Drivers of Labor Force Participation in Advanced Economies: Macro and Micro Evidence*
Author: Mr. Francesco Grigoli, Zsoka Koczan, and Petia Tapalova
Publication Date:
25 Jun 2018
eISBN:
9781484364581
Language:
English
Keywords:
WP; workforce
Download PDF (557.6 KB)

Despite signicant headwinds from population aging in most advanced economies (AEs),labor force participation rates show remarkably divergent trajectories both across countriesand across diferent groups of workers. Participation increased sharply among prime-age womenand, more recently, older workers, but fell among the young and prime-age men. This pa-per investigates the determinants of these trends using aggregate and individual-level data.We finnd that the bulk of the dramatic increase in the labor force attachment of prime-agewomen and older workers in the past three decades can be explained by changes in labor mar-ket policies and institutions, structural transformation, and gains in educational attainment.Technological advances such as automation, on the other hand, weighed on the labor supplyof prime-age and older workers. In light of the dramatic demographic shifts expected in thecoming decades in many AEs, our fndings underscore the need to invest in education andtraining, reform the tax system, reduce early retirement incentives, improve the job-matchingprocess, and help individuals combine family and work life in order to alleviate the pressures from aging on labor supply.

Abstract

Despite signicant headwinds from population aging in most advanced economies (AEs),labor force participation rates show remarkably divergent trajectories both across countriesand across diferent groups of workers. Participation increased sharply among prime-age womenand, more recently, older workers, but fell among the young and prime-age men. This pa-per investigates the determinants of these trends using aggregate and individual-level data.We finnd that the bulk of the dramatic increase in the labor force attachment of prime-agewomen and older workers in the past three decades can be explained by changes in labor mar-ket policies and institutions, structural transformation, and gains in educational attainment.Technological advances such as automation, on the other hand, weighed on the labor supplyof prime-age and older workers. In light of the dramatic demographic shifts expected in thecoming decades in many AEs, our fndings underscore the need to invest in education andtraining, reform the tax system, reduce early retirement incentives, improve the job-matchingprocess, and help individuals combine family and work life in order to alleviate the pressures from aging on labor supply.

1 Introduction

Population growth in advanced economies (AEs) is slowing, life expectancy is rising, and the number of elderly is increasing steeply. As these trends gather pace, the United Nations projects that by the middle of this century, total population will be shrinking in almost half of AEs and individuals of what is currently considered working age will be supporting close to double the number of elderly that they do now (Figure 1, Panels 1 and 2). Unless more people participate in labor markets, aging could slow AEs’ growth, and, in many cases, undermine the sustainability of their social security systems (Clements et al., 2015).

Figure 1:
Figure 1:

Demographic Transition in AEs

Citation: IMF Working Papers 2018, 150; 10.5089/9781484361528.001.A001

Source: Authors’ calculations.Notes: Based on a sample of 36 AEs. The black lines denote the averages, shaded dark grey areas denote the interquartile ranges for AEs, and light grey shaded areas indicate projections.

Even though population aging already exerts pressure on labor supply, aggregate labor force participation rates evolved quite differently across AEs.1 In half of these economies, the aggregate participation rate actually increased since the global financial crisis of a decade ago, which coincided with an acceleration of the demographic transition (Figure 1, Panel 3). Headline numbers also hide stark differences in the participation rates of different groups of workers. For example, prime-age male participation rates declined almost everywhere, while female participation rates increased. More recently, older workers became increasingly likely to remain in the labor force longer, while the participation of the young fell.

What underlies these diverging trajectories across countries and for different workers? Various forces are likely at play. Differences in the exact timing and pace of demographic transition could explain some of the divergence. However, the disparity in participation trends across specific groups of workers suggests a potentially important role for policies and institutions that influence the decisions of individuals to join, remain in, or reenter the labor force. Another factor could be differences in exposure and resilience to global forces such as technological advances and trade that may have depressed long-term demand for workers with certain skill sets (Acemoglu and Autor, 2011; Autor and Dorn, 2013; Goos et al., 2014; Autor et al., 2016; IMF, 2016b; IMF, 2017a). Identifying and ranking the key drivers of participation across population groups is necessary in designing policies that could enable those willing to work to do so and counteract the forces of aging.

Motivated by these considerations, in this paper, we investigate the key drivers of changes in aggregate participation rates and the attachment of various groups of workers to the labor force in a large sample of AEs using both aggregate data from the past three and a half decades and individual-level data since the beginning of the 2000s. We assess the relative importance of cyclical and structural changes in the economy, labor markets policies and institutions, and policies targeting specific groups of workers, namely women, older individuals, and migrants. Importantly, we examine whether shifts in the demand for certain types of labor due to the automation of routinizable tasks weighed on labor force participation of affected workers, using two complementary empirical approaches. First, we leverage the cross-country heterogeneity in the initial mix of employment across occupations to analyze whether declines in the relative price of investment led to greater reductions in participation rates in economies where a larger share of occupations were routinizable and hence subject to automation. Second, we rely on individual level data on current and past employment from 24 European countries during 2000–16 to estimate whether individuals whose occupation is more easily routinizable have a higher likelihood of being out of the labor force. We further examine whether labor market policies and other country characteristics can help attenuate the link between routinizability of occupation and subsequent labor force detachment.

Our findings suggest that policies and institutions, such as the tax benefit system, active labor market programs, and policies that encourage specific groups to participate, together with structural changes and gains in educational attainment, account for the bulk of the dramatic increase in the labor force attachment of prime-age women and older workers in the past three decades. On the other hand, technological advances—the automation of tasks where labor is easily substitutable by capital—weighed on the participation rates of most groups of workers. Encouragingly, we find that higher spending on active labor market programs and education is associated with a lower likelihood that an individual previously employed in a routinizable sector or occupation drops out of the labor force. This likelihood is also significantly lower in urban areas, pointing to the importance of accessing diverse pools of employers in minimizing the adjustment costs associated with technology-induced structural transformation.

