A systematic review of causes of recent increases in ages of labor market exit in OECD countries.

Ages of labor market exit have increased steadily since the late 1990s in OECD countries, but with continuing population aging, there are calls for further stimulation of labor force participation at older ages. Social scientists have extensively studied causes of variation in retirement timing between individuals and across countries, but have paid less attention to causes of variation over time. This study systematically reviews evidence of causes of increases in ages of labor market exit over the past 30 years in OECD countries. Two goals are pursued: first, to provide an overview of the retirement domains that have been subject to investigation; second to compare studies with respect to the magnitude of change in retirement behavior that they attributed to different causes, in different contexts. Nineteen studies were reviewed. Available evidence articulates itself around four domains: inter-cohort changes in labor force participation of women (3 studies), educational attainment (3 studies) and lifetime wealth (1 study), and changes to social security systems (16 studies). Determinants in all domains explain a significant amount of past increases in ages of labor market exit, though figures attributable to similar determinants vary between studies and across countries. Evidence suggests that further postponement of labor market exit may depend on further increases to normal retirement ages and more limited access to early retirement programs, but also on further increases in educational attainment and the continued integration of women in the labor market. However, a large share of the past increases in ages of labor market exit remains unexplained; therefore, other factors such as those related to work and organizational characteristics deserve further research.

Introduction

Since the late 1990s in Organization for Economic Co-operation and Development (OECD) countries, labor force participation at older ages has increased steadily. While less than half of the adults aged 55 to 64 were active in the labor market in 1996, nearly two-thirds of them were active in 2016 [1] (Fig 1). This constitutes a reversal of the historical trend, as labor force participation at older ages had previously always been constant or declining [2]. Now, for the first time in history, each younger birth cohort can expect to exit the labor market at a later age than previous cohorts. For example, an average worker in 1996 could expect to exit at approximately age 62, whereas the same worker in 2016 could expect to exit at age 64 [3]. Increases in ages of labor market exit are pervasive, since they have occurred in all OECD countries (despite variation in timing of onset and magnitude), and they have affected both men and women of all socio-economic classes[4].

Fig 1

Labor force participation at ages 55 to 64 (left axis) and average labor market exit age (right axis) for OECD countries.

Labor force participation rates are calculated as the total of people in the labor force divided by the total population ages 55 to 64. [1] The average effective age of labor market exit is based on changes in labor force participation rates and are therefore not affected by the proportion of people working. [3].

These developments have caught the attention of researchers in social sciences and recent studies have investigated the causes for the increases in ages of labor market exit in OECD countries. Synthesizing the increasing amount of evidence on what causes increases in ages of labor market exit over time could prove highly valuable in the context of ongoing population aging[5]. As the proportion of older people in the population increases and the proportion of younger people stagnates or declines, scientists and policy makers alike are calling for further stimulation of labor force participation at older ages[6,7]. Therefore, we aim to systematically review the available scientific evidence of the causes for increases in ages of labor market exit in OECD countries in recent decades.

Reviews have summarized the evidence on the causes for variation in retirement timing between individuals[8-10]. Results showcase high agreement on the determinants of such variation; these are summarized in Table 1 referring to four retirement domains (individual, job, family, and socio-economic). Information contained in this table will serve as a reference for the remainder of this article.

Table 1

Overview of retirement domains to which current evidence on causes of variation between individuals in ages of labor market exit belong.

Adapted from Wang & Schultz 2010, Fisher, Chaffee & Sonnega 2016 and Scharn et al. 2018.

INDIVIDUAL	JOB	FAMILY	SOCIO-ECONOMIC
• Demographic characteristics (age, education)• Personality, needs, motivations and values• Knowledge, skills and abilities• Attitudes towards work and retirement• Health and lifestyle• Employment history• Income, wealth and health insurance	• Job characteristics• Age stereotypes and norms, diversity and discrimination• HR policies and practices• Employer provided pension plan• Training/skill development opportunities	• Caregiving responsibilities• Partnership status and relationship’s quality• Partner’s retirement status	• Social norms about retirement• Macroeconomic conditions• Social security systems

Another important strand of literature investigated causes for variation in retirement behavior between countries at one point in time [11-13]. This work concentrated on the incentives created by social security in inducing retirement at specific ages and developed the concept of implicit tax on work. The finding that a higher implicit tax on work correlates with lower retirement ages prompted countries to introduce changes to their social security systems with the aim of encouraging later retirement [3]. These changes likely played a role in the recent increases in ages of labor market exit and are summarized in Table 2 [11].

Table 2

Classes of changes brought to social security systems over the last decades in OECD countries.

Adapted from Börsch-Supan & Coile 2018.

• Change to retirement age or in years of contribution required [early or normal]

• Change to programs allowing partial retirement

• Change to the generosity of social security benefits

• Change to the actuarial adjustment of social security benefits [early or delayed claiming]

• Change to earnings tests

• Change to pension plans [e.g. defined benefits to defined contribution]

• Change to early retirement, disability insurance and unemployment insurance programs

Though causes for differences in retirement timing between individuals as well as between countries were already reviewed, we are not aware of any study that reviewed causes for changes in retirement behavior over time. By filling this gap, this systematic review contributes to the current state of knowledge in two main ways. Firstly, it summarizes and synthesizes the available evidence on factors that affect changes in ages of labor market exit over time. As a result, it also identifies knowledge gaps in our understanding of what causes increasing ages of labor market exit, thereby providing guidance for future research on this topic. Secondly, it compares studies with respect to the magnitude of change in retirement behavior that was attributed to different causes, in different contexts. Effects are presented individually for each cause, referring to the context in which they were estimated. Assessing whether systematic differences emerge between causes and the contexts in which they were studied may prove instrumental in informing policy on how to further increase ages of labor market exit in the context of population aging.

We reviewed articles that aimed at explaining why ages of labor market exit have increased in the last decades in OECD countries. In the studies reviewed, changes in the age of labor market exit were measured in the form of either changing rates of labor force participation (LFP) or changing retirement probabilities. LFP in older age groups captures both employment and retirement patterns within the group. Retirement probabilities, on the other hand, are obtained by following working individuals over a period and recording retirement occurrences. In comparison to changes in LFP, changes in retirement probabilities over time reflect variation in retirement behavior more closely. For ease of interpretation, working life expectancies are often calculated from retirement probabilities, referred to as the effective retirement age (ERA). ERA can be contrasted with the normal retirement age (NRA), which is set by law and determines the age at which full pension benefits are granted. In the studies reviewed here, changes in retirement behavior were most of the time measured in the form of changes in LFP or ERA. In the remainder of this article, changes in retirement behavior will thus be referred to as changes in LFP/ERA.

We first identify domains for which evidence is available regarding changes in LFP/ERA. Then, we present results regarding the amount of change that is attributable to each domain, as estimated in each study and by country and gender, if applicable. Studies were divided into two groups according to the approach that was taken for explaining change over time in LFP/ERA: the first group considers differences in LFP/ERA between two points in time, while the second one considers differences between groups that were differently affected by external factors such as pension reforms. Results that were extracted from the first group of studies are proportions of change in LFP explained by a specific factor as well as total change in LFP during the period under study, while results that were extracted from the second group of studies are regression outputs (e.g., coefficients) that give the ceteris paribus effect of an exogenous change on retirement behavior. This review is limited to OECD countries, which share similarities regarding social security systems, population structure, and trends in age of labor market exit. This review is further limited to studies that address national populations or population subgroups (e.g., men of a specific age range during a period), and thus studies of narrow samples (e.g., professional groups) were excluded.

Materials and methods

Database search

We systematically searched databases EconLit, PubMed, and Web of Science. The same search terms, adjusted for syntax requirements, were used in each database. Four strings of words were identified and combined to use during a single search within titles. The first string contained the words raise, labor force participation, and old age, the second raise and retirement age, the third extension and working life, and the fourth delay and retirement. The terms that formed each string were combined using the Boolean operator AND, and equivalent expressions were used where applicable using OR (e.g., delay OR postponement of retirement) (S2 Table).

Inclusion criteria

We included peer-reviewed research articles, published in English since the year 2000, which analyzed the general population of OECD countries (excluding studies of specific subgroups, e.g. occupational groups). Studies were included when they assessed retirement behavior (not retirement intentions), and aimed to explain changes in LFP/ERA over time (not between individuals or across countries). Research designs must have included quantifiable changes in retirement behavior (e.g., proportion employed, effective retirement age) as outcome variables over well-defined periods, or birth-cohorts, and age-groups, while explanatory variables must have denoted changes over time in any explanatory factor. No criteria were applied concerning the longitudinal or cross-sectional design of studies.

Selection procedure

Searches were conducted simultaneously and independently by three researchers (i.e., J.O.M., K.T., and M.B.) on 13 February 2019. Duplicates were removed. Screening was performed in three rounds: first based on titles, then on abstracts and to finish on full texts. Criteria regarding article formats and populations were applied throughout; those regarding a study’s purpose were applied from the second round forward, and those regarding research designs during the last round only. During each round, two of the three researchers (i.e., J.O.M., K.T., and M.B.) reviewed studies independently. In case of disagreement regarding inclusion, the third researcher made the final decision. Additional articles were considered based on reference lists of the studies that passed the second round of screening as well as on expert knowledge (Fig 2).

Fig 2

Decision tree for article inclusion.

Data extraction

Data regarding a study’s design, results and methodology were retrieved manually from each article. Information on study design included population of interest (e.g. married men), country, age and year ranges, the dependent and main independent variable(s) and their measurement (e.g. percentage points, year, probabilities). Information on results included, if available, changes due to a cause of interest in labor force participation rates, retirement ages or retirement probabilities, or otherwise coefficient values in regression output. Values were extracted concerning each predictor of interest, separately for men and women and each country, if applicable. Values pertaining to the preferred specification were extracted, if indicated, or averaged over the different specifications otherwise. Information on methodology included the dataset(s) used, number of observations, the statistical model used, and the strategy adopted for tracking causality. To facilitate interpretation, we consider two main classes of results. The first one includes results from studies designed to explain differences in LFP between two points in time. Since the distance between these points in time may vary considerably, results are presented in terms of yearly change in LFP. The second class includes results from studies designed to explain differences in retirement behavior between two groups that were affected differently by external factors such as pension reforms. This class is further subdivided into four subclasses denoting different outcome variables: LFP rates, retirement ages, retirement probabilities and hours worked. Articles were deemed as having made effort towards tracking causality if they used statistical models controlling for potentially spurious correlation and included instruments controlling for endogenous relations, for example by using control groups not affected by the independent variable of interest. The PRISMA statement checklist was referred for the review process (S3 Table).

Results

Six hundred eighty-nine studies were identified through database searches, of which 511 remained after duplicates were removed. Studies were then considered for inclusion based on selection criteria (see Materials and Methods). Information included in article titles allowed us to exclude 386 studies, and information in abstracts and texts allowed us to exclude another 108. Following this procedure, 17 studies were selected for assessment. The reference lists of these studies were checked, and experts in the field were consulted regarding missing studies, which added 31 more. Following assessment of these studies, 19 were included in the final selection (Fig 2).

Articles concentrated on 11 countries, all of which are located in Europe or North America. Most assessed single countries, but two included cross-national comparisons. They focused most often on the United States (6 studies), followed by Germany (5 studies). Austria, Sweden, and the United Kingdom were studied two times each, and Belgium, Canada, Denmark, Estonia, Spain and Switzerland were studied once. Studies covered periods of varying lengths. The shortest period covered one and a half years (2007–2008.5) while the longest covered 17 years (1988–2005). Other studies followed birth-cohorts. The oldest cohort was born in 1928 and the youngest in 1951. Outcomes were measured among women slightly less than half of the time, and the studies considered varying age ranges. Some considered changes in retirement behavior in a narrow age group (e.g., 62 to 63), while others considered groups of greater width (e.g., up to 20 years). Most studies concentrated on age groups strictly before or both before and after the NRA, while few considered age groups strictly after the NRA. Studies analyzed data from national registers, labor force surveys, and surveys that were representative of national populations (e.g., Health and Retirement Study). Excepting two [12,13], all articles assessed causal relationships between explanatory variables and an outcome (S1 Table).

Domain and outcome coverage

Table 3 shows studies classified regarding the explanatory and outcome variables contained in their analyses. Explanatory variables covered 4 of the 18 domains identified in Table 1, including partner’s retirement status (3 studies), demographics (3 studies), income, wealth, and health insurance (1 study), and social security systems (16 studies). Studies that investigated changes to social security systems were further broken down according to the categories identified in Table 2. Seven studies considered the effects of increases in LFP/ERA in terms of changes to early retirement and related programs. Five considered changes to statutory retirement ages or the number of years of required contribution for early or normal retirement. Four investigated changes to pension plans over time (i.e., defined benefits to defined contributions) and the same number considered changes to actuarial adjustments of benefits or earnings tests. Three studies estimated the effects of changes to benefits generosity over time, usually in the direction of less generous benefits (two of three studies). Two studies estimated the effect of removing earnings tests, and one investigated changes to regulations regarding partial retirement. Some studies examined the combined effect of multiple changes to social security systems and therefore appear in different categories (to be hereafter reffered to as “extensive” reforms). Regarding outcome variables, changes in retirement behavior were measured ten times in terms of LFP, six times in terms of ERA, two times in terms of retirement probabilities and one time in terms of hours worked.

Table 3

Domain and outcome coverage among reviewed studies.

	Type of outcome
Domains	Labor force participation rates	Effective retirement age	Retirement probabilities	Hours worked	No. of results
Partner’s retirement status	Blau & Goodstein (2010) Pérez et al. (2020) Schirle (2008)				3
Demographic characteristics (education)	Blau & Goodstein (2010) Larsen & Pedersen (2017) Schirle (2008)				3
Income, wealth and health insurance	Blau & Goodstein (2010)				1
Social security systems, including:					16
Retirement age / years of contribution (early or normal)	Blau & Goodstein (2010) Dejemeppe et al. (2015) Gustman & Steinmeier (2009)	Mastrobuoni (2009) Puur et al. (2015)			5
Partial retirement programs	Dejemeppe et al. (2015)				1
Benefits’ generosity	Dejemeppe et al. (2015) Staubli & Zweimüller (2013)		Hanel & Riphahn (2012)		3
Actuarial adjustment of benefits	Blau & Goodstein (2010) Gustman & Steinmeier (2009)	Berkel & Börsch-Supan (2004)	Buchholz et al. (2013)		4
Earnings tests	Gustman & Steinmeier (2009)			Disney & Smith (2002)	2
Pension plan	Hurd & Rohwedder (2011)	Friedberg & Webb (2005) Qi et al. (2018)	Buchholz et al. (2013)		4
Early retirement, disability insurance and unemployment insurance programs	Dejemeppe et al. (2015) Staubli & Zweimüller (2013) Staubli (2011) Hanel (2010)	Berkel & Börsch-Supan (2004) Bönke et al. (2018)	Buchholz et al. (2013)		7
No. of results	10	6	2	1

Differences in LFP/ERA between two points in time

Five studies considered differences in LFP between two points in time and aimed at explaining it in terms of change in one or more independent variables. Table 4 presents an overview of these studies’ designs while Fig 3 presents the amount of yearly change in LFP that was observed (full bars), broken down by the amount that was explained by each variable of interest (lower part), and the amount that was not explained by each variable of interest (upper part). Levels concern the mean annual change over the whole period of observation to improve comparability as year ranges vary. In total, results were available for 20 effects covering eight countries and six classes of predictors. These predictors included changes in inter-cohort educational attainment, LFP of women, normal retirement age, delayed retirement credits, lifetime earnings, and extensive social security reforms. The proportion of change in LFP attributable to different predictors varies strongly between studies. The effect of educational attainment varies from 0.004 pp per year (women in Germany in [13]) to 0.21 pp per year (men in the United States in [14]). Likewise, the proportion of change in LFP attributable to change in LFP of women varies from 0.04 pp per year (men in the United States in [15]) to 0.43 pp per year (men in Spain in [16]).

Table 4

Overview of the first group of studies which investigated differences in LFP between two points in time.

Author	Year	Countries	Subpopulation	Age range	Year range	Effect(s) studied	Details	Causal
Blau	2010	United States	All men	55–69	1988–2005	Delayed retirement credits (DRC)	Introduction of credits for delayed retirement past the NRA over the period 1987 to 2005	Yes
						NRA	Increase of normal retirement age from age 65 to 65.5
						Lifetime earnings (LE)	Increases in total lifetime earnings
						LFP women	Increases in labor force participation of women
						Educational attainment	Increases in inter-cohort educational attainment
Dejemeppe	2015	Belgium	Initially employed men and women	50–59	2004–2013	Extensive reform (ER)	Reduction in employers' social security contributions for workers aged 50–56	Yes
							Stricter admissibility criteria to early retirement
							Higher age of admissibility to early retirement
							Easier access to partial retirement
							Increase in the generosity of retirement benefits
Larsen	2017	Denmark, Germany & Sweden	All men and women	65–69	2004–2013	Educational attainment	Increases in inter-cohort educational attainment	No
Pérez	2020	Spain	All men who live with a partner	55–64	1995–2016	LFP women	Increases in labor force participation of women	Yes
Schirle	2008	Canada, United Kingdom & United States	All married men	55–64	1994–2005	Educational attainment	Increases in inter-cohort educational attainment	Yes
						LFP women	Increases in labor force participation of women

LFP=Labor force participation rate; MR=Multiple reforms; NRA=Normal retirement age; DRC=Delayed retirement credits; LE=Lifetime earnings. M=Men; W=Women.

Fig 3

Differences in LFP between two points in time and proportion attributable to specific factors.

Figures correspond to the yearly change in LFP (or LFP equivalent), by gender and country. Results are grouped according to the explanatory variables that are indicated at the bottom of the graph. Articles from which results were retrieved are referred to by the first author’s name (in parentheses). Numbers above bars indicate the total observed change; numbers below refer to the explained part. For example, in a study by Larsen among men in Denmark, the observed yearly change in LFP was 1.44 pp, of which 0.13 pp was explained by increases in educational attainment. Details about the studies’ designs are provided in Table 4. Details about the calculations made are presented in S1 Table. LFP=Labor force participation rate; MR=Multiple reforms; NRA=Normal retirement age; DRC=Delayed retirement credits; LE=Lifetime earnings. M=Men; W=Women. * The authors note a lack of statistical power to draw firm conclusions.

Differences in LFP/ERA between groups

Fourteen studies investigated differences in LFP/ERA between groups. These aimed at identifying the ceteris paribus effect of an exogenous change on retirement behavior, for example brought about by a reform of the social security system. Table 5 presents an overview of the studies’ designs while their results are illustrated in Fig 4, where each panel refers to a different type of outcome (LFP, ERA, retirement probabilities and hours worked). Six types of reforms in social security systems are considered: extensive pension reform, pension plan, disability benefits, early retirement age, earnings tests, benefits reduction and NRA. Differences among subgroups with regards to the way that they were affected by reforms of social security systems vary greatly, stretching from nearly null effects (men and women in Germany in[17]; women in Germany in[18]; women in Sweden in[19]) to considerable ones, sometimes above 20 pp (pension reform affecting men and women in Austria in[20]; the removal of earnings tests for men in the UK in[21]). Once again, there does not seem to be one particular strategy which consistently brings about similar changes in LFP/ERA, or one particular context that saw stronger increases.

Table 5

Overview of the second group of studies which investigated differences in LFP/ERA between age- or cohort-groups, grouped by type of outcome studied.

Author	Year	Countries	Subpopulation	Age range	Year range	Effect(s) studied	Details	Causal
Outcome: Labor force participation rates
Gustman	2009	United States	Married men	65–67	1992–2004	Extensive reform (ER)	Increase of normal retirement age from age 65 to 65.17	Yes
							Introduction of credits for delayed retirement past the NRA over the period 1987 to 2005
							Reduction (1990) and elimination (2000) of the implicit tax on earnings past the normal retirement age
Hurd	2011	United States	Initially employed men and women	61–68	1992–2004	Pension plan	Decrease in the proportion of workers with pension plan with defined benefits and increase in the proportion with pension plan with defined contribution	Yes
Staubli	2011	Austria	Men and women working in private sector	55–56	1994–1999	Disability benefits	Increase of availability to disability benefits from age 55 to 57	Yes
Staubli & Zweimüller	2013	Austria	Men and women working in private sector	57–64 (men) and 52–59 (women)	2000–2010	Early retirement age	Increase of early retirement age from age 60 to 62 (men) and age 55 to 58.2 (women)	Yes
Outcome: Hours worked per week
Disney	2002	Great Britain	All men and women	60–74 (men) and 55–69 (women)	1986–1994	Earnings tests	Abolition of the earnings tests in 1989	Yes
Outcome: Retirement ages
Berkel	2004	Germany	All men and women	55–70	1984–1997	Extensive reform (ER)	0.3% benefit reduction per month of early retirement	Yes
							0.5% pension increase per month of work past the normal retirement age
							Restricted access to disability pension
Bönke	2018	Germany	All men	63–65	2004–2012	Benefit reduction	0.3% benefit reduction per month of early retirement	Yes
Friedberg	2005	United States	Men and women initially in full employment	63–65	1983–2015	Pension plan	Decrease in the proportion of workers with pension plan with defined benefits and increase in the proportion with pension plan with defined contribution	Yes
Hanel	2010	Germany	Initially employed men and women	55–67	1995–2002	Extensive reform (ER)	0.3% benefit reduction per month of early retirement	Yes
							Gradual increase in early retirement ages over period 1997–2005
Mastrobuoni	2009	United States	All men and women	62–65	1989–2007	NRA	Increase of normal retirement age from age 65 to 65.67	Yes
Puur	2015	Estonia	All women	unspecified	2002–2011	NRA	Increase of normal retirement age from age 58.5 to 61.5	No
Qi	2018	Sweden	Initially employed men and women	60–67	1997–2011	Pension plan	Shift in pension plans from defined benefits to notional defined contribution	Yes
Outcome: Retirement probabilities
Buchholz	2013	Germany	Initially employed men and women	60–70	1984–2007	Extensive reform (ER)	0.3% benefit reduction per month of early retirement	Yes
							0.5% pension increase per month of work past the normal retirement age
Hanel & Riphahn	2012	Switzerland	Initially employed or unemployed men and women	62–64	2000–2005	Benefit reduction	3.4% reduction age 62 year 2000	Yes
							6.8% reduction age 62 year 2001–2004
							3.4% reduction age 63 year 2005

Fig 4

Differences in LFP/ERA between groups attributable to specific explanatory variables, by type of outcome (panels A-D), gender, specification and country.

Results are grouped in panels according to outcome measures. They are further grouped by the explanatory variables indicated at the bottom of the graph. Articles from which results were retrieved are referred to by the first author’s name (between parentheses). For example, in a study by Berkel among men in Germany, the effect of treatment (pension reform) measured as a difference in retirement age between treatment and control group was 1.8 year. Further details about the studies’ designs are provided in Table 5. Details about how the figures were extracted from the articles are available in the appendix. Pens. ref=Extensive pension reform; Pens. plans= Change in pension plans; Disab. benef.= Restricted disability benefits; ERA= Increased early retirement age; Earnin. tests= Removal of earnings tests; Benefit red.= Benefit reduction; NRA = Increase of NRA. M=Men; W=Women.

Discussion

The majority of the reviewed articles considered changes to social security systems as a cause for the recent increases in LFP/ERA. Extensive reforms in Germany and Austria induced strong increases in LFP/ERA over periods ranging between seven and twenty years [18,20,22], though effects tended to be weaker among women in Germany [18,22]. A key element of the reforms in both countries was the introduction of financial penalties for claiming benefits before the NRA. In the United States, changes to social security introduced in the 1990s and 2000s also contributed to increases in LFP/ERA. The combined effect of the introduction of delayed retirement credits, the removal of earnings tests, and increases to the NRA brought about a 2 pp increase in LFP among men ages 65–67 [23], while the increase in NRA induced increases in LFP of 0.05 pp per year among men ages 55–69 [15] or a postponement of retirement of 0.5 years per year of increase in NRA among those ages 62–65 [24]. The phasing in of delayed retirement credits was responsible for a 0.06 pp increase per year in LFP among American men ages 54–69 [15].

The decrease in the proportion of workers having defined benefits pension plans and the increase in the proportion of those having defined contribution plans in the United States (or with a notional defined contribution in Sweden) caused considerable increases in LFP among men, but weaker ones among women [19,25,26]. The removal of earnings tests past the NRA induced strong increases in the number of hours of work among men ages 65–69 and women ages 60–64 in the United Kingdom [21]. Other changes to social security systems included a reduction of benefits generosity linked to early retirement among German men [27] and Swiss women [18] and the tightening of admissibility criteria to disability benefits among Austrian men [28]. These interventions were more targeted as they affected narrower age groups but induced clear changes in retirement behavior among these groups.

Changes to social security systems clearly induced prolonged labor force participation in recent decades, but they could not explain all recent increases in LFP/ERA. Studies that assessed increases in LFP in terms of changes to social security systems explained, at best, half of the increases [28], though most attributed 30% to 40% of the increases to such changes [15,19,20,29]. Four studies investigated the effect of factors not related to social security systems on increases in LFP/ERA. Changes in educational attainment across birth-cohorts of men increased LFP by 0.01 to 0.21 pp per year, depending on the country and the study [13-15]. Only one study [13] considered changes in educational attainment over cohorts of women, finding a contribution of 0.004 to 0.18 pp over the 2004 to 2013 period, depending on the country. Increases in labor force attainment of women induced increases in LFP of married men that ranged from 0.04 pp per year in the United States [15] to 0.43 in Spain [16].

Research gaps

One article combined the effects of changes to social security systems (i.e., higher NRA, actuarial adjustment, benefits generosity), changes in educational attainment of successive cohorts, and increases in labor force attainment of married women, explaining approximately 73% of the observed increases in LFP [15]. Other articles assessed fewer predictors, explaining much smaller proportions of the increases. There are other factors which are conceptually likely to have affected LFP in recent years that have yet to have been subject to empirical analyses (see Table 1). For example, despite ongoing increases in older adults’ health and their ability to work [30,31], no study assessed the influence of this trend on ages of labor market exit. Similarly, individuals with more complex career paths tend to exit the labor market at higher ages [32] but it is not clear whether the de-standardization of careers has contributed to increased LFP at older ages [33]. Changes in organizations, such as improved accommodative HR practices for older workers and progress with counteracting age discrimination at work [34] might have contributed to increases in ages of labor market exit. Also, employers are more likely to employ older workers as they increasingly recognize their value and are getting more experienced with dealing with an aging workforce [35]. Decreasing physically demanding jobs, improved quality of work, and the rise of flexible work arrangements might also have facilitated higher ages of labor market exit [36]. Changing societal norms regarding work at later ages and retirement may have also contributed to increases in LFP, as younger cohorts have different attitudes, preferences, and expectations regarding work and retirement than older cohorts [37]. Finally, following the Great Recession of 2007–2008, studies could investigate factors like decreased coverage of private pension plans, higher debt load among older adults, and decreased returns on private savings which decreased disposable income and may have forced older adults to delay retirement [38].

Future prospects

As of 2017, increases to the NRA were planned in close to half of OECD countries, including, in some countries, automatic links to changes in life expectancy [3]. Given the evidence reviewed in this paper, further increases in LFP/ERA are to be expected following these changes. In contrast, other types of changes to social security systems may become more difficult to implement. Reductions to the generosity of retirement benefits and tightening of admissibility criteria must be implemented carefully since they might negatively influence economic wellbeing at older ages. Other changes, such as those pertaining to actuarial treatment of pension benefits, changes from defined benefits to defined contribution schemes, and elimination of earnings tests, can be implemented only once, and thus countries that have already implemented such changes cannot benefit from them in the future. Finally, the effects of other relevant factors might be slowly diminishing. For example, future increases in educational attainment across birth-cohorts will be milder than those that have prevailed until now [39]. The labor force participation of women is approaching that of men in many countries, and when the labor force attainment of married women starts leveling, the effect on the retirement timing of men might begin to wear off. However, factors such as more accommodating HR practices for older workers, improving quality of work, and changing societal norms may start playing a larger role.

Limitations

Some limitations inherent to this review should be mentioned. The studies reviewed here were performed in countries that differ in their level of education, health, occupational structure, age composition, or social security systems. These factors may interact with reforms or processes, leading eventually to different effects for LFP/ERA. Additionally, studies concerned different time frames, different populations (e.g. labor or marital status), as well as different age groups or birth cohorts. Because of this, a more quantitative treatment of the results (e.g. meta-analysis) was not feasible, and results should be interpreted bearing the particularities of each study in mind. Furthermore, though studies covered eleven countries, nine out of nineteen studies strictly focused on either the United States or Germany, and only 11 countries of the total of 36 OECD countries were covered. Therefore, our conclusions mainly apply to the countries covered by the reviewed studies, and care should be taken when extrapolating the findings to other OECD countries.

Conclusions

The last 30 years provided social scientists with the opportunity to gain direct evidence on what causes individuals to postpone retirement. This systematic review shows that although several studies investigated causes of recent increases in ages of labor market exit, the variety of topics on which they concentrated remains limited. In countries for which evidence was available, increasing the NRA and limiting access to and the generosity of early retirement programs, among other changes to social security systems, contributed to higher ages of labor market exit and increasing labor force participation at older ages. In the same countries, changes in the patterns of labor force participation among married women and in educational attainment across birth cohorts seem to have played similar roles, though evidence is less robust. Policies that aim at increasing ages of labor market exit should thus consider modifying the incentives created by social security in inducing retirement at specific ages, but also promoting education and life-long learning, and facilitating the integration of female workers to the labor market. Other factors such as changes in population health, the substantive nature of work, the role of HR practices, and norms and attitudes towards work at older ages may have been equally powerful in increasing ages of labor market exit, but evidence on the role they played is missing. Increasing the scope of evidence to other potential causes of increases in ages of exit from the labor market, as well as to more countries, will provide scientific grounds for stimulating further increases in ages of labor market exit in OECD countries.

Studies overview: Data source and methods.

(DOCX)

Click here for additional data file.

PRISMA 2009 checklist.

(DOCX)

Click here for additional data file.

Description of calculations, results reported in Fig 3.

(DOCX)

Click here for additional data file.

Description of calculations, results reported in Fig 4.

Effects (i.e. regression coefficients) were extracted directly from articles where available and appear in column “Effect”. Estimates for separate groups appear in the columns “Value”; in such cases, it is the difference between groups that appears under “Effect”.

(DOCX)

Click here for additional data file.

Full strings used in searches.

(DOCX)

Click here for additional data file.

3 Dec 2019

PONE-D-19-28928

A systematic review of causes of recent increases in ages of labor market exit in OECD countries

PLOS ONE

Dear Dr. Boissonneault,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

1. The topic is relevant and hence worthy of investigation, currently the paper lacks some important issues that should be addressed in a future version

2. The reviewers did not consider the review systematic since only 18 papers were reviewed and there is a large body of the literature on the topic that should be included. Also, it would be important to developed a more quantitative analysis. One strong suggestion was to use a meta-analysis to study the problem.

3. Introduction should state more clearly the contribution of the paper

4. The concluding section should state more practical policy conclusion

5. The conclusions and discussion are based on a set of few heterogenous studies. Thus, some conclusions are reached only by looking at results and it is necessary to perform a more quantitative study.

6. it is hard to justify ruling-out cross-sectional studies since they can contain equally-valid causal claims about the causes of labor-market choices, so I think the rationale for this needs to be more explicit.

7. The paper should note that there is earlier research that examines the role of job characteristics using linked survey and register data. https://doi.org/10.1016/j.jeoa.2019.02.001

8. paper needs some revision - text and structure

9. Please, see more detailed comments below.

We would appreciate receiving your revised manuscript by Jan 17 2020 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

We look forward to receiving your revised manuscript.

Kind regards,

Bernardo Lanza Queiroz, Ph.D

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

1. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information.

2. PLOS requires an ORCID iD for the corresponding author in Editorial Manager on papers submitted after December 6th, 2016. Please ensure that you have an ORCID iD and that it is validated in Editorial Manager. To do this, go to ‘Update my Information’ (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field. This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager. Please see the following video for instructions on linking an ORCID iD to your Editorial Manager account: https://www.youtube.com/watch?v=_xcclfuvtxQ

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: No

Reviewer #3: Partly

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: No

Reviewer #3: N/A

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: Comments

1. The revised introduction should state the contribution to the literature.

2. Three authors independently conducted the search for relevant research. Is the use of three authors a standard procedure in the literature?

3. Could the studies be weighted by the quality of research design (e.g. cross-sectional vs. panel data estimations)?

4. Six studies in the analysis focus on Germany (page 7). This is a large proportion compared to the economic size of German. Are the results that are presented in the paper globally representative?

5. What was the criteria that the study evaluated “causality”?

6. Older individuals who work and are not retired are those who are the healthiest and probably also the most motivated to work. This may cause potential problems when comparing the results from different countries, because the health status of workers approaching the retirement age is not identical in all countries. For example, the health status of workers maybe better in the countries with universal provision of health care services.

7. The effects can be different in different institutional settings. Is this point relevant for the analysis?

8. The paper should note that there is earlier research that examines the role of job characteristics using linked survey and register data. https://doi.org/10.1016/j.jeoa.2019.02.001

9. The concluding section should state more practical policy conclusions.

Reviewer #2: Report on the paper: “A systematic review of causes of recent increases in ages of labor market exit in OECD countries”

Summary and overall appraisal

According to the authors, the aim of the paper is, on the one hand, to provide an overview of the retirement domains into which current evidence falls, thus identifying research gaps; and, on the other hand, to assess the proportion of change in ages of labor market exit attributable to each investigated cause.

The paper does not have a high technical complexity from a statistical point of view. In fact they do not carry out any statistical analysis by themselves and their conclusions rely on the statistical analysis (mainly econometric analysis) from other research works.

Although the topic studied is relevant and hence worthy of investigation, currently the paper lacks some important issues that should be addressed in a future version of it, in order to reach a publishable standard, from my point of view. Thus, I encourage the authors to take into account seriously the comments detailed below.

Comments

1. As the own authors point out, one of their basic goals is to perform a “systematic review”. In this vein, I miss some relevant references on this literature, for example:

Pérez, C., Martín-Román, Á., & Moral, A. (2015). The impact of leisure complementarity on the labour force participation of older males in Spain. Applied Economics Letters, 22(3), 214-217.

The authors should also check some of the references contained in it. Another interesting paper for a country not belonging to the OECD, but containing literature on this topic is:

Queiroz, B. L., & Souza, L. R. (2017). Retirement incentives and couple’s retirement decisions in Brazil. The Journal of the Economics of Ageing, 9, 1-13.

2. In addition, there is a fresh paper investigating thoroughly one of the domains that the authors identify. Despite being published after the bibliographical database was collected (February 2019), I think it could be also included in a revised version of the paper:

Pérez, C., Martín-Román, Á., & Moral, A. (2019). Two decades of the complementary leisure effect in Spain. The Journal of the Economics of Ageing, forthcoming.

3. Tables 4 and 5 constitute the core of the paper. The authors carry out a descriptive study by reviewing some of the literature on the topic. However, I am afraid that they are comparing a set of very heterogeneous results. I think no clear conclusion can be obtained by just looking at these numbers without a quantitative treatment of them. Particularly, I deem that with such analysis they cannot “assess the proportion of change in ages of labor market exit attributable to each investigated cause” as they state when defining their goals.

4. Following the argument elaborated in my previous comment, I would suggest the authors to perform a meta-analysis. The information contained in tables 4 and 5 might be a promising starting point to do that.

5. I would also suggest the authors to review the writing and to rewrite some parts of paper in a more structured manner.

Reviewer #3: Summary:

Overall, the paper presents a rough meta-analysis of papers addressing causes of labor-market decisions at older ages. The paper is well-written and was an interesting read. The paper also makes a noble attempt at summarizing the state of knowledge in an extremely important and policy-relevant area. However, I identified a few concerns with the paper.

Concerns:

1. I am concerned that Column 7 in Table 4 is not recording the "change attributable to explanatory variables" but rather effect sizes. An inspection of the Staublil and Zewimuller (2013) paper seems to just be indicating coefficients rather than changes explained by (the 9.8 and 11.0 are the coefficients of regressions estimated by the paper, not the fraction of change attributable to the explanatory variable). The same seems true of the Qi et al (2018) paper, based on a quick look at an early version of that paper. If this is true, then I'm not sure what the difference is in the papers in Table 4 vs. those in Table 5.

2. I'm not sure if there are other similar meta-analyses done that attempt explicitly what the paper is doing (summarizing literature and addressing gaps). If there are any, they should be mentioned. If not, the authors should make this explicit.

3. The authors seem to be interested in addressing impacts on retirement ages over time as noted on line 84 and on lines 122-124, to the exclusion of cross-sectional studies. That the authors seem to be interested only in studies with a longitudinal dimension seems a bit odd, since I would assume that some external validity concerns are common to almost all studies. Why are cross-sectional findings not relevant? Is this concern about econometrics or something else? To put things simply, I find it hard to justify ruling-out cross-sectional studies since they can contain equally-valid causal claims about the causes of labor-market choices, so I think the rationale for this needs to be more explicit.

Much more minor issues:

3. The discussion on lines 49 through 57 seems a bit awkwardly framed. The authors seem to imply that labor force exit ages are rising, however yet also address issues with what would happen if these ages didn't rise. The discussion could be more consistently framed in a way that raises a need for the findings of the paper.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

Reviewer #3: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org. Please note that Supporting Information files do not need this step.

Submitted filename: letter to authors.pdf

Click here for additional data file.

Submitted filename: letter to editor.pdf

Click here for additional data file.

30 Jan 2020

Editor

1. The topic is relevant and hence worthy of investigation, currently the paper lacks some important issues that should be addressed in a future version

We would like to thank the editor and three anonymous reviewers for their constructive comments. Below, we respond to each of the issues that were raised by the reviewers. Changes in the manuscript are highlighted.

2. The reviewers did not consider the review systematic since only 18 papers were reviewed and there is a large body of the literature on the topic that should be included. Also, it would be important to developed a more quantitative analysis. One strong suggestion was to use a meta-analysis to study the problem.

We would like to address separately: (1) the issue of the number of articles included in our review; and (2) the importance of developing a more quantitative analysis.

(1) Number of articles

Systematic review is a method that ‘seeks to systematically search for, appraise and synthesize research evidence, often adhering to the guidelines on the conduct of a review’ (Grant and Booth 2009, p. 102). The Prisma protocol, which we followed for this review, does not specify a lower boundary for the number of studies that should be included in a systematic review, and neither do other protocols that we are aware of. Other systematic reviews related to retirement reviewed a similar number of studies (e.g., Scharn et al. 2018 reviewed 20; Barnett et al. reviewed 19) or even fewer (Cloostermans et al. reviewed 4). Perhaps, in our original submission, we did not clearly enough describe the goal of our review and the inclusion criteria for the systematic review. Our review is strictly focused on empirical articles that offered an explanation of changes in labor force participation / ages of labor market exit in OECD countries, published since the year 2000. We purposely excluded studies that concentrated on variation in retirement behavior among individuals and between countries at one point in time. These have been summarized in other review articles that we refer to extensively (i.e., Fischer et al. 2017; Scharn et al. 2019). In the revised manuscript, we have modified the introduction section to state the goal of our systematic review more clearly. We have also improved the description of our inclusion criteria to be more explicit. Finally, we assessed the articles that were proposed by the reviewers, and deemed one study relevant for inclusion in our review (i.e. Pérez, Martín-Román & Moral 2019).

(2) Importance of developing a more quantitative analysis

A meta-analysis consists of statistically combining “the results of quantitative studies to provide a more precise effect of the results” (Grant and Booth 2009, p. 98). We agree that a more quantitative analysis would be helpful and we considered it at the earlier stages of work. Unfortunately, a meta-analysis is not an option for this systematic review. The main reason for this is that the methodologies and results presented in our review are, as one reviewer correctly points out, relatively disparate. There exists ample discussion about the kind of results that a meta-analysis should include, but the mainstream knowledge is to only combine studies that include comparable results. Considering the literature that did perform such analysis, narrowly defined dependent and independent variables are usually included. In our case, independent variables denoted considerably varying realities, e.g. change in educational attainment, change in the proportion of workers who contribute to defined contribution schemes, or change in the normal retirement age. Furthermore, studies included in our review had various designs, for example some considered married men ages 55 to 64 in the 1990s in the United States, while others considered women ages 62 to 64 in the 2000s in Switzerland. Methodologies varied considerably too, with some studies using probit models and others relying on a shift-share analysis, for example. As a result, a meta-analysis was not feasible for this article.

We nevertheless made substantial efforts in order to make the results more comparable across studies. First, quantitative results were originally presented in tables, alongside the context in which they were arrived at. These quantitative results are now presented in graphs, which facilitate comparison between studies. Second, results that were originally presented in Table 4 referred to periods of varying lengths, making comparisons across studies difficult. In the new version, results were converted into annual change, thus again improving comparison between studies. Third, results from Table 5 referred to different metrics but no clear distinction was made between studies according to the metric used. The new version presents results from different studies by grouping them according to the way that the outcome variable was measured, which make comparisons between studies with similar outcomes easier. Fourth, results were further grouped according to classes of predictors in order to improve comparisons. New tables were made which accompany the figures and give important information about the context in which studies were performed. Text in the methods and results sections was adjusted to reflect these changes. In sum, although a meta-analysis appeared impossible to perform, we believe that results in the new manuscript can be more easily compared across studies.

Furthermore, we reformulated the objective of the review stated in the introduction to reflect better the analyses that are reported in the results section. The way that the second goal of the study is formulated was slightly changed and now reads “Secondly, we quantitatively assess the amount of increases in ages of labor market exit that can be attributed to each investigated cause. Effects are presented individually for each cause, referring to the context in which they were estimated.”

Finally, we have added a separate limitation section to the discussion, where we discuss how designs and methodologies of the articles that we reviewed vary considerably, warranting caution when interpreting the results.

3. Introduction should state more clearly the contribution of the paper

We have extensively revised the introduction so that it now states the contribution of the paper more clearly (paragraph 5 in the introduction).

4. The concluding section should state more practical policy conclusion.

We have revised the discussion and conclusion sections to clarify the implications for policy makers.

5. The conclusions and discussion are based on a set of few heterogenous studies. Thus, some conclusions are reached only by looking at results and it is necessary to perform a more quantitative study.

The issue of performing a more quantitative analysis was addressed above, under comment #2.

6. it is hard to justify ruling-out cross-sectional studies since they can contain equally-valid causal claims about the causes of labor-market choices, so I think the rationale for this needs to be more explicit.

We did not a priori rule out cross-sectional studies from our systematic review. However, we explicitly focus on studies that explain changes in ages of labor force exit over time. Conceptually, cross-sectional studies could have been included in our review if they satisfied the inclusion criteria, but none did. Nevertheless, we follow the reviewer’s advice and revised the Methods section to clarify our inclusion criteria, also mentioning that we did not rule out cross-sectional studies.

7. The paper should note that there is earlier research that examines the role of job characteristics using linked survey and register data.

The potential inclusion of this research was assessed following our initially established criteria. However, it did not satisfy the criterion stating that studies should aim at explaining change over time in retirement behavior (see also comments under comment #2).

8. paper needs some revision - text and structure

The paper underwent a thorough revision of its text and structure.

9. Please, see more detailed comments below.

The following addresses each detailed comment separately.

Reviewer #1

1. The revised introduction should state the contribution to the literature.

We have extensively revised the introduction section so that it now states the contribution of this systematic review more clearly (paragraph 5 in the introduction).

2. Three authors independently conducted the search for relevant research. Is the use of three authors a standard procedure in the literature?

While no strict guidelines are formulated in the PRISMA protocol, a survey of several systematic reviews confirms that the use of three authors is a standard procedure for articles search and inclusion. For example, in Scharn et al. (2018), two authors independently screened titles and abstract, and a third one was consulted if consensus was not reached.

3. Could the studies be weighted by the quality of research design (e.g. cross-sectional vs. panel data estimations)?

All included studies are longitudinal (i.e. panel data), so we cannot compare them with cross-sectional findings. In the new version we specify in the Methods part that studies with either cross-sectional or panel data were considered for inclusion. We chose not to weight studies on any other criteria, as any such criterion and the weight would be relatively arbitrary and add unnecessary complexity.

4. Six studies in the analysis focus on Germany (page 7). This is a large proportion compared to the economic size of German. Are the results that are presented in the paper globally representative?

Germany, as well as the United States are indeed overrepresented in our results. We have added text to the discussion section (under limitations) to emphasize this issue. While single study findings cannot directly be applied to other countries, we think that this review has considerable value as it is the first to summarize available evidence from different country contexts.

5. What was the criteria that the study evaluated “causality”?

We have revised the Methods section to clarify our criteria for determining whether a study made efforts towards tracking causality: “Articles were deemed as having made effort towards tracking causality if they used statistical models controlling for potentially spurious correlation and included instruments controlling for endogenous relations, for example by using control groups not affected by the independent variable of interest.”

6. Older individuals who work and are not retired are those who are the healthiest and probably also the most motivated to work. This may cause potential problems when comparing the results from different countries, because the health status of workers approaching the retirement age is not identical in all countries. For example, the health status of workers maybe better in the countries with universal provision of health care services.

We agree with this point, which addresses that the included studies use different populations and refer to different institutional contexts. On the one hand, this is a limitation and we include it in the revised discussion section, under limitations. On the other hand, the goal of this study was to provide an overview of finding from different settings, even if full comparability is impossible.

7. The effects can be different in different institutional settings. Is this point relevant for the analysis?

This point is related to the previous one. The revised discussion section now mentions this as a limitation.

8. The paper should note that there is earlier research that examines the role of job characteristics using linked survey and register data.

The potential inclusion of this research was assessed following our initially established criteria, but it unfortunately did not satisfy the one stating that studies should aim at explaining change over time in retirement behavior (see also editor’s comments under point #2).

9. The concluding section should state more practical policy conclusions.

We have revised the discussion and conclusion sections to clarify the implications for policy makers.

Reviewer #2

Summary and overall appraisal

According to the authors, the aim of the paper is, on the one hand, to provide an overview of the retirement domains into which current evidence falls, thus identifying research gaps; and, on the other hand, to assess the proportion of change in ages of labor market exit attributable to each investigated cause.

The paper does not have a high technical complexity from a statistical point of view. In fact they do not carry out any statistical analysis by themselves and their conclusions rely on the statistical analysis (mainly econometric analysis) from other research works.

Although the topic studied is relevant and hence worthy of investigation, currently the paper lacks some important issues that should be addressed in a future version of it, in order to reach a publishable standard, from my point of view. Thus, I encourage the authors to take into account seriously the comments detailed below.

Comments

1. As the own authors point out, one of their basic goals is to perform a “systematic review”. In this vein, I miss some relevant references on this literature, for example:

Pérez, C., Martín-Román, Á., & Moral, A. (2015). The impact of leisure complementarity on the labour force participation of older males in Spain. Applied Economics Letters, 22(3), 214-217.

The authors should also check some of the references contained in it. Another interesting paper for a country not belonging to the OECD, but containing literature on this topic is:

Queiroz, B. L., & Souza, L. R. (2017). Retirement incentives and couple’s retirement decisions in Brazil. The Journal of the Economics of Ageing, 9, 1-13.

We have revised the manuscript so that our inclusion criteria are clarified in the introduction. The articles mentioned in this comment and their references were also checked using these criteria, but did not qualify for inclusion in our systematic review (see also reply to editor’s comment #2).

2. In addition, there is a fresh paper investigating thoroughly one of the domains that the authors identify. Despite being published after the bibliographical database was collected (February 2019), I think it could be also included in a revised version of the paper:

Pérez, C., Martín-Román, Á., & Moral, A. (2019). Two decades of the complementary leisure effect in Spain. The Journal of the Economics of Ageing, forthcoming.

Thank you for this suggestion. This article satisfied our criteria and was therefore included in the revised manuscript. It was probably published after our initial search, which is why it was not included in the previous version.

3. Tables 4 and 5 constitute the core of the paper. The authors carry out a descriptive study by reviewing some of the literature on the topic. However, I am afraid that they are comparing a set of very heterogeneous results. I think no clear conclusion can be obtained by just looking at these numbers without a quantitative treatment of them. Particularly, I deem that with such analysis they cannot “assess the proportion of change in ages of labor market exit attributable to each investigated cause” as they state when defining their goals.

4. Following the argument elaborated in my previous comment, I would suggest the authors to perform a meta-analysis. The information contained in tables 4 and 5 might be a promising starting point to do that.

We address points 3 and 4 jointly as they are very similar. Please see our reply to the editor’s comment #2. To summarize, we did not perform a meta-analysis because the results are not sufficiently comparable. However, we transformed some of the results into more comparable metrics and improved their presentation to make them more comparable.

5. I would also suggest the authors to review the writing and to rewrite some parts of paper in a more structured manner.

We have thoroughly revised and rewrote parts of the paper in a more structured manner.

Reviewer #3

Overall, the paper presents a rough meta-analysis of papers addressing causes of labor-market decisions at older ages. The paper is well-written and was an interesting read. The paper also makes a noble attempt at summarizing the state of knowledge in an extremely important and policy-relevant area. However, I identified a few concerns with the paper.

Concerns:

1. I am concerned that Column 7 in Table 4 is not recording the "change attributable to explanatory variables" but rather effect sizes. An inspection of the Staublil and Zewimuller (2013) paper seems to just be indicating coefficients rather than changes explained by (the 9.8 and 11.0 are the coefficients of regressions estimated by the paper, not the fraction of change attributable to the explanatory variable). The same seems true of the Qi et al (2018) paper, based on a quick look at an early version of that paper. If this is true, then I'm not sure what the difference is in the papers in Table 4 vs. those in Table 5.

We extensively revised our presentation of the results (new tables 4-5 and figures 3-4). The terminology used was checked thoroughly to reflect the actual measures. Tables 4-5 in the revised manuscript may now refer to different studies than in the original manuscript. More specifically, Table 4 now summarizes articles studying differences in LFP between two points in time, while Table 5 summarizes articles studying differences in LFP/ERA between age- or cohort-groups, distinguishing between different types of outcome variable. We have further revised the results and discussion sections to reflect this different approach.

2. I'm not sure if there are other similar meta-analyses done that attempt explicitly what the paper is doing (summarizing literature and addressing gaps). If there are any, they should be mentioned. If not, the authors should make this explicit.

We are not aware of a systematic review with same aims as ours. We now mention this explicitly in the fifth paragraph of the introduction.

3. The authors seem to be interested in addressing impacts on retirement ages over time as noted on line 84 and on lines 122-124, to the exclusion of cross-sectional studies. That the authors seem to be interested only in studies with a longitudinal dimension seems a bit odd, since I would assume that some external validity concerns are common to almost all studies. Why are cross-sectional findings not relevant? Is this concern about econometrics or something else? To put things simply, I find it hard to justify ruling-out cross-sectional studies since they can contain equally-valid causal claims about the causes of labor-market choices, so I think the rationale for this needs to be more explicit.

We agree that longitudinal and cross-sectional studies can contain equally valid causal claims about the causes of labor market choices. As mentioned above (editor’s comment #6), we did not a priori rule out cross-sectional studies. It is important to understand, however, that our focus was not on labor market choices in general, but on change over time in ages of labor force exit. Conceptually, cross-sectional studies could have been included in our review if they satisfied this criterion, but our search did not yield any such study. We have revised the Methods section to clarify our inclusion criteria. It also mentions that cross-sectional studies were not explicitly ruled out.

Much more minor issues:

4. The discussion on lines 49 through 57 seems a bit awkwardly framed. The authors seem to imply that labor force exit ages are rising, however yet also address issues with what would happen if these ages didn't rise. The discussion could be more consistently framed in a way that raises a need for the findings of the paper.

We agree with this point and have thoroughly revised this part of the discussion section, emphasizing the findings of the paper.

References

Barnett, I., van Sluijs, E. M., & Ogilvie, D. (2012). Physical activity and transitioning to retirement: a systematic review. American journal of preventive medicine, 43(3), 329-336.

Cloostermans, L., Bekkers, M. B., Uiters, E., & Proper, K. I. (2015). The effectiveness of interventions for ageing workers on (early) retirement, work ability and productivity: a systematic review. International archives of occupational and environmental health, 88(5), 521-532.

Fisher, G. G., Chaffee, D. S., & Sonnega, A. (2016). Retirement timing: A review and recommendations for future research. Work, Aging and Retirement, 2(2), 230-261.

Grant, M. J., & Booth, A. (2009). A typology of reviews: an analysis of 14 review types and associated methodologies. Health Information & Libraries Journal, 26(2), 91-108.

Pérez, C., Martín-Román, Á., & Moral, A. (2020). Two decades of the complementary leisure effect in Spain. The Journal of the Economics of Ageing, 15, 100216.

Scharn, M., Sewdas, R., Boot, C. R., Huisman, M., Lindeboom, M., & Van Der Beek, A. J. (2018). Domains and determinants of retirement timing: A systematic review of longitudinal studies. BMC public health, 18(1), 1083.

Submitted filename: reply_reviewers_final.docx

Click here for additional data file.

11 Mar 2020

PONE-D-19-28928R1

A systematic review of causes of recent increases in ages of labor market exit in OECD countries

PLOS ONE

Dear Dr. Boissonneault,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

We appreciate the authors’ receptiveness to earlier critiques and their inclusion of changes and the authors’ receptiveness to earlier critiques and their inclusion of changes. However, based on the reviewers comments and my own reading, I believe the paper needs some additional adjustments.

In general, there is an disagrement by how  systematic is the review. There are some  miussing relevant references on this literature.

the descriptive analysis do not lead to the conclude that the paper  assess “the proportion of change in ages of labor market exit attributable to each investigated cause”,

one suggestions is to incorporate a broader literature and perform a meta-analysis

the paper does not carry out any statistical analysis but, the conclusions rely on the statistical analysis. So, we believe this is an important adjustment to the paper.

The detailed comments by Reviewer #2 and # 3 are organized in the attached document.

We would appreciate receiving your revised manuscript by Apr 25 2020 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

We look forward to receiving your revised manuscript.

Kind regards,

Bernardo Lanza Queiroz, Ph.D

Academic Editor

PLOS ONE

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

Reviewer #2: (No Response)

Reviewer #3: (No Response)

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: No

Reviewer #3: Partly

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: No

Reviewer #3: No

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: No

Reviewer #2: Yes

Reviewer #3: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: I am happy with the revised version of the paper. I like the research question, the structure of the paper, the quality of writing, and the way the authors describe their empirical proceeding and results. Most importantly, the authors have addressed all the issues stated in my referee report for the first version appropriately.

Reviewer #2: Report on the paper: “A systematic review of causes of recent increases in ages of labor market exit in OECD countries” Revision 1.

First of all, I would like to acknowledge that the authors have carried out an important review work. Thus, I would like to congratulate them for that. I strongly believe that the paper has improved significantly in this second round. However, I still have some important remarks to make.

1. Now I understand the label “systematic review”. The authors explain that it is the name of a research technique. In any case, the number of articles used is still very small to draw any sensible “general” conclusion. Despite the fact that they make use of arguments from authority to justify this sample size, I cannot see how they can make any inference for the OECD countries. Particularly with such an unbalanced sample of countries (e.g. six studies for Germany and so on). I align myself here with reviewer 1.

2. As for the quantitative analysis, the authors claim that “a meta-analysis was not feasible for this article”. I do not agree with that. It is true, however, that they should define an appropriate dependent variable (this the first and key decision to be made) and a set of independent variables, many of them defined as dummy variables. In order to discard the meta-analysis, they state that “their independent variables denoted varying realities”, “studies included had various designs” and “methodologies varied considerably too”. I wonder if such “flaws” do not impede to perform a wise qualitative analysis. My point of view is different: a good meta-analysis could overcome that issues.

3. As regards the quantitative analysis again, the authors allege to have performed some quantitative analysis. I do not think so. For instance, they state: “quantitative results were originally presented in tables”. In my view, what they call quantitative results are only some figures extracted from other research works. I do not want to appear as a maximalist quantitative researcher, but the truth is that PLOS ONE is a journal that specifically encourages the use of a sound statistical analysis. In fact, one of the questions I have to answer as a reviewer is: “Has the statistical analysis been performed appropriately and rigorously?” To be honest there is a lack of statistical analysis. There is only a use of statistics from a descriptive perspective.

4. After reading the authors’ response to my first comment in the previous round, I am seriously questioning the inclusion criteria followed. If Perez et al. (2015) does not qualify for inclusion, that criteria should be revised. I affirm this because this article practically mimics the work by Schirle (2008), which is included.

Reviewer #3: I would supplement the data included in the paper with exactly what point estimates and what results were used from each paper.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

Reviewer #3: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org. Please note that Supporting Information files do not need this step.

Submitted filename: review_comments.pdf

Click here for additional data file.

20 Mar 2020

See attached document for a complete response to all comments from the three reviewers and the editor.

Submitted filename: round2_reply_reviewers_final.docx

Click here for additional data file.

3 Apr 2020

A systematic review of causes of recent increases in ages of labor market exit in OECD countries

PONE-D-19-28928R2

Dear Dr. Boissonneault,

We are pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it complies with all outstanding technical requirements.

Within one week, you will receive an e-mail containing information on the amendments required prior to publication. When all required modifications have been addressed, you will receive a formal acceptance letter and your manuscript will proceed to our production department and be scheduled for publication.

Shortly after the formal acceptance letter is sent, an invoice for payment will follow. To ensure an efficient production and billing process, please log into Editorial Manager at https://www.editorialmanager.com/pone/, click the "Update My Information" link at the top of the page, and update your user information. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, you must inform our press team as soon as possible and no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

With kind regards,

Bernardo Lanza Queiroz, Ph.D

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

13 Apr 2020

PONE-D-19-28928R2

A systematic review of causes of recent increases in ages of labor market exit in OECD countries

Dear Dr. Boissonneault:

I am pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

For any other questions or concerns, please email plosone@plos.org.

Thank you for submitting your work to PLOS ONE.

With kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Bernardo Lanza Queiroz

Academic Editor

PLOS ONE