You are here

Article Text

Article menu

Download PDFPDF

Other topics
Adolescent mental health and earnings inequalities in adulthood: evidence from the Young-HUNT Study
  1. Miriam Evensen1,
  2. Torkild Hovde Lyngstad2,
  3. Ole Melkevik1,
  4. Anne Reneflot1,
  5. Arnstein Mykletun1
  1. 1Division of Mental Health, Norwegian Institute of Public Health, Oslo, Norway
  2. 2Department of Sociology and Human Geography, University of Oslo, Oslo, Norway
  1. Correspondence to Miriam Evensen, Division of Mental Health, Norwegian Institute of Public Health, P.O. Box 4404 Nydalen, Oslo 0403, Norway; miev{at}fhi.no

Abstract

Background Previous studies have shown that adolescent mental health problems are associated with lower employment probabilities and risk of unemployment. The evidence on how earnings are affected is much weaker, and few have addressed whether any association reflects unobserved characteristics and whether the consequences of mental health problems vary across the earnings distribution.

Methods A population-based Norwegian health survey linked to administrative registry data (N=7885) was used to estimate how adolescents' mental health problems (separate indicators of internalising, conduct, and attention problems and total sum scores) affect earnings (≥30 years) in young adulthood. We used linear regression with fixed-effects models comparing either students within schools or siblings within families. Unconditional quantile regressions were used to explore differentials across the earnings distribution.

Results Mental health problems in adolescence reduce average earnings in adulthood, and associations are robust to control for observed family background and school fixed effects. For some, but not all mental health problems, associations are also robust in sibling fixed-effects models, where all stable family factors are controlled. Further, we found much larger earnings loss below the 25th centile.

Conclusions Adolescent mental health problems reduce adult earnings, especially among individuals in the lower tail of the earnings distribution. Preventing mental health problems in adolescence may increase future earnings.

  • MENTAL HEALTH
  • INEQUALITIES
  • LONGITUDINAL STUDIES
  • EMPLOYMENT
  • Social and life-course epidemiology

https://doi.org/10.1136/jech-2015-206939

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    Introduction

    Health, especially during formative years in childhood and adolescence, is strongly related to socioeconomic status in adulthood.1 A large literature documents that physical health in early life is crucial for later life outcomes such as education and labour market outcomes.2–4 A growing body of literature have also shown negative consequences of childhood and adolescent mental health problems5 ,6 and some studies indicate that the adverse consequences of mental health problems exceed those of physical health and affect a wider array of outcomes.7 ,8 There is also an emerging consensus that adolescence constitutes a particularly important developmental period, due to the rapid biological, psychological and social changes that unfold during the teenage years.9 ,10 This makes adolescence a particularly sensitive and consequential period in the transition to adulthood.

    Concern for the adverse consequences of mental health problems is augmented by the fact that mental health problems are generally widespread. In developed countries, estimates show that internalising and externalising problems, the two most common forms of mental health problems, affect as many as 20% of children and adolescents.11 ,12 The high prevalence and an early onset makes mental illness a leading cause of disability worldwide.13

    To date, several studies have found adolescent mental health problems to be associated with adverse labour market outcomes.14–18 There are however some limitations with these studies. First, some studies have investigated employment measured very early in adult life.16 ,17 To what extent these effects extend to permanent labour market outcomes, enduring over the life course is less established. Studies have also relied on retrospective questions about child and adolescent health, which may put them at risk for bias due to reverse causality.14 ,19 Further, a key challenge in the literature is to distinguish between the impact of mental health problems on adverse life outcomes and confounding factors predisposing an individual to mental health problems and contributing to poor labour market outcomes.

    Sibling comparison designs constitute a powerful way of controlling for unobserved family factors, as siblings share family environment and, on average, half of their genotype. Yet, only a few studies have addressed the potential impact of adolescent health on adult outcomes by simultaneously taking into account unobserved confounders by means of sibling fixed effects. Effects of mental disorders on income and employment have been found among Swedish men using military enlistment data.20 Strong effects of mental health problems on household income have also been found in the USA, although the study relied on retrospective data on childhood health.18 The studies most similar to ours used data from a health survey among adolescents and found large effects of self-reported attention deficit hyperactivity disorder (ADHD)21 and symptoms of depression22 on labour market outcomes. While the aforementioned studies indicate that child and adolescent health are associated with adverse labour market outcomes, all studies share a focus on average effects of mental health problems on earnings. There may be variation in employment flexibilities across the earnings distribution which may influence the size of the potential penalties. Focusing on average effects may thus mask variations in earnings losses associated with mental problems.23

    Our aim in this study is twofold: first, we examine the mean impact of several mental health problems on earnings in early adulthood. While we refer to mental health problems throughout, it should be mentioned that by this we mean symptoms of self-reported mental health problems rather than clinical diagnoses. By linking adolescent mental health information to register data on earnings in adulthood, we assure the direction of the associations. Further, we use sibling fixed-effects models that controls for shared factors at the family level to reduce potential bias from unobserved variables. Second, we use a recently developed quantile regression method to examine whether mental health problems have differential effects at different parts of the earnings distribution.

    Methods

    Study design and sample

    We use information from a population-based health survey, the Young-HUNT Study, compiled in 1995–1997. The study took place in Nord-Trøndelag County in Norway. A comprehensive questionnaire was completed by 8949 adolescents, aged 13–19 years, during class hours. All participants or their parents, if the child was younger than 16, gave their written consent to participate in the study. The response rate was 90%.24 Data from the survey was further merged to population registries covering information about annual earnings and sociodemographic variables for parent and child. These linkages were approved by the Regional Committee on Ethics in Medical Research. Since our follow-up is register-based there is no loss to attrition.

    The data set contains information on parents and siblings, thus allowing us to identify siblings for use in sibling fixed-effects models. There are 7885 observations with valid measures on all mental health problems and earnings. Within this sample, there are 2637 full siblings.

    Measures

    Table 1 provides summary statistics of means and SDs for all variables used in the analyses separately for the full sample and the sibling sample.

    View this table:
    Table 1

    Descriptive statistics of variables used in analysis

    Adult earnings

    Our outcome measure is adult earnings. Adult earnings consist of annual tax reports on pretax wages and income from self-employment, in fixed prices. It is measured for all years when the adolescent aged 30–36 years, up to 2013. Earnings are inflated to 2013 levels using the Norwegian consumer price index. Earnings are measured in natural logs, thus the coefficients will be interpreted as percentage changes. Individuals with zero income are excluded (5%).

    Adolescent mental health

    We include four measures of mental health problems: internalising problems, attention problems and conduct problems, in addition to a total score based on the three measures. We measure internalising problems as symptoms of anxiety and depression, a five-item scale based on the Hopkins Symptoms Checklist-25 (SCL-25). The 5-item version (SCL-5) asks if the respondents during the last 14 days have been ‘feeling fearful’, ‘have felt nervousness or shakiness inside’, are ‘feeling hopeless about the future’, ‘have felt blue’ and ‘worrying too much about things’. Each question is measured on a four-point scale from not bothered (1) to very bothered (4). The five-item scale has shown to be a reliable measure and correlates highly with the SCL-25.25 We use the mean values of SCL-5 for respondents with valid scores on three or more of the items. Our measures of attention and conduct problems were calculated using items from a module on school adjustment. Questions from this module have been used in several studies, although in different ways.26 ,27 The attention problems score included the statements ‘cannot sit still’ and ‘have difficulties concentrating’. The statements that made up the conduct problems dimension were: ‘are reprimanded by the teacher’, ‘argue with the teacher’, ‘get into a fight’ and ‘skip school’. The statements were measured in a never (1) to very often (4) format, and we use the mean values on these measures. For the total score, we use mean values summarised for internalising, attention and conduct problems for each individual. Further, all mental health indices were z-standardised (mean=0, SD=1).

    Covariates

    We also include information on immigrant background, birth year and number of siblings, and whether the individual was the first-born child of his or her mother. The family structure variable indicated whether the respondent had experienced parental divorce or separation at the time of survey screening. Parental education is measured using information on five levels of attainment level for the parent with the highest qualifications when the respondent was aged 16. From the survey, we can also identify which schools the adolescents attended at the time of screening, which enables us to control for stable school characteristics by including school fixed effects.

    Statistical analyses

    We start by examining the relationship between adolescent mental health problems and adult earnings using ordinary least squares (OLS) regression. We estimate the following school fixed-effects model separately for each of the mental health problem indicators:Embedded Image 1where Log(earnings)is is the natural logarithm of earnings of individual i in school s. The model includes a school fixed effect, θs, which controls for all time-invariant school-level factors common to individuals who attended the same school at the time of screening and a vector of observed individual and family-level characteristics, Xis, listed above. The key parameter to be estimated is β, which estimates the relationship between the mental health indicators in adolescence and adult earnings. While Eq. (1) will control for important individual, family-level and school-level confounding factors, there could also be important unobserved family-level determinants that could confound the estimated relationship.

    To further test the robustness of the mental health–earnings relationship, we turn to the sibling sample. For this subsample, we estimate the school fixed-effects models from Eq. (1) as well as the following sibling fixed-effects model:Embedded Image 2where Log(earnings)if refers to the outcome of sibling i in family f. The inclusion of the sibling fixed effects, µf, limits the comparison to variation between siblings within the same family, thus controlling for all time-invariant characteristics shared by siblings at the family level. Zif refers to the observed individual-level characteristics that vary between siblings. Importantly, Eq. (2) will indicate whether the results obtained from Eq. (1) were driven by unobserved family-level factors that affect mental health problems and earnings.5 ,6

    While OLS models estimate the effects of mental health problems on average earnings, we are also interested in how the effects of mental health problems vary across the earnings distribution. Unconditional quantile regression (UQR) models enable us to evaluate the effect of mental health on earnings at different positions throughout the earnings distribution (eg, 5–95th centile) by replacing the outcome variable (ie, log earnings) with the recentred influence function (RIF).28 ,29

    Results

    Earnings results from linear regression models

    Table 2 shows results for the association between mental health problems and log earnings from OLS regression models with school fixed effects and with sibling fixed effects. We present results separately for our measures of mental health problems. In the school fixed-effects model, we adjust for the respondent's sex, whether the parents were divorced or separated at screening, whether the respondent was the first-born child of his or her mother, number or siblings, immigrant background, parental educational level, birth cohort and school fixed effects using the full sample and the sibling sample. In the sibling fixed-effects model, we adjust for the respondent's sex, whether the parents were divorced or separated at screening, whether the respondent was the first-born child of his or her mother and birth cohort using the sibling sample.

    View this table:
    Table 2

    Estimated relationship between mental health status and log earnings from ordinary least squares (OLS) regressions

    In the school fixed-effects models, the results show that all mental health problems significantly reduce earnings. Starting with the full sample, the estimated coefficient shows that a one SD increase in total mental health problems reduces earnings by about 6.5% (ie, e–0.067–1=–0.065, which is the formula (eb–1) we use throughout to transform the log coefficients into percentage changes). Turning to the separate indicators, the estimated associations are slightly stronger for internalising problems than for attention and conduct problems. Narrowing the focus to the sibling sample, we find a similar pattern for our measures of externalising problems, although coefficients are smaller in size. For internalising problems, the earnings reduction is almost halved from an estimated 6.1% to 3.1%.

    The sibling fixed-effects models show that the point estimates obtained are comparable with those found in the school fixed-effects models for both samples with respect to total mental health problems, internalising problems and attention problems, although SEs are much larger. In contrast, point estimates for conduct problems are substantially reduced.

    Overall, we find that mental health problems reduce earnings, although the magnitude of the associations seems modest. However, when we include sibling fixed effects, only the estimates for total problems and attention problems remain significant. The explained variation in adult earnings (R2) is substantially higher in the sibling fixed-effects models.

    Earnings results from UQR models

    Figure 1 presents results for log earnings using UQR to examine whether the relationship between mental health problems and adult earnings varies over the earnings distribution. For each of the mental health indicators, the panels in the figure plot estimated coefficients of mental health indicators on log earnings (with a 95% CI) in five centile intervals from the 5th centile to the 95th centile. Each model is estimated for the full sample with adjustments as in table 2, model 1.

    Figure 1
    Figure 1

    The relationship between adolescent mental health problems and log income in young adulthood using unconditional quantile regression (UQR) with 95% CIs. All models control for year of birth, sex, parental education, parental divorce, number of siblings, whether the child was the first born to his or her mother, immigrant background and school fixed effects.

    For all mental health indicators, there are substantially larger earnings penalties related to mental health problems in the lower part of the earnings distribution. Figure 1 shows that the estimated impact of mental health problems on earnings is progressively stronger below, approximately, the 25th centile. If we focus on the estimated effects at the 10th centile, a one SD increase in the total symptoms of mental health problems reduces earnings by about 15%. Looking at the specific mental health problems, we find that internalising problems impose the strongest negative effect on earnings, resulting in a 16% reduction, whereas attention problems and conduct problems are related to 9% and 8% reductions in earnings, respectively.

    Taken together, these results provide evidence of a considerably stronger relationship between mental health problems and adult earnings among individuals found in the lowest part of the earnings distribution. Importantly, the estimated average effect of mental health problems on log earnings obtained from OLS regression therefore masks considerable heterogeneity across the earnings distribution.

    Discussion and conclusions

    This study examines the association between several mental health problems in adolescence and earnings in adulthood. We find that all mental health problems reduce earnings. To examine whether the association between mental health problems and earnings was confounded by characteristics within families, we estimated models with sibling fixed effects, which controls for time-invariant factors shared between siblings. The results from the sibling fixed-effects models generally confirm the results from school fixed-effects models for total problems; however, looking at the specific indicators of mental health problems, attention problems seem to be most consistently associated with lower earnings. For internalising problems, the effects remain in the same range, although the coefficients are no longer significant when we include sibling fixed effects. Effects of conduct problems are substantially reduced in the sibling fixed-effects models, which suggest that conduct problems are correlated with unobserved factors at the family level that also lowers the individuals' earnings. Further, we found substantially larger effects in the lower part of the earnings distribution. For example, the OLS specification shows that a one SD increase in our overall measure of mental health problems reduces average earnings with 7% compared with a 15% reduction at the 10th centile in the UQR model.

    Our results are in line with previous research which also indicates that adolescent mental health problems reduce adult earnings.20–22 Moreover, we show that the estimated average effect of mental health problems on earnings obtained from OLS regression conceals considerable variation across the earnings distribution. Thus, assessments of average outcomes for adolescents with poor mental health may be useful, but do not paint the full picture of how this health problem affect life chances. If adolescents have managed to acquire skills, qualifications and jobs to reach median income levels or above, they are less affected than if they rank in the lower end of the distribution.

    There are several ways mental problems may affect labour market outcomes. One possibility is that mental health affects productivity, concentration or motivation. However the pattern with larger losses at the bottom of the earnings distribution could imply that those at the median or above have some (unobserved) characteristics, such as self-esteem, motivation or other traits which lets them overcome some of the difficulties related to adolescent mental health problems. Another explanation may be that employers' willingness to accommodate those with mental health problems depend on job characteristics and pay.23 ,30

    Our study has several strengths. We use data from a population survey with a register-based follow-up, not plagued with attrition. Further, our study is one of a few that employ a sibling design, which enables us indirectly to rule out all confounding factors at the family level that are shared among siblings. Thus, it represents an improvement in the understanding of causal pathways between adolescent mental health and adult economic outcomes. Moreover, the use of UQR models shows the importance of considering differential effects across the earnings distribution.

    Yet, the findings should also be considered in light of several limitations. First, the study might have benefitted from using diagnostic information about mental illnesses that provide more detailed information about severity and clinical relevance. In contrast, screener questions asked to all adolescents, as used here, tap directly into the symptoms apparent in a population and are less susceptible to biases in diagnosis (eg, if resourceful parents are more able to get their children proper medical help).6 ,31 Second, although sibling comparisons constitute a useful strategy for reducing omitted variable bias, they can only account for unobserved confounders that are stable within families. We are therefore not able to control for confounding child-specific characteristics that may vary within families, such as bullying in school or genetic endowments.32 Relatedly, although the study is well powered, the likelihood of type II error is markedly higher in sibling fixed-effects models.

    Given these considerations, our study shows that adolescent mental health problems reduce adult earnings, although at a greater degree in the lower tail of the earnings distribution. Preventing and treating adolescent mental health problems may positively influence adolescents' later-life economic well-being.

    What is already known on this subject

    • Child and adolescent mental health problems are associated with adverse labour market outcomes in adulthood.

    • Few studies have examined whether the association is causal, and it is unclear where in the labour market adversities occur.

    What this study adds

    • This study is one of a few to consider the long-term association between adolescent mental health problems and earnings, and whether the association is confounded by unobserved family characteristics.

    • This study is the first to consider whether the long-term associations between adolescent mental health problems and earnings vary across the earnings distribution.

    • The findings suggest that adolescent mental health problems reduce adult earnings. These penalties are larger in the lower tail of the earnings distribution, which suggests that conventional models of average effects conceal important variation across the distribution.

    Acknowledgments

    The Nord-Trøndelag Health Study (The HUNT Study) is a collaboration between HUNT Research Centre (Faculty of Medicine, Norwegian University of Science and Technology NTNU), Nord-Trøndelag County Council, Central Norway Health Authority and the Norwegian Institute of Public Health.

    References

    1. ↵
      1. Case A,
      2. Paxson C
      . Causes and consequences of early-life health. Demography 2010;47(Suppl):S65–85. doi:10.1353/dem.2010.0007
      OpenUrlCrossRefPubMedWeb of Science
    2. ↵
      1. Black SE,
      2. Devereux PJ,
      3. Salvanes KG
      . From the Cradle to the Labor Market? The Effect of Birth Weight on Adult Outcomes*. Q J Econ 2007;122:409–39.
      OpenUrlAbstract/FREE Full Text
    3. ↵
      1. Case A,
      2. Fertig A,
      3. Paxson C
      . The lasting impact of childhood health and circumstance. J Health Econ 2005;24:365–89. doi:10.1016/j.jhealeco.2004.09.008
      OpenUrlCrossRefPubMedWeb of Science
    4. ↵
      1. Currie J,
      2. Stabile M,
      3. Manivong P, et al
      . Child health and young adult outcomes. J Hum Resour 2010;45:517–48. doi:10.1353/jhr.2010.0013
      OpenUrlCrossRefWeb of Science
    5. ↵
      1. Fletcher JM
      . Adolescent depression and educational attainment: results using sibling fixed effects. Health Econ 2010;19:855–71. doi:10.1002/hec.1526
      OpenUrlPubMedWeb of Science
    6. ↵
      1. Currie J,
      2. Stabile M
      . Mental health in childhood and human capital. The problems of disadvantaged youth: an economic perspective. University of Chicago Press, 2007:115–48.
    7. ↵
      1. Goodman A,
      2. Joyce R,
      3. Smith JP
      . The long shadow cast by childhood physical and mental problems on adult life. Proc Nat Acad Sci 2011;108:6032–7. doi:10.1073/pnas.1016970108
      OpenUrlAbstract/FREE Full Text
    8. ↵
      1. Delaney L,
      2. Smith JP
      . Childhood health: trends and consequences over the life-course. Future Child 2012;22:43. doi:10.1353/foc.2012.0003
      OpenUrlCrossRefPubMed
    9. ↵
      1. Viner RM,
      2. Ross D,
      3. Hardy R, et al
      . Life course epidemiology: recognising the importance of adolescence. J Epidemiol Community Health 2015;69:719–20doi:10.1136/jech-2014-205300.
      OpenUrlFREE Full Text
    10. ↵
      1. Patel V
      . Reducing the burden of depression in youth: what are the implications of neuroscience and genetics on policies and programs? J Adolesc Health 2013;52:S36–38. doi:10.1016/j.jadohealth.2012.04.016
      OpenUrl
    11. ↵
      1. Thapar A,
      2. Collishaw S,
      3. Pine DS, et al
      . Depression in adolescence. Lancet 2012;379:1056–67. doi:10.1016/S0140-6736(11)60871-4
      OpenUrlCrossRefPubMedWeb of Science
    12. ↵
      1. Lee FS,
      2. Heimer H,
      3. Giedd JN, et al
      . Adolescent mental health—Opportunity and obligation. Science 2014;346:547–9. doi:10.1126/science.1260497
      OpenUrlAbstract/FREE Full Text
    13. ↵
      1. Gore FM,
      2. Bloem PJN,
      3. Patton GC, et al
      . Global burden of disease in young people aged 10–24 years: a systematic analysis. Lancet 2011;377:2093–102. doi:10.1016/S0140-6736(11)60512-6
      OpenUrlCrossRefPubMedWeb of Science
    14. ↵
      1. Ettner SL,
      2. Frank RG,
      3. Kessler RC
      . The impact of psychiatric disorders on labor market outcomes. Ind Labor Relat Rev 1997;51:64–81. doi:10.1177/001979399705100105
      OpenUrlCrossRefWeb of Science
    15. ↵
      1. Fergusson DM,
      2. Woodward LJ
      . Mental health, educational, and social role outcomes of adolescents with depression. Arch Gen Psychiatry 2002;59:225–31. doi:10.1001/archpsyc.59.3.225
      OpenUrlCrossRefPubMedWeb of Science
    16. ↵
      1. Veldman K,
      2. Reijneveld SA,
      3. Ortiz JA, et al
      . Mental health trajectories from childhood to young adulthood affect the educational and employment status of young adults: results from the TRAILS study. J Epidemiol Community Health 2015;69:588–93. doi:10.1136/jech-2014-204421
      OpenUrlAbstract/FREE Full Text
    17. ↵
      1. Egan M,
      2. Daly M,
      3. Delaney L
      . Childhood psychological distress and youth unemployment: evidence from two British cohort studies. Soc Sci Med 2015;124:11–7. doi:10.1016/j.socscimed.2014.11.023
      OpenUrlCrossRefPubMed
    18. ↵
      1. Smith JP,
      2. Smith GC
      . Long-term economic costs of psychological problems during childhood. Soc Sci Med 2010;71:110–5. doi:10.1016/j.socscimed.2010.02.046
      OpenUrlCrossRefPubMed
    19. ↵
      1. Smith JP
      . The impact of childhood health on adult labor market outcomes. Rev Econ Stat 2009;91:478–89. doi:10.1162/rest.91.3.478
      OpenUrlCrossRefPubMed
    20. ↵
      1. Lundborg P,
      2. Nilsson A,
      3. Rooth D-O
      . Adolescent Health and Adult Labor Market Outcomes. J Health Econ 2014;37:25–40. doi:10.1016/j.jhealeco.2014.05.003
      OpenUrlCrossRefPubMed
    21. ↵
      1. Fletcher JM
      . The effects of childhood ADHD on adult labor market outcomes. Health Econ 2014;23:159–81. doi:10.1002/hec.2907
      OpenUrlCrossRefPubMed
    22. ↵
      1. Fletcher J
      . Adolescent depression and adult labor market outcomes. South Econ J 2013;80:26–49. doi:10.4284/0038-4038-2011.193
      OpenUrlCrossRef
    23. ↵
      1. Marcotte D,
      2. Wilcox-Gök V
      . Estimating earnings losses due to mental illness: a quantile regression approach. J Ment Health Policy Econ 2003;6:123–34.
      OpenUrlPubMed
    24. ↵
      1. Holmen TL,
      2. Bratberg G,
      3. Krokstad S, et al
      . Cohort profile of the Young-HUNT Study, Norway: a population-based study of adolescents. Int J Epidemiol 2014;43:536–44doi:10.1093/ije/dys232.
      OpenUrlAbstract/FREE Full Text
    25. ↵
      1. Tambs K,
      2. Moum T
      . How well can a few questionnaire items indicate anxiety and depression? Acta Psychiatr Scand 1993;87:364–7. doi:10.1111/j.1600-0447.1993.tb03388.x
      OpenUrlCrossRefPubMedWeb of Science
    26. ↵
      1. Torvik FA,
      2. Rognmo K,
      3. Ask H, et al
      . Parental alcohol use and adolescent school adjustment in the general population: Results from the HUNT study. BMC Public Health 2011;11:706. doi:10.1186/1471-2458-11-706
      OpenUrlCrossRefPubMed
    27. ↵
      1. De Ridder KA,
      2. Pape K,
      3. Johnsen R, et al
      . Adolescent health and high school dropout: a prospective cohort study of 9000 Norwegian adolescents (The Young-HUNT). PLoS ONE 2013;8:e74954. doi:10.1371/journal.pone.0074954
      OpenUrlCrossRefPubMed
    28. ↵
      1. Fortin N,
      2. Lemieux T,
      3. Firpo S
      . Unconditional quantile regression. Econometrica 2009;77:953–73. doi:10.3982/ECTA6822
      OpenUrlCrossRefWeb of Science
    29. ↵
      1. Killewald A,
      2. Bearak J
      . Is the motherhood penalty larger for low-wage women? A comment on quantile regression. Am Sociol Rev 2014;79:350–7. doi:10.1177/0003122414524574
      OpenUrlAbstract/FREE Full Text
    30. ↵
      1. Chatterji P,
      2. Alegria M,
      3. Takeuchi D
      . Psychiatric disorders and labor market outcomes: evidence from The National Comorbidity Survey-Replication. J Health Econ 2011;30:858–68. doi:10.1016/j.jhealeco.2011.06.006
      OpenUrlCrossRefPubMed
    31. ↵
      1. Kessler RC,
      2. Zhao S
      . The prevalence of mental illness. In: Horwitz AV, Scheid TL, eds. A handbook for the study of mental health. Cambridge University Press, 1999:58–78.
    32. ↵
      1. Frisell T,
      2. Öberg S,
      3. Kuja-Halkola R, et al
      . Sibling comparison designs: bias from non-shared confounders and measurement error. Epidemiology 2012;23:713–20. doi:10.1097/EDE.0b013e31825fa230
      OpenUrlCrossRefPubMedWeb of Science

    Footnotes

    • Twitter Follow Miriam Evensen at @miriamevensen

    • Contributors ME planned the study, performed the statistical analyses and wrote the paper. THL contributed to the interpretation of results, commented on drafts and details in the research design. AR and OM contributed to the interpretation of results and commented on drafts. AM contributed to planning of the study, interpretation of results and commented on drafts. All authors have approved the final draft of the manuscript.

    • Funding The research is funded by a grant from the Norwegian Research Council (grant 213744).

    • Competing interests None declared.

    • Ethics approval The Regional Committee for Medical Research Ethics in Norway approved this study.

    • Provenance and peer review Not commissioned; externally peer reviewed.