable and mediators between the job categories. First, it shows that there are a significant differences in the mean preferred retirement age between the job categories (F = 56,81; p < 0,01). Herein, employees with a higher intellectual or executive occupation score the highest (M = 64,07; SD = 2,17). Thereafter the employees within the category ‘other non-manual work’ (M = 63,78; SD = 2,02). It is notable that that category has a higher mean than the intermediate intellectual or executive employees (M = 63,44; SD = 1,99), whereas it was expected that the latter was higher on the job category range of retirement age preferences. Subsequently, the ANOVA-analysis shows that both the categories ‘skilled executive manual work’ (M = 62,83; SD = 2,06) and ‘semi-skilled or un‘semi-skilled manual work’ (M = 62,66; SD = 2,24) have a preference to retire earlier than the non-manual work categories. Also a post-hoc tests makes it explicit that the difference between each job category is significant (p < 0,05). This is in line with the third hypothesis, wherein it is expected that employees with higher intellectual non-manual work prefer to retire later than employees with lower intellectual manual labor. Finally, it is notable that all the job categories have an average preference for a retirement age that is below the age of 65, which was the common basic old pension age (AOW-leeftijd) for employees born before April 1952.

Table 6. Job categorical differences for the dependent and mediation variables Preferred

retirement age (x100.000) Wealth € Health Late-career work disengagement Job satisfaction

Job category Mean SD Mean SD Mean SD Mean SD Mean SD

Higher intellectual or executive 64,07 2,17 2,23 1,64 3,37 0,87 2,48 0,68 5,55 1,01

Intermediate intellectual or executive

63,44 1,99 1,59 1,43 3,23 0,87 2,74 0,68 5,29 1,05

Other non-manual work 63,78 2,02 1,37 1,39 3,13 0,83 2,87 0,68 5,29 0,98

Skilled executive manual work 62,83 2,06 1,50 1,34 3,12 0,82 3,07 0,62 5,32 1,03

Semi-skilled or unskilled

manual work 62,66 2,24 1,28 1,31 3,06 0,87 3,22 0,65 5,19 1,01

N 4999 4999 4999 4999 4999

F 56,81** 62,33** 17,24** 138,47** 16,40**

Why are there differences in retirement age preferences between job categories? : the explanatory role of wealth, health, late-career work disengagement, and job satisfaction