Our study contributes to the vast literature on the determinants of labor force participation in three distinct ways. First, it considers a wider set of factors that shape individuals’ decisions to work, including policies and institutions as well as the role of automation and structural transformation. While numerous studies leveraged cross-country heterogeneity to examine the role of policies on participation and employment outcomes of men and women in AEs (see Blanchard and Wolfers, 2000; Genre et al., 2005; Bertola et al., 2007; Bassanini and Duval, 2006; Bassanini and Duval, 2009; De Serres et al., 2012; Murtin et al., 2014; Gal and Theising, 2015), to the best of our knowledge, ours is the first study to estimate the effect of technological progress on participation in a cross-country setting.2 The role of migrant integration policies also received relatively little attention in the literature. Second, we combine the cross-country empirical findings with evidence from individual-level data to shed further light on the role of characteristics such as education and exposure to technological advances in workers’ participation decisions. Finally, we re-visit earlier evidence on the effects of labor market policies on participation of different groups of workers (see Jaumotte, 2003; Genre et al., 2010; Blau and Kahn, 2013; Cipollone et al., 2013; Thévenon, 2013; Dao et al., 2014; Christiansen et al., 2016 for cross-country analyses of female labor force participation and employment and Blöndal and Scarpetta, 1999 and Duval, 2004 for cross-country analyses of retirement decisions) in a significantly larger estimation sample. The inclusion of more recent data allows us to re-assess the validity of previous findings in a period which witnessed significant shifts in the participation behavior of some workers, such as sizable increases in the participation of older workers, the decline in participation among the young, and the plateauing of female participation gains.

The rest of the paper is organized as follows. To set the stage, Section 2 describes the key patterns of labor force participation in AEs over the past three decades. Section 3 discusses various factors likely to affect labor force participation, while Section 4 outlines the empirical strategy. Section 5 discusses the results based on aggregate and individual-level data. Section 6 concludes.

2 Patterns of Labor Force Participation

An investigation into the long-term trends of aggregate labor force participation and the workforce attachment of individual groups of workers in AEs reveals several striking patterns.3 Over the past 30 years, the aggregate average labor force participation rate in AEs as a group barely changed (Figure 2, panel 1). However, the group aggregate masks significant differences in the experience of individual countries. While in a large share of AEs aggregate labor force participation in 2016 was within a couple of percentage points of what it was in 1985, several countries saw significant increases in the workforce attachment of their populations, with aggregate participation rates gaining more than five percentage points in countries such as Germany, Korea, Spain, and the Netherlands (Figure 2, panel 2). Moreover, the distribution of participation rates across AEs narrowed remarkably.

Figure 2:
Figure 2:

Labor Force Participation Rates by Gender and Age

(Percent)

Citation: IMF Working Papers 2018, 150; 10.5089/9781484361528.001.A001

Source: Authors’ calculations.Notes: Solid lines denote medians, dotted lines denote population-weighted averages, and shaded areas denote interquartile ranges. In panels 5 to 8, solid lines in blue and red denote medians for men and women, respectively.

Even more striking is the divergence in the trends in labor force participation of different groups of workers (Figure 2, panels 3 and 8, and Figure 3). Across AEs, the share of women who are employed or actively looking for work increased by close to 10 percentage points over the past three decades. Gains in female participation were substantially larger in countries where women were historically less likely to be part of the workforce, a convergence that significantly narrowed the dispersion in women’s participation across AEs since 1985. Conversely, participation rates of men, which are significantly higher and tend to be much more similar across countries, declined almost across the board. For the median AE, the participation rate among men was more than 6 percentage points lower in 2016 than in 1985. These divergent trends narrowed gender gaps.

Figure 3:
Figure 3:

Labor Force Participation Rates in Selected Countries and Groups

(Percent)

Citation: IMF Working Papers 2018, 150; 10.5089/9781484361528.001.A001

Source: Authors’ calculations.Notes: Based on a sample of 33 AEs. The solid (dashed) [dash-dotted] lines denote averages for the United States (Europe) [other AEs]. Light red (blue) lines denote women (men). Other AEs include Australia, Canada, Japan, and Korea.

Significant differences also exist in how participation rates evolved across individuals of different ages (Figure 2, panels 5–8). The young (between ages 15–24) are significantly less likely to be part of the labor force in 2016 than in 1985, with similar trends observed for men and women. To a significant extent, declining labor force attachment reflects the secular trend toward greater investment in human capital and higher school enrollment rates.4 In fact, the share of “idle” youth—defined as those who are neither employed, unemployed, or enrolled in school—is quite small and has been stable since the early 2000s.5 Given the increase in the returns to schooling in many AEs, the decline in participation among the young could partly reflect an expected response to economic incentives (Krueger, 2017).

Participation rates of older men and women (aged over 55), on the other hand, increased significantly since the mid 1990s, following decades of steady decline.6 The increase is particularly pronounced for the 55–64 age group, but in the past decade, even individuals older than 65 remained in the labor force longer. For men, the observed increase in work force attachment at older ages likely reflects reduced retirement rates amid stable or slightly declining labor force participation at younger ages. For women, the observed increase can be associated with a growing pool of workers reaching those ages as well as changes in retirement behavior. The gains in participation among older workers should be viewed in the context of significantly longer lives. Life expectancy at birth increased by about seven years and at age 50 by over five years since 1985, prompting many countries to adopt policies to encourage longer working lives through later retirement.

