Abstract
This study estimates the causal effect of paid vacation on health. Using register data on the universe of central government employees in Sweden, I exploit an age-based rule stipulated in the collective agreement covering these employees. I achieve identification by combining a regression discontinuity with a difference-in-differences design to control for time-invariant differences between consecutive birth cohorts and isolate the true effect at two separate discontinuities at ages 30 and 40. The main results indicate that an increase of three paid vacation days at age 30 and four days at age 40 do not cause significant changes in health, as proxied by visits to specialized outpatient care, inpatient admissions, and long-term sick leaves. These findings challenge the anecdotal view of additional paid vacation days as an adequate means to improve workers’ health.
Similar content being viewed by others
Notes
Paid annual leave is the sum of paid vacation days and paid public holidays. Whereas paid public holidays (e.g., May Day) typically take place on the same date every year, a worker can more or less freely distribute paid vacation days across the year. Whether workers stay at home or travel while on vacation is not considered in this study.
To give a few examples, the Finnish government intended in 2015 to cut vacation days for civil servants from 38 to 30 days but instead reduced their vacation pay by 30% for the years 2017 to 2019. In 2012, the Irish government capped paid annual leave for civil servants at 32 days; previously, it could amount to 40 days in some cases. In Saudi Arabia, vacation was capped at 30 days for civil servants in 2016. In Switzerland, a proposal for an extension of the minimum vacation entitlement from 4 to 6 weeks was voted down in a referendum in 2012, due to concerns about negative consequences for the economy’s competitiveness.
The content of the health-related argument for paid vacation has changed over time from the so-called protection motive to the recreation motive in Sweden. The foremost reason for the introduction in the 1930s was to protect workers from “wear and tear” and work-related accidents. Nowadays, vacation should give workers time for rest, recuperation, and relaxation to maintain health as well as an opportunity to pursue own interests during leisure time (Arbetsgivarverket 2009; Ericson and Eriksson 2015).
A small related literature focuses on the causal effects of changes in the regular weekly working hours on workers’ health while holding earnings constant. Berniell and Bietenbeck (2017) find that longer working hours have negative effects on health, and Lepinteur (2019) finds positive effects of shorter working hours on job and leisure satisfaction. However, reductions in the regular weekly working hours provide workers with additional leisure time every week of the year, whereas additional vacation days provide additional leisure time during certain days of the year. The way workers are “treated” with additional leisure time is thus different.
By way of comparison, Lepinteur (2019) finds statistically positive effects ranging from 7 to 15% of a standard deviation of shorter weekly working hours on job and leisure satisfaction.
This pattern has been dubbed “the vacation cycle”, which consists of a vacation effect (i.e., the boost in well-being and health during a vacation) and a fadeout effect (i.e., the positive outcomes vanish swiftly as vacationers return to their routine environment) (Kirillova and Lehto 2015).
The original agreement was called ALFA Cirkulär 1997:A 4. The subsequent renewals were ALFA Cirkulär 1998:A 8, ALFA Cirkulär 2001:A 8, ALFA Cirkulär 2002:A 5, ALFA Centrala avtal 2005:4, ALFA Centrala avtal 2008:1, and ALFA ALFA-T Centrala avtal 2011:4. The rules on vacation entitlement have not been altered in any of the renewals.
Since October 1, 2007, about 40% of all employees covered by the ALFA and represented by the trade union Saco-S have the possibility to negotiate different terms on the number of paid vacation days with their employer on an individual basis. An analysis by Saco-S has shown that only a tiny fraction of employees (0.1% of all central government employees in 2010 and 0.3% of all in 2011) actually made use of this possibility (Saco-S and Arbetsgivarverket 2014). As a robustness check, I run the analysis only including the years before 2007.
Note that vacation days correspond to workdays in the standard 5-day workweek stipulated by the ALFA. A 40-year-old employee thus receives exactly 7 weeks of paid vacation per year.
If an employee uses all vacation days for a year but quits her employment before the end of the year, she has to pay back vacation days in proportion to the length of employment.
There are three exceptions. Vacation days can be paid out in money if an employment lasts for fewer than 3 months. They must be paid out in money if an employee quits her employment and still has vacation days left. If an employee due to sickness or other reasons could not take all vacation days during a year, then these days are saved for later, but if the total number of saved days exceeds 40 days (35 days since 2011), the exceeding days are paid out in money.
For instance, a 45-year-old part-time central government employee who works 4 h Monday to Friday (i.e., 50% part-time) gets 35 vacation days per year. If the same employee would work 8 h on Mondays and Tuesdays, 4 h on Wednesdays, and not work on Thursdays and Fridays, then the vacation entitlement is 17.5 days (= 35 days * 50% employment) per year instead.
For instance, with a gross monthly salary of SEK 25,000 (USD 2650), the vacation pay is comprised of the ordinary salary per workday of about SEK 1136 (USD 120) (assuming 22 workdays per month) and the supplement of SEK 110 (USD 12) (= 0.44% * SEK 25,000). Three (four) additional vacation days at age 30 (40) would thus increase annual earnings by a mere 0.11% (0.15%). By contrast, annual working time decreases by 1.35% (1.83%) at age 30 (40), assuming 250 workdays per year and that the full vacation entitlement is used.
Approximately 230,000–250,000 people or 5–6% of the Swedish working population were employed in the central government sector in the period 1997–2011 (Statskontoret 2015). Note that employees in (wholly or partly) state-owned enterprises and foundations are not central government employees.
It is possible to distinguish them from a small group of employees who are working in special government agencies called statliga affärsverk. They are covered by another collective agreement that mandates similar though not identical rules on vacation entitlement.
This requirement does not exclude employees with poor health, as employees cannot be dismissed due to, e.g., long and repeated sick leaves. Staff turnover has been fairly stable at least since 2006. About 10–13% of those employed in a certain year quit their job in the following year (including due to retirement) (Statskontoret 2015).
It is not possible to distinguish between acute and planned admissions in the data.
In total, 12.3% of all cases in the register in 1997 are classified based on ICD-9, which I recode accordingly. In all subsequent years, more than 99.9% of all cases have a valid ICD-10 code.
Between January 1, 1997, and March 31, 1998, the Social Insurance Agency paid out sickness benefits after a sickness period of 28 calendar days, and between July 1, 2003, and December 31, 2004, after 21 calendar days.
The first day of a sickness period is not remunerated. From days 2 to 14, the employer is obliged to pay sick pay equal to 80% of the ordinary salary. The sickness benefits paid out by the Social Insurance Agency as of day 15 amount to slightly less than 80% of the ordinary salary, but there is a cap on the maximum benefits per day. In 2008, sickness benefits paid out by the Social Insurance Agency became time-limited to 1 year in normal cases and a maximum of 914 days in exceptional cases. I address the potential impact of this policy change in a robustness check.
The most common diagnoses (based on two-digit ICD-10 codes) at ages 30 and 40 for outpatient care are follow-up examination after surgery (Z09), abdominal and pelvic pain (R10), supervision of normal pregnancy (Z34), and special examination of persons without complaint or reported diagnosis (e.g., blood pressure, allergy tests, pap test, mammography) (Z01). For inpatient care, they are childbirth (O80+O81+O82), abdominal and pelvic pain (R10), gallstones (K80), and acute appendicitis (K35). For sick leave, they are disorders related to severe stress (F43), depression (F32), back pain (M54), and maternal care for other conditions predominantly related to pregnancy (O26).
As opposed to inpatient admissions, some outpatient visits might be used for preventive purposes and thus reflect more of an input to health. Excluding outpatient visits which might be partly preventive (diagnosis group with ICD-10 code Z00–Z99) does not affect the results (available upon request).
Every outpatient visit and every day spent at a hospital is subject to a patient fee (about SEK 200 (USD 21) and SEK 100 (USD 11), respectively), but a ceiling limits the total amount a patient has to pay during a 12-month period.
Table 6 in the Appendix provides an overview of the ten most common occupational groups in the central government sector for the period 2001–2011. College, university, and higher education teaching professionals (which encompass academic staff and PhD students) are the most common group as most major universities and university colleges are run as government agencies.
For instance, Plug (2001) shows that maturity differences within the class room influence school performance and earnings in the Netherlands, a country where people of the same birth cohort attend the same class, just as in Sweden.
Fredriksson and Öckert (2014) show that parental education is continuous across the December–January threshold for the entire Swedish native population born 1935–1955. This indicates that Swedish parents did not time the birth of their children.
Note that the peak in March to May is about 9 months after the period when Swedes take most of their vacation days; see panel (a) in Fig. 7 in the Appendix.
Panel (a) in Fig. 6 in the Appendix also shows that the number of observations is almost linearly increasing from age 27 to 30. This is related to two issues. First, many university graduates, who constitute a substantial share of central government employees, enter into employment at that age. Second, there are fewer observations aged 31 and increasingly fewer for every additional year below that age, as the register data are less complete for cohorts born after 1980. I address the latter issue in a robustness check.
A related concern is that foreign-born central government employees lack reliable information on date of birth; see, e.g., Torun and Tumen (2017). These employees are excluded in panels (c) and (d) in Fig. 6 in the Appendix, yet the patterns are very similar to panels (a) and (b), indicating no concerns. The results from the density test differ neither from the full sample; see panels A2 and B2 of Table 7 in the Appendix. In a robustness check, I nevertheless exclude foreign-born employees.
This is the finest granulation of age available in the data.
The treatment indicator is defined as:
$$ T_{ijt}=\begin{cases} 1 \ \text{if }\ (age_{c}-24) < age_{ijt} < (age_{c}-12) \ \text{ or } \ age_{ijt} > age_{c} \\ 0 \ \text{if }\ age_{ijt} < (age_{c}-24) \text{ or } (age_{c}-12) < age_{ijt} < age_{c}. \end{cases} $$The counterfactual indicator is defined as:
$$ C_{ijt}=\begin{cases} 1 \ \text{if} \ age_{ijt} > (age_{c}-12) \\ 0 \ \text{otherwise}. \end{cases} $$An alternative approach would be to use a local linear regression with a triangular kernel (instead of a rectangular kernel as in my case) to put more weight on observations closer to the thresholds. I follow however the suggestion by Lee and Lemieux (2010) and vary the size of the bandwidth instead.
The estimated regression model with quadratic trends (d = 2) on each side of the thresholds is:
$$ Health_{ijt} = \sum\limits_{b=0}^{d}(\alpha_{b} age_{ijt}^{*b}) \!+ T_{ijt}\sum\limits_{b=0}^{d}({\upbeta}_{b} age_{ijt}^{*b}) + C_{ijt}[\sum\limits_{b=0}^{d}(\gamma_{b} age_{ijt}^{*b}) + T_{ijt}\sum\limits_{b=0}^{d}(\delta_{b} age_{ijt}^{*b})] + \zeta_{j} + \eta_{t} + \epsilon_{ijt}. $$(2)I do not consider higher dimensional trends, as suggested by Gelman and Imbens (2019).
The minimum 1-month bandwidth corresponds to a comparison of means between employees born in December in a certain year and employees born in January in the following year.
This could stem from the fact that the earnings development is more erratic at younger ages.
There are also no representative Swedish surveys or European surveys that include Sweden that contain suitable vacation-related information.
Goerke et al. (2015) find an association between trade union membership and days of paid vacation taken in Germany. Union membership is not a concern among central government employees in Sweden, as everyone is covered by the ALFA collective agreement irrespective of actual membership.
On July 1, 2008, a small earmarked lump sum that can be used to pay for dental care was introduced. The year one turns 30 the annual lump sum is halved from SEK 300 (USD 32) to SEK 150 (USD 16). It is doubtful that this reduction leads to swift changes in dental health that in turn affect the health measures used in this study. In a robustness check, I exclude the years affected by this policy. At age 40, women in Sweden receive their first invitation letter for breast cancer screening (which used to come along with a small patient fee of SEK 150 (USD 16)). This routine started in a small part of Sweden in 1986 and had been gradually extended, but in 2005 the whole country was still not covered (Hellquist et al. 2011). This might have an impact on the health measures in this study, as positive test results are subject to further analysis and might lead to treatment at a hospital. As a robustness check, I carry out the analysis separately for women and men as well as diagnoses related to breast cancer screening and breast cancer treatment.
When standard errors are clustered at the 24 distinct values of the running variable, they decrease markedly and the point estimate for outpatient visits at age 40 is rendered significant at the 10% level. As pointed out by Kolesár and Rothe (2018), clustering standard errors by the running variable leads to biased confidence intervals. They recommend conventional heteroscedasticity-robust standard errors if the bandwidth is small. The use of heteroscedasticity-robust standard errors or unadjusted standard errors yields virtually identical confidence intervals as the clustered standard errors in Table 3 (results available upon request). It should also be noted that the standard errors for outpatient visits in Table 3 are smaller than of the two other health measures, as this measure has more variation.
Here, I am forced to rely on a sample including the years 2001 to 2011, as data on specialized outpatient care are not available prior to 2001.
As discussed by, e.g., Kling et al. (2007) and Hoynes et al. (2016), aggregating multiple outcome measures can improve statistical power. The summary index is the arithmetic average of the z-scores of the three health measures. The z-score is calculated by subtracting the mean of a health measure from the observed outcome and dividing the difference by the standard deviation of the health measure. I use the mean and the standard deviation of the control group (aged 27 or 29) in calculating the z-score separately around the actual and the counterfactual threshold. Thus, all three components of the index have mean 0 and standard deviation 1 for the control group. I re-normalize the summary index to have standard deviation 1.
Another conceivable approach to increase statistical power is to use employees in non-government sectors as controls. However, it is impossible to identify workers that are covered by a certain collective agreement outside the central government sector unambiguously in the SIP database. In addition, other collective agreements often contain an increase in the vacation entitlement at age 40 but give employees the choice to decline it and receive a salary increase instead.
An alternative approach would be to choose a small bandwidth (to reduce bias) and at the same time increase the size of the control group (to increase precision), as done by, e.g., Carneiro et al. (2015). Using a bandwidth of 6 months and defining observations around the two thresholds at ages 27–28 and ages 28–29 or around the three thresholds at ages 26–27, 27–28, and 28–29 (and at analogous thresholds at age 40) as the control group, I also obtain insignificant results (results available upon request).
The concentration of vacation taken during the summer is due to a provision in the Annual Leave Act, which entitles every worker to at least 4 weeks of uninterrupted vacation leave during June to August.
The results for the ten least common groups (which include circulatory diseases, ICD-10 code I00–I99) are overwhelmingly insignificant (results available upon request).
One threat to the RD-DID design is the start of the breast cancer screening program for women at age 40. However, for outpatient visits at age 40, there are no significant estimates (neither in the female subsample nor in the full sample) for the diagnosis group Z01, which includes routine mammography, the diagnosis group Z12, which includes examination for breast cancer, the more general diagnosis group Z00–Z13, which comprises persons encountering health services for examination and investigation, and the diagnosis group C50, which encompasses breast cancer treatment. For inpatient admissions, there are virtually no observations with diagnosis codes Z01 and Z12 and only about 100 female observations with diagnosis codes Z00–Z13 and C50 in the 4-year age interval around age 40, but the estimates for the latter two diagnosis groups are insignificant as well.
http://www.oecdbetterlifeindex.org/topics/work-life-balance/ (accessed November 12, 2018)
Part-time workers cannot be identified in the SIP database.
References
Altonji JG, Usui E (2007) Work hours, wages, and vacation leave. ILR Rev 60(3):408–428
Arbetsgivarverket (2009) Om semester - En vägledning för statliga arbetsgivare. Arbetsgivarverket, Stockholm
Bell D, Otterbach S, Sousa-Poza A (2012) Work hours constraints and health. Ann Econ Stat (105/106):35–54
Berniell I, Bietenbeck J (2017) The effect of working hours on health, IZA Discussion Paper
Carneiro P, Løken KV, Salvanes KG (2015) A flying start? Maternity leave benefits and long-run outcomes of children,. J Polit Econ 123(2):365–412
Chikani V, Reding D, Gunderson P, McCarty CA (2005) Vacations improve mental health among rural women: The Wisconsin Rural Women’s Health Study. Wis Med J 104(6):20–23
de Bloom JG, Sabine A. E., Kompier MAJ (2009) Do we recover from vacation? Meta-analysis of vacation effects on health and well-being. J Occupat Health 51(1):13–25
de Bloom JG, Sabine AE, Kompier MAJ (2012) Effects of short vacations, vacation activities and experiences on employee health and well-being. Stress Health 28(4):305–318
de Bloom JG, Sabine AE, Kompier Michiel AJ (2013) Vacation (after-) effects on employee health and well-being, and the role of vacation activities, experiences and sleep. J Happin Stud 14(2):613–633
Eaker ED, Pinsky J, Castelli WP (1992) Myocardial infarction and coronary death among women: Psychosocial predictors from a 20-year follow-up of women in the Framingham Study. Am J Epidemiol 135(8):854–864
Ericson B, Eriksson K (2015) Semesterlagen - med kommentar, 5th edn. Studentlitteratur, Stockholm
Expedia (2014) Expedia 2014 vacation deprivation study, Expedia
Fakih A (2014) Vacation leave, work hours, and wages: new evidence from linked employer–employee data. LABOUR 28(4):376–398
Fredriksson P, Öckert B (2014) Life-cycle effects of age at school start. Econ J 124(579):977–1004
Gelman A, Imbens G (2019) Why high-order polynomials should not be used in regression discontinuity designs. J Bus Econ Stat 37(3):447–456
Goerke L, Jeworrek S, Pannenberg M (2015) Trade union membership and paid vacation in Germany. IZA J Labor Econ 4(1):17
Grembi V, Nannicini T, Troiano U (2016) Do fiscal rules matter?. Amer Econ J Appl Econ 8(3):1–30
Grossman M (1972) On the concept of health capital and the demand for health. J Polit Econ 80(2):223–255
Gump BB, Matthews KA (2000) Are vacations good for your health? The 9-year mortality experience after the Multiple Risk Factor Intervention Trial. Psychosom Med 62(5):608–612
Hellquist BN, Duffy SW, Abdsaleh S, Björneld L, Bordás P, Tabár L, Viták B, Zackrisson S, Nyström L, Håkan J (2011) Effectiveness of population-based service screening with mammography for women ages 40 to 49 years: evaluation of the Swedish Mammography Screening in Young Women (SCRY) cohort. Cancer 117(4):714–722
Hesselius P, Nilsson J, Johansson P (2009) Peter Sick of your colleagues’ absence. J Eur Econ Assoc 7(2-3):583–594
Hoynes H, Schanzenbach DW, Almond D (2016) Long-run impacts of childhood access to the safety net. Am Econ Rev 106(4):903–34
Hummels D, Munch J, Xiang C (2016) No pain no gain: the effects of exports on effort, injury, and illness. NBER Working Paper
Kühne S, Schneider T, Richter D (2015) Big changes before big birthdays? Panel data provide no evidence of end-of-decade crises. PNAS 112 (11):E1170
Kühnel J, Sonnentag S (2011) How long do you benefit from vacation? A closer look at the fade-out of vacation effects. J Organ Behav 32(1):125–143
Kim D (2019) Does paid vacation leave protect against depression among working Americans? A national longitudinal fixed effects analysis, Scandinavian Journal of Work. Environ Health 45(1):22–32
Kirillova K, Lehto X (2015) An existential conceptualization of the vacation cycle. Ann Tourism Res 55:110–123
Kling JR, Liebman JB, Katz LF (2007) Experimental analysis of neighborhood effects. Econometrica 75(1):83–119
Kolesár M, Rothe C (2018) Inference in regression discontinuity designs with a discrete running variable. Am Econ Rev 108(8):2277–2304
Lalive R (2008) How do extended benefits affect unemployment duration? A regression discontinuity approach. J Econ 142(2):785–806
Lee DS, Card D (2008) Regression discontinuity inference with specification error. J Econ 142(2):655–674
Lee DS, Lemieux T (2010) Regression discontinuity designs in economics. J Econ Lit 48(2):281–355
Leonardi M, Pica G (2013) Who pays for it? The heterogeneous wage effects of employment protection legislation. Econ J 123(573):1236–1278
Lepinteur A (2019) The shorter workweek and worker wellbeing: evidence from Portugal and France. Labour Econ 58:204–220
McCrary J (2008) Manipulation of the running variable in the regression discontinuity design: a density test. J Econ 142(2):698–714
Plug EJS (2001) Season of birth, schooling and earnings. J Econ Psychol 22(5):641–660
Saco-S and Arbetsgivarverket (2014) Partsgemensam utvärdering av RALS-T - 2:a utvärdering 2014-05-06, Saco-S and Arbetsgivarverket
Schnitzlein DD (2012) Extent and effects of employees in Germany forgoing vacation time. DIW Econ Bullet 2(2):25–31
SOU (2001) Arbetstiden - lag eller avtal, Statens offentliga utredningar (SOU Swedish Government Official Reports) 2001:91. Arbetsmarknadsdepartementet, Stockholm
Statskontoret (2015) Den offentliga sektorn i korthet 2015. Statskontoret, Stockholm
Strandberg TE, von Bonsdorff M, Strandberg A, Pitkälä K, Räikkönen K (2017) Associations of vacation time with lifestyle, long-term mortality and health-related quality of life in old age: The Helsinki Businessmen Study. Eur Geriat Med 8(3):260–264
Torun H, Tumen S (2017) The empirical content of season-of-birth effects: An investigation with Turkish data. Demogr Res 37(57):1825–1860
Acknowledgments
I thank Jan Bietenbeck, Petter Lundborg, Otto Sevaldson Lillebø, N. Meltem Daysal, and Aline Bütikofer, as well as seminar and conference participants, at Lund University, KORA in Copenhagen, ESPE 2017, EALE 2017, the University of Amsterdam, the Erasmus University Rotterdam, and the SNS Job Market 2018 for their comments and suggestions. I also thank the Centre for Economic Demography at Lund University for granting me access to their SIP database. I greatly appreciate the help of Patrik Spånning Westerlund and Ewa Nordgaard at Arbetsgivarverket who helped me get hold of all collective agreements.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The author declares no conflict of interest.
Additional information
Responsible editor: Shuaizhang Feng
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
Rights and permissions
About this article
Cite this article
Hofmarcher, T. The effect of paid vacation on health: evidence from Sweden. J Popul Econ 34, 929–967 (2021). https://doi.org/10.1007/s00148-020-00789-z
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00148-020-00789-z