Background
Increasing caregiving needs for family members has created pressure on prime-age workers. Combined with the ageing population, the demand for care related to illness and disability by relatives mean more of the workforce may have to consider caring needs (Bauer and Sousa-Poza, Reference Bauer and Sousa-Poza2015). ‘Informal caregivers’ provide care generally without payment (Yoo et al., Reference Yoo, Bhattacharya and McDonald2004). In contrast to formal care, informal caregivers usually have a close relationship with the recipient: for example, siblings and adult children. Informal caregiving is considered a desirable option to meet support needs from several perspectives; these caregivers may be preferred by recipients relative to formal arrangements especially during severe acute illnesses. Caregivers may also feel a personal sense of responsibility to look after loved ones rather than defer to strangers (Fine, Reference Fine2012) though this may depend on the individual’s needs and the available alternatives. Although men are starting to play an important role due to shifting social gender roles, the vast majority of informal caregivers are women who increasingly attempt to juggle caring with labour force participation (Carmichael et al., Reference Carmichael, Hulme and Sheppard2008).
Evidence to date has found that caregiving may limit caregivers’ ability to remain employed. Dating back to the 1990s, a substantial body of literature has suggested that caring responsibilities negatively impact employment in some way (Gautun and Bratt, Reference Gautun and Bratt2017; Jacobs et al., Reference Jacobs, Van Houtven and Laporte2015; Francesca et al., Reference Francesca, Ana and Jérôme2011; Carmichael et al., Reference Carmichael, Hulme and Sheppard2008; Bolin et al., Reference Bolin, Lindgren and Lundborg2008; Berecki-Gisolf et al., Reference Berecki-Gisolf, Lucke and Hockey2008). A recent systematic review found that only informal caregivers heavily involved in caregiving are significantly more likely to withdraw from the labour force, (Lilly et al., Reference Lilly, Laporte and Coyte2007) while other caregivers had a similar level of employment to non-caregivers. This is potentially due to some degree of self-selection into or out of caregiving roles (Bauer and Sousa-Poza, Reference Bauer and Sousa-Poza2015; Leigh, Reference Leigh2010). However, caring responsibilities are generally found to negatively impact employment, which may be more pronounced among those with less human capital (Crespo and Mira, Reference Crespo and Mira2014; Berecki-Gisolf et al., Reference Berecki-Gisolf, Lucke and Hockey2008; Henz, Reference Henz2006). Further, women specifically are more likely to leave the labour force after they start caring for an ill or disabled relative (Pavalko and Henderson, Reference Pavalko and Henderson2006). Beyond the time use, emotional stress and the unpredictability of caring duties can limit productivity in the workplace to the point that people may be forced to reduce or stop employment (Leigh, Reference Leigh2010). In addition to losing current and future financial well-being through various impacts on retirement and pension benefits (McGarry, Reference McGarry2006), a conflict in identity and reduced mental health can result.
Relative to other age groups, older workers face increasing expectations for caring for ageing family members and may be more prone to leave employment. Spousal caregiving is the most frequent type of caregiving among this group (Carmichael et al., Reference Carmichael, Hulme and Sheppard2008), and is associated with higher stress particularly among middle aged and older women (Wu and Penning, Reference Wu and Penning2015). The transition to spousal caregiving tends to entrust older workers with adopting a new caregiving role that can feel foreign. Recently, Cadar (Cadar et al., Reference Cadar, Murray and Head2016) demonstrated in a longitudinal study that full-time workers above age 50 were at a higher risk to leave employment than those not providing care, though it may be related to length of the anticipated caregiving need. Further compounding the issue, OECD projections suggest that statutory retirement ages will increase (Finnish Centre for Pensions). For example, Denmark’s retirement age is expected to reach aged 72 years by 2050. Going forward, caregiving among older workers would thus hinge on balancing longer work obligations with family roles. Though research on caregiving and employment in the older age group is limited, understanding older workers’ employment and the policies that may impact their ability to balance caregiving with paid work is pertinent.
Paid family medical leave policies aim to support a balance between work and caregiving. They are designed to provide income during needed periods of absence. These policies may reduce exit from paid employment and/or the labour force when accompanied by job protections (Berecki-Gisolf et al., Reference Berecki-Gisolf, Lucke and Hockey2008; Francesca et al., Reference Francesca, Ana and Jérôme2011), particularly where the leave and caregiving intensity is time-limited. Most OECD countries do provide some duration of leave for caregiving, although it is not always paid (Bauer and Sousa-Poza, Reference Bauer and Sousa-Poza2015). By lessening the opportunity cost of caregiving and potentially reducing disruption to labour force participation (Sarasa and Billingsley, Reference Sarasa and Billingsley2008), paid family medical leave policies ideally help adults to provide care, minimize income loss and remain in the labor force. The material cost of caregiving is high when it affects employment. In 1997, the economic value of caregiving was estimated at $196 billion in the U.S. alone (Arno et al., Reference Arno, Levine and Memmott1999). The impact of leaving employment for the individual can be non-recoverable (Pavalko and Artis, Reference Pavalko and Artis1997).
Although paid family medical leave policies have the potential to increase employment, their effects are underexplored in the literature. The relationship between caregivers and working is as complex as the nature of the job and the person’s family and social roles. These roles compete for time, commitment and energy (Marks, Reference Marks1977), and their totality can become too demanding forcing a change or an undesirable prioritization (Goode, Reference Goode1960; Marks and MacDermid, Reference Marks and MacDermid1996). Decisions about commitments to caregiving and work are thus intertwined (Pavalko and Artis, Reference Pavalko and Artis1997). On the one hand, the intensity and demand of the caregiving role during an acute period of need can impact the ability to remain employed, while, on the other side, the options to transition caregiving to formal services or maintain the caregiving role through flexible work arrangements are also influential in preventing leaving employment entirely. Indeed, workplaces that provide flexible working arrangements are more successful in retaining women in employment after they start caregiving (Pavalko and Henderson, Reference Pavalko and Henderson2006). Caregivers are also repeatedly found to have reduced working hours (Lilly et al., Reference Lilly, Laporte and Coyte2007), suggesting this option helps to reconcile the logistical difficulty of competing time use for employment and caring (Leigh, Reference Leigh2010). Paid leave policies may help to buffer against leaving employment where they help individuals during a period of intensity with high caregiving demands, but they may be insufficient where caregiving needs are longer term, flexible working conditions or other transitional services are inaccessible, or if jobs are not adequately protected. Despite potentially variable effects, the impact of paid family medical leave policies on employment is not well understood (Börsch-Supan et al., Reference Börsch-Supan, Brandt and Hunkler2013; Verbakel, Reference Verbakel2014; Gautun and Bratt, Reference Gautun and Bratt2017).
To begin to address this gap, this quasi-experimental study used data from the Survey of Health, Ageing and Retirement in Europe (SHARE) to estimate the impact of introducing paid family medical leave policies on the probability of working among older European men and women.
Methods
Design
To estimate the impact of paid family medical leave policies, we compare the observed probability of working in countries that introduced or extended a paid family medical leave policy to those that did not, using a difference-in-differences design. In this design, respondents in countries that reformed a family medical leave policy (‘treated’ countries) would have access to different benefits after the policy reform, which might affect their probability of working. The counterfactual is estimated using a set of control countries that did not introduce a paid family medical leave policy during the study period. As in a traditional randomized controlled trial design, the difference-in-differences design does not require that all individuals in the sample are susceptible to, or would be impacted by, the policy. We can determine the overall population-level (‘intention-to-treat’) impact of the policy without detailed information on those most likely to uptake it.
When the assumptions of the difference-in-differences design are met, it is possible to estimate a causal policy effect. A difference-in-differences design is valid if the control countries experienced a similar trend (but not necessarily level) in the probability of working prior to the treated countries’ policy reform and there are no other coincident factors that affect trends in the probability of working over the same time period (Dimick and Ryan, Reference Dimick and Ryan2014). If these assumptions are met, the controls are assumed to represent the counterfactual for the probability of working in treated countries had they not introduced the policy. The estimated effect consists of the difference of the difference between the probability of working pre and post policy reform in the treated versus control countries. The impact of the policy on the probability of working is therefore estimated as the absolute change in the probability of working in treated countries vis-à-vis the change in control countries.
Sample
We used data from SHARE, a population-based cohort study designed to be representative of the adult population aged 50 years and older in selected European countries (www.share-project.org). At baseline (2004 in most countries), all age-eligible members and spouses in sampled households were selected for in-person interviews, conducted with the assistance of a Computer Assisted Programming Interview application. Survey topics related to health, income, and social networks. The full details on the SHARE survey are available elsewhere (Börsch-Supan et al., Reference Börsch-Supan, Brandt and Hunkler2013). In most countries, SHARE respondents have been followed-up every two years, with refresher samples to maintain initial sample sizes and representativeness. Additional countries joined over time.
Wave 3 of SHARE included a special questionnaire, SHARELIFE, that collected detailed information on the life-course, including childhood, partners, housing, and financial and employment history (Börsch-Supan, Reference Börsch-Supan2017; Antonova et al., Reference Antonova, Aranda, Brugiavini, Cavapozzi, Pasini and Trevisan2017). All interviews were conducted in 2008-09 except for respondents in Ireland who were interviewed in 2009-11. The employment history began when the respondent left full-time education. Although respondents were interviewed only once, they provided a yearly account of their life histories. To collect life histories, the study used a focused Events History Calendar technique which was developed based on the natural structure of autobiographical memory (Belli, Reference Belli1998). In the electronic version of this technique, dates for life events such as marriages and jobs are graphically displayed on a life grid that the respondent and interviewer gradually filled in together. Within the application, respondents could look up general country events to provide reference points (Schröder, Reference Schröder2011). A similar method was used for collection in the English Longitudinal Study of Ageing (Scholes et al., Reference Scholes, Medina and Cheschire2009), which also served as the basis for other retrospective life history surveys (Kendig et al., Reference Kendig, Byles and O’Loughlin2014).
SHARELIFE included 12,528 men and 15,964 women across 14 countries. We restricted the sample to individuals born in 1958 or earlier, which in practice resulted in excluding spouses of primary survey participants who were born later (n=554). This left 12,482 men and 15,456 women eligible for analyses. We used data starting in 1990 (Figure 1).
Eligible Sample
Measures
The treatment variable of interest was the introduction of a policy granting paid family medical leave of at least six months. Current data on policies regarding leave for private sector employees to care for sick family members for each sampled country were provided by the University of California Los Angeles World Legal Rights Data Centre, and data on past leave policies starting in 1995 were collected by McGill University’s PROSPERED research program (https://www.prosperedproject.com/). We defined paid family medical leave as time-off from work that can be used to care for adult family members with serious or general health problems. While some countries allow this leave to be used for any relative, including children, spouses and parents with any health problem, some limit eligibility based on the severity of illness, degree of relation, adult vs child status, and employee’s length of tenure and/or contributions to the social insurance program. We did not include policies that were only available for end-of-life care or only for sick children, but did not distinguish policies based on restrictions regarding severity of disease, tenure and/or contributions requirement or degree of relation. Further details on countries’ policies are in Supplemental File 1, Table 1 (Supplementary Materials).
Our primary outcome was work status, coded as working or not. Work status was not directly reported, but derived from the self-reported start and end dates of jobs based on the retrospective Job Episodes Panel constructed from the SHARELIFE survey (Antonova et al., Reference Antonova, Aranda and Pasini2014; Brugiavini et al., Reference Brugiavini, Cavapozzi, Pasini and Trevisan2013). The start and end dates of jobs were determined based on the life history calendar method described above. For each year within those boundary dates, which were considered ‘job spells,’ a person was recorded as working part-time or full-time. If there was no reported job, the person was coded as not working. Given the derivation of the working variable and the emphasis on major life events in the data collection methodology, we believe most workers would report job start and end dates that cover a period of leave, such that those on paid leave would be considered as ‘working.’ (Brugiavini et al., Reference Brugiavini, Cavapozzi, Pasini and Trevisan2013; Antonova et al., Reference Antonova, Aranda and Pasini2014; Schröder, Reference Schröder2011) There may be some variability, as highlighted in the Limitations. Employment indicators from SHARELIFE have previously been used: for example, to study labour market outcomes (Wahrendorf, Reference Wahrendorf2015; Antonova et al., Reference Antonova, Aranda, Brugiavini, Cavapozzi, Pasini and Trevisan2017) and employment trajectories (Hedel et al., Reference Hedel, Mejía-Guevara and Avendaño2016).
Using these variables from SHARELIFE, the data were structured at the individual-level, with a work status indicator for each year starting at 1990 until 2008. This in effect created a time series of employment by year for each subject, with respondents nested within country. Subjects in treated countries were given a 1 on a time-dependent policy indicator variable for the post-policy period, while all other time points and subjects were coded as 0.
Treated and Control Countries
Countries eligible to be in the treatment group had to participate in the SHARELIFE Job Episodes Panel, have introduced a paid family medical leave policy of at least six months in duration between 1995 and 2007 and have at least one suitable control country. Countries to be eligible as controls did not introduce a paid leave policy that could affect the probability of working between 1995 and 2007.
Two countries met the criteria to be treated countries: Belgium (1998) and Denmark (2002). With the 2002 adoption of Act No. 359 of 2002 and Act No. 397 of 2002, Denmark gave the right to absence from work for up to 6 months for workers to take care of severely ill or disabled relatives while receiving payments from the municipalities. In 2016, the full-time caregiver allowance was approximately 2900 € (DKK 21,546) per month, not conditional on a certain length of contribution to social insurance schemes. In Belgium, the 1998 policy allowed providing care to seriously ill adult relatives as well as children. This leave could last three months but was expandable up to 12 months and was conditional upon a minimum of 12 months of contributions to the social insurance scheme. The employees on leave were to be compensated with flat rate payments regardless of salary scale; in 2017 this amount was 818 €.
Apart from Ireland and Netherlands, the remaining countries in the survey were eligible as controls. Although Ireland introduced 65 weeks of paid leave to care for seriously ill family members in 2001, we could not find a suitable control group because it simultaneously experienced a period of rapid economic growth associated with a marked reduction in corporate tax rates which could confound results. Netherlands did pass a paid leave policy in the studied time period; however, it provided less than six months of paid leave, meaning the country did not meet our criteria to be a treated or a control country. Notably, France reformed an existing policy in 2001/2 to provide universal assistance to people aged 60 and over who are dependent on others to care for them. While this policy did not meet our criteria to be a treated country, France remained a candidate control country because we did not expect its policy change to substantially influence the average probability of working beyond workers in the personal services sector undertaking formal caregiving. Thus, the existence of the policy change did not result in any violation of the assumptions. We did however conduct a sensitivity analysis to ensure this policy change in France was not driving any observed effect.
After the exclusions, 10 countries included in SHARELIFE were eligible to be controls because they either had consistently the same policy or no adult family medical leave policy over the same time period (France, Sweden, Austria, Czech Republic, Germany, Greece, Italy, Poland, Spain, and Switzerland). We estimated the impact of the reforms occurring in treated countries separately and all analyses were stratified by gender. After eliminating Netherlands and Ireland (n=3062), 11,091 men and 13,785 women who responded to the survey in the remaining countries were eligible across 12 countries (Figure 1).
Control countries were selected for each treated country from the pool of eligible countries based on whether they met the key difference-in-differences design assumption of having a similar pre-policy trend in employment. The selection of control countries was done separately for men and women. In both treated countries and by gender, we identified the functional form of the age-adjusted pre-policy trends in the proportion of respondents working by modeling and comparing trends using: (1) indicator terms for each pre-policy year, (2) a linear term for year, and (3) linear and quadratic terms for year. We determined the appropriate functional form by testing whether parameter(s) for year in each model were significantly different from 0 at the p=0.05 level. After we determined the appropriate functional form, control selection was based on comparing the age-adjusted pre-policy trends in the proportion of respondents working. Specifically, we examined whether the assumption of parallel trends was tenable by testing for differences in pre-policy trends between the treated and candidate control countries using an F-test (‘chunk test’). Countries were removed until this F-test suggested that the trends for remaining control countries were not significantly different from the treated country (we relaxed this strict definition in sensitivity analysis, described below). We visually inspected the trends in the remaining countries, and removed those that were not parallel (Supplemental File 2, Supplementary Materials).
Analysis
We estimated the impact of introducing paid adult medical leave on the probability of working using a linear probability model with cluster-robust standard errors. As our sample consisted of older adults, we expected to see secular trends in our outcome, i.e. the average probability of working would diminish over time naturally. To adjust for this, we included linear and quadratic terms for calendar year that would allow the relationship between the probability of working and time to vary. Further, we adjust for age using a categorical variable that would allow the probability of working to vary flexibly within each category only.
The model for individual i in country c at time t took the general form in (1).
$${Y_{ict}} = \alpha + \beta {x_{ct}} + {\gamma _c} + \delta year_{t} + \tau year_t^2 + \sum \nolimits_{a = 1}^4 {\theta _a}age_{i} + {\varepsilon _{ict}}$$
where β is the parameter of interest, i.e. the effect of the policy reform on the probability of working, Y
ict
. We included fixed effects for country
${\gamma _c}$
and linear and quadratic terms for calendar year,
$\delta $
and
$\tau $
. The age categories (30-39, 40-49, 50-59, 60-69, 70+) were included in the fully adjusted model and standard errors were clustered at the individual level (Esarey and Menger, Reference Esarey and Menger2018). The estimated coefficient β measures whether there is any change, on average, in the probability of working in the treated country before vs. after its reform relative to corresponding changes in control countries.
Sensitivity Analysis
We assessed the robustness of our main findings using several approaches. First, we estimated ‘lead’ and ‘lag’ effects. This involved re-assigning the year of the policy change as two years prior and two years after the year the change actually occurred, respectively. The primary analysis was then re-run with the artificial policy year to detect if there was i) evidence of a policy effect before the policy was actually passed, which would represent evidence against a causal policy effect, or ii) whether any policy effect appeared and/or persisted into the post-policy period. Second, we examined the sensitivity of our main findings to the composition of the control group. Alternate control groups were created by relaxing and tightening the p-value criterion of 0.05 used to select controls, and by using quadratic parallel trends rather than linear where it was deemed appropriate. Third, we included additional variables that could predict work status, and explored the impact of retirement by i) including individuals’ eligibility for ‘normal’ retirement in a given country-year; ii) restricting to only non-retired people; and iii) excluding those who retired before policies changed. Though our design is not invalidated by the inclusion of people who are retired, this sensitivity analysis sought to determine whether the effect estimates changed when restricting the sample to those most likely to be affected by the policy. Fourth, we conducted a post-hoc sensitivity analysis excluding France as a control as it was ultimately selected as an appropriate control country for Belgium. Finally, we included the sampling weights provided by SHARE. Further details regarding the sensitivity analyses are provided in Supplemental File 3 (Supplementary Materials).
All analyses were done in RStudio and Stata 14 (StataCorp, 2015). The full code for the primary and sensitivity analyses and instructions on replicating our results are available at www.github.com/deepajag/paid-leave, and on https://dataverse.harvard.edu/dataverse/3po.
Results
Sample Description
To estimate the effect of introducing paid family leave in Belgium, three control countries were selected based on linear pre-policy trends for analyses of women (Greece, France and Italy) and men (France, Spain and Switzerland) (Figure 2). For Denmark, we identified two countries with parallel pre-policy trends for each gender, i.e. Greece and Sweden for men and Austria and Sweden for women (Figure 3). We found little evidence for differences in age-adjusted pre-policy trends in each treated country and its selected control group based on both visual examination of parallel trends and the F-test for joint significance of the effect of treatment status and year on the proportion working (Supplemental File 2, Supplementary Materials). Thus, we ensured that our study met one of the key assumptions of the difference-in-differences approach both visually and statistically.
Pre-policy employment trend in Belgium vs Controls
Pre-policy employment trend in Denmark vs Controls
The proportion working across all countries declined between the pre- and post-policy periods. Absolute declines in the proportion working were larger in magnitude for men than women. The mean age across all included countries in 1990, the first year of our study, was 48 years and was relatively consistent across countries (Table 1). The breakdown by country is available in Supplemental File 1, Table 2 (Supplementary Materials).
Sample Description
i) France, Greece, Italy; ii) France, Spain, Switzerland; iii) Austria, Sweden; iv) Greece, Sweden
Main Results
There was limited evidence that the introduction of paid family medical leave had an impact on the proportion of the population working (Table 2). In Belgium, the policy introduction was associated with 1.2 (95% CI -0.4 to 2.8) and 1.9 (95% CI 0.1 to 3.7) percentage point reductions in the proportions of women and men working, relative to control countries, respectively. In Denmark we found little evidence of a policy effect (percentage point difference = 0.8 [95% CI -1.4 to 3.0] for women and percentage point difference = 1.0 [95% CI –1.1 to 3.1] for men relative to control countries). The full regression results are available in Supplemental File 1, Table 3 (Supplementary Materials).
Main results: Effect of policy introduction on proportion working
i Fixed effects for country and square terms for year and age category. The effect estimates represent the mean percentage point difference in proportion working in treatment vs control countries (before vs. after the policy change)
Sensitivity Analyses
The effect estimates were similar when the policy year was shifted to two years before and two years after the true policy year, suggesting that any changes in the proportion working were not likely attributable to the introduction of paid family medical leave (Figure 4). Additional sensitivity analyses, including the examination of different control countries, selecting controls based on non-linear time trends for the outcome, and including sampling weights, were consistent with our main findings. The estimates were also consistent when the primary model was adjusted for the age of retirement eligibility, restricted only to the non-retired population, eliminating France as a control, and when including other country-level characteristics. Excluding those who had retired before policy changes resulted in the stronger magnitudes for the estimated effects, but did not alter conclusions (Supplemental file 3, Supplementary Materials).
Lead and lag effects
Discussion
Overall, we did not find consistent evidence that the introduction of paid family medical leave policies in Belgium and Denmark affected the proportion of older adults working. Our sensitivity analyses were consistent with the main estimates.
The broader context of the policies and workplaces may have limited the policies’ impact. The availability and affordability of other services could support these policies to improve employment if they are sufficient to allow a transition back to employment. The strength of formal arrangements in Denmark, for example, has allowed elderly people to utilize these services instead of relying on family (Rostgaard and Szebehely, Reference Rostgaard and Szebehely2012). However these arrangements in addition to flexible working arrangements may not be enough for caregivers. Second, caregivers may stay in employment by compromising elsewhere or because the financial tradeoff is imbalanced. Caregivers have been previously hypothesized to reduce leisure time instead of employment, or to maintain employment to afford care, keep insurance, and/or provide a distraction (Jacobs et al., Reference Jacobs, Van Houtven and Laporte2015; Agree and Glaser, Reference Agree and Glaser2009). Third, there is also the possibility that some caregivers were more likely to cease working after taking up the leave. In sensitivity analyses that explored our findings in relation to retirement indicators we did find that the estimated effect was stronger (though less precise), which could support such a theory.
There are also alternate outcomes that the policies could affect rather than employment, which could provide avenues for future research. Paid family medical leave policies may have a primary impact on income rather than employment (Berecki-Gisolf et al., Reference Berecki-Gisolf, Lucke and Hockey2008). Workers may substitute a choice of unpaid leave with paid leave, rather than choose to leave employment altogether pre-policy. The financial strain associated with unpaid leave (Lester, Reference Lester2005) means that paid leave policies may still have positive effects related to income, but we were unable to study the extent to which such substitution may occur. Firms may have also changed their practices in response to the policy change. By creating fewer jobs that may otherwise have been available to this older group of people who would be more likely to make use of the policy, the average employment probability may decline (Ahn and Yelowitz, Reference Ahn and Yelowitz2015). This phenomenon could potentially offset any gains or explain some of our effects that were in a negative direction in Belgium. Finally, the policies may have only affected certain subgroups for which we would not be able to detect an impact. Previously, only those providing the most intense levels of care have been found to leave employment (Lilly et al., Reference Lilly, Laporte and Coyte2007), and thus the probability of leaving may only be improved among them. Studying these subgroups could shed light on this question.
This study has limitations associated with the data and study design. Data on work histories were collected retrospectively based on self-report, raising concerns that information could be remembered incorrectly. This is likely not different across countries and previous work has found recall bias in this survey was minimal (Mazzonna and Havari, Reference Mazzonna and Havari2011). Our estimate may be biased towards the null due to variability in the way respondents reported their job status when respondents were on leave, though we expect this to be minimal as the working variable was constructed based on the starts and end dates of jobs. It is also possible that we may not have captured time-varying factors within treated countries that could affect the probability of working, a key assumption on our study design. An independent policy was introduced in the Flanders region of Belgium in 2001 with a cash benefit of 130 Euros/month; though, given that it is not enough to cover formal care options, it would likely have minimal influence on the general probability of staying employed. We could not find other major initiatives over our study’s time frame that could explain our findings. To mitigate the potential for residual confounding, we also selected controls based on parallel trends, and conducted several sensitivity analyses with country-level economic factors to test the robustness of our preferred specifications. Inherently, the difference-in-differences method is biased towards the null like the analogous randomized controlled trial design. However, this does not explain our findings because we were still unable to see a pattern in lead/lags that was consistent with a policy effect. Finally, the size of the working population within the study samples may also have reduced our ability to detect small policy effects precisely and forced us to treat part- and full-time status together, despite the potential for switching between the two.
Conclusion
We found little evidence that paid family medical leave policies had an important impact on the proportion of older adults working in Denmark or Belgium. In estimating the impacts of these policies on employment, we addressed an important gap in the literature: namely, whether these policies which provide income during leave also impact employment. In the absence of other changes regarding the availability and affordability of long term care services, paid family leave was not sufficient to increase average employment. Further research should examine the impact of paid family leave when combined with the availability of affordable long term care. This evidence base will ultimately help to inform policy objectives to support those caring for elderly parents, spouses, other kin and partners in the context of a rapidly ageing population.
Acknowledgements
This study was supported by the Foundation Grant “Development epidemiology: identifying evidence-based interventions for improving population health and promoting health equity (FRN 148467) and the Canada Research Chairs program (AN). Support was also provided by UCLA Public Health under Service Agreement Number 1915 UA132. All opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the view of the Sponsors. Strumpf and Harper were supported by Chercheur-boursier Junior 2 awards from the Fonds de recherche du Québec – Santé.
This paper uses data from SHARE Waves 1, 2, 3 (SHARELIFE), 4, 5 and 6 (DOIs: 10.6103/SHARE.w1.610, 10.6103/SHARE.w2.610, 10.6103/SHARE.w3.610, 10.6103/SHARE.w4.610, 10.6103/SHARE.w5.610, 10.6103/SHARE.w6.610) see Börsch-Supan et al. (Reference Börsch-Supan, Brandt and Hunkler2013) for methodological details.
The SHARE data collection has been primarily funded by the European Commission through FP5 (QLK6-CT-2001-00360) FP6 (SHARE-I3: RII-CT-2006-062193, COMPARE: CIT5-CT-2005-028857, SHARELIFE: CIT4-CT-2006-028812) and FP7 (SHARE-PREP: N°211909, SHARE-LEAP: N°227822, SHARE M4: N°261982). Additional funding from the German Ministry of Education and Research, the Max Planck Society for the Advancement of Science, the U.S. National Institute on Aging (U01_AG09740-13S2, P01_AG005842, P01_AG08291, P30_AG12815, R21_AG025169, Y1-AG-4553-01, IAG_BSR06-11, OGHA_04-064, HHSN271201300071C) and from various national funding sources is gratefully acknowledged (see www.share-project.org).
This paper uses data from the generated Job Episodes Panel (DOI: 10.6103/SHARE.jep.600) see Brugiavini et al. (Reference Brugiavini, Cavapozzi, Pasini and Trevisan2013) and Antonova et al. (Reference Antonova, Aranda and Pasini2014) for methodological details. The Job Episodes Panel release 6.0.0 is based on SHARE Waves 1, 2 and 3 (SHARELIFE) (DOIs: 10.6103/SHARE.w1.600, 10.6103/SHARE.w2.600, 10.6103/SHARE.w3.600).
Declaration of interest
We have no conflicts of interest to declare.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S0047279420000264