Among prime-age workers, the most notable pattern is diverging trends of the labor force attachment of men versus women. The small decline in participation rates of prime-age men, which remains very high and varies little across countries, was more than offset by the dramatic entry of prime-age women in the labor force, leading to overall gains in the participation rates of prime-age workers in most AEs.7 The United States is a notable exception to this overall pattern (Figure 3). Compared to other AEs, the decline in the prime-age male labor force participation rate was particularly steep in the past decade.8 While in most other AEs an increasing share of women joined the labor force, the prime-age female labor force participation rate in the United States plateaued in the late 1990s, and has been on a declining trend since the Global Financial Crisis.

Because labor force participation patterns could reflect significant shifts in the characteristics of the prime-age populations such as education, fertility, marriage, and immigration status, Figure 4 provides a more granular picture of the changes in the participation of subgroups since 2000 for most AEs (panels 1 and 5) and advanced European economies (panels 2 to 4 and 6 to 8).9 With the notable exception of relatively low-educated women, the rise in female labor force participation is remarkably widespread. Across Europe, single and married women, those with young children (below the age of 5), older children (below the age of 15) or no children, natives and immigrants are significantly more likely to be employed or looking for work in 2016 than in 2000. For prime-age men, the decline in participation was the deepest for those with the lowest educational attainment. Across all remaining groups, there was a small decline or stagnation for the median AE, suggesting that changes in population characteristics towards groups with lower participation, such as the falling share of married prime-age single men, were sizable. The United States stands out, with particularly deep declines for both women and men in the prime-age category across all levels of educational attainment.

Figure 4:
Figure 4:

Labor Force Participation Rates of Prime-Age Men and Women by Demographic Characteristics, 2000 and 2016

(Percent)

Citation: IMF Working Papers 2018, 150; 10.5089/9781484361528.001.A001

Source: Authors’ calculations.Notes: Dark blue and dark red denote year 2000, light blue and light red denote year 2016. In panels 4 and 8, dark bars show data for 2004 instead of 2000. Panels 1 and 5 are based on data from most AEs, while panels 2 to 4 and 6 to 8 are based on data from advanced European economies. Panels 3 and 7 report statistics for married individuals. Young children are those below the age of 5; older children are those age 5-15. Level of educational attainment is defined according to the International Standard Classification of Education (ISCED). Primary education contains ISCED 2011 levels 0-2; secondary education contains ISCED 2011 levels 3-4; and tertiary education contains ISCED 2011 levels 5-8.

Although the fall in labor force participation of prime-age men appears small sized for the median AE, it is worrisome for several reasons. First, the decline is broad-based. Second, since prime-age men are still the largest segment of the labor force in AEs and have traditionally been the main income-earners for their families, even a small decline in their labor supply could have sizable macroeconomic consequences.10 Finally, detachment from the labor force during an individual’s peak productive time is associated with lower happiness and life satisfaction for men (Winkelmann and Winkelmann, 1995; Knabe and Rätzel, 2011; Lucas et al., 2004; Krueger, 2017), poorer health and higher mortality (Eliason and Storrie, 2009; Gerdtham and Johannesson, 2003; Sullivan and Von Wachter, 2009), and depressed employment prospects (Arulampalam et al., 2000; Arulampalam et al., 2001).

Interesting insights can be gleaned from the reasons prime-age workers give for being out of the labor force. Figure 5 uses data from millions of workers surveyed across 26 countries in Europe to break down the non-participants into those who are students, retired, those who are not retired but never worked before, and those who were employed before but dropped out. It further breaks down the last group of nonparticipants based on the reason they reported for their detachment from the labor force.

Figure 5:
Figure 5:

Subgroups of the Inactive, 2000 and 2016

(Percent)

Citation: IMF Working Papers 2018, 150; 10.5089/9781484361528.001.A001

Source: Authors’ calculations.Notes: Reported statistics are estimated from a random sample of 10,000 respondents per country per year. Retired includes early retirement.

Comparing the responses of prime-age men and prime-age women points to important gender differences in reasons for inactivity: for instance, women are still more likely to drop out of the labor force to look after children, while a higher fraction of men report illness and disability as reasons for not being employed. The responses also suggest that a non-negligible share of those who are out of the labor force may be “involuntarily inactive”: they used to work but stopped due to economic (demand-side) factors, rather than a personal decision. Individuals reporting being dismissed from their last job as a reason for inactivity can be seen as a lower bound for this group.11

Involuntary nonparticipants drop out disproportionately from certain sectors of the economy (Figure 6, panel 1). Wholesale and retail trade, manufacturing, mining and quarrying, and utilities together account for over half of the involuntary inactive, even though less than one third of active workers (including employed and unemployed) are attached to these sectors. Excess involuntary inactivity—measured as the difference between the inactive individuals attached to a sector as a share of all nonparticipants and the active workers attached to the same sector as a share of the labor force—tends to be concentrated precisely in sectors that have a greater share of routine jobs that are vulnerable to automation (Figure 6, panel 2).

Figure 6:
Figure 6:

The Role of Exposure to Routinization

Citation: IMF Working Papers 2018, 150; 10.5089/9781484361528.001.A001

Source: Authors’ calculations.Notes: Reported statistics are estimated from a random sample of 10,000 respondents per country per year over the period 2000–16. In panels 1 and 3, active includes employed and unemployed and involuntarily inactive refers to people inactive due to dismissal. For the inactive, sector or occupation are those of last employment. In panel 2, excess involuntarily inactive refers to the difference between the shares of active and involuntarily inactive.

These stylized facts provide suggestive evidence of the potential harm of technological progress to participation rates of certain types of workers, an issue this paper examines in greater detail. They also highlight potentially important income distributional consequences of involuntary inactivity. Displacement of workers tends to occur disproportionately among lower and middle-skilled occupations (Figure 6, panel 3) and vulnerability to routinization is especially pronounced in the middle and lower parts of the income distribution (Figure 6, panel 4).

3 Drivers of Labor Force Participation: Conceptual Framework

Numerous interrelated factors influence individuals’ decisions to supply labor at various points in their life as they assess the expected return to market work relative to non-participation. Individual characteristics, such as gender, educational attainment, previous occupation, and household structure, would clearly shape that decision, as these determine potential earnings in the market place relative to the return to household work.

But labor market programs, institutions, and non-economic factors that govern individuals’ prospects of finding (or retaining) a job and the relative benefit from working can also affect participation behavior. Some of these policies, such as the tax-benefits system, directly impact the incentives to supply labor, while others, such as wage setting institutions, can shape supply indirectly through reduced labor demand. For example, an increase in the labor tax wedge could reduce the incentives to work or seek employment, both by reducing net wages and suppressing labor demand by firms due to the increase in labor costs. However, the net effect of higher taxes on labor supply is ambiguous, as individuals may respond by working more to maintain their income. Conversely, active labor market programs that support jobseekers in finding vacancies may induce individuals to join the labor force and prevent those who temporarily lose employment from becoming permanently detached. Cultural attitudes toward people’s role in society are also important, as they determine the disutility of market work, for example, from violating social norms or personally held beliefs (Fernández, 2013).

Policies tailored to address the challenges faced by specific workers can also influence their labor supply decisions. For example, provision of childcare, as well as family-friendly policies that bring more flexibility to work arrangements, make it easier for women to combine paid employment and motherhood and may discourage transitions out of the labor market.12 For older workers, financial incentives embedded in pension systems and other social transfer programs are important considerations in retirement decisions. With migration accounting for more than half of the growth in population in AEs in the past decade (see Koczan and Hilgenstock, 2018c), the ability of migrants to join the local labor force is becoming increasingly important for aggregate labor force participation rates. Policies that enable immigrants’ swift integration into labor markets, such as provision of working rights, access to language and activation programs, and the like, can help them overcome their many disadvantages, including lack of information, poor access to informal networks, lack of transferable skills and qualifications, and low language proficiency (Aiyar et al., 2016).

Long-lasting changes in the demand for workers’ skills could also influence individuals’ workforce attachment. For example, the secular expansion of the services sector in many AEs (IMF, 2018) may have created significant employment opportunities for women, who are seen to have a comparative advantage in services, thus raising female participation.13 On the other hand, technological progress that enabled routine jobs to be automated may have reduced the demand for less-skilled labor in AEs and made certain jobs obsolete. While these global developments benefit the economy as a whole, and create new opportunities in other sectors, workers may be unable to take advantage of these opportunities due to lack of relevant skills and training, preferences, hardship involved in relocating geographically, or a perceived inadequate return compared with their previous earnings.

Participation decisions are also shaped by even more short-lived changes in labor demand, such as those due to cyclical fluctuations (Elsby et al., 2015). The rise in unemployment during recessions may lead some workers to drop out of the labor force permanently. Diminished job prospects during recessions may also induce students to remain in school longer or lead parents (women especially) with young children to stay at home instead of seeking jobs.14

4 Empirical Strategy

We use two complementary empirical approaches to tease out the relative importance of the various drivers of labor force participation for different groups of workers. First, we estimate crosscountry panel regressions to disentangle the influence of labor policies, technology and other factors on the participation of different population segments. While the potential set of drivers is large, we focus on the variables most commonly discussed in the policy debate: the tax benefit system, activation policies, including those for migrants, wage-setting institutions, the role of structural changes, and exposure to routinization. The cross-country panel approach has the advantage of capturing the general equilibrium effects of various drivers, and quantifying their role in a unified framework. However, the measurement of policies is often imperfect and the identification of causal impacts can be problematic.

Second, alongside the analysis of macro data, we examine individual-level data from 24 European economies. This data allow for a deeper look at the effect of individual characteristics, including the extent to which (past) occupation can be automated, on workforce attachment, and the potential for policies to shape this relationship.

4.1 Aggregate Analysis

The aggregate analysis examines the historical relationship between the participation of individual groups of workers and potential drivers since 1980 across 23 economies, which were classified as AEs for the entire period.15 In line with the literature, we employ a reduced-form specification of labor force participation that relates the participation rate of specific groups of workers to factors that may affect the decision to supply labor.16 Our paper, however, expands considerably the temporal coverage of the analysis, capturing the last decade during which significant changes in participation occurred. Our focus on the effects of long-lasting shocks to labor demand, such as those stemming from technological advances, and on migrant integration policies is also new.

While the potential set of drivers of labor force participation is large, the analysis, guided by the conceptual framework described in Section 3, focuses on factors that can be measured relatively consistently across countries and over time, and are most commonly discussed in policy debates. 17 In particular, we estimate the following equation:

LFPi,tg=βX,gXi,tg+βD,gDi,t+βGAP,gGAPi,t−1+βZ,gZi,t+πig+τtg+ɛi,tg(1)

where LFP denotes the participation rate of worker group g in country i in year t; X represents the set of policies and institutions (some of which are specific to group g) including the labor tax wedge, the generosity of the unemployment benefits system, public expenditure on active labor market programs, the restrictiveness of migration policy, union density, the degree of coordination in wage setting, policies that help reconcile work inside and outside the household (i.e., public spending on childcare and education as a share of GDP, the proportion of employees working part-time, and the number of weeks of job-protected maternity leave), retirement incentives (proxied by the statutory retirement age and by the generosity of pension schemes). In the baseline specification, the generosity of pension schemes is measured as old-age and incapacity spending as percent of GDP, purged from fluctuations due to cyclical and demographic factors (share of the older population in different age groups and health status, proxied by life expectancy), that may mechanically generate a negative correlation with the labor force attachment of older workers. Conceptually more-appropriate measures of incentives for early retirement, such as the change in net pension wealth from an additional year in the labor force, or pension replacement rates, would severely restrict the sample, but are examined in robustness tests. In equation 1, D denotes a set factors that may shift the demand for worker group g, as the exposure to technological progress, the size of the service sector compared to the industrial one, and urbanization; GAP is the cyclical position of the economy proxied by the output gap; Z includes other determinants of labor supply, as educational attainment. Finally, πi and τt are country- and time-fixed effects, respectively, which control for all differences across countries that are constant over time, and all shocks that are common to all countries.

We follow the literature in the construction of labor market policy variables and structural characteristics used in the analysis. 18. Relative to previous studies, however, we employ two novel measures to capture (i) the impact of technological progress, and (ii) the role of migration policies. We measure exposure to technological progress as the interaction between the country’s exposure to automation through its initial occupational mix and the relative price of investment as in Dao et al. (2017). To do so, we first compute the country’s exposure to automation by assigning routinizability scores to each occupation following Autor and Dorn (2013), and aggregating these to a time-invariant country-level score using the initial employment distribution across occupations as weights. We then interact this country score with the average relative price of investment across AEs. The intuition behind this measure is that a decline in the relative price of capital goods, driven by global technological progress, would induce firms to substitute labor for capital. However, the labor market consequences of this process would be larger in countries where a larger fraction of workers have occupations in which labor can be easily substituted by capital, such as occupations in which many tasks performed by workers are routine and can be automated.

We also construct a new measure to capture policies supportive of migrants’ integration in labor markets. Using the DEMIG POLICY database maintained by the International Migration institute, we focus on major changes in policies guiding the post-entry rights or other aspects of migrants’ integration (see also De Resende, 2014). These changes are cumulated starting in 1980 to construct an index of the restrictiveness of migration policy for each country, with a higher value denoting more restrictive policies.

As discussed previously, we use a simple cross-country panel framework to estimate the sensitivity of labor force participation of various groups of workers to the set of potential drivers. The groups comprise young workers (ages 15–24), prime-age men (25–54), prime-age women (25–54), and older workers (ages 55 and over); an additional equation is estimated for a group encompassing all workers of ages 15 and over. Results from panel unit root tests suggest that the time series of labor force participation rates for different age groups are trend stationary. Limited data availability for some of the explanatory variables also precludes from employing a dynamic specification, which, in the presence of country-fixed effects, would return biased estimates (Nickell, 1981). Some of the evidently endogenous variables, such as the output gap and trade openness, are included in the specification with a one-year lag. Given the complex correlation structure of the error term with dependence across countries, autocorrelation due to the slow-moving nature of the dependent variable, and heteroskedasticity, we correct the standard errors using the Driscoll and Kraay (1998) method to make statistical inferences.19

To quantify the role played by each of the explanatory variables, we calculate the contributions from each regressor to changes in participation of group g between year t and t’ as:

Ci,t,t′S,g=β^S,g(Si,t′g−Si,tg)(2)

with S = [X, D, GAP, Z] and where Ci,t,t′S,g is the contribution of variable S.

It is important to emphasize from the outset that, in this paper, we seek to identify patterns and correlations rather than to establish causality between various policies, structural, and individual characteristics on the one hand and labor force participation on the other. Changes in national labor market policies and institutions may reflect the evolution of societal and cultural attitudes toward work that influence observed trends in labor supply beyond their impact on policies. For example, the evolution of social norms toward more egalitarian gender roles may induce both family legislation and higher female labor force participation. Female labor supply shifts may also create political support for more family-friendly policies, leading simultaneously to higher female employment and greater parental leave rights (Olivetti and Petrongolo 2017). Yet, by providing a rich description of the cross-national and overtime patterns of labor force participation and its association with a broad set of drivers, the analysis aims to offer valuable guidance on potential areas for policy action.

4.2 Individual-Level Analysis

We complement the cross-country analysis by examining evidence on labor force participation decisions from millions of individuals in Europe. The use of micro data offers important advantages relative to the cross-country analysis discussed so far: it allows for a deeper exploration of individual and household-level determinants of participation, it also mitigates the endogeneity bias arising from omitted variables and reverse causality in regressions relying on aggregate data, and it allows zooming in on the impact of technology and the extent to which policies can help offset its effect on individuals’ decision to drop out of the labor force.20

The analysis relies on the European Labour Force Survey for 24 AEs during 2000–16.21. Due to the extremely large size of the survey, we draw a random sample of 10,000 individuals per country per year, which forms the basis of our empirical investigation. We then estimate logit models of the following form:

Φ(Sj=1)=βCCi+βHHj+βRRj+πc+υr+τt+ɛj(3)

where Φ is the probability function; S is a dummy variable for whether the individual j is in or out of the labor force; C includes a set of individual characteristics such as age, gender (for the age 55 and over group), whether the individual was born in the country or abroad, whether they live in an urban or rural area, and their highest level of completed education (up to lower secondary, upper secondary or tertiary education); H includes measures of family composition, such as number of children, number of other employed adults in the household, and whether the individual lives in a household of a single adult without children (the baseline category), a single adult with children, and a couple without or with children; R is the routinization score of an individual’s current occupation (if currently employed) or last occupation (if currently unemployed or inactive); πc, υr, and τt are country-, region-, and year-fixed effects. We cluster standard errors at the country-year level.

5 Results

5.1 Aggregate Data

The results in Table 1 indicate that education, cyclical and long-lasting shifts in labor demand, and labor market policies are strongly associated with participation rates. However, there are significant differences in the strength of association of workforce attachment to these factors across groups of workers.

Table 1:

Drivers of Labor Force Participation Rates

article image
Source: Authors’ calculations.Notes: All specifications include country and year fixed effects. Driscoll-Kraay standard errors are reported in parentheses. ***, **, and * indicate statistical significance at 1, 5, and 10 percent, respectively.

In line with economic theory, education is a powerful predictor of labor force participation. An increase in the share of workers with secondary and especially tertiary education is associated with significantly higher participation, particularly for prime-age women and older workers. Higher education is also positively associated with participation of prime-age men, but to a smaller degree, in line with the much smaller variability in their participation rates, as shown in Figure 4.22

For most groups of workers, participation rates depend on the state of the business cycle. As expected, the association is significantly higher for those more marginally attached to the workforce, such as the young and women. The results also confirm that structural transformation that may shift the demand for certain types of workers affects their labor market involvement. A relative increase in service sectors employment is typically followed by the entry of prime-age women into the labor force, while urbanization brings gains in the participation of all groups, potentially by exposing them to a larger set of job opportunities.

Conversely, while technological change can benefit the economy as a whole, and create new opportunities in other sectors, it may not be fully benign from the point of view of some workers. A decline in the relative price of investment is associated with lower participation rates in countries where the initial occupation mix is tilted toward routine-task occupations, highlighting the difficulties of workers displaced by automation in finding alternative employment (see also Koczan and Hilgenstock, 2018a, and Koczan and Hilgenstock, 2018b for sub-national evidence on the United States and Europe). This finding is consistent with the role of technological progress, along with varying exposure to routine occupations, in the decline in the labor share in AEs documented in IMF (2017b) and Dao et al. (2017), and the sizable employment losses in the United States’ local labor markets with greater exposure to robots as documented in Acemoglu and Restrepo (2017).

Participation rates are also responsive to labor market programs and institutions. In particular:

  • The tax-benefit system has a robust relationship with participation rates. Higher labor tax wedges and more generous unemployment benefits are associated with lower labor force attachment for most groups of workers, in line with findings in the cross-country literature on their effect on employment (see, for example, Gal and Theising, 2015, and references therein). The negative relationship between participation rates and the generosity of unemployment benefits, measured as the gross benefit replacement rate, is consistent with (1) the positive correlation found in cross-country data between generosity of unemployment benefits and unemployment levels, which could depress participation through a discouragement effects, and (2) the fact that in many countries the unemployment insurance system provides a path to early retirement for older workers.

  • Policies specifically geared toward improving the job-matching process are generally associated with stronger participation rates.

    • – Higher public spending on active labor market programs tends to raise the share of young and prime-age women working or seeking employment.

    • – The analysis also indicates that policies that encourage the integration of migrants can help boost prime-age workers’ participation, with more pronounced effects on women. The positive association likely reflects the success of these policies in narrowing the sizable participation gaps between native and immigrant workers, which are especially wide for women (see Koczan and Hilgenstock, 2018c). However, other channels are possible. A more migrant-friendly policy stance may bring in more immigrants. Although migrants have a lower propensity to work than natives when they arrive, they are more likely to be prime-age than the native population and may boost aggregate participation rates through compositional shifts. Several recent studies also emphasize the complementarity of migrants’ skills to those of the native population, which helps boost natives’ labor market outcomes, especially women’s.23 The negative association between more friendly migration policies and youth labor force participation is not surprising since integration measures include providing migrants with access to education and training, which could lead to a higher inflow of foreign students and increase school enrollment of non-native youth.

  • Women’s willingness to work or seek employment is significantly influenced by policies that help them reconcile work inside and outside the household. Consistent with the findings of a large body of literature, we find that better access to childcare, longer maternity leave, and greater flexibility in work arrangements are associated with higher female labor force participation. 24

  • For older workers, incentives for retirement have a powerful effect on labor force attachment. 25 Raising the statutory retirement age is associated with delayed exit from the labor market, while greater pension scheme generosity seems to encourage early retirement. The latter finding is robust to using conceptually more appropriate, but less widely available, measures of incentives for early retirement, such as the implicit tax on continued work, or pension replacement rates.

  • Finally, the evidence on the role of wage-setting institutions—unionization, and the degree of wage bargaining coordination—is mixed. Higher coordination of wage setting is associated with greater labor force participation for most groups of workers, consistent with the idea that more coordinated bargaining systems may lead to wage moderation during downturns as unions internalize the potentially detrimental effects that excessive wage pressure may have on overall employment (Soskice, 1990, and Bassanini and Duval, 2006). However, the correlation between unionization and participation is less robust to changes in the sample, or the inclusion of other policies.

Putting policies, education, structural shifts, and technology together, we examine the contributions of these factors to changes in participation rates between 1995 and 2011 in Figure 7. Supportive policies and educational gains were key factors behind the dramatic increase in the participation of prime-age women and older workers, with structural transformation contributing positively as well. On the other hand, technological advances weighed on participation for all groups of workers, except the young.

Figure 7:
Figure 7:

Average Contributions to Changes in Participation Rates, 1995–2011

(Percentage points)

Citation: IMF Working Papers 2018, 150; 10.5089/9781484361528.001.A001

Source: Authors’ calculations.Notes: Other AEs include Australia, Canada, Japan, and New Zealand.

For the young (and to a certain extent prime-age male workers), a significant share of the decline in participation is attributed to a common component across AEs, captured by the time effects in the regressions. This common factor could reflect the common influence of global forces, such as technological progress or globalization, concurrent changes in policies, structural transformations, or other factors that may affect labor supply decisions across the advanced world, including changing returns to education, rising life expectancy, or common scars from the global financial crisis. For older workers, the latter might have delayed retirement, as captured in the positive common component, due to suppressed returns on retirement savings as global interest rates fell, losses in financial wealth, and potentially higher indebtedness.

Comparing how the various factors relate to participation changes across geographic regions can shed light on the reasons behind their (sometimes) divergent trends. For example, the analysis reveals that the striking difference in the participation trend for women in the United States relative to the average European trend can be attributed to the more supportive policy changes in Europe as well as the larger gains in educational attainment among prime-age European women. The factors behind the rise in participation among older workers are very similar across all regions: gains in education, structural transformation, and the introduction of policies that discourage early retirement.26 However, the reason why prime-age men and the youth in the United States became so much more disconnected from the labor market than their European counterparts remains somewhat puzzling, as evidenced by the sizable residual in the decomposition of the change. Many hypotheses have been put forth for this decline that are specific to the United States and can, consequently, not be evaluated in a cross-country setting, such as the role of rising disability and opioid usage, higher incarceration rates, and improved leisure technology.27, 28

5.2 Individual-Level Data

We report the results of the logit estimations of equation (3) in Table 2. In line with the aggregate findings, the analysis points to large and significant effects of higher education on participation. Having tertiary education roughly doubles the odds of being active in the labor market relative to having up to lower secondary education, with somewhat larger effects for women. Living in urban areas also increases participation rates, likely on account of having access to a more diverse labor market with more opportunities. Natives are also more likely to participate than immigrants.

Table 2:

Determinants of Being in the Labor Force

article image
Source: Authors’ calculations.Notes: Logit regressions are based on a random sample of 10,000 respondents per country per year from 19 countries. The table reports exponentiated coefficients. All specifications include country, region, and year fixed effects. The base category for education is “up to lower secondary education”, for family composition the base category is “one adult without children”. Standard errors clustered at the country-year level. ***, **, and * indicate statistical significance at 1, 5, and 10 percent, respectively.

Family composition has a considerable influence on the decision of an individual to work or seek employment, although there are large gender differences. Relative to the baseline category of being the only adult in a household without children, being part of a couple and having children is associated with higher participation of men, but lower participation of women. Similarly, more children are associated with lower participation of women, but higher participation of men, consistent with the historical allocation of work across genders within a household. Interestingly, the presence of other employed adults in the household is associated with a higher likelihood of being active, likely pointing to common local labor market effects. These findings should, however, be treated as associations rather than causal effects as labor supply decisions and family composition are likely jointly decided. 29

Finally, in line with the country-level results, the micro analysis points to significant negative effects of exposure to routine tasks on the probability of being part of the labor force. Working/having worked in an occupation that is more vulnerable to routinization is associated with lower odds of participation. This effect is especially pronounced for workers of ages over 55. The effects are both statistically and economically significant: a unit change in routinization scores roughly corresponds to the difference in the routinization score of technicians and the routinization score of managers. Whereas about 87 percent of prime-age male managers are active, about 84 percent of prime-age male technicians are in the labor force—the difference in their routinization scores alone can explain about one third of this 3 percentage-point difference in participation rates.30

Can policies help those vulnerable to losing their jobs to technology remain active in the labor market? To answer this question, we examine whether various country-level labor market policies, such as spending on active labor market programs or employment protection, can offset some of the negative effect of routinizability on participation. We augment the logit model described above with an interaction between the routinization score and the relevant policy measure. Table 3 reports the effect of a unit change in the routinization score, estimated at the 25th and 75th percentiles of the distribution of policies (in other words, in countries with relatively low versus relatively high spending on active labor market programs, and the like).

Table 3:

Effects of Policies on the Relationship between Participation and Routinization

article image
Source: Authors’ calculations.Notes: Logit regressions are based on a random sample of 10,000 respondents a country a year of 19 countries, controlling for individual characteristics, household composition, year-, country- and region-fixed effects as well as the output gap, trade openness, and the service to industry employment ratio. Standard errors clustered at the country-year level. * (+) denotes effects are significantly different from each other at the 10 percent level for men (women).

Policies can offset at least some of the negative association between routinization and participation. In particular, higher spending on active labor market programs seems to attenuate the link between participation and routinizability of occupation. The negative association between routinizability and participation is about one-third as large in countries at the 75th percentile of active labor market spending as in countries at the 25th percentile. Disaggregated data on different active labor market programs suggest that the finding is driven by spending on training, which mitigates some of the negative effect for prime-age women.31

While policies appear to help somewhat in offsetting the effects of routinization for prime-age workers, the negative effects of routinizability are larger for older workers, and policies also provide less of an offset.

5.3 Robustness

In Tables 4 to 8, we report numerous robustness checks for the cross-country panel regressions. Specifically, Table 4 contains the estimated coefficients for the regression of the young, Table 5 of the prime-age male workers, Table 6 of the prime-age female workers, Table 7 of older workers, and Table 8 of the aggregate participation rate. Each table shows the results from the baseline specification, and establishes its robustness to alternative measures, specification, error structure, among others.

Table 4:

Drivers of Youth (Ages 15–24) Labor Force Participation Rates, Robustness

article image
Source: Authors’ calculations.Notes: All specifications include country and year fixed effects. Column (1) reports the baseline estimation results; column (2) reports the results after applying the logistic transformation to the dependent variable; column (3) reports the estimates from a seemingly unrelated regressions (SUR) estimation of a 4-equation system (one for each group); column (4) shows the results using the Beck-Katz estimator; column (5) reports the estimates with heteroskedasticity and autocorrelation consistent (HAC) standard errors, without the correction for cross-sectional dependence; column (6) shows the results with the Newey–West correction for the standard errors; column (7) shows the results based on a sample of 5-year averages; column (8) reports the results dropping global financial crisis (GFC) years 2008 and 2009 from the sample; column (9) reports the coefficients when the sample includes countries that became AEs after 1980; column (10) shows the results when the lag of the output gap is replaced with the lag of the unemployment rate; and column (11) reports the median coefficient from a distribution of estimates obtained by dropping one country at a time from the sample. Driscoll–Kraay standard errors are reported in parentheses in columns (1),(2), (7)–(11); bootstrapped standard errors are reported in parentheses in column (3); HAC standard errors assuming a panel-dependent correlation structure are reported in column (4). Column (11) reports the 10th and 90th percentile of the estimated coefficients in parentheses. **, and * indicate statistical significance at 1, 5, and 10 percent, respectively.
Table 5:

Drivers of Prime-age Men (Ages 25–54) Labor Force Participation Rates, Robustness

article image
Source: Authors’ calculations.Notes: All specifications include country and year fixed effects. Column (1) reports the baseline estimation results; column (2) reports the results after applying the logistic transformation to the dependent variable; column (3) reports the estimates from a seemingly unrelated regressions (SUR) estimation of a 4-equation system (one for each group); column (4) shows the results using the Beck-Katz estimator; column (5) reports the estimates with heteroskedasticity and autocorrelation consistent (HAC) standard errors, without the correction for cross-sectional dependence; column (6) shows the results with the Newey–West correction for the standard errors; column (7) shows the results based on a sample of 5-year averages; column (8) reports the results dropping global financial crisis (GFC) years 2008 and 2009 from the sample; column (9) reports the coefficients when the sample includes countries that became AEs after 1980; column (10) shows the results when the lag of the output gap is replaced with the lag of the unemployment rate; and column (11) reports the median coefficient from a distribution of estimates obtained by dropping one country at a time from the sample. Driscoll–Kraay standard errors are reported in parentheses in columns (1),(2), (7)–(11); bootstrapped standard errors are reported in parentheses in column (3); HAC standard errors assuming a panel-dependent correlation structure are reported in column (4). Column (11) reports the 10th and 90th percentile of the estimated coefficients in parentheses. **, and * indicate statistical significance at 1, 5, and 10 percent, respectively.
Table 6:

Drivers of Prime-age Women (Ages 25–54) Labor Force Participation Rates, Robustness

article image
Source: Authors’ calculations.Notes: All specifications include country and year fixed effects. Column (1) reports the baseline estimation results; column (2) reports the results after applying the logistic transformation to the dependent variable; column (3) reports the estimates from a seemingly unrelated regressions (SUR) estimation of a 4-equation system (one for each group); column (4) shows the results using the Beck-Katz estimator; column (5) reports the estimates with heteroskedasticity and autocorrelation consistent (HAC) standard errors, without the correction for cross-sectional dependence; column (6) shows the results with the Newey–West correction for the standard errors; column (7) shows the results based on a sample of 5-year averages; column (8) reports the results dropping global financial crisis (GFC) years 2008 and 2009 from the sample; column (9) reports the coefficients when the sample includes countries that became AEs after 1980; column (10) shows the results when the lag of the output gap is replaced with the lag of the unemployment rate; and column (11) reports the median coefficient from a distribution of estimates obtained by dropping one country at a time from the sample. Driscoll–Kraay standard errors are reported in parentheses in columns (1),(2), (7)–(11); bootstrapped standard errors are reported in parentheses in column (3); HAC standard errors assuming a panel-dependent correlation structure are reported in column (4). Column (11) reports the 10th and 90th percentile of the estimated coefficients in parentheses. ***, **, and * indicate statistical significance at 1, 5, and 10 percent, respectively.
Table 7:

Drivers of Older Worker (Ages 55 and Over) Labor Force Participation Rates, Robustness

article image
Source: Authors’ calculations.Notes: All specifications include country and year fixed effects. Column (1) reports the baseline estimation results; column (2) reports the results after applying the logistic transformation to the dependent variable; column (3) reports the estimates from a seemingly unrelated regressions (SUR) estimation of a 4-equation system (one for each group); column (4) shows the results using the Beck-Katz estimator; column (5) reports the estimates with heteroskedasticity and autocorrelation consistent (HAC) standard errors, without the correction for cross-sectional dependence; column (6) shows the results with the Newey–West correction for the standard errors; column (7) shows the results based on a sample of 5-year averages; column (8) reports the results dropping global financial crisis (GFC) years 2008 and 2009 from the sample; column (9) reports the coefficients when the sample includes countries that became AEs after 1980; column (10) shows the results when the lag of the output gap is replaced with the lag of the unemployment rate; and column (11) reports the median coefficient from a distribution of estimates obtained by dropping one country at a time from the sample. Driscoll–Kraay standard errors are reported in parentheses in columns (1),(2), (7)–(11); bootstrapped standard errors are reported in parentheses in column (3); HAC standard errors assuming a panel-dependent correlation structure are reported in column (4). Column (11) reports the 10th and 90th percentile of the estimated coefficients in parentheses. **, and * indicate statistical significance at 1, 5, and 10 percent, respectively.
Table 8:

Drivers of Aggregate Labor Force Participation Rates, Robustness

article image
Source: Authors’ calculations.Notes: All specifications include country and year fixed effects. Column (1) reports the baseline estimation results; column (2) reports the results after applying the logistic transformation to the dependent variable; column (3) shows the results using the Beck-Katz estimator; column (4) reports the estimates with heteroskedasticity and autocorrelation consistent (HAC) standard errors, without the correction for cross-sectional dependence; column (5) shows the results with the Newey–West correction for the standard errors; column (6) shows the results based on a sample of 5-year averages; column (7) reports the results dropping global financial crisis (GFC) years 2008 and 2009 from the sample; column (8) reports the coefficients when the sample includes countries that became AEs after 1980; column (9) shows the results when the lag of the output gap is replaced with the lag of the unemployment rate; and column (10) reports the median coefficient from a distribution of estimates obtained by dropping one country at a time from the sample. Driscoll–Kraay standard errors are reported in parentheses in columns (1),(2), (6)-(9); HAC standard errors assuming a panel-dependent correlation structure are reported in column (3); heteroskedasticity and autocorrelation robust standard errors are reported in parentheses in column (4); and Newey-West corrected standard errors are reported in column (5). Column (10) reports the 10th and 90th percentile of the estimated coefficients in parentheses. **, and * indicate statistical significance at 1, 5, and 10 percent, respectively.