Family matters: The effects of parental unemployment in early childhood and adolescence on subjective well-being later in life

University of Groningen

Family matters

Nikolova, Milena; Nikolaev, Boris N.

Published in:

Journal of Economic Behavior & Organization

DOI:

10.1016/j.jebo.2018.05.005

IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from

it. Please check the document version below.

Document Version

Publisher's PDF, also known as Version of record

Publication date:

2021

Link to publication in University of Groningen/UMCG research database

Citation for published version (APA):

Nikolova, M., & Nikolaev, B. N. (2021). Family matters: The effects of parental unemployment in early

childhood and adolescence on subjective well-being later in life. Journal of Economic Behavior &

Organization, 181, 312-331. https://doi.org/10.1016/j.jebo.2018.05.005

Copyright

Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).

Take-down policy

If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.

Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.

ContentslistsavailableatScienceDirect

Journal

of

Economic

Behavior

and

Organization

journalhomepage:www.elsevier.com/locate/jebo

Family

matters:

The

effects

of

parental

unemployment

in

early

childhood

and

adolescence

on

subjective

well-being

later

in

life

Milena

Nikolova

_,

_Boris

_N.

_Nikolaev

a Faculty of Economics and Business, University of Groningen, Nettelbosje 2, 9747AE Groningen, The Netherlands b The Brookings Institution, Washington, DC 20036, USA

c IZA, Schaumburg-Lippe-Str. 5-9, 53113 Bonn, Germany

d John F. Baugh Center for Entrepreneurship, Hankamer School of Business, Baylor University, 1450 S. 4th Street Waco, TX 76706, USA

a

r

t

i

c

l

e

i

n

f

o

Article history: Received 27 July 2017 Revised 12 May 2018 Accepted 15 May 2018 Available online 26 May 2018

JEL Codes: I31 J01 J65 Keywords: Life satisfaction Parental unemployment Company closures Life-cycle analysis

German socio-economic panel

a

b

s

t

r

a

c

t

We are the first to examine how parental unemployment experienced during early-, mid- and late-childhood affects adult life satisfaction. Using German household panel data, we find that parental unemployment induced by plant closures and experienced during early (0–5 years) and late (11–15 years) childhood leads to lower life satisfaction at ages 18– 31. Nevertheless, parental unemployment can also have a positive effect depending on the age and gender of the child. Our results are robust even after controlling for local un- employment, individual and family characteristics, parental job loss expectations, financial resources, and parents’ working time when growing up. These findings imply that the ad- verse effects associated with parental unemployment experienced at a young age tend to last well into young adulthood and are more nuanced than previously thought.

© 2018 Elsevier B.V. All rights reserved.

1. Introduction

Thenegativeeffectsofunemploymentforthosedirectlyaffectedbyitarewell-documented.Unemploymentisnotonly associatedwithworselabormarketoutcomes(e.g.CouchandPlaczek,2010;DavisandWachter,2011),butalsowitha de-clineinphysicalandmentalhealth(e.g.EliasonandStorrie,2009;GerdthamandJohannesson,2003;PaulandMoser,2009), chronicstress(Baumetal.,1986),lowerlifesatisfaction(seeWinkelmann(2014)forareview),andcanevenleadtosuicide (AvdicandChevalier,2016;Millneretal., 2013).Thesenegativeeconomic andpsychological effectscan persistevenafter unemployedpeoplereturntowork (e.g.Clarketal., 2001;CouchandPlaczek, 2010;Winkelmann, 2014).1 _Unemployment

canalsoincreasethelikelihoodoffamilydissolution(e.g.,CharlesandStephens,2004;DoironandMendolia,2012)andmay

∗ _{Corresponding author at: Faculty of Economics and Business, Global Economics and Management, University of Groningen, Nettelbosje 2, 9747AE,}

Groningen, The Netherlands.

E-mail addresses: m.v.nikolova@rug.nl (M. Nikolova), boris_nikolaev@baylor.edu (B.N. Nikolaev).

1 Regional unemployment can also indirectly affect individual well-being, as even people who have jobs may worry about the possibility of becoming

unemployed in the future, fear crime and social tension, or feel bad about the unfortunate fate of those around them ( Di Tella et al., 2003; Frey & Stutzer, 2002 ).

https://doi.org/10.1016/j.jebo.2018.05.005

havenegativewell-beingspillovereffectstoother familymemberssuchasco-habitingspousesandchildren (e.g.Nikolova andAyhan,2018;PowdthaveeandVernoit,2013).

While a large literature in economicshas explored the relationship betweenfamily characteristics inearly childhood andadultoutcomessuch asacademic achievementandearnings (e.g.Cunha andHeckman,2008; Goodmanet al., 2011; Hilger,2016;Oreopoulos etal., 2008), thereremains a dearth ofstudies abouttheeffects ofchildhood circumstances on psychologicalwell-beinglaterinlife.2_{Thus,muchlessisknownabouthowparentalunemploymentexperiencedasachild}

impactsone’slifesatisfactionoverthelifecourse.

Inthispaper,we offertwocontributions tothislineofresearch.Thefirstoneissubstantive.Specifically,toour knowl-edge, we present the first evidence on the relationship betweenexogenous parental unemployment experienced during childhoodandyoungGermans’lifesatisfactionlaterinlife.Thus,weaddtotheemergingliteratureonthelong-termeffect offamilycircumstancesduringchildhoodonpsychologicalwell-beingacrossthelifecourse(Layardetal.,2014;Frijtersetal., 2014).Adistinctivefeatureofourresearchisthatweconsiderthepossibilitythatthepsychologicalconsequencesofparental unemployment differ across the different stages of child development(Gauvain and Cole, 2004; Piaget, 1971; Vygotsky, 2009).Thus, we systematicallyexamine theeffectofparentalunemploymentatages 0–5,6–10,and11–15,whichallows ustostudywhetherthereisalong-termeffectonchildren’spsychologicalwell-beingdependingonthetiming ofparental unemploymentinchildhood.We alsoempiricallydemonstratethat therelationship betweenparentalunemploymentand lifesatisfactionisheterogeneousandnuanceddependingonthechild’sandparent’sgender.

Oursecond contributionismethodological. We userich longitudinalhousehold-level informationavailable inthe Ger-manSocio-EconomicPaneltoidentifythelong-termconsequencesofparentaljoblessnessusingexogenousparental unem-ploymententry. Specifically, our identificationstrategy relieson company closures, which havebeen widely used in the literaturetocausallyinvestigatevariousconsequencesofunemployment(e.g.ChadiandHetschko,2017;Kassenboemerand HaiskenDeNew,2009;NikolovaandAyhan,2018;Schmitz,2011).Furthermore,bycontrollingfor(lagged)parentaljobloss expectationsandarichsetofindividual,parental,andfamilycharacteristics,wecantestwhetherfirmclosuresare unantic-ipatedandaddressthelimitationthat onlyaself-selectedsampleofparentswillremaininthefirmuntilitcloses.Ideally, wewould liketohave alarge administrativedatasetandinformation onmasslayoffs, whichcan beviewed asanatural experimentandstrengthenthecausal interpretationofour findings(ToddandWolpin, 2003).However, even ifavailable, administrativedataare unfeasibleinourcaseastheydonotcontain informationonpsychological well-being.Instead,we relyonplantclosingsasa sourceofexogenous variationinparentaljoblessness.Thus, ourfindings canbe interpretedas causalunder theassumption that parental unemploymentdueto company closuresis unrelatedto family characteristics thatweeitherdonotcontrol fororcannotmeasure.We testtherobustnessofourfindings indirectlyby usingdismissals andvolutnary parentalunemploymentanddirectybycontrolling fortheBig 5personalitytraitsofbothparents andtheir children.

Wefurthercomplementourempiricalworkbydrawingontheoriesfromeconomicsandpsychologythatemphasizethe criticalrole ofchildhoodexperiences fordevelopmental outcomesoverthelife course(e.g.BeckerandTomes,1986;Elder andConger,1995).Wearguethattheoreticallythelong-termconsequencesofparentalunemploymentcanbebothpositive andnegative.Ontheonehand,parentalunemploymentcannotonlysignificantly limitfamilies’abilitytoinvestin essen-tialresources (e.g.high-qualityeducation,healthcare,orhousing) that cancreatesafeandcognitively-enhancing learning environment(KalilandZiol-Guest,2008; CouchandPlaczek, 2010;DavisandWachter,2011),butalsoincrease children’s exposuretostressandaffectsocialinteractionswithinthefamily.Inturn,ahigherexposuretostresscanimpairchildren’s cognitive andemotionaldevelopment andcauseenduring systematicinflammatory response, which canlead to negative developmental cascade throughoutthe life course, thus increasing the risk of psychopathologyin later life (Danese and Baldwin, 2017). At the same time, unemployed parents mayincrease the time spent in childcare, which can have posi-tiveconsequencesforthechild’shumancapitaldevelopmentandwell-beinglaterinlife.Parentalunemploymentcanalso makesome childrenmore resilienttoadversity,which,in turn,can leadto greatersuccessandthushigherpsychological well-beinginadulthood.

Ourmainfinding isthat young adultswho experienceparentalunemployment inchildhoodare morelikely toreport lowerlevelsoflife satisfactioninadulthood (atages18–31)ifparentalunemploymentoccurredatages0–5or11–15. We explainthisbysuggestingthatthepsychologicalcostofparentalunemploymentmaybehigherforyoungchildrenbecause negativeexperiencesaccumulateoverthelifecourseandstressfullifeeventsearlyinlifehavestrongereffectonoutcomes later in life (Arbeit, 2013; DiPrete and Eirich, 2006). Older children, on the other hand,may feel pressure to take more responsibilityinthefamily(Arbeit,2013)andstartdevelopingabstractthinking(ages11–15),whichcanmakethemmore

2 Notable exceptions include _{Layard et al. (2014) , Frijters et al. (2014) , and Clark and Lee (2017) . First, using the British Cohort Study (1970), Layard}

et al. (2014) examine which childhood characteristics predict income, educational attainment, employment and partnership status, self-reported health, emotional health, and life satisfaction at the age of 34, with life satisfaction being the over-arching well-being measure that the authors consider. The paper finds that the child’s emotional health and conduct are the most important predictors of adult life satisfaction. The paper also documents that income and education are among the least important determinants of life satisfaction at the age of 34. Second, mainly utilizing information from the 1958 National Child Development Survey, Frijters et al. (2014) further conclude that socio-economic status variables up to the age of 16 are poor predictors of adult life satisfaction (at the ages of 33, 42, 46, and 50). Child conduct and social problems predict some portion of adult life satisfaction. Finally, relying on the Wisconsin Longitudinal study Clark and Lee (2017) conclude that circumstances at age 18 are good predictors of well-being (happiness and eudaimonia) later in life, even after controlling for contemporaneous covariates. All three studies show that childhood characteristics predict only a minor percent of the variation in life adult life satisfaction.

awareofsocialstigmaassociatedwithparentalunemploymentortheirdisadvantagedsocio-economic status(Gauvainand Cole,2004; Piaget,1971). Atthesametime, thereissome heterogeneityaccordingto thechild’sgender andwhetherthe motheror the fatherbecame jobless.Boys, forexample, tend tobe more negatively affected by parentalunemployment compared to girls, possibly because atyounger ages girls show better emotional self-regulation (Matthews et al., 2009). Finally,wealsofindthatparentaljoblessnesscanevenhavepositiveconsequencesforsome youngadultslikelyduetothe parentalinvestmentandthequalityofthetimeunemploymentparentsspendwiththeirkids.Theseresultsarerobusteven aftercontrollingfortheaveragestateunemploymentrateandhouseholdincomeduringchildhood,suggestingthat weare notpickinguptheconsequencesofgrowingupinarecessionorinpoverty.Furtherrobustnesschecksconfirmthevalidity ofourmainfindings.

Adoptingalifecourseperspectiveoffamilyunemploymentallows ustodemonstratethattheintergenerational psycho-logical costsof unemploymentare morenuanced thanpreviously thought.Giventhat a life courseperspective can show thepointintimeatwhichpublicpolicyinterventionscanbemosteffective(Layardetal.,2014),ourfindingssuggestthat programs targetingunemployedparentscanalsohelp toalleviate thepersistentpsychologicalburdenonchildren.Finally, becausesubjectivewell-beingisassociatedwithpositiveoutcomesinmanylifedomains(DeNeveetal.,2013),such child-hoodpolicyinterventionscanpotentiallyprovidelife-longbenefits.

2. Literatureonthewell-beingconsequencesofparentalunemployment

Severalrecentstudies havedocumented the contemporaneous spillovereffects ofparentalunemployment onthe well-being ofco-resident children. For example,using British panel data, Powdthavee andVernoit (2013) show that parental joblessnessispositive forchildren’shappinesswhenthechildisup to11yearsoldbutnegativeorinsignificantforolder co-resident children.3 _{A similar studyfor Germany finds that the life satisfaction of male children aged 17–25 declines}

following their father’s unemployment (Kind and Haisken-DeNew, 2012). Bubonya et al. (2017) also reveal that parental unemploymentworsensco-residentfemale children’smentalhealthinAustralia.Whilethesestudiesexaminethe contem-poraneouseffectofparentalunemployment,westillknowverylittleabouthowparentalunemploymentinthepastaffects children’spsychologicalwell-beingoverthelifecourse.Weproposethatparentalunemploymentexperiencedinchildhood couldimpactadultlifesatisfactionthroughbothpositiveandnegativechannels.

To our knowledge, no other paper investigates how exogenous parental unemployment during childhood affects the life satisfaction of young Germans later in life. Nevertheless,our study ismost similar in its contribution toa studyby Ermischetal.(2004),whoexaminehowparentaljoblessnessduringchildhoodaffectsarangeofoutcomes,including psy-chologicaldistress(measuredusingtheGHQ-12),educational attainment,inactivity,earlychildbearing,andsmoking. Rely-ingonsiblingfixed-effectsestimatorsandBritishHouseholdPanelStudydata,theauthorsfindthatparentalunemployment duringchildhoodincreasestheprobabilityofexperiencingpsychologicaldistresslaterinlife,especiallyifitoccurredduring theagesof11–15.4 _{However,ourstudydiffersinimportantwayssuchastheidentificationstrategy (i.e.relying on}

exoge-nous variationin parental unemployment), thechoice of the outcome variable(s), the dataset,andthe years of analysis. We alsoacknowledgethat Pinger(2016) studieshowparentalunemploymentatage16 affectseducational outcomesand demonstrates that parentalunemploymentis unassociated withthe adultchild’s mentalhealth (buthas asmall positive effectonphysicalhealth).

3. Mechanisms

Twotheoreticalperspectiveselucidatetheeconomicandpsychologicalimpactsofearlychildhoodexperienceson devel-opment andwell-being inlater life. The first perspective—largely advanced by economists—is basedon the “investment” modelproposedby BeckerandTomes (1986).Accordingthismodel,joblessnesscandiminishfamilies’economicabilityto invest inthe necessary resources topromote children’sachievement andwell-being later inlife (Brooks-Gunnand Dun-can, 1997; Cunha etal., 2010;Duncan et al., 1998; Heckman andCarneiro, 2003).5 _{More specifically, economic setbacks}

3 The authors also show that five consecutive years of paternal unemployment (from the age of 11–15) reduces girls’ happiness with life by 1.4 points (on

a 1–7 scale), while five consecutive years of maternal unemployment reduces boys’ life satisfaction by 2.4 points (on a 1–7 scale). Furthermore, a working paper by Powdthavee and Vernoit (2012) looks at the happiness of children who are 11–15 years old resulting from three continuous years of parental unemployment. The authors find that the father’s unemployment over the past three years reduces happiness with life by about 0.3 points on a scale of 1–7. Maternal unemployment has no such associated effects.

4 A related paper by _{Powdthavee (2014) exploits British panel data to examine how childhood characteristics mediate the relationship between unem-}

ployment and life satisfaction and mental health for adults aged between 16 and 29 years. He finds that while the coefficient estimate on own unemploy- ment is negative and statistically significant in both the life satisfaction and mental health regressions, that on father’s unemployment at ages 11 and 15 is positive and statistically significant for both life satisfaction and mental health, while maternal unemployment is negative but insignificant in both regres- sion. Maternal unemployment has small and negative effects on the life satisfaction of men aged 16–29 but positive and significant effects on women aged 16–29. The paper by Powdthavee (2014) differs from our research in that it examines life satisfaction at much earlier ages (with most of the respondents being in their teens), does not rely on exogenous unemployment entry, and uses UK panel data.

5 Consistent with this economics perspective, a large body of literature shows that involuntary-displaced workers suffer substantial income loses that are

persistent over the life course ( Couch and Placzek, 2010; Davis and Wachter, 2011; Farber, 2005 ). Empirical estimates suggest an immediate 33% earning loss and a loss as high as 15% even six years after displacement, with a cumulative lifetime earning loss of roughly 20% ( Brand, 2015; Couch and Placzek, 2010 ).

cansignificantlyinhibitfamilies’abilitytoaffordhigh-qualityeducation,properhealthcare,housinginsafeneighborhoods, nutritiousfood, clothing,toys, games, orbooks(Yeungetal., 2002), whichare criticalforcreating asafe andcognitively rich learningenvironment (KalilandZiol-Guest, 2008). Because braindevelopmentinearly childhood requires psychoso-cial stimulation and energy(Aboud andYousafzai, 2015), livingin an economically-depressed environment—which often accompaniesfamilyunemployment—cansignificantlyimpairchildren’spsychologicaldevelopmentoverthelifecourse.6 _For

instance,children who receivea generous amount ofattentionandverbal andcognitive stimulation by sensitive and re-sponsivecaregiverstendtoberelativelymoreadvancedinallrealmsofdevelopment(LambandAhnert,2007;Phillipsand Lowenstein,2011).

Alack ofeconomic resources mayalso lead tofeelings ofenvy,jealousy,or perceptionsofunfairness aschildren en-gageinsocialcomparisonswiththeirmoreeconomically-fortunatepeers(ClarkandOswald,1996;Ferrer-i-Carbonell,2005; Clarketal.,2008; deBotton, 2008).Suchfeelingsofrelativedeprivation canarisebecausechildren feelthatthey lack re-sourcesandqualitiesthataredeemedsociallydesirable(Easterlin,1995;Festinger,1954;Frank,2005).Forinstance,previous studiesdocumentthatchildrenofdisplacedworkerstendtoreportlowerself-esteem,experiencebehavioralproblems,and aremorelikely todropout orbeexpelledfromcollege(Johnson etal., 2012;StevensandSchaller,2011).Giventhat low socio-economicstatuscorrelateswithpoorphysicalhealth(MatthewsandGallo, 2011),enduring systematicinflammatory responseinearly childhoodcansignificantly increasethe riskofpsychopathologyinlaterlife, includingsymptomsof de-pressionandpsychosis(DaneseandBaldwin,2017;Khandakeretal.,2014).

Furthermore,socialstigmaassociatedwithparentalunemployment,lowsocio-economicstatus,orexcessiverelianceon publicassistancecandecreasesocialinteractionswithpeers,teachers,orparents(Rubinetal.,2009;Schneideretal.,2000). Inturn,moresocially-withdrawnchildrenareatgreaterrisk ofexperiencingsocio-emotionaldifficultieswhilegrowingup, includingloneliness,anxiety,depression,lowself-esteem,andself-blame(Rubinetal.,2009).Inadditiontothese internaliz-ingproblems,childhoodwithdrawalmayalsoleadtopeerdifficultiesincludingrejection,victimization,andpoorfriendship quality(Ohetal., 2008;Rubinetal.,2002; Rubinetal.,1990).Thesenegativeoutcomescanhavelong-termconsequences andpersistintoadolescenceandadulthood(AsendorpfandDenissen,2006;Rubinetal.,2009).

Thesecond theoreticalperspectiveisbasedontheviewthat“familystress” duetounemploymentcanspilloverto chil-drenandsignificantlyimpairtheirpsychologicaldevelopmentoverthelifecourse(e.g.ElderandConger,1995).A substan-tialbody ofliterature in thesocial sciencessupports thisview by showing that involuntaryunemploymentis associated withadeclineinphysicalandmentalhealth(e.g.EliasonandStorrie,2009;GerdthamandJohannesson,2003;Kuhnetal., 2009;Marcus,2013;SullivanandVonWachter,2007;Wanberg,2012),7_{chronicstress(Baumetal.,1986),andlowerlevels}

ofsubjectivewell-being(Clark andOswald,1994;GerlachandStephan, 1996;KassenboehmerandHaisken-DeNew,2009; WinkelmannandWinkelmann,1995;1998).Empiricalstudiesineconomicsevensuggestthatjoblessnessdepressesmental healthandlowerslifesatisfactionmorethananyothersinglefactorattheindividuallevel(PowdthaveeandVernoit,2013). Withinthisperspective,increasedfamilystressduetojoblessnesscandisruptthestructureoffamilyrelationsandthus inhibitchilddevelopment.Unemploymentnotonlyhasspillovereffectsatthecouplelevel(NikolovaandAyhan,2018),but itcanalsoinhibithealthyfamilyfunctioningandincreasethelikelihoodofdivorce(e.g.CharlesandStephens,2004;Doiron andMendolia,2012;Hansen,2005;Regeetal.,2007).Inturn,maritalconflictcannegativelyaffectchildren’sdevelopment andpermanentlyscarthem(e.g.CummingsandDavies,2011;Gruber,2004).

Finally,joblessness hasbeen previously linked to an increasedprevalence of risky behaviorssuch assmoking, heavy drinking,andtheuseofpsycho-activedrugs(seeHenkel,2011forareview),whichcanalsoaffectparenting.Psychological stressassociatedwithunemploymentcanalsoinhibittheemotionalwarmthofparents andlead tomoredisengagedand punitiveparentingpractices (Kessler etal.,1989). Inturn,ineffectiveparentingcan leadto poordevelopmentaloutcomes andadjustmentinchildrenoverthelifecourse(PhillipsandLowenstein,2011).

Weviewthesetwo perspectivesascomplementaryratherthanasmutuallyexclusive.Thus,parentaljoblosscanaffect children’sfuture psychological well-being,partly through limitingaccess toessential resources (suchas educationanda safeandcognitively-enhancingenvironment)andpartlyby increasingchildhoodexposuretostress, whichcanbe interde-pendent.Inturn,higherexposuretostressinchildhoodcan causeenduringsystematicinflammatory response,whichcan then impair cognitive andemotional developmentof key behavioraldomains (Daneseand Baldwin, 2017). Consequently, these“hiddenwounds” canlead todevelopmentalcascadesthrough childhoodandadolescenceandsubstantiallyincrease theriskofpsychopathologyinlaterlife.

6 Consistent with the investment perspective, an emerging body of literature in economics also shows that parental job loss affects children’s long-

term outcomes such as college attainment, earnings, and own unemployment ( Hilger, 2016; Oreopoulos et al., 2008; Pan and Ost, 2014 ). For example,

Pinger (2016) demonstrates that parental unemployment decreases German adolescents’ probability of completing upper secondary education and that the majority of the effect is attributable to the decreased resources, cognitive performance, and academic confidence. Oreopoulos et al. (2008) find that Canadian children of displaced fathers earn about 9% less compared to similar children whose fathers did not become displaced. Pan and Ost (2014) show that parental job loss between the ages of 15–17 reduces college enrollment by 10 percentage points. Studies emphasize the deleterious effects of the father’s joblessness, while maternal unemployment is often found to be much less detrimental educational outcomes later in life ( Rege et al., 2011; Stevens and Schaller, 2011 ). Note that Mäder et al. (2015) show that while there is a correlation between parental unemployment during childhood and the adult child’s unemployment status, the relationship is not causal.

7 Some studies find no causal effects of unemployment on physical health ( _{Browning et al., 2006; Kuhn et al., 2009; Schmitz, 2011 ). However,}

Ofcourse,children’sexperiencesfollowingparentalunemploymentare likelytobe heterogeneous.Previoustheoretical andempiricalstudiessuggestthatunemploymentcanalsohaveapositiveeffectonchildren’spsychologicalwell-being(e.g., VandellandRamanan,1992;MooreandDriscoll,1997;Howardetal.,1999;PowdthaveeandVernoit,2013).Several mecha-nismscanexplainthispotentialpositiveeffect.First,parentsfacevariousconstraintsandtrade-offswhenmaximizingtheir utilityfunction.Ontheonehand,timespentworkingcanincreasefamilyincomewhichcanpositivelyaffectparents’ability toinvestinessentialresourcessuchashigh-qualityeducation,health,orhousingthatcancreateasafeandcognitively-rich learningenvironment(KalilandZiol-Guest,2008;CouchandPlaczek,2010;DavisandWachter,2011).Ontheother hand, time spentearninganincomemeanslesstimespent withchildren. Becausechildcareisoftenviewedasamajorinputin the productionfunction of humancapital development, especiallyat an early age(BeckerandTomes, 1986; Powdthavee andVernoit,2013;Ermisch andFrancesconi, 2013), aninsufficientparentaltime investmentcanhavea negativeeffecton thechildren’scognitiveandpsychologicaldevelopmentinlater life.Thus,onepossiblemechanismthroughwhich jobless-ness canpositively influencechildren’spsychological well-beingcould be that unemployedparents allocatemoretime to childcare.Insupport ofthis,Knabeetal.(2010) reportthat unemployedpeople inGermanyspendalmost twiceasmuch time inchildcareastheir employedcounterparts, whichisalsotheonlyactivitythatthe joblessenjoymorecomparedto employees.However, thispositiveeffectcan dependonthe qualityofcare andthe timing ofunemployment,withyoung children likely benefitingmore fromparental unemploymentcompared toolder children. Previousresearch, forinstance, suggestthatthepositive(contemporaneous)effectofparentalunemploymentfoundatearlieragesiseithernullornegative forolderkids(PowdthaveeandVernoit,2013).

Second, studiesonresiliencesuggest thatchildren maydevelopcoping strategiesandsupport structuresasa resultof previousexperienceofadversity,whichcanallowthemtominimizethenegativeeffectsofstressfuleventssuchasparental unemployment(Howardetal., 1999;Masten andNarayan, 2012). Onthe onehand,children of unemployedparents may becomemore motivatedto stayinschoollonger anddowell toavoidtheir parents’misfortunelaterinlife. Ontheother hand,suchchildren maydevelop pessimisticattitudestowards theirownchancesofsuccessbasedontheir parents’labor marketexperience,whichcanlowertheirmotivationandleadtodisengagementfromschoolorwork(Barlingetal.,1998). Thisrelationship, however,is morelikelyto manifestitself ifparentalunemployment occursamongolderchildren when theydevelopindependentandabstractthinking(GauvainandCole,2004;Piaget,1971;Vygotsky,2009).

Finally, both ofthesemechanisms can depend on theunemployed parent’s gender, whereby maternalunemployment maybe positively associatedwiththeadultchild’s life satisfaction,while paternalunemploymentmayhavethe opposite effect.Accordingtothegenderidentityhypothesis(AkerlofandKranton,2000),peoplederive“identityutility” from adher-ing tothe socialnorms relatedtoone’s gender. Forexample,across cultures,menare typicallyviewed asmore assertive andfocused on material success(Hofstede,2001).Forret etal.(2010) find that menwithchildren are farmore likelyto perceive unemploymentasa defeatduetosocietalexpectationsthat they shouldbe providingfortheirfamilies. Women, ontheother hand,tendtohaveidentitiesasmothersandaswives. Forexample,women aremorelikelytoaccept lower payingjobsinordertoachieveabetter work-familybalance(Valcour etal., 2007).Specifically,women withchildren may feelgreaterresponsibilityfortheemotionalwell-beingoftheirfamily(GreenhausandFoley,2007)andmayprefertowork part-time (Booth andvanOurs, 2008). Because majority ofchildcare responsibilities are carriedby women regardlessof labor forcestatus(HochschildandMachung,2003),women mayseeunemploymentasanopportunity tofulfilladditional family-relateddutiesandinvestingintheirchildren’sdevelopment.Thus,uponbecomingunemployed,men– traditionally viewedas“breadwinners” – may avoidhouseworkandspending time withthe children, whileunemployedwomen may substitute market work withhousehold work (Groganand Koka, 2013), includingincreasing parenting time, which may havelong-term well-beingbenefits.8 _{Theseeffects maybe furtheraffected by thechild’s gender. Girls, forexample, may}

be moreeffectiveincopingwithadverse lifesituations inearlylife becausethey havehigherlevelsofinhibitorycontrol, emotionalunderstanding,prosocialandinternalizingbehaviorthanboys(Matthewsetal.,2009).

4. Empiricalstrategy

For each child i belonging to family j, the effect ofparental unemployment on life satisfaction later in life Y can be modeledas(e.g.,Francesconietal.,2010;LangandZagorsky,2001):

Yi j =

β

λ

α

ε

whereUP_{isabinaryindicatorforparentalunemployment(atages0–5,6–10,11–15),whichtakesthevalueof1ifatleast} one ofthe parents experienced parental unemploymentdueto firm closureand0 ifboththe motherandfather remain in theinitial labor marketstate. In some specifications,we separately examinepaternal andmaternaljoblessness dueto companyclosings.Xisavectorofthechild’ssocio-demographicandhouseholdcontrols,includingthefather’sandmother’s characteristics,asexplainedbelow.Theerrortermcomprisesbothafamily-specificpart

α

ε

8 Given these traditional social roles regarding child rearing and breadwinning, it is not surprising that some previous studies on the topic have fo-

cused primarily on estimating the effects of maternal unemployment on children’s educational and cognitive outcomes (e.g., Bernal, 2008; Ermisch and Francesconi, 2013 ).

Ourempirical strategyrelies oncomparing children withfathers andmothers workingasprivate employees (or non-workingmothers),someofwhomexperienceanexogenousunemploymentshockduringchildhood(treatmentgroup),while othersremainintheinitiallabormarket condition(comparisongroup).Tomakethetreatmentandcomparisongroupsas similaraspossible,weensurethat allparentsinitiallyhavethesamelabormarketstatus,wecontrolforparentaljobloss expectations,andanumberofadultchild-level,parent-levelandfamily-levelcharacteristics,whichshouldfurthermitigate selectionissues.

Given that parental unemployment occurred in childhood, while life satisfaction is measured in adulthood, reverse causalityisnot a concern.The main identificationthreat is that theeffectof parentalunemploymenton life satisfaction later inlife may be biaseddueto an unobservable orunmeasurable causal factor(i.e., theerror termiscorrelated with parentalunemployment).Forexample,parentalmentalillnessorriskybehaviorssuchasdruguseduringchildhoodmaybe associatedwithparentalemployment status,thefamilyenvironment,parentalinvestments inthe children,andchildren’s later-in-lifewell-being outcomes. Conversely, parents who are likelyto hold onto their jobs mayalsobe more likely to takegood careoftheir children(Ermisch etal.,2004).Furthermore,asdiscussedabove, ifunemploymentis voluntaryor expected,parents, andespeciallymothers,mayview joblessnessasan opportunityto increasetheir parentaltime invest-ments.

Weidentify Eq.(1)using unemploymententrydue toplant closings,which not onlyprovides exogenous variation in unemploymentthat is uncorrelated withunobservable variables atthe individual andfamily level,but isalso a reason-ableproxy forinvoluntaryunemployment(Chadi, 2010). Thus,our mainassumption isthat companyclosureshappen in-voluntarilyandindependent offamilycircumstances andhouseholdcharacteristicsthat simultaneouslyinfluence parental unemploymententryandthechild’slifesatisfactionlaterinlife.

Nevertheless,evenifparentscannotinfluencecompanyclosures,theymayexpectthem,andthusdecidetostayorlook foranotherjob.Companyclosuresrarely happenovernightandthus parentswithspecific characteristicsmightchoose to stay untilthe firm goes out of business,rather than findnew employment, likelybecause havehad goodreasons to do so(Kassenboemer andHaisken De-New, 2009).For example,women orworkers withhightenure andhighfirm-specific knowledgemaybelessflexibleinchangingtheirjobs(KassenboemerandHaiskenDe-New,2009).Inthiscase,anynegative effectsofparentalunemploymentonsatisfactionlaterinlifethatweidentifymaybeoverestimated.

Similarlyto Nikolova andAyhan(2018) andMarcus (2013), we addressthis concernby controlling forthe lagged ex-pectationsof bothparents to losetheir job inthe nexttwo years.We note that we relyon observationaldata, andthus wecannot eliminateallsources ofendogeneity.Yet,thecombinationofrelyingonplantclosuresandconditioningonthe (lagged)joblossexpectations,parentalandfamilycharacteristics,should inprinciple allowustorecoverthecausaleffect ofparentalunemploymentonlifesatisfactionlaterinlifeundertheassumptionmentionedabove.Thus,ourestimatescan beinterpretedasthelocalaveragetreatmenteffect(LATE).

Giventhatlifesatisfactionisanordinalvariable,weshouldtechnicallyestimateEq.(1)usinganordinallogitorprobit re-gression.Sinceignoringtheordinalityofthedataholdslittleconsequencefortheendresult(Ferrer-i-CarbonellandFrijters, 2004; Frijters andBeatton, 2012), we followthe commonpractice inthe literature to estimate the subjectivewell-being regressionsusingpooledordinaryleastsquares(OLS)andGeneralizedLeastSquares(GLS)randomeffectsestimators.9 _Our

preferredspecificationsarethemodels estimatedusingrandomeffects,whichnetouttheinfluenceofunobservablesthat areuncorrelatedwiththeexplanatoryvariables.Thisdecisionisalsosupportedbythe Breusch–PaganLagrangemultiplier test,whichfavorstheuseofarandomeffectsestimatoroverOLS.Sinceourkeyindependentvariables– parental unemploy-mentandmostofourcontrols– suchasgender, birthorder,migrationbackground,localunemploymentconditions– are exogenous,thisassumptionmaynotbeunreasonable.Allspecificationsuserobuststandarderrors,clusteredatthelevelof theadultchild.Wealsoofferspecificationswhichcontrolforparentalandthechild’spersonalitytraitsandothersensitivity tests,whicharedescribedinSection7.

5. Data,analysissample,andvariables

Cohortstudies are arguablyoptimalto studylife-cyclewell-beingasthey provide longtime spans,a large numberof observations,andreal-timemeasurementofkeysocio-demographicandfamilyvariablesratherthanretrospectiveaccounts (Clark,2014).GivenalackofGermancohortdata,weutilizeinformationfromtheGermanSOEP(Version32.1).Since1984, theSOEP is a representative panel ofall householdmembers aged 17andolder (Wagner etal., 2007) andcontains rich longitudinalinformation on well-beingand labor market characteristics,income,household composition andfinances, as wellaspersonalbiography.10_{Respondentsarefollowediftheymovetootherhouseholds,whichallowsustoincludeinour}

analysesadultchildrenwholeavetheparentalhome.

9 Because we have multiple measurements of subjective well-being later in life (at ages 18–31), ideally, we would have liked to use fixed effects (within-

child) estimators to net out time-invariant unobservables but cannot as most variables in our analyses are time-invariant.

10 We do not use the mother and child questionnaires in the SOEP as they were only introduced in 2003 and the oldest children are now only 14 years

SeveralfeaturesmaketheSOEPaparticularlyopportunedatasourceforouranalyticalpurposes.11_{First,akeyadvantage}

isthatsince1985,theSOEPhasaquestiononthereasonsforjobterminationandsince1991,theanswershaveincludeda categoryfor“placeofworkclosed,” (KassenboehmerandHaiskenDeNew,2009),whichallowsusto createtheexogenous unemployment entrytreatment variable.12 _{Second, because the panel has followed parents and their children for about}

quarterofacentury,we caninvestigatethelong-termwell-beingconsequences ofparentalunemploymentthathappened overthe entirechildhoodstage.The longitudinalinformationalsolets usmeasurehouseholdvariablessuch ashousehold incomeovertheentirechildhood,whichmitigatesissuesassociatedwithtransitoryshocksandreportingormeasurement error(Siedler, 2011).AsinSiedler(2011),conditioningonhouseholdincomeallows ustobetter distinguishthelong-term consequencesofparentalunemploymentfromanyconsequencesofgrowing upinpoverty.Finally,all information regard-ing the parents comes from their own contemporaneous responses rather than from the children retrospectively, which minimizesreportingerrors(Siedler,2011).

To studyhow parental unemploymentduring childhood influences the life satisfaction of young adults,we first link mothers andfathersusing thepartner identifierin SOEP andthenmatchchildren to both oftheir biological oradoptive parentsusingthemotherandfatheridentifiervariables.Mothersandfatherscanbeco-habitingorformallymarried.Atthis step,weensure thatchildren andtheir parentsinouranalysissampleareall SOEPrespondents.Wealsorequirethatthe young adultsinour sample(i) are bornin1984 orlater; (ii) haveparentallabor market informationover thechildhood periods (0–5, 6–10, 11–15years); (iii) andhave lived withboth of their adoptive or biological parents over their entire childhood(from0to15years).

Thefirsttworestrictionsensurethatinformationontheparents(age,education,income,labormarketstatus)and fam-ilycircumstances can beobserved overthe entirechildhood.All parentalinformationcomes fromthemotherandfather themselvesrather thanfromthechildren’sreports. Sincewe are measuringa spillovereffect,thethird conditionensures thatchildrenandparentslivedinthesamehouseholdduringchildhood,sothataspillovercanoccur. Adaptingdefinitions fromMarcus(2013)andNikolovaandAyhan(2018),ourtreatmentgroupcompriseschildrenwhoseparentslosetheir jobs duetoplantclosuresandareregistered unemployedattheGermanEmploymentOffice(Arbeitsamt).Wedefine three sep-aratetreatment groups, namely children whoseparents experienced plantclosings whenthe children were (i) 0–5years old;(ii)6–10 yearsold and(iii)11–15years old.Inall threecases,the comparisongroupcomprises parents whoremain continuously employed (ormothers continuously out ofthe workforce) over the respectivechildhood period (0–5,6–10, 11–15years).In allthree cases,thetreatment groupcomprises ofparentswho switch fromtheinitial labormarket state (fatherswhoinitiallyworkasprivateemployeesandmotherseitherworkasprivateemployeesorareoutoftheworkforce) tobeingjoblessduetoplantclosure.13_{Eitherparentcanexperienceunemploymentinthetreatmentgroup.Parentsinthe}

treatment groupcan inprinciple experienceunemploymentduetoplantclosures morethanonce, thoughthisonly hap-pensina handfulofcases.In separatespecifications,we alsosplittheparentalunemploymentvariableinto paternaland maternalunemployment.Ourchoiceforthechildren’sagesduringtreatmentfollowsthedevelopmentpsychologyliterature, accordingtowhichchildren’scognitiveandemotionaldevelopmentoccursinstages(GauvainandCole,2004;Piaget,1971; Vygotsky,2009).Becauseyoungerchildrencandoverylittlefororbythemselves,theyrelyexclusivelyonotherpeoplefor socialinteraction(GauvainandCole,2004).Asthey growolder,however,theybecomeincreasinglymoreindependentand self-sufficient.Forinstance,accordingtoPiaget(1971),childrenenteraformaloperationalstagewhentheybecome11years old,whichismarkedbytheabilitytothinkabstractlyandmanipulateideasandhypotheticalsituationsnotyetexperienced. Theseyearsinthechild’slifealsotendtobemuchmoreimpressionableintermsofpsychologicaldevelopmentandpossibly havestrongerwell-beingeffectsinlaterlife.

Ourmaindependentvariableisthegrown-upchild’slifesatisfaction,measuredatages18–31onascaleof0(completely dissatisfied) to10(completely satisfied).Sincewe restrict thesampleto thosebornin1984orlater, adultchildren’sages rangefrom 18to 31.We also includea batteryofindividual, paternal,andfamily-level conditioning variables,which are basedonthecontrolvariablesincludedinsimilarstudies(e.g.,Francesconietal.,2010;Siedler,2011).Inall specifications, followingGuilianoandSpillimbergo(2013),we controlfortheannualstateunemploymentratefacedasachild,averaged overtheyears0–5,6–10,and11–15respectively,tonetout thelong-termlife satisfactionconsequencesofgrowingupin a recession.14 _{Wealso controlforyearof birthdummies, whichcapturecommonly-shared experiences such aseconomic}

shocksandtechnologicalprogress,aswellascohort-specificunobservablecharacteristics(GuilianoandSpillimbergo,2013).

11 Ours is not the first study to utilize the SOEP when studying childhood circumstances and later-in-life outcomes. For example, _{Siedler (2011) uses the}

SOEP to investigate how parental unemployment during childhood affects young adults’ right-wing attitudes. Moreover, Francesconi et al. (2010) explore the relationship between childhood family structure and educational outcomes later in life using the SOEP.

12 The answer related to company closure was excluded from the SOEP questionnaires in 1999 and 20 0 0.

13 The treatment group does not change if we also allow fathers to be out of the workforce or working in the period before they experience unemploy-

ment. The “private employee” category excludes pensioners, civil servants, the self-employed and those in military/community service and education. To capture the working-age population, we restrict parental ages to range between 18 and 67. Unemployed persons due to all other reasons other than plant closings are excluded from the treatment group and also from the comparison group.

14 The historical state unemployment data are from the German Employment Agency, Bundesagentur für Arbeit, Arbeitslosigkeit im Zeitverlauf 2016. The

data before 1991 are for West Germany only and respondents with missing state information are assigned the national average unemployment rate during their respective childhood ages (0–5, 6–10, and 11–12). Research shows that macroeconomic conditions experienced in childhood influence political values and trust later in life ( Giuliano and Spilimbergo, 2013; Hörl et al., 2016 ). In addition, Bertoni (2015) demonstrates that having experienced hunger as a child affects adult life satisfaction.

We furtherinclude agedummies forthe grown-upchild, the adultchild’s educational attainment, whetherthe adult childstilllivesintheoriginalhousehold,gender,thechild’smigrationbackground,stateofbirth,whetherthechildisborn inEast/WestGermany,mother’sandfather’seducation,mother’sandfather’sageatwhichthechildwasborn,mother’sand father’smigrationbackground,SOEPsampleindicator,15_{surveyyeardummies,thesizeofthelocalitywherethechildgrew}

up,state of residence,and locationin East/WestGermany.We alsocontrol forbirth ordersince previous studiessuggest thatfirst-bornchildrentendtohavegreatereconomicsuccessthantheir siblingsinlaterlife (seeBlack,2017forareview oftheevidence).

Additionalcontrolsincludehouseholdincomequintileattherespectiveagesduringchildhood(0–5,6–10,11–15),16

aver-agehouseholdsizeattherespectiveagesduringchildhood,thenumberofsiblingsaveragedovertherespectiveagesduring childhood,theaveragesizeofthehomeduringtherespectiveagesduringchildhood,thecumulativeunemployment experi-enceofthemotherandthefatherduringtherespectiveagesduringchildhood,andeachparent’scumulativeunemployment experienceduration(inyearsandmonths)duringtherespectiveages.17_{AsinSiedler(2011),otherthanage,education,}

res-idenceinthe originalhousehold, andstate, wedo notcontrol foranycontemporaneouslife satisfactiondeterminantsfor thegrown-upchild such asadult income,adultmaritalstatus, adultunemploymentstatusand othersasthese variables aredeterminedaftertheparentalunemploymentandassucharepotentialoutcomesoftheparentalunemployment.These “bad controls” introduceselection bias andassuch should beavoided asargued inAngrist andPischke (2009),andalso inBertoni (2015) in thecontext ofthe life satisfactionliterature.18 _{FollowingSiedler (2011),we include theadult child’s}

educational attainment inthe main specifications,but neverthelessreport results withoutthis control in Table2, Model (2).Furthermore,whilechildhoodincomeandparentalunemploymentarelikelyhighlycorrelated,asinSiedler(2011),we includethiscontrolbecause,fromapolicyperspective,itisimportanttodistinguishtheeffectsofparentaljoblessnessfrom thoseofgrowingupinpoverty.

Finally,toavoidbiasfromdroppingobservationsduetomissingdata,forallvariablesincludedintheanalyses,wecreate anadditionalmissingdataindicator.Continuousvariablesarethussplitintodummyvariablesbeforetheadditionalmissing dataindicator iscreated.This“noinformation” categoryhasnointerpretationbutmerelyhelpstopreserve thenumberof observations.SummarystatisticsfortheanalysissamplesareshowninAppendixTablesA1–A3.

6. Results

6.1. Mainresults

Table1presentstheresultsfromourbaselinespecification.Inseparatemodels,weregresslifesatisfactionatages18–31 onparentalunemploymentduetoplantclosureduringtheagesof0–5,6–10,11–15.19_{Eachrowinthetablecontainsthe}

resultsofaseparateregression,dependingonthetimingofparentalunemployment(atthechild’sages0–5,6–10,11–15). Ourpreferredspecificationisthe one relying onthe GLSrandom effectsestimator (Models(3)−6)).Forcomparison pur-poses,wealsoshowspecificationsusingpooledOLS((1)–(3)).Forbothestimators,wereport thefindings bysubsequently addingcontrols,withmodels(3)and(6)beingthemostconservativeastheycontain(lagged)parentaljoblossexpectations controls. Allspecificationscontrol forconfounderssuch asgrowing up ina recession,characteristics ofthe child,mother andfather,andModels(3)–(4)and(5)–(6)arealsoindependentofchildhoodincomeandlivingconditions.

Table1demonstratesthatparentaljoblossduringearlychildhood(0–5ages)andearlyadolescence(11–15)negatively affectsadultlifesatisfaction,butthatduringmiddlechildhood(6–10)doesnotseemtomatter.Incolumn(6),ourpreferred specification, the coefficient estimates for parental joblessness at ages 0–5 and 11–15 remain negative and (marginally) statisticallysignificantevenifwecontrolforhouseholdincomequintileduringchildhood,thechild’seducationalattainment, andanumberofother householdcharacteristicssuch ashomesize, householdsize, thenumberofsiblings,andparental job loss expectations. Specifically, parental joblessness experienced atages 0–5 lowers young adults’ life satisfaction by about0.59pointsonaverage, whichisabout8% of thesamplemean.Similarly, thelong-term scarringeffect ofparental unemploymentexperiencedatages11–15isonaverageabout0.34points(ona0–10)scale,whichrepresentsabout5%of thesamplemean.Indeed,theseeffectsmayappearsmallinmagnitudebutareeconomicallymeaningful,consideringthat theyarelong-termeffects,andinsomecases,arefeltseveraldecadesafterthenegativeexperience.However,thedifference betweentheestimatesatages0–5and11–15isnotstatisticallysignificant(p-value₌0.79).20

15 Since its inception in 1984, the SOEP has added several different samples, such as that of East Germany in 1990, a migrant sample, and several

refreshments.

16 Post-government household income is in 2011 real terms.

17 This latter variable is provided by the SOEP in the person-generated data files and it reflects the total length of unemployment based on unemployment

experiences from the calendar dataset, which contains monthly information since the respondent entered the SOEP and biographical information from the biography questionnaires (SOEP Group, 2017). As such, this variable is subject to misreporting and recall bias ( Akerlof and Yellen, 1985; Jürges, 2007 ) and likely under-reports unemployment. Nevertheless, we include it in the analysis as it captures the overall parental unemployment “climate” when the respondent was growing up.

18 Results using these “bad” controls, which are not systematically different from our main estimates, are also available in _{Table A4 in the appendix.} 19 The explained variation in our regressions is relatively low, in the range of a few percent, which is not uncommon in life-cycle subjective well-being

regressions ( Clark, 2014 ).

Table 1

Baseline results, parental unemployment during childhood and life satisfaction later in life.

DV: Life satisfaction of young adults (1) (2) (3) (4) (5) (6) OLS OLS additional

covariates

OLS add. cov. & parental job loss expectations

RE RE additional covariates

RE add. cov. & parental job loss expectations Parental unemployment Mean life sat. 18–31 N. unemp. parents N β R 2 _{β R} 2 _{β R} 2 _{β R} 2 _{β R} 2 _{β R} 2 Child’s age 0–5 7.479 66 3,168 −0.709 ∗ _{0.075 −0.560 0.096 −0.548 0.109 −0.687} ∗ _{0.066 −0.571 0.084 −0.590} ∗ _0.098 (0.367) (0.373) (0.367) (0.366) (0.362) (0.345) Child’s age 6–10 7.386 149 5,168 0.116 0.062 0.267 0.093 0.321 0.098 0.101 0.056 0.251 0.085 0.279 0.087 (0.221) (0.232) (0.236) (0.239) (0.242) (0.241) Child’s age 11–15 7.365 122 6,927 −0.323 ∗ _{0.058 −0.261 0.084 −0.183 0.093 −0.530} ∗∗∗_{0.050 −0.399} ∗∗ _{0.074 −0.344} ∗ _0.082 (0.174) (0.177) (0.183) (0.196) (0.189) (0.187) Source: Authors’ calculations based on SOEP, V 32.1.

Notes: RE = Random effects. Robust standard errors in parentheses, clustered at the adult child’s level. The table shows the coefficient estimates of three different regressions whereby the focal independent variable is parental unemployment due to plant closure at the respective ages. The focal independent variable is coded as 1 if either parent became unemployed due to plant closure and 0 if both parents remained continuously employed or if the mother remained outside the workforce during the respective ages. The dependent variable is the life satisfaction of the child measured using all non-missing observations between the ages of 18 and 31. Models (1)–(3) are estimated using OLS and Models (4)–(6) are estimated using GLS random effects. Models (1) and (4) include controls for the annual cumulative state unemployment rate at the respective ages, age dummies, year of birth, gender, child’s migration background, whether the child is first born or not, the state of birth, whether the child is born in the East/West of Germany, mother’s and father’s education, mother’s and father’s age at which the child was born, mother’s and father’s migration background, SOEP sample, survey year, state of residence and location in the East/West of Germany. Models (2)–(3) and (5)–(6) include additional controls for household income quintile at the respective ages during childhood, the adult child’s educational attainment, whether the adult child still lives in the original household, the size of the locality where the child grew up, household size at the respective ages during childhood, the number of siblings during the respective ages during childhood, the size of the home during the respective ages during childhood, the unemployment experience of the mother and the father during the respective ages during childhood, and each parent’s current unemployment experience duration. Finally, models (3) and (6) add controls for each parent’s (lagged) job loss expectations.

∗∗∗_{p < 0.01.} ∗∗_{p < 0.05.} ∗_{p < 0.1}

The negative consequences of parental unemploymentfor young children could be due to the long-term psychologi-cal trauma dueto stressful experiences atan early age. Specifically,the theory of cumulative disadvantage suggeststhat iftwochildren experiencethesamenegativeshock,thechildthat experiencesit ata youngeragewillberelativelymore impactedcomparedtothechildexperiencingitatanolderage(Arbeit,2013;DiPreteandEirich,2006).Because disadvan-tagesover the life courseaccumulate overtime, experiencing setbacksearly in life stronglyinfluences outcomesin later life (Arbeit, 2013). Eventhough thedifference betweenthecoefficient estimatesatages0–5and11–15is statistically in-significant,ourresultspresentsuggestiveevidenceconsistentwiththisperspectiveasthenegativeconsequencesofparental unemploymentforlifesatisfactionlaterinlifeappeartobelargerforthosewhowere0–5thanforthosewhowere11–15 when the unemploymentepisodeoccurred. They are alsosimilar to the finding inErmisch et al.(2004) who show that children who experiencedparental joblessnessatan earlierage(0–5 years) ismorelikely tolead toa lower educational attainment,higherchanceofeconomicinactivity,greaterdistress,andsmokinglater inlife,comparedtochildrenwho ex-periencedparentalunemploymentatlaterages.Suchearlydistressinachild’slifelikelyleadtodisruptionsinthecognitive, social,andfamilyenvironment.Inaddition,afurtherexplanationforthenegativeconsequencesassociatedwith experienc-ing parentaljobloss atages11–15isthat olderchildrenmayfeel morepressuretotake responsibilityin thehome after their parents are no longerworking(Arbeit, 2013).21 _{Furthermore,atolderages children start developingabstract}

think-ing(i.e.reasoningintermsofhypotheticalsituations), whichcanmakethemmoreawareofsocial stigmaassociatedwith parentalunemploymentortheirdisadvantagedsocio-economicstatus(GauvainandCole,2004;Piaget,1971).

6.2. Channelsandheterogeneityanalyses

InSection3,wediscussanumberofchannelsandmechanismsthroughwhichparentalunemploymentexperienced at anearlyagecanaffectadultlifesatisfaction.Inthissection,wetestempiricallytowhatextentthesechannelsaccountfor ourfindings. The firstsets oftestsfor ourmechanismsare presentedin Table1,whereby inModels (2)–(3) and(5)–(6), respectively,we introduceadditionalcontrol variablesto thebaselinespecifications (1)and(4).Wewill focusthe discus-sionontherandomeffectsspecificationsinModels(3)–(6).AddingadditionalcovariatesinModel(5)– householdincome quintilewhengrowingup,theadultchild’seducationalattainment,andhouseholdcharacteristics– reducesthemagnitudes ofthe coefficientestimatescompared tothosein Model(4).Forexample,thecoefficient estimateon parental unemploy-mentatages11–15fallsfrom−0.53inModel(4)to−0.40inModel(5).Therefore,householdcharacteristicsappeartobe onepossiblechannelthroughwhichunemploymentreducestheadultchild’slifesatisfaction.Inaddition,inModel(6),we

21 The result that parental unemployment at ages 11–15 is negatively associated with life satisfaction later in life is in line with findings in

Table 2 Channels.

(1) (2) (3)

No childhood income controls No adult education outcomes With controls for mother’s & father’s working hours

Child’s age 0–5 −0.588 ∗ _−0.584 ∗ _−0.598 ∗ (0.347) (0.343) (0.347) N 3,168 3,168 3,168 R 2 _{0.097 0.098 0.0978} Child’s age 6–10 0.285 0.274 0.276 (0.236) (0.242) (0.240) N 5,168 5,168 5,168 R 2 _{0.089 0.088 0.0887} Child’s age 11–15 −0.337 ∗ _−0.347 ∗ _−0.359 ∗ (0.185) (0.186) (0.186) N 6,927 6,927 6,927 R 2 _{0.081 0.079 0.084}

Source: Authors’ calculations based on SOEP, V 32.1.

Notes: RE = Random effects. Robust standard errors in parentheses, clustered at the adult child’s level. Each column shows the coefficient estimates of three different regressions whereby the focal independent variables are maternal and paternal unemployment due to plant closure at the respective ages. The focal independent variables are coded as 1 if the mother (father) became unemployed due to plant closure and 0 if both parents remained continuously employed or if the mother remained outside the workforce during the respective childhood ages. The dependent variable is the life satisfaction of the child measured using all non-missing observations between the ages of 18 and 31. All models are estimated using GLS random effects. All models include the full set of covariates listed in Table 1 , including controls for parental job loss expectations. Model (1) excludes childhood income controls during the respective ages. Model (2) excludes controls for the adult child’s educational attainment. Model (3) includes controls for mother’s and father’s working hours during the respective ages.

∗∗∗_{p < 0.01,} ∗∗_{p < 0.05,} ∗_{p < 0.1.}

alsoconditiononeach parent’sjob loss expectations,whichshould furthermitigateunobserved heterogeneity issuesand enhance thecausal interpretationof ourresults. Thecoefficient estimates remainrelatively stablein thesespecifications, withthecoefficientestimatesonparentaljoblessnessatages0–5and11–15beingmarginallystatisticallysignificant.

Furthermore,inTable2,Model(1),webuildonspecification(6)fromTable1andremovethechildhoodincomecontrols. ThemotivationbehindthischeckisthatcontrollingforchildhoodincomeasinModels(5)and(6)couldbealready reflect-ingthechannelrelatedtofinancialresourcesandtheabilitytoinvestinthechildrendescribedinSection3.Thus,wewould expectthat removingtheincomecontrolwouldincrease theabsolutevalue ofthemagnitudeofthecoefficientestimates. Nevertheless,thefindingsinTable2,Model(1)remainverysimilartothoseinModel(6)inTable1.Inaddition,inModel (2)in Table 2, we also remove the adulteducational attainment to explore whetherthisis the channel linking parental unemploymentinchildhoodandadultlifedissatisfaction.Theintuitionisthatkidsfromhomeswhereparentswere unem-ployedmayreceivelesseducation,whichwouldlowertheiropportunitiesinlifeandthusleadtolifedissatisfactionlaterin life.TheresultsinModel(2)remaincomparabletothoseinModel(1)andtothebaselinespecificationpresentedinModel (6),Table 1.Finally,in Model (3)inTable 3, we include an additionalcontrol variable – theaverage numberofworking hoursforeachparentduringtherespectivechildhoodages.Iftheeffectofparentalunemploymentisdrivenbychangesin workingtimeortimespentathomewiththechildren, thenwewouldexpect thattheresultsinModel(3)todecreasein absolutevalueorbecomeinsignificant.22_{Whatwe findinsteadisthattheresultsdonotchange.Ofcourse,workinghours}

isacrudeproxyforthequalityoftheparentingtimeandfuturestudiesshouldfurtherinvestigatethischannel.

Finally,additionalfindings(availableuponrequest)demonstratethatparentaljoblossduringchildhoodisunassociated withtheadultchild’sjoblossinsecurity,suggestingthatthechannelmayindeedbethepsychologicaltrauma(scarring ef-fect)ratherthanjobinsecurity(scaringeffect).Allinall,thesechecksimplythatpsychologicalburdenofexperiencingsocial stigmaandtakingresponsibilityatayoungageappeartolastwellintoyoungadulthoodandareindependentofchildhood income,theparents’workinghourswhengrowingup,andtheadultchild’seducationalattainment.Theseconclusions cor-roboratethefindingsregardingthecontemporaneouseffectsofparentalunemploymentonchildren’slifesatisfactioninthe UK(PowdthaveeandVernoit,2013) andsuggestthattheeffectswe documentrepresentthelong-termpsychologicalcosts ofpastparentalunemployment.

InTable3,we expandourbaseline analyses toexplore ifthe effectswe reportdepend onwhetherthemotherorthe fatherbecamejoblessandwhetherboysorgirlsaredisproportionatelyaffectedbychildhoodadversitycausedby unemploy-ment.Allspecificationsrelyonrandom effectsestimatorswiththemostconservativelistofcovariates,includingparental joblossexpectations.Theseresultsshouldbeinterpretedassuggestiveevidenceasouralreadysmalltreatmentsamplesare furtherdecreasedbysplittingthemalongparentalandchildgender.

First,inColumns(1)and(2),wefindthatwhilepaternalandmaternalunemploymentatages6–10areunassociatedwith lifesatisfactionlaterinlife,maternalunemploymentatages0–5andpaternalunemploymentatages11–15areparticularly harmfultoyoungadults’lifesatisfaction.Specifically,maternalunemploymentatages0–5lowersadultlife satisfactionby about1.4points, whichrepresentsabout19% ofthesamplemeanof7.5.Paternal unemploymentatages 11–15decreases

Table 3

Maternal and paternal unemployment during childhood and life satisfaction later in life.

DV: Life satisfaction of young adults All Females Males

(1) (2) (3) (4)

OLS add. cov. & parental job loss exp.

RE add. cov. & parental job loss exp.

RE add. cov. & parental job loss exp.

RE add. cov. & parental job loss exp. Mom unemployed: child’s age 0–5 −1.338 ∗∗∗ _−1.416 ∗∗∗ _−0.983 ∗∗∗ _−2.714 ∗∗∗

(0.322) (0.332) (0.380) (0.483)

Dad unemployed: child’s age 0–5 0.478 0.321 −0.176 1.259 ∗∗

(0.383) (0.376) (0.484) (0.632)

I-test for equality of coeff. ( p -value) 0.0 0 0 0.0 0 0 0.130 0.0 0 0

N 3,168 3,168 1,498 1,670

R 2 _{0.114 0.103 0.203 0.165}

Mom unemployed: child’s age 6–10 0.380 0.335 0.775 ∗∗∗ _−0.451

(0.263) (0.262) (0.286) (0.525)

Dad unemployed: child’s age 6–10 0.031 0.124 0.115 0.256

(0.500) (0.537) (0.735) (0.691)

F-test for equality of coeff. ( p -value) 0.534 0.723 0.405 0.410

N 5,168 5,168 2,528 2,640

R 2 _{0.098 0.089 0.144 0.136}

Mom unemployed: child’s age 11–15 0.077 0.013 0.246 −0.601

(0.290) (0.282) (0.290) (0.532)

Dad unemployed: child’s age 11–15 −0.464 ∗∗ _−0.622 ∗∗∗ _{−0.213 −1.026} ∗∗∗

(0.211) (0.224) (0.313) (0.391)

F-test for equality of coeff. ( p -value) 0.149 0.096 0.320 0.517

N 6,927 6,927 3,336 3,591

R 2 _{0.094 0.086 0.145 0.122}

Source: Authors’ calculations based on SOEP, V 32.1.

Notes: RE = Random effects. Robust standard errors in parentheses, clustered at the adult child’s level. Each column shows the coefficient estimates of three different regressions whereby the focal independent variables are maternal and paternal unemployment due to plant closure at the respective ages. The focal independent variables are coded as 1 if the mother (father) became unemployed due to plant closure and 0 if both parents remained continuously employed or if the mother remained outside the workforce during the respective childhood ages. The dependent variable is the life satisfaction of the child measured using all non-missing observations between the ages of 18 and 31. Model (1) is using OLS and Models (2)–(4) are estimated using GLS random effects. All models include the covariates listed in Table 1 . N unemp. dads = 34 at ages 0–5, 22 at ages 6–10, 59 at ages 11–15. N unemp. moms = 32 at ages 0–5, 127 at ages 6–11, and 65 at ages 11–15. The table also shows the p -value of an F -test of equality of coefficients for mom unemployed and dad unemployed at the respective ages.

∗∗∗_{p < 0.01,} ∗∗_{p < 0.05,} ∗_{p < 0.1.}

adultlifesatisfactionby 0.6points,onaverageabout8%ofthesamplemean.Thus,boththetimingandwhichtheparent becomesunemployedmatterforlifesatisfactionofyoungadults.

Second,furthersplittingtheresultsbygender(Models(3)–(4))demonstratesthatmaternalunemploymentatages0–5 isharmfultothelife satisfactionofbothadultmenandwomen,buttheeffectisstrongeramongboys,being−2.7points ona scaleof0–10,whichissubstantial. Onepossibleexplanation forthisfinding canbe thatgirlsgenerallyshow higher levels of inhibitorycontrol, emotional understanding andexpression, more prosocial andinternalizing behavior, and less externalizingbehaviorsthanboysatanearlyage(Matthewsetal.,2009;Changetal.,2011).Inturn,betteremotional self-regulationenablegirlstomoreeffectivelycopewithadversityassociatedwithparentalunemploymentinearlylife.Thisis alsosupportedbyourfindingthatthepsychologicalcostsrelatedtopaternalunemploymentatages11–15areconcentrated amongboys,withtheaverageeffectbeing−1pointsonaverage.OneadditionalnotableresultthatemergesfromTable3is that parentalunemploymentcould alsohavepositive consequencesforthelifesatisfactionofadultchildren. Forexample, maternalunemploymentduringtheagesof6–10ispositivelyassociatedwithyoungwomen’slifesatisfaction,likelybecause ofthepositiveimpactoftheadditionalparentingtimethatunemployedmothersinvestintheirfemaleoffspring.Inaddition, paternaljoblessnessattheagesof0–5isbeneficialforthewell-beingofsons.Thiscould beduetothesystematicbiases parentsshowwhenforcedtochoosebetweenspendingonsonsanddaughters,withmothersfavoringdaughtersandfathers consistentlyfavoringsons(Nikiforidisetal.,2018).

Next,wefurtherexamineatwhatpointoftheadultagedistributionparentalunemploymentjoblessness during child-hoodismostlikelytomanifestitself(Table4).Becauseoftheprolongedtransitiontoadulthood,especiallyinthedeveloped world, whichscholarshavestarteddescribing asa separate developmentalstage (e.g.,seeArnett, 2007), we focusonthe effectofinvoluntaryparentalunemploymentonbothyoungadults(ages18–20and21–23) andemergingadults(ages24– 31). In Panel A,we show thefindings for theanalysis sampleoverall, whereby the samplecomposition is not thesame throughoutthespecifications.ThefindingsinPanelAsuggestthatparentalunemploymenthurtsmorewhenchildrengrow olderratherthan whenthey are youngadults.It appearsthat theresultsare mainlydriven by youngadultsaged 24–31, and,tosomeextent,youngadultsaged18–20. Thismaybebecausesocietalorownexpectationsforsuccessmayincrease withage,whichmayawakenthememoryaboutchildhoodexperiencesandincreasethepressuretoavoidanothereconomic misfortune.InPanel B,we repeatthe sameanalysisbutonlyforthe sameindividualswho weobserve atallthree adult ages– 18–20,21–23and24–31.TheseresultsaresubstantivelysimilartothefindingspresentedinPanelBandshowthat thepsychologicalcostsofchildhoodparentalunemploymentareconcentratedamonggrownupchildrenatages24–31.The

Table 4

Parental unemployment during childhood and life satisfaction later in life, by adult child’s age.

Ages 18–20 Ages 21–23 Ages 24 and older DV: Life satisfaction of young adults (1) (2) (3)

RE add. cov. & parental job loss expectations

add. cov. & parental job loss expectations

add. cov. & parental job loss expectations

Panel A: All observations in analysis sample, unbalanced panel

Child’s age 0–5 −0.527 −0.476 −1.203 ∗ (0.362) (0.530) (0.675) Overall N 1,712 930 526 R 2 _{0.117 0.199 0.295} Child’s age 6–10 0.181 0.515 0.514 (0.253) (0.334) (0.362) Overall N 2,701 1,477 990 R 2 _{0.131 0.107 0.205} Child’s age 11–15 −0.385 ∗ _{0.065 −0.728} ∗ (0.204) (0.270) (0.409) Overall N 3,622 1,973 1,332 R 2 _{0.107 0.111 0.198}

Ages 18–20 Ages 21–23 Ages 24 and older DV: Life satisfaction of young adults (1) (2) (3)

RE add. cov. & parental job loss expectations

RE add. cov. & parental job loss expectations

RE add. cov. & parental job loss expectations

Panel B: Only the same individuals

Child’s age 0–5 −0.602 −0.707 −1.077 (0.575) (0.757) (0.681) Overall N 681 673 618 Number of individuals 206 206 206 R 2 _{0.237 0.266 0.293} Child’s age 6–10 0.369 0.549 0.603 ∗ (0.410) (0.411) (0.357) Overall N 1171 1169 1094 Number of individuals 310 310 310 R 2 _{0.219 0.188 0.215} Child’s age 11–15 0.539 0.121 −0.768 ∗∗ (0.354) (0.306) (0.390) Overall N 1579 1562 1449 Number of individuals 425 425 425 R 2 _{0.175 0.197 0.203}

Source: Authors’ calculations based on SOEP, V 32.1.

Notes: RE = Random effects. Robust standard errors in parentheses, clustered at the adult child’s level. The table shows the coefficient estimates of three different regressions whereby the focal independent variables are maternal and paternal unemployment due to plant closure at the respective ages. Panel A comprises all individuals in the analysis sample broken down according to their adult ages. Panel B holds the sample composition constant and only shows the regressions for the same individuals who provided information at ages 18–20, 21–23, and 24 and older. Note that the number of individuals in Panel B is the same across the three models. The focal independent variables are coded as 1 if the mother (father) became unemployed due to plant closure and 0 if both parents remained continuously employed or if the mother remained outside the workforce during the respective ages. The dependent variable is the life satisfaction of the child measured using all non-missing observations between the ages of 18 and 31. All models are estimated using GLS random effects. All models include the covariates listed in Table 1 .

∗∗∗_{p < 0.01,} ∗∗_{p < 0.05,} ∗_{p < 0.1.}

findingsfrombothPanelsAandBimplythatparentalunemploymentexperienced atages11–15lowerstheyoungadult’s lifesatisfactionatages24–31byabout0.73–0.77points,whichconstitutesabout10%ofthesamplemean.

7. Robustnesschecksandlimitations

We test the robustnessofour results both directly andindirectly. First,our indirect checksinclude regressions using voluntaryparental unemploymentdueto own resignationora mutualagreement withthe employer.23 _{As expected, the}

regressionsusingvoluntaryparentalunemploymentarestatisticallyinsignificantthroughoutthespecifications(Column (1) inTable5).Bycontrast,thoserelyingonanotherinvoluntaryunemploymentmeasure– dismissals– showthesamepattern asthemainresultsusingplantclosures(Column(2)inTable5).Themaindifferencebetweendismissalsandplantclosures isthatthelatterprovidea moreexogenous sourceofvariationinparentalunemployment.Parentalunemploymentdueto dismissalsatages 11–15is associatedwitha0.33 declineinlife satisfactionatages 18–31,whichisalsoabout5% ofthe samplemean.ThefindingsinTable5highlighttheimportanceofdistinguishingbetweenvoluntaryandinvoluntaryparental unemploymentandsupportourmainfindingthatinvoluntaryandunexpectedunemploymentatages11–15resultsin long-termdeclinesinlifesatisfactionforchildrenwhohadtolivethroughit.