Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Do Family Values Shape the Effectiveness of Labour-to-Work Policies? An Analysis of a Recent Parental Leave Policy Reform in Germany Mireia Borrell-Porta
1
Dofamilyvaluesshapetheeffectivenessoflabour‐to‐workpolicies?Ananalysisof
arecentparentalleavepolicyreforminGermany
MireiaBorrell‐Porta1
Early draft – please do not circulate
Abstract
Theeffectivenessofpolicyreform isnot independentof familyvalues.To test thisargument,
thispaperexplorestheeffectofaparentalleavepolicyreforminGermany(2007)onthereturn
toworkofmotherswhoholddifferentfamilyvalues.Thepolicyincentivizedanearlierreturnto
work by reducing the paid parental leave subsidy from two to one year.Although itwould
encompassdeliveringchildcare,thereturnwasexpectedtogeneratean incomeeffect inthe
secondyearwhichwouldmainlyaffect low‐incomemothers.Thepaperreliesonaregression
discontinuitystrategytoestimatetheimpactofthepolicyanditshowsthatthemagnitudeof
the incomeeffect issubjecttothe familyvaluesheldbymothersaswellastotheireducation
attainment.Specifically,thepaperfindsthathighly‐educatedmothersupholdingliberalfamily
valuesaremorewilling toaccelerate theirreturn toworkafter thepolicyreformandhence
benefitfromthelabour‐to‐workpolicy.
I. Introduction
Severalscholarshavedevelopedtheideathatfamilyvaluesshapeeconomicdecisions.Reher
(1998)givesahistoricalaccountofthedifferencesinfamilytiesacrossEuropetohighlight
their relevance for social and economic outcomes. Specifically, he suggests that much of
socio‐economicdisparitieswillremaininplacedespitesimilarmodernizationprocesses,and
that familystructure isarelevantdriver.Esping‐Andersen(1990,1999)acknowledgesthe
role of the family in certain conservative welfare states of southern Europe, a core trait
1 PhD Candidate at the European Institute, London School of Economics and Political Science.
Email: [email protected]
2
which–forsomeauthorssuchasNaldini(2003)andLeibfried(inFerge&Kolberg,1992)‐
sets thebasis foradistinct typeofwelfare regime. Ferrera’spaper(1996)goes further in
characterizingthe‘southernmodelofwelfarestate’,highlightingthefamilyroleasa“social
clearinghouse” in southernEurope.Although taking amore empirical approach, economic
scholars(seeforinstanceAghionetal.,2011;Aghionetal.,2010;Alesinaetal.,2010;Algan&
Cahuc, 2007) have too made the case for a key role of family values on labour market
institutions and policies. They hypothesise the co‐evolution between values, demand for
regulation (behaviour) and policy design. Values, it is argued, influence the demand for
certain types of regulation, which, at the same time reinforce certain attitudes. In this
context twopossible equilibria emerge; a ‘good’ equilibriumwith low levels of regulation
andhighlevelsofsocialcapitalanda‘bad’equilibriumwiththeoppositeresults.
Thispaper ismotivatedbythe latterandattempts togoonestep furtherbyanalysingthe
interactionbetweenapolicy reformand family values and their impact onpreferences to
work.Iaimatshowingevidencethatasocialpolicyreformdependsontheexistingfamily
valuestobeeffective.
Theempiricalstrategyisthatofexaminingtheimplementationofaparentalleavepolicyin
Germanywhichaimedatacceleratingthepaceofreturntoworkaftermaternityinacontext
where different family values co‐exist. The paper relies on a regression discontinuity
strategytoestimatetheimpactofthepolicyanditshowsthatitseffectivenessissubjectto
thefamilyvaluesheldbymothersaswellastotheireducationattainment.
Thepapercontributestoaliteraturewhichusesthebroaderterm‘culture’andlooksatits
interaction with economic outcomes. Specifically, Barro and McCleary (2003), Tabellini
(2010),Guiso,etal., (2004,2009), IchinoandMaggi (2000)amongmanyothershavecast
light on the multiple ways culture can affect social preferences and economic outcomes.
However,thereiscontroversyonthedeterminismofthe‘cultural’argument.Whereassome
authorsperceivecultureandsocialpreferencesas ratherstaticandslow‐movingconcepts
(see for instanceRoland, 2004), others – such asRaquel Fernandez ‐ explicitly reject this
notion of culture and preferences as something “irrational, static or slow
3
changing”(Fernandez et al., 2004, p. 4) and argue that their change “depends on the
environment broadly speaking, including the opportunities which determine individual’s
learning pace, their interactions with others, and particular historical experiences”(p. 4).
Hence,policiescaninfluencetheenvironment.Similarly,thispapercontributestothedebate
onpolicyconvergenceataEuropeanlevel.Althoughthechoiceofpolicyisultimatelyleftto
member states, there is littledoubt that theseEU recommendationshave an effect on the
policy direction adopted by each state(Featherstone & Radaelli, 2003). These policies,
however, and following the argument stated above,may have different results depending
not only on the existing institutional framework but also on its cultural framework and,
more precisely, on the family values in place (Fernandez, 2007, p. 306). These ones,
therefore,cannotbeignoredatthetimeofdesigninglabourmarketandsocialpolicies.
The rest of the paper is organized as follows. Section II describes the policy reform and
section IIIgivesanaccountof thedataandthemethodologyused.Section IVpresents the
results,followedbythediscussioninsectionV.Finally,sectionVIconcludes.
II. The2007policyreform
ThenumerousparentalpolicyreformsthattookplaceinGermanyinthepastthreedecades
reflect a conflictive equilibriumbetween the traditional breadwinnermodel and thedual‐
earner‐carer model. The introduction of maternal leave dates back to 1979, when the
coalition of Social‐democrats and Liberal devised a maternal leave policy grounded on
mothers’ health and well‐being issues (Leitner, 2010), which, in the eyes of the policy‐
makerscouldhavepromisingeffectsinincreasingemploymentratesamongmothers.The
policy implemented a six months paid maternal leave period which enabled formerly
employedmotherstoreceiveacappedearnings‐relatedbenefit.Thisbenefitdidnotaccount
forpartners’earnings,andtargetedformerlyemployedmothers,whichsuggestsadeparture
from the predominant existing breadwinner model. However, part‐time work was not
included,giventhatitwasnotinlinewithhealthgoalsofthepolicy.
4
In1986thecoalitionofChristian‐democratsandLiberalspromotedasecondreform,which
revertedbacktotheoldbreadwinnermodel(Leitner,2010).Firstly,thepre‐existingcapped‐
earningsbenefitwassubstitutedbyaflat‐ratebenefitavailabletobothemployedandnon‐
employedmothersaswellasfathers.Thebenefit,however,wassolowthatitdidnotattract
fathers. Secondly, breadwinner’s earnings were taken into account and could reduce the
benefit (i.e. it was a means‐tested benefit). Thirdly, the paid maternal leave period was
increasedfirstlytotenmonthsandlateron,in1993,totwoyears.Additionally,theoverall
leaveperiodwasextended to threeyears. Implicitly themodelwas thereforepromotinga
breadwinnermodel,thetraditionalfamilymodel(Leitner,2010).Part‐timing,however,was
permittedupto18‐19hoursperweek.In2000theSocialdemocratsandtheGreenparty2,
implemented another reform which slightly departed from the breadwinner model
(Fleckenstein,2011),acknowledgedthe individualrighttoparental leavebyallowingboth
parentstotaketheleavesimultaneously,althoughthebenefitremainedameans‐testedone.
It also allowedpart‐timeworkup to 30hours perweek and it included thepossibility of
havingahigherflat‐ratebenefit ifthebenefitspanwasreducedfromtwotooneyear.The
impactofthereform,however,wasweakenedbythelackofinstitutionalchildcarefacilities.
In2005theso‐calledRed‐Greencoalitionmadeanattempttotacklethisissuebypassinga
lawwhichcommitted to theexpansionof childcare facilities forchildren less than3years
old.
Finally, in2007, theSocial‐democrats andChristian‐Democrats tookabig step forward to
shift from a breadwinnermodel towards a dual earner‐carermodel (Fleckenstein, 2011).
The reform – called Elterngeld ‐ replaced the flat‐rate benefit with a wage‐replacement
benefit up to 67% of earnings beforematernity leave, funded by the federal government
throughpublictaxation(Blum,2012).Acapof€1800andaminimumof€300wassetand
thenon‐employedwere entitled to thisminimum. Importantly, the reformalsodecreased
thebenefitspanfromtwotooneyearsanddevotedresourcestotheexpansionofchildcare
places.
2 Thesepartieswereencouragedbythe1996EUdirectiveonparentalleave(96/34/EC)
5
Theexpectedbenefitsofthe2007parentalleavereform
The design of the policy suggests that low‐income mothers are the group which should
experiencealargerchangeintheirworkbehaviour.Beforethepolicytheywereentitledtoa
maximumof€300,whereasafterthepolicytheyareentitledto67%oftheirpre‐maternal
earningswithaminimumof€300,asubstantialincreaseofthebenefit.Inthesecondyear,
however, by design, employed low‐income mothers experience a total decrease of the
benefit.High‐incomeemployedmothers,instead,donotseetheirincentivesmuchchanged
bythepolicy,especiallyinthesecondyear.Beforethepolicyhigh‐incomemothersdidnot
receiveanybenefitsotheincentivestoreturntoworkaftermaternityleavewerehigh.After
thepolicythissituationchangesandtheyreceive67%oftheirearningsduringthefirstyear
(withacapof€1800)andnothinginthesecondyear.
Theargumentof thepaper is that theseexpectationsare likely tobe influencedby family
valuesinthecaseofaworkingmother.AsBorkstatesinhispaper(2011),attitudestowards
workingmothers inGermanyhavebeenrathernegativeovertheyears,especially inWest
Germany. A termhas been coined ‐ ‘Rabenmütter’ (ravenmother) ‐ to designateworking
motherswithyoungchildren.Fleckenstein(2011)makesasimilarpointinhispaperwhen
he argues that, despite a decline in traditional family values, ‘West Germany remains
relatively conservative by international standards’ (p. 548). With the aim of testing the
argumentempirically,Ithereforeexpectthepolicytoworkonlyforthosemotherswhohold
liberalfamilyvalues.
III. Dataandmethods
InthissectionIshow,firstly,theintuitionbehindtheempiricalstrategyfollowed.Secondly,I
explainthechoiceofthemethodused:theRegressionDiscontinuityDesign(RDD).Thirdly,I
describe the nature of the variables, treatment and control, placing special attention to
familyvalues.Finallythesectioncontainssomedescriptivestatisticsandsomepreliminary
evidence.
6
Empiricalstrategy
The paper examines the effects of the 2007 policy on the decision to return‐to‐work for
motherswithdifferent familyvalues.Thepaper runsa seriesof logit specificationsof the
followingtype:
P(Yit=1) = α + β1tit + β2fvj + β3titfvj+ β4Xi +ε;
whereYit is theprobabilityofpreferringa fast return toworkaftermaternity fora given
individual i. It takes value1 if themother states that shewill goback towork as soonas
possibleorwithinoneyear,and0ifshestatesthatitwilltaketwoormoreyearstogoback
towork.titstandsforthetreatmentgroup,thatis,theindividualsaffectedbythepolicy,and
isatimedummyvariablewhichtakesvalue0iftheobservationbelongstotheperiodbefore
thepolicywasimplemented(1stJanuary2007)and1afterthepolicy.Dataspacesfrom2005
until2009,thatis,twoyearsbeforethepolicyandthreeyearsafterthepolicy.Thechoiceof
yearsaccountsforapotentialdelayinimplementation.fvjisaproxyofthefamilyvaluesof
eachindividual,whichareassigneddependingonthecountryofmigration(seesubsection
below for an explanation of the construction of the variable family values). titfvj is the
interaction between the time dummy and the family values. Finally,Xit includes a set of
individual characteristics as controls. This is themain specification corresponding to the
hypothesisof thepaper. InadditionIalsorunothersimilarspecificationswhereI test the
effectofthepolicyondifferentsubgroupsofpeopleaccordingtotheirincomeandeducation.
In these cases the interaction termhappensbetweenoneof the control variables and the
timedummy.
Methodology
The method used in this paper is the Regression Discontinuity Design (RDD) with a
Difference‐in‐Differences (DD) specification. The RDD method is used to estimate causal
7
effectsofaninterventionbyexaminingcomparableobservationsbeforeandafterthecut‐off
point. Itreliesontheassumptionthattheinterventionisrandomlyassignedandtherefore
observationsaroundthecut‐offpointarecomparable.Observationsbeforethecut‐offpoint
(the implementation of the policy) can then be treated as a control group (Green et al.,
2009).ThepolicyinterventionanalysedinthispaperseemssuitedtoaRDDmethod.Firstly,
theinterventioncanbearguedtoberandomlyassigned,giventhatthetreatmentwouldbe
available for all new‐mothers from 1 January 2007. The cut‐off point, therefore, did not
dependonany individualcharacteristicsof themother,onlyonthebirthrateof thechild.
Althoughitcanbearguedthatmotherscouldhaveattemptedtochangetheirbehaviourand
delaymaternity, this argument is ratherweakdue to the speed of the legislationprocess.
ThemainfeaturesofthereformwerediscussedinMay2006,draftedinJune, the lawwas
passedinSeptember2006anditbecameeffectiveon1January2007(Kluve,2009).Figure1
supportsthisargumentbyshowingthatthemonthlynumberofbirthratesdidnotchange
significantlyfrom2005to2007.
Figure1:Birthratesacrossyears
Source:,BergemannandRiphahn(2011),fromtheGermanFederalStatisticaloffice
Theobservationsintheanalysisarethereforedividedbetweenyears2005and2006,which
belongtothecontrolgroup,andyears2007to2009,whichbelongtothetreatmentgroup3.
3Seethesection‘Discussion’forananalysisofthechoiceofthenumberofyearstaken.
8
Moreover,IuseaDDspecification–reflectedintheinteractionterm‐comparingdifferent
subgroupsofindividualsaccordingtotheirfamilyvalues,educationorincome.
ThechoiceoftheRDDmethodhasanumberofadvantagesworthmentioning.Tostartwith,
intheanalysisofthepapertheoverallRDDchoicesubstitutesthedifference‐in‐difference‐
in‐differencesmethod(DDD)–whoseresultsarealwaysmoredifficulttointerpret.Inthis
paper this is a great advantage because, aside from a control and treatment group that
allowsmetoinfercausality,Ialsoneedacontrolandtreatmentgrouptoassesstheimpactof
familyvalues(i.e.theimpactofthepolicydependingonvaluesneedsacomparisonbefore
andafterandbetweentraditionalandliberalfamilyvalues).Secondly,inaRDDcontextthe
control group is probablymore similar to the treatment group than in aDD context. And
thirdly, the RDD specification reduces the probability of havingmeasurement errors: my
dependentvariableisbasedonaquestionintheGSOEPdatabaseaboutthe ‘willingnessto
return to work’ (see data and descriptive statistics section for more information), which
leadstoafourcategoriesresponsewhichIcodeasbinary:fastreturnorslowreturn.This
choicewouldnothavebeenpossiblewithaDDDspecification,giventhatmycontrolgroup
wouldhaveprobablybeenworkingmothers(childlessorwitholderchildren),forwhomit
makesnosensetoaskaquestiononthewillingnesstoreturntowork(giventhattheyare
already working). Alternatively, an option for the DDD would have been to look at the
numberofhoursworkedbymotherswithone‐yearoldchild,andcomparethosebeforeand
after thepolicy,withworkingmothersorwomenas a control variable, forwhich Iwould
alsolookatnumberofhoursworked.Thedata,however,makesitdifficulttoknowtheage
of the child as it only provides data for the mother, and this could possibly lead to
measurementerrors.
The identification strategy follows Fernandez (2007) to evaluate the impact of the policy
reform on the decision to return‐to‐work across individuals with different family values.
Thisapproachusesmigrantstoovercometheproblemofdisentanglingtheeffectsofculture
or values from economic and institutional environment. Migrant groups face the same
institutionalandeconomicenvironmentofthenativeindividualsinthecountryofresidence
9
but they are assumed to preserve, to a certain extent, family values of their country of
ancestry. With this approach I then analyse the effects of the policy for different female
migrants groups, comparing their outcome within these groups and between them and
nativegroups(EastandWestGermans).Thebasisforclusteringtheindividualsindifferent
groups will be family values. That is, I cluster them according to whether they hold
traditional or liberal family values (family values can also be – and will be – rated on a
continuum).Inordertoavoidproblemsofreversecausality(i.e.currentattitudesmayhave
been influencedbyprevious economicoutcomes), I use aproxy for current family values,
whicharevaluesexpressedbyindividualsinthemigrant’scountryofancestryinprevious
years.
Dataanddescriptivestatistics
IusetheGermanSocio‐economicpaneldata(GSOEP)4,alongitudinaldatasetrunningyearly
since1984until2011(thelatestwave)whichinterviewsallthemembersofthehousehold,
newcomersandfollowstheleaversinnewhouseholds.TheGSOEPhasgraduallyincreased
its sample up to nine times, with some of these samples being focused onmigrants (see
Appendix1forarelationoftheexistingsamples).Intotal–from1984to2011‐itcontains
around600.000observations.ForthepresentanalysisIselectwomenwhoworkandhave
hadachildinoneoftheyearsfrom2005to2009.Toknowwhethertheyhadachild,thereis
a questionwhich asks ‘Hasyour family situationchangedafterDecember31,200X?’ (200X
belongs to n‐2, i.e. if the questionnaire belongs to year 2008, the question will refer to
December 31, 2006). One of the answers is ‘Yes, had a child’ and for each answer the
respondentisaskedwhetherthiswasinyearnorn‐1(i.e.inthequestionnairebelongingto
year2008, theoptionsare:2007and2008).Given that the interviewshappen indifferent
monthsoftheyearforeachrespondent,itcanbethecasethattheyareaskedthisquestion
4ThedatausedinthispaperwereextractedusingtheAdd‐OnpackagePanelWhizv4.0(Oct2012)forStata.PanelWhizwaswrittenbyDr.JohnP.Haisken‐DeNew([email protected]).ThePanelWhizgeneratedDOfiletoretrievetheSOEPdatausedhereandanyPanelwhizPluginsareavailableuponrequest.Anydataorcomputationalerrors in thispaperaremyown.Haisken‐DeNewandHahn (2010)describePanelWhiz indetail.
10
beforetheyhavehadachild(e.g.therespondentisinterviewedinJanuary2007andshehas
achildinDecember2007).Toavoiddroppingwomenwhohaveactuallyhadachild,Irely
ontheanswersfromyearn‐1.Inthecaseofmultiplebirths,Ihaveonlykepttheobservation
belonging to last birth.After droppingmissing observations and coding all the variables I
need,Iendupwithatotalof491observations.
Mydependent variable is basedon the followingquestion from theGSOEPquestionnaire:
‘Whenapproximately,wouldyouliketostartwithpaidemployment?’Theanswerscanbe:‘1)
Assoonaspossible,2)Nextyear,3) In thenext two to fiveyears,4) In thedistant future, in
more than fiveyears’. I code themas fast return (dummy=1) if the answer is ‘as soonas
possible’or‘nextyear’andslowreturnifotherwise(dummy=0).Ithereforehaveabinary
dependentvariablewhichistheintentiontoreturntowork.Appendix2explainsindetailthe
codingofthedependentvariable.
Myindependentvariable–individualfamilyvalues‐is,asstatedabove,proxiedbysocietal
familyvaluesexpressedbyindividualsinthemigrant’scountryofancestryintheyear2000.
Thesocietalfamilyvaluesareconstructedasfollows:Itaketheanswersfromthefollowing
question fromtheWVS:Doyouagreewith the followingstatement?:Aworkingmothercan
establish justaswarmandsecurearelationshipwithherchildrenasamotherwhodoesnot
work’andrunanindividuallogitregression,withthisquestionbeingthedependentvariable
and my main independent variable being country dummies. The base region is ‘West
Germany’. The country coefficients then tell me the likelihood that an individual from a
certaincountryorregionwillagreewiththepreviousstatementcomparedtoanindividual
fromWest Germany. I control for age, age squared, size of town, marital status, sex and
education.Thecountrydummycoefficientswillbetheinstrumentusedformyindependent
variable. All coefficients happen to be statistically significant (most of them at 1%
significancelevel)exceptforPortugal.Figure2depictstheresults.
11
Figure2:effectsofcountryoforiginon‘workingmother’acceptance
Inmyanalysisthesevalueswillbeusedasacontinuousvariableaswellasabinaryvariable.
Ithereforecreateadummyvariablewhichisadichotomizationofthevariablefamilyvalues.
TocarryoutthisdichotomizationIhavecalculateditsmeanvalueandIhavecategorizedall
observations below themean as traditional family values and all observations above the
meanasliberalfamilyvalues.
Both family values and the dummy variable are then assigned to each ofmy observation
depending on theirmigrant origin. The GSOEP database allowsme to knowwhether the
observation has ‘nomigrant background’, a ‘directmigrant background’ (first‐generation
migrant)oran‘indirectmigrantbackground’(second‐generationmigrant)andittellsmethe
countryoforigin for thosewhoare categorizedasmigrants. SeeAppendix2 fordetailson
howIhavecodedthecountryoforiginforeachmigrant.ToseparateEastfromWestGerman
observations,Ihaveusedtheinformationfromthevariable‘sample’.SampleCincludesonly
observations from East Germany. I have also used the variable ‘sample’ to drop some
observationswhichbelongedtosampleswherethemotherwasGermanbutthefatherwasa
migrant.Thatis,Ihavedroppedtheobservationsforwhichthemigrationbackgroundofthe
12
mother was ‘no migration background’ and it belonged to sample B (foreigners), D
(Immigrants)orF (Innovation)whereat leastonememberof thehousehold isamigrant.
The following table, Table 1, shows the migration background and the number of
observations.
Table1:Countryoforiginofobservations
Countryoforigin
Nomigrationbackground
Directmigrationbackground
Indirectmigrationbackground
TOTAL
EastGermany 234 234WestGermany 125 125Turkey 10 20 30Russia 11 0 11Greece 1 5 6Poland 6 0 6Macedonia 2 3 5Croatia 4 0 4Italy 2 1 3Rumania 3 0 3Ukraine 3 0 3Portugal 2 0 2Serbia 2 0 2USA 2 0 2Belarus 1 0 1Brazil 1 0 1Canada 1 0 1China 1 0 1CzechRepublic 1 0 1France 1 0 1Hungary 1 0 1Iran 1 0 1Philippines 1 0 1Slovakia 1 0 1TOTAL 359 71 32 462Source:ownelaborationbasedonGSOEPNote:Ihavedroppedtheobservationswithmissinginformationonanycrucialvariabletopursuetheanalysis.
Several controls are included in the regression. Household income is coded as 0 if it is
greaterthan€2000/monthand1ifitislowerthanthisthreshold.Thereasontocodehigh
incomeas0andlowincomeas1–whichmightseemunusual–isjustifiedbythefactthat
thepolicywasexpectedtoinfluencemainlylowincomemothers,andtherefore,this isthe
‘treatment’ group that I am interested in. In order to dichotomise it I have calculated the
meanvalueof incomeand Ihave treatedall observationsbelow themeanas low income.
Education,measured in ‘yearsof education’ is another control variable. It hasbeen coded
13
bothasacontinuousvariableandasabinaryvariable.Inthelattercase,thetwocategories
are low and high education, with the threshold being A‐levels. Low education, therefore,
includesobservationswithnoA‐levels,thatis,withlessthantwelveyearsofeducation.High
educationincludesobservationswiththirteenormoreyearsofeducation.Someregionsin
GermanyneedtwelveyearsofeducationtoachieveA‐levels,whereasothersneedthirteen.
Given thisdifference,whenusingeducationasabinaryvariable,observationswith twelve
years of education have been dropped. Another control is marital status. This one takes
value0ifthemotherisnotmarriedand1ifsheismarried.Finally,Iuseregionaldataonthe
percentage of zero to three year‐old babies that attend childcare as a proxy for regional
childcareavailability.
Table2presentssomebasicdescriptivestatisticsofthedatabeforeandafterthepolicy.The
data reveals that there are no statistically significant differences in the means of the
variables between the two periods (before and after the policy). Themeans for the ‘fast
return’ variable, the dependent variable, indicates that in both periods thewillingness to
quicklyre‐enterthejobmarketislow.Thedataalsoshowsthattheaveragematernalageis
32yearsold,andthatmostoftheobservationsinbothperiodsaremarried.Withregardsto
years of education, the average is 13 years, which is the A‐levels threshold. Average
household income amounts to approximately €2.700 and average family values aremore
traditional.
Table2:Descriptivestatistics
Number of children born
Variables mean SD mean SD
fast return (0/1) 0,45 0,5 0,48 0,5
maternal age 31,6 6,87 31,7 5,32
married (0/1) 0,67 0,47 0,7 0,46
years of education 13,14 2,65 13,26 2,69
household income 2644 1419 2764 1212
family values 0,35 0,48 0,29 0,45
childcare 0,17 0,15 0,16 0,14
Note: t‐tests indicate no statistically significant difference between subgroups at 1% and 5% levels.
Before the policy reform
2005‐2006
187 303
2007‐2009
After the policy reform
14
IV. Empiricalanalysisandresults
Tables3to5presenttheresultsoftheempiricalanalysis. Inallregressionsthedependent
variable is the probability of preferring a fast return to work after maternity. I start by
looking at the entire sample. Given that the policy was mainly targeted at shaping the
behaviour of low‐income mothers, I also show the results of the effect of the policy for
mothers living in low‐incomehouseholds.Table3 is therefore relevant toprovideuswith
information on how well the policy reached its target, which will then motivate the
introductionoffamilyvalues.
Thebaselineregression(col.1)regressestheprobabilityofpreferringafastreturntowork
on a time dummy variable and a set of individual characteristics, which include marital
status,educationasacontinuousvariable,incomeandaproxyofchildcareavailability.The
resultsshowthatthepolicyhashadapositiveeffectonshapingpreferencestowardsafast
return to work, although strictly speaking the effect is small (of almost 2%) and not
statistically significant. Instead, the four controls appear tobe statistically significant.One
year increase ineducation increases theprobabilityofpreferringa fast return toworkby
almost3%,significantat1%level.Beingmarrieddecreasestheprobabilityofpreferringfast
returntoworkbyabitmorethan10%,significantat10%level.Onepossibleinterpretation
isthatmarriedmothersarelikelytohaveothersourcesof income(e.g. fromthehusband)
which allow them to delay the return to work. An increase of one percent in childcare
availabilityleadstoanimpactonthepreferencesofafastreturntoworkof32%,significant
at10%level.Havingalowhouseholdincomealsodecreasestheprobabilityoffastreturnto
workbyapproximately16%,significantat1%level.
An interaction between the policy (time dummy) and income is added in a difference‐in‐
differencefashion(col.2).Thisinteractionisinterestingbecause,aspointedoutinsectionII,
thepolicydesignmakesitmoreexpensiveforlowincomehouseholds‐whowereentitledto
the parental leave means‐tested benefits before the policy ‐ to stay at home. These
householdswouldseeafterthepolicyaprobableincreaseofthebenefitinthefirstyearof
childbirthfollowedbyatotaleliminationofthebenefitinthesecondyear.Itwastherefore
15
expectedthatitwouldgenerateanincomeeffectinthesecondyearwhichwouldincentivise
lowincomemotherstoreturnfastertowork.Theresultsdepictedincolumn2,however,do
notquitesupportthisexpectation.Thecoefficientofthevariableincomesuggestthatbefore
thepolicya lowincomemotherwas24%less likely toprefera fastreturntoworkthana
high incomemother, a resultwhichmotivates the introductionof thepolicy.Nonetheless,
theinteractiontermtellsusthatthissituationdidnotchangesignificantlyafterthepolicy.
Althoughafterthepolicyalowincomemotherwas14%morelikelytopreferafastreturnto
work than before the policy, its effect appears to be statistically insignificant at 10%
significance level, implying that the preference of fast return to work for low income
individualsdidnotchangeafterthepolicyreform.Thisresultisrobustevenwhenonlythe
individualswith the lowest income5 in the scale are taken into account and its income is
interactedwiththepolicy.Theresultisshownincolumn3,whichshowsthattheinteraction
term remains statistically insignificant. Both in column 2 and 3 income, education and
maritalstatusaresignificant,withsimilareffectsasincolumn1,whereastheavailabilityof
childcareisonlysignificantincolumn2.Thisfirstanalysis,onthebasisoftheresultsinTable
3,suggeststhatthepolicyhasnothadavisibleeffect.
5Ihavetakenthefirstquartileofincome.Thatis,Ihavedichotomizedthevariable,withthelowestquartilecodedas1andtherestcodedas0
16
Table3:Averagemarginaleffects:effectofthepolicyandeffectofthepolicyinteractedwithlowincomemothers,entiresample.
Variables (1) (2) (3)
timedummy 0.017 ‐0.024 ‐0.017 (0.738) (0.685) (0.766)
education(c) 0.029*** 0.029*** 0.033*** (0.002) (0.002) (0.000)
maritalstatus ‐0.104* ‐0.106* ‐0.096* (0.061) (0.056) (0.087)
Childcare 0.319* 0.299* 0.247 (0.078) (0.099) (0.173)
Income ‐0.158*** ‐0.243*** (0.007) (0.006)
time*income 0.143 (0.196)
income,lowestpercent. ‐0.171* (0.061)
time*lowestpercent.income 0.135 (0.238)
Observations 403 403 403pvalinparentheses
***p<0.01,**p<0.05,*p<0.1 Nonetheless, it is plausible that the effect of the policy is concealed by some kind of
heterogeneitywithin the population.One plausible source of heterogeneitywhich goes in
line with the hypothesis of the paper is family values. These ones are likely to play a
significant role in shaping the outcomeof aparental leavepolicy and thus, the lattermay
appearineffectiveiffamilyvaluesarenottakenintoaccount.TheresultsinTable4suggest
that this might be the case, although they are not very robust. Family values, when
categorized as a binary variable (family values b, col. 4), seem to matter for the policy
effectivenessat10%significance level.After thepolicy,motherswith liberal familyvalues
are21%more likely toprefera fastreturn towork thanbefore thepolicy.This isnot the
case for mothers with more traditional values, as the insignificance of the time dummy
coefficientshows.Nonetheless,thisresultisnotrobustwhenfamilyvaluesaremodelledasa
17
continuousvariable(familyvalues(c)),shownincolumn5.Heretheinteractionbetweenthe
time dummy variable and family values is insignificant, as it is the time dummy and the
family values variable. The only significant variables are marital status, income and
education.
Table4:Averagemarginaleffects:impactofthepolicyinteractedwithfamilyvalues,impactofthepolicyinteractedwitheducation;entiresample.Impactofincomeandvalues,highly‐educatedsample.
Variables (4) (5) (6) (7) (8) (9)
timedummy ‐0.050 ‐0.072 ‐0.023 0.028 ‐0.007 ‐0.245 (0.431) (0.554) (0.786) (0.719) (0.938) (0.201)
maritalstatus ‐0.128** ‐0.132** ‐0.090 ‐0.097 ‐0.130 ‐0.126 (0.029) (0.024) (0.137) (0.238) (0.130) (0.141)
Childcare 0.379* 0.258 0.281 0.080 ‐0.031 0.003 (0.092) (0.230) (0.163) (0.764) (0.926) (0.992)
Education1 0.028*** 0.029*** 0.094 (0.004) (0.003) (0.296)
time*educ 0.093 (0.400)
Income ‐0.171*** ‐0.167*** ‐0.158** ‐0.170 ‐0.120 ‐0.115 (0.007) (0.009) (0.014) (0.234) (0.229) (0.249)
time*income 0.166 (0.367)
familyvalues(b) ‐0.130 ‐0.065 (0.166) (0.648)
time*values(b) 0.211* 0.267* (0.054) (0.088)
familyvalues(c) ‐0.003 ‐0.073 (0.954) (0.507)
time*values(c) 0.065 0.227* (0.390) (0.068)
Observations 364 364 340 206 191 191pvalinparentheses
***p<0.01,**p<0.05,*p<0.1
18
Another source of heterogeneity might be education. The results on the previous tables
suggest that education is a highly influential variable for the probability to return fast to
work,bothinstatisticandeconomicterms6.Moreover,educationisrecognizedbyandlarge
in the literature as one of the most decisive factors in influencing female labour market
participation (see, for example, OECD 2004:4). I would then expect education to play a
positive role in the effectiveness of policy, because highly‐educated individuals are more
likely tobeawareof thebenefits for theirworkingcareerofreturning faster towork.The
effectsofeducation,however,maybe limitedbytwodifferent factors. Firstly, itwouldbe
logical to think thatonly thehighly‐educatedmotherswith low incomereact to thepolicy
(given that the policy is targeting low income individuals). Figure 3 shows that this
hypothesis is worth testing given that although there is a positive correlation between
incomeandeducation,thehighertheeducationlevel,themorevarianceincomeexperiences.
Secondly,andfollowingthemainargumentstatedinthispaper,theeffectofeducationmight
fadewhen family values are taken into account.Ahighly‐educatedmother coming froma
strongfamilytraditionandamarkedpreferenceforhomechildrearingislikelytopreferto
stay at home with the child for a longer period of time, regardless of any economic
incentives.
6 Marital status is also very relevant, although less statistically significant than education.Unfortunately, the possibility of analysing the policy only for married or lone mothers isunfeasibleduetothereductionofthesamplesize.
19
Figure3:Correlationbetweenhouseholdincomeandeducation,entiresample.
Column6ofTable4 starts bypresenting the impact of the policy and its interactionwith
educationontheprobabilityofafastreturntowork.Theinteractiontermsuggeststhatthe
effectof thepolicy is insignificantboth forhighand low‐educatedmothers,controlling for
marital status, childcare availability and income. Column 7 uses the restricted sample of
highly‐educatedmothers7and testswhether the low‐income–highly‐educated individuals
react to the policy. Again, the interaction term between time dummy and level of income
showsthatthisisnotthecase.Inotherwords,itappearsthathighly‐educatedmotherswith
lowincomeexhibitthesamepreferencestoreturnfasttoworkbeforeandafterthepolicy.
Column8and9testthehypothesisinwhichonlythosehighly‐educatedmotherswithliberal
valuesreacttothepolicy.Theresultsconfirmthehypothesis,regardlessofwhetherfamily
valuesaretreatedasabinary(column8)orcontinuous(column9)variable.Accordingtothe
interaction termsof themodels,motherswith liberal familyvaluesarebetween22%and
27%morelikelytopreferafastreturntoworkafterthepolicy.Theresult issignificantat
10%significancelevel.Thisisnotthecaseformotherswithmoretraditionalvalues,asthe
7Athree‐wayinteractionbetweenthethreevariableswouldbepossible,althoughitwouldresultincoefficientsdifficulttointerpret.
02
000
400
06
000
800
01
0000
5 10 15 20Amount Of Education Or Training In Years
dincome Fitted values
20
insignificance of the time dummy coefficient in bothmodels shows. Therefore, it appears
thateducationisrelevanttotheeffectivenessofthepolicy,butitisnotsufficienttochange
mother’s behaviour. The relevance of family values reveals that it is only those highly‐
educated mothers coming from a liberal family tradition who are more likely to be
influencedbythepolicyreformandadjusttheirdecisiononre‐enteringthelabourmarketat
afasterpace.
Thepredictedprobabilitiesofpreferringafastreturntoworkforhighly‐educatedmothers
holdingdifferentfamilyvaluesareshowninTables5and6.Whenclassifyingfamilyvalues
dichotomously,(Table5),theprobabilityofpreferringafastreturntoworkisaround50%
formothersholdingtraditionalfamilyvalues,bothbeforeandafterthepolicy(assuggested
abovewith the timecoefficients fromTable4 small change inprobabilitybeforeandafter
thepolicyisinsignificantformotherswithtraditionalfamilyvalues).Withregardtomothers
withliberalfamilyvalues,theprobabilityofpreferringafastreturntoworkamountto46%
beforethepolicy,whereasitincreasesupto71%afterthepolicy.
Table5:Predictedprobabilitiesofpreferringafastreturntowork,dichotomousfamilyvaluesandhigh‐educatedmothers’sample
Before the policy After the policy
Family values = 0 0,53 0,52
Family values = 1 0,46 0,71
Table6:Predictedprobabilitiesofpreferringafastreturntowork,dichotomousfamilyvaluesandhigh‐educatedmothers’sample
Before the policy After the policy
Family values = 0,5 0,57 0,44
Family values = 1 0,54 0,52
Family values = 1,5 0,50 0,60
Family values = 2,0 0,46 0,67
Family values = 2,5 0,43 0,74
Note:thetableonlydepictsfivevaluesoffamilyvalues,butthevaluesrangefrom5to25.
21
Beforeturningtothediscussion,andgiventhestrongimpactofeducationontheanalysed
policy and the significant interaction of education and family values, there is the need to
confirm that it is education thatmatters, andnot other correlatedomittedvariables.Two
maincandidatesaretobeconsidered:incomeandoccupation.
With regard to the first one – income– it is reasonable to think that it is correlatedwith
education, and that therefore only low income mothers with liberal family values (as
opposedtothosehighly‐educatedwithliberalfamilyvalues)reacttothepolicy.Column10
fromTable7restrictsthesampletothelow‐incomemothersandinteractsthetimedummy
withfamilyvalues.Theresultsrejectthehypothesisandshowthatlowincomemotherswith
liberalfamilyvaluesdonotsignificantlyreacttothepolicy.
Occupation is the other variablewhich is likely to be correlatedwith education. I expect
mothers in occupationswhich donot involvemuch physical effort to bewilling to return
fastertowork.Thecorrelationbetweenoccupationandourdependentvariableamountsto
24%8.Ihavethenincludedoccupationinthemodelandinteractitwiththepolicy,asshown
in column 11. The results suggest that occupation is significant at 10% level before the
policy,andthatwhite‐collarworkersarethreetimesmorelikelytoexperienceafastreturn
toworkthanblue‐collarpeers.Thisoutcomedoesnotchangeafterthepolicyinasignificant
way. This result, however, is not robust to the inclusion of marital status and education
controls,asshownincolumn12.Inotherwords,onceeducationisincludedinthemodel,the
effect of occupation disappears. This suggests, therefore, that education stands as a
significantvariable.
8Ihavecodedoccupationasabinaryvariablethattakesvalue0ifitis‘blue‐collar’and1ifitis‘white‐collar’. I have dropped the ‘self‐employed’, ‘apprentices’ and ‘civil servant’ categories,sinceitwashardertoclassifythemaccordingtothephysicaleffortrequired.
22
Table 7: Average marginal effects: impact of the policy on low-income sample, effect of the policy with occupation.
Variables (10) (11) (12)
timedummy 0.099 0.025 0.101 (0.370) (0.899) (0.616)
familyvalues(b) 0.121 (0.424)
time*values(b) 0.093 (0.596)
maritalstatus ‐0.140 0.064 (0.118) (0.414)
education(c) 0.043*** 0.027* (0.005) (0.077)
childcare 0.085 0.659*** (0.795) (0.009)
occupation 0.266* 0.200 (0.086) (0.218)
time*occupation ‐0.070 ‐0.119 (0.745) (0.581)
Observations 125 195 184pvalinparentheses
***p<0.01,**p<0.05,*p<0.1
23
V. Discussion
This paper has attempted to testwhether family values act as a filter to the success of a
familypolicy.Theresultshaveshownthatthepolicyhasbeeneffectiveforhighly‐educated
motherswithliberalfamilyvalues.Onthisbasis,thissectiongivesanoverviewoftheresults
and turns to adiscussionof anumberofmethodological anddata issues arising from the
analysisaswellaspolicyimplications..
Resultsandrobustnesschecks
Thepolicyanalysed in thispaperwasdesignedtoaccelerate thereturn toworkmainlyof
motherswith low income.Nonetheless, the results from theprevious section suggest that
thepolicydidnotsignificantlychangethepreferencesofthisgrouptoreturnfasttowork.
Instead,theyshowthattwoothervariables–educationandfamilyvalues–arerelevantto
explainwhoreactedtothepolicy.
Previous literature has always pointed out at education as a key variable to understand
different labour‐market outcomes. Indeed it is reasonable to think that highly‐educated
mothers aremore aware of the benefits of returning faster towork for their professional
careers.Theresultsof thisanalysis,however,addtothe literatureshowingthateducation
alone isnotasignificant factor for thesuccessof thepolicy. Itappearsthatamonghighly‐
educated mothers, upholding liberal values increases the probability of preferring a fast
return to work by around 23% to 27%. This is not the case amongmothers frommore
traditionalvaluesbackgrounds,forwhichnoevidenceisfoundofachangeinpreferencesto
return towork. This result is robust to the specification of family values as a continuous
variable,anditissignificantat10%level.Thisfindingisinlinewiththehypothesisofthe
paper and suggests that highly‐educatedmothers with traditional family valuesmay still
prefertochildrearingathomefora longerperiodanddelaytheirreturntowork; i.e.that
financial incentives are less relevant. Instead, highly‐educatedmotherswith liberal family
valuesmayprefertore‐enterthejobmarketassoonaspossible.
24
Some robustness checks are carried out to validate the argument. One alternative
explanationtotheoneofferedwhichwouldrulefamilyvaluesoutcouldbethatitisactually
highly‐educatedand low incomemotherswho react to the policy ( as opposed to highly‐
educatedwithliberalfamilyvaluesbackground).Thismightbe,Iargue,becausewithinthe
highly‐educated theremightbe lowandhigh income individuals, forwhomthepolicyhas
income effects which work in opposite directions, as explained in section II. But even
restricting the sample to the highly‐educated and calculating the impact of the policy
dependingonincomedoesnotrevealanyimpactofeducationtogetherwithincomeonthe
probabilityoffastreturntowork.
Another alternative explanation could be that the power of education is overstated in
detriment to other correlated variables with potentially equal power, such as income or
occupation.Withregard to income,onemight think that it isnothighly‐educatedmothers
withliberalvaluestheoneswhoreactbettertothepolicybutlow‐incomemothers(atwhich
thepolicywastargeted)withliberalfamilyvalues.Ifthisisthecase,theimplicationswould
be that thepolicywouldhavepartially succeed, although the significanceof valueswould
pointtotheneedtotakefamilyvaluesseriouslyifthepatternsofreturntoworkaretobe
changed.Results,however,donotsupportthisargument;lowincomemotherswithliberal
familyvalueshavenotsignificantlyreactedtothepolicy.Turningtooccupation,itwouldbe
logical to think that mothers with less physically demanding jobs return faster to work.
Althoughresultspointoutattheimportanceofoccupationfortheprobabilityoffastreturn
beforethepolicy,thissignificancedisappearsonceIcontrolforeducation.
Policyimplications
Several policy implications stem from these findings. In order to enhance policy
effectiveness,ashiftfromeconomicincentivestowardseducationisneeded.However,asthe
findingsreveal,notallhighly‐educatedmothersreactsimilarlytothepolicy.Itisonlyhighly‐
educatedmotherswith liberalvalueswhoaremore likely tochangetheirpreferencesand
accelerate theirreturn toworkafter thepolicy.Thissuggests that familyvaluesshouldbe
25
takenonboardforpolicydesign.Inotherwords,whattheresultsareimplyingisthatsimilar
socialandlabour‐marketpoliciesarelikelytohavedifferenteffectsdependingonthefamily
valuesofthesociety inwhichthepolicy is implemented.Asnotedinthe introduction, this
result is of great significance for the process of European integration and in the present
European context. Although the design and implementation of labour‐market and social
policiesfallwithinthestateresponsibility,theEuropeanUnionisencouragingconvergence
ofcertainpoliciesandoutcomesrelatedtochildcareand,morebroadly,tosocialcare.Thisis
even more relevant in current times, when the current economic crisis has led some
countries to adopt labour‐market policies on the basis of success in other European
countries.Althoughsomeofthesepoliciesmightbeneeded,itsimplementationincountries
where family ties and values are different may lead to unexpected outcomes. Notably,
ineffectivenessofsuchpoliciesindifferentcountriesmaysubsequentlybeseenasanissue
of implementation failure or of lack of administrative capacity, whereas, as the results
suggest,differencesineffectivenessmayinfactbedrivenbycountryheterogeneityinfamily
values.
These results, however, provide an alternative picture to other findings. Bergemann and
Riphahn(2011)analysethe2007policyreformandfindthattherehasbeenanimpactfor
low‐income mothers at 10% significance level. They highlight a substantially higher
propensityofEastGermanwomentoreturnfastertowork.KluveandTamm’spaper(2009),
whichalsoanalysesthesamepolicyreform,takenon‐employedmotherswithintheirsample
aswell,andconcludethattheimpactofthereformhasonlybeenpositiveforEastGermany
mothersandwomenwhowerenotpreviouslyemployed.TherelevanceoftheEastGerman
mothersinbothpapersmayreflect,assuggestedinthispaper,differencesinfamilyvalues.
Tofullyunderstandthereasonswhytheresultsoftheseotherpapersdifferfromthisone,
thepresentpapercouldcarryoutdifferentrobustnesscheckswhicharespelledout inthe
nextsection.
26
Methodologyanddataissues
The choice of RDD has some advantages which have already been stated in section III.
Nonetheless, RDD also suffers from some drawbacks which are worth discussing. One of
themandprobablythemostimportantisthetrade‐offbetweenthenumberofobservations
andtheaccuracyofthecontrolandtreatmentgroup.IntheRDDspecification,thecloserthe
observationsaretothecut‐offpoint,thelowertheriskthatthetreatmenteffectsuffersfrom
omitted variable bias(Green et al., 2009). On the other hand, having a narrow timeframe
leads to fewer number of observations, which increases the sampling variability. In this
paperthetimeframehasbeentwoyearsbeforeandthreeyearsafterthepolicyinorderto
allowanacceptablenumberofobservations.Thischoice,however,ismadeattheexpenseof
increasing the bias. At least two alternatives exist to correct for thepotential bias: one of
themwould be to increase the number of control variables. I could control for childcare
facilities,tenureatthejob,macroeconomiccontrolssuchasemploymentlevels,numberof
childrenathome,and theLänderamongothers.However,addingcontrolshas tobedone
withcare,giventherelativelysmallnumberofobservationsIhave,especiallyinthehighly‐
educatedsample.Anotheroption is tocarryoutaDDDanalysisandcomparetheoutcome
with that of theRDDanalysis.Despite the above‐mentioned setbacksofDDD,using ithas
severaladvantages.Firstly,itwouldallowmetoenlargethetimeframeandasaconsequence
increasethenumberofobservations.Secondly,andgiventhatIwouldhaveacontrolgroup
for thewhole period, I could avoidmany controls andwin degrees of freedom. Finally, it
wouldallowme toexpandmyanalysis and test forpotential ‘delay’ effects and long‐term
effectsofthepolicy.Withregardtothe‘delay’effects,itmightbethecasethatittakessome
time for the individuals to adjust to the policy, something that the DDD analysis could
account for by omitting the observations closer to and past the cut‐off point. This is not
possiblewiththeRDDanalysis,giventhatittakespreciselyobservationsclosetothecut‐off
point. DDDcanalsotest forwhetherthe long‐termeffectsof thepolicyaredifferent from
the short‐termones. For example, this paper suggests that the policy does notworkwith
individuals with low‐education and traditional values. However, it might be that family
27
valuesareshapedby thepolicyandexperimentashift in themedium‐longrun.This isan
effectthatDDDismorelikelytocapture,whereasRDDdoesnotallowfortesting.Another
relatedeffectthatRDDcouldpotentiallybemissing–althoughthetestsinthedatasection
suggest that it is not – is an anticipation effect. DDD analysis would complement the
mentionedtestsbyomittingtheobservationscloserandprevioustothecut‐offpoint.
Someofthevariablesincludedintheanalysis–especiallythedependentvariableandfamily
values variable – deserve special attention. With regard to the dependent variable, one
alternativeistotake‘actualreturntowork’asopposedto‘willingnesstogobacktowork’.
This, however,has twomaindrawbacks. Firstly thismeasure ismuchmore influencedby
factorssuchas thehazardsof findinga joborotherpersonal factors thanthequestionon
‘willingness to return towork’. Secondly, although thepanel characteristics of thedataset
wouldhaveallowedmetofollowindividualsacrosstimeandthereforetracethetimethatit
takesforthemtogobacktowork,thefactthattheindividualdataisyearly–asopposedto
monthly–andthatinterviewstakeplaceindifferentmonthsforeveryindividualandevery
year,wouldhaveprobablyleadtomeasurementerrorprobleminmydependentvariable.
Family values areworth discussing, not only because of its salience in thepaper but also
becauseofthemanypotentialwaysofmeasuringit.Inordertoreinforcetheresultsofthe
paperIcouldcarryoutalternativeanalysiswhereotherwaysofproxyingfamilyvaluesare
used.Theapproachusedinthispaperfollowsseveralpaperswhichhaveexamined‘culture’
and‘familyvalues’,andwhichusuallyusequestionsfromdatabasessuchastheWorldValue
Survey(WVS),EuropeanValueSurvey(EVS)ortheInternationalSocialSurveyProgramme
(ISSP).Therearemultiplequestionsandstatementsinthesedatabasesthatrefertovalues,
and this paper chose a statement from theWorld Value Survey – ‘aworkingmother can
establishasecureandwarmrelationshipwithherchildasamotherwhodoesnotwork’ ‐
whichwasstronglylinkedtochildcare.Otherquestionsorstatements,however,couldalso
be used as proxies and therefore test the robustness of my results. One that it is a very
28
relevant statement to childcare is the following: ‘A pre‐school child suffers if themother
works’.Theonlysetbackof thisstatement is thereducednumberofcountrieswhichhave
been asked the question. Yet, it would still be an interesting robustness check. Another
alternative‐andonethat fittheepidemiologicalapproachusedinthepaper‐wouldbeto
takethefemalelabourforceparticipationofthemigrantscountryofancestryinthepast.It
is assumed that this figure reflects the institutions and values of the country of ancestry.
Therefore,ifitcorrelateswiththefemalelabourforceparticipationofthesecondgeneration
migrantsintheanalysedcountry(Germanyinthiscase)it isagoodmeasureofthevalues
andattitudesofmigrantswhichinfluencetheirwillingnesstore‐enterthejobmarketaswell
asthetimeandcircumstances.Anotherpertinentissuetodiscussistheroleofthepartner’s
family values in shaping women’s attitudes towards work. Addressing this issue could
broadentheunderstandingoffemalelabour‐marketdecisionsandproviderobustnesstothe
results. The idea is based on the assumption that working decisions are not made by
individualsbutbyhouseholdunits,andthepartnerisakeycomponent,whosefamilyvalues
and attitudes towardswomen’swork can have asmuch influence toworking patterns of
mothers as their own values. A similar issue arises when the parents of the observed
individualcomefromdifferentcountriesofancestry,andthereforehavepotentiallydifferent
familyvalues.Toknowhowtheseonesare transmitted to the individual isadifficult task
and any decision involves some assumptions. So far, this paper has taken a conservative
approach by including only the second generation individuals whose parents come from
same country. Nonetheless, finding a way to include the individuals with parents from
differentcountriesofancestrycouldrevealinterestingresults.
To sumup, exploring both differentmethodologies and avenues to quantify family values
wouldbeagoodexercisetotesttherobustnessoftheresultsandtounderstandwhatfactors
are relevant fora childcarepolicy tobeeffectiveandaccelerate return toworkaswell as
boostingfemaleemploymentlevels.
29
VI. Conclusion
In this paper I suggest that the impact of labour‐market related policies is dependent on
familyvalues.Theempiricalresultsof thepolicyanalysedpartiallyconfirmthisargument,
findingthattheimpactofafamilypolicywhichattemptedtoacceleratethereturntoworkof
mothershasonlybeensignificantforhigh‐educatedmotherswithliberalfamilyvalues.The
mainconclusionsarethereforethreefold:firstly,itisrevealedthateconomicincentivesare
notsufficienttoacceleratereturnofmotherstowork.Secondly,thepapergoesinlinewith
the already existing evidence of the relevant role of education in shaping labour‐market
outcomes, andmore specifically, in shaping thepaceof return towork.At the same time,
however,thepapersuggeststhateducationisnotasufficientfactortosecurethesuccessof
thisfamilypolicy,andthatfamilyvaluesplayasignificantrole.
Policy implications can be drawn from these findings. Firstly, policies directed towards
accelerating the return towork ofmothers should concentrate not only on the economic
incentives, but they should consider other factors such as the level of education of the
mother. Secondly, the inclusionof family values increases theunderstandingof individual
behaviourwhenitcomestodecisionsaboutworkandleisureandcanthusimprovepolicy
effectiveness and ultimately labour‐market outcomes. Thirdly, the findings contribute
significantlytothedebateonpolicytransferabilitydebateintheEuropeanUnion.Thepaper
suggests that similar social and labour‐marketpolicies are likely tohave adifferent effect
dependingonthefamilyvaluesofthesocietyinwhichthepolicyisimplemented.
Finally, thepaperhas provideda discussiononmethodology andotherdata issueswhich
canenhancetherobustnessofresults.RDDhasproventobeagoodmethodalthoughthere
are some shortcomings which could be overcome with a complementary analysis of the
policyusingDDD.Furthermore,ananalysisofthekeyvariableofthepaper–familyvalues–
reveals that thereareotherwaysofproxying familyvalueswhichwouldbe interesting to
examine.
30
Thereisstillalongwaytogotofullyunderstandhowfamilyvaluesandpolicyinteract,but
hopefully this paper has provided some substantial evidence of the relevance of family
valuestolabour‐marketrelatedpolicies,whicharecurrentlyatthecentrestageofEuropean
politics.
31
Appendix
Appendix1
GSOEPsamplesrelation
Source:SOEPSamplesOverview–2011/Wave28 Appendix2–Codingofvariables
Dependentvariable
Two problems arise in the coding of the dependent variable. Firstly, given that the
interviewsweredone indifferentmonthsof theyears, thereexists thepossibility that the
questionisaskedbeforethewomenhashadachild,inwhichcasetheanswerwouldbe‘no
apply’.GiventhatIhaveinformationonwhethertheyhavehadachild(seeabove),ifthisis
thecaseIcheckattheanswerofthisquestionforthenextyear.Secondly,theremightbea
biggapbetweenthebirthdateandtheinterviewdate,especiallyinthecasewhereweget
theanswer to thedependentvariable fromthe followingyear.Thisposesan inconvenient
for thecodingof thedependentvariable if theