Physical functioning as a predictor of retirement: Has its importance changed over a thirty-year period in Sweden?

The role of health as a predictor of retirement has been recognized for a long time: studies in the 1970s found health-related work limitations to be predictive of early retirement (Bixby 1976), and since then, studies have consistently demonstrated the importance of health for retirement. A systematic review of 44 studies, including Swedish data, showed that self-rated health, mental health problems, chronic diseases, and musculoskeletal disorders all independently increased the risk of early retirement (van Rijn 2014).

Functional health status affects the timing of retirement; mobility limitations and musculoskeletal pain are two examples of such health issues. Mobility limitations have been found to be predictive of early work exit (Rice et al. 2011). Mobility limitations often represent a pre-clinical stage of disability and are associated with severe disability and high health care expenditures (Heiland et al. 2016; Musich et al. 2018). Studies have found that 10–20% of individuals aged 50–64 have mobility limitations (Gardener et al. 2006; Heiland et al. 2016; Nilsson et al. 2010) and that these limitations increase with age (Fritzell et al. 2007). Swedish studies have shown that women were more likely to report mobility limitations compared to men from the 1960s and onwards; however, both gender differences and the total proportion of older people with mobility limitations have decreased (Parker et al. 2008). It has also been established that there are socioeconomic differences in mobility limitations, where individuals of lower class experience earlier onset (Nilsson et al. 2011a, b) and greater risk of limitations (Avlund et al. 2004; Fritzell et al. 2007).

Musculoskeletal pain has consistently been identified as a predictor for early and disability retirement (Härkäpää 1992; Haukka et al. 2015; Kamaleri et al. 2009; Øverland et al. 2012; van Rijn 2014). A greater number of pain sites on the body have been associated with reduced self-reported physical and mental work ability, the anticipation that work ability will deteriorate, the feeling of being unable to continue working in one’s current job, and thoughts about retiring early (Miranda et al. 2010). The number of pain sites independently predicts disability pension retirement (Haukka et al. 2015; Kamaleri et al. 2009) and decreased functional ability (Kamaleri et al. 2008). Lower back pain, an important component of musculoskeletal pain, is the leading cause of years lived with disability in Europe, but the age-standardized point prevalence has decreased slightly from 1990 to 2017 (Wu et al. 2020). The prevalence of musculoskeletal pain increases with age (Rustøen et al. 2005), with female sex (Breivik et al. 2006; Schmitz and Lazarevič 2020; Wranker et al. 2016), and with lower socio-economic status (Dorner et al. 2018; Todd et al. 2019).

Over the last few decades, the labour market has been subject to major changes driven by globalization and technological development. The nature of occupations has gradually shifted from factory work to service work (Baruch 2015; Parker and Agahi 2013), and from more physically strenuous to sedentary jobs (Johansson et al. 2018). Accordingly, the class structure of the labour market has changed in many respects since the 1980s. In Sweden, the share of unskilled manual workers has decreased from 32% in 1985 to 20% in 2015.

Advantages and disadvantages accumulate over the course of life; for example, there is evidence that exposure to adverse working conditions over a long period of time contributes to a decline in physical functioning (Mänty et al. 2016; Nilsen et al. 2017). Factors in the physical work environment, e.g. heavy lifting, repetitive movements, and loud noises, have shown a strong association with disability pension and early retirement (Sundstrup et al. 2018). Adverse working conditions are more predominant in manual occupations than non-manual. Older adults who have held manual occupations, have a low level of education or low income, are more likely to experience health problems and die at a younger age than older adults who have worked in non-manual occupations, have a higher level of education, or have higher incomes (Lennartsson et al. 2018; Rehnberg and Fritzell 2016). Thus, because of the changing nature of occupations, from manual to non-manual and from physically strenuous to sedentary, the importance of physical functioning as a predictor for retirement is likely to have diminished.

The labour market is gender segregated: women are more likely to have lower-status jobs, work in the public sector in fields such as education, health care and caregiving, have lower wages, work part-time, and as a consequence, have lower pensions compared to men (Edge et al. 2017). Physical working conditions for many sectors have changed for the better, especially the male-dominated sectors. The change is, however, less notable in the typically female-dominated occupations such as caregiving, health care, and retail (Westerlund 2005). The typically female-dominated occupations involve frequent contact with people and meeting other people’s needs. Such tasks often have high psychosocial demands, they are more difficult to plan and control compared to tasks involving machinery (SOU 2015), and they are associated with higher stress levels (Anxo et al. 2014). Moreover, women are typically worse off regarding pain, multimorbidity, poor self-rated health, chronic health conditions, activity limitations, and depression (Schmitz and Lazarevič 2020). Women take more sick leave from work, report poorer health, and retire earlier compared to men (Anxo et al. 2014). Taken together, the female-dominated labour market has not seen as much change for the better, and pathways towards early retirement and disability pension are closing. This may lead to women working additional years despite health limitations.

Sweden is an interesting case to study when it comes to determinants of retirement, partly because of the high employment rate among both women and men, and the high retirement age in an international comparison (OECD 2021), but also because Sweden has already implemented various pension reforms that other countries have on the agenda. Flexible retirement age was introduced in 2003 in Sweden, offering individuals increased freedom to choose when to retire. At the same time, other reforms have closed exit pathways out of the labour market, for example criteria for disability benefits have become stricter, the partial pension scheme has been abolished, and there are stronger financial incentives to stay in the labour market. Sweden has undertaken major policy reforms, and the spillover effects of such reforms have the potential to impact health. As many countries face major pension reforms in the near future, examples from Sweden are of more general interest. A description of major reforms in Sweden since the 1980s that possibly influenced retirement behaviour of the cohorts under study and thus are of significance to this study can be found in Supplementary material.

Since the 1960s, women’s labour force participation has increased dramatically in Sweden (König and Lindquist 2016). The employment rate among women aged 55–64 has increased from 54% in the early 1980s to almost 76% in 2019, while in the same period the employment rate among men has remained stable, just below 80% (Statistics Sweden 2020). The employment rate for people aged 55–64 is notably lower in Europe today: 62% for women and 72% for men on average. The average age for leaving the labour market, restricted to those in the labour force at the age of 50, was calculated at 63.3 years for women and 64.3 years for men in 2016. This reflects a substantial increase from 61 and 63.2 years in 1981 for women and men, respectively (Swedish Pensions Agency 2017). Over 80% of people retire at or before the age of 65 (Swedish Pensions Agency 2020).

There is a bulk of evidence regarding the significance of good health and physical functioning for labour force participation. However, little is known about whether the importance of good physical functioning for employment has changed along with the shifting labour market and generally improved health among older people in the past decades. The main aim of this paper is therefore to explore whether the importance of physical functioning as a predictor for retirement has changed over a three-decade period, using nationally representative data from Sweden. Because of the gender segregation on the labour market, women’s worse physical functioning compared to men, stricter eligibility criteria for disability pension, and smaller technical advancements within female-dominated occupations, the second aim is to investigate whether developments differ for women and men.

We hypothesize that (1) physical functioning loses its predictive power for retirement during the study period and (2) the decrease in predictive power will be larger among women.

The study is based on the Swedish Level of Living Surveys (LNU) from 1981, 1991, 2000 and 2010, and the linked Longitudinal integrated database for health insurance and labour market studies (LISA). LNU is a longitudinal study that started in 1968 and has since then been repeated at six-to-ten-year intervals. It is based on face-to-face interviews with a representative sample of the Swedish population in the age range of 18–75 years. Each consecutive wave also includes a new random sample of young people and new immigrants in addition to the original sample, thus ensuring that each wave remains representative of the total adult population from a cross-sectional perspective (Fritzell and Lundberg 2007). The response rate of LNU has varied between 82.4% (1981), 79.1% (1991), 76% (2000), and 61% (2010). LISA is a national income register covering all residents in Sweden older than 16 years (Ludvigsson et al. 2019). Data on retirement were obtained from LISA and linked to the survey data using personal identification numbers assigned to each Swedish resident.

This is a trend study with four baseline waves (T₀ 1981, 1991, 2000 and 2010) with two-year follow-ups for each wave T₁ (see Fig. 1). For example, participants in the T₀ LNU 1981 wave are followed using income register data in the two consecutive calendar years T₁ 1982–1983. All study participants who were aged 50–67 and employed (defined as more than 50% of their total individual annual income coming from employment) at baseline (T₀) were included in the analysis, resulting in a total of 3890 cases over the study period (T₁). Due to internal missing values, the sample size for the fully adjusted analysis with mobility limitations was n = 3862, and with musculoskeletal pain n = 3870.

Logistic regression models were used to study the association between physical functioning and retirement for men and women separately by period. These models were adjusted for covariates (age, occupational-based social class, adverse physical working conditions, and job demands). Since there are differences between men and women in labour market involvement, we ran all analyses separately for men and women. Because of the unobserved heterogeneity that can vary across the compared samples, groups, or points in time, the more traditional odds ratios cannot be compared for models across groups, samples or time points (Mood 2010). Therefore, estimates are reported as average marginal effects (AMEs) with 95% confidence intervals (CI)s. AMEs facilitate interpretability and comparability across models for groups, samples or time points (Mood 2010). An AME can be interpreted as the average difference in the probability (0–1) of the outcome depending on the value of the independent variable, that is, it is the average change in probability of retirement (the outcome) when mobility limitations or musculoskeletal pain (the exposures) increase by one unit. Finally, we use predictive margins (PM) to illustrate the probability of retirement over the period by mobility limitations on the one hand, and musculoskeletal pain on the other. PMs are easier to interpret when presenting group differences than regression coefficients (Graubard and Korn 1999), especially in the presence of interaction terms. The PMs show the probability of retirement for all levels of the exposure variable while holding other variables in the model constant, while the AMEs use one level of the exposure variable as a reference category and show the discrete change from the reference category and if the difference is significant. All analyses were weighted to adjust for differences in response rates by some standard background variables, such as sex, age and education.

We use four waves of nationally representative data, spanning three decades. The participation rate was relatively high in all waves, ensuring generalizability to the Swedish population. In addition, we made use of a register-based outcome not liable to subjective recall error to assess retirement timing (Eyjólfsdóttir et al. 2021). Using annual register data has its limitations; one can estimate whether, in a given calendar year, income from pensions succeeded income from employment. As income from pension is generally lower than income from employment, a person must have more than six months of pension income to surmount the income from employment. This leads to the risk of determining a retirement age that is one year higher than the actual age of labour market exit for people retiring late in the calendar year. This might introduce information bias. However, a paper by Eyjólfsdóttir et al. (2021) examining different definitions of retirement age, using the same materials as the current study, did not find significant differences between self-reported retirement age and retirement age attained by this current definition, nor by using different cut-offs than 50%. Another limitation is that we employ the same definition of retirement over the three decades under study, although there have been many reforms during the period that have made the retirement process more flexible over time. Therefore, the definition might capture the timing of retirement better in some periods than in others. Moreover, retirement behaviour could be more influenced by economic context and pension reforms than by health status in specific periods, resulting in overestimation of the importance of health for retirement. An additional limitation might be that there are few respondents retiring in the two years following the interview, possibly creating statistical power issues in the interaction analysis. On the other hand, the follow-up time needs to be short for the health indicators to be relevant as predictors of retirement.

As populations age, the financial sustainability of many welfare states and pension systems is threatened. In response, most European countries are implementing reforms in order to delay exit from the labour market and thus enhance fiscal sustainability (European Commission 2021). In this study, we found that the importance of physical functioning for labour market participation has declined in Sweden over a thirty-year period. These findings are promising, as physical functioning might be of less hindrance for labour force participation among older workers and thus for postponement of retirement. However, we also found that those who already have functional limitations are less likely to retire in the most recent time period, especially women. This might indicate that because of financial constraints some groups are stuck on the labour market despite poor health. In efforts to raise retirement age and increase labour force participation among older workers, policymakers should recognize the heterogeneity among older workers. Options for disability insurance and part-time retirement for older workers in adverse working conditions or for those with poor health are important in order to not enlarge health inequalities in later life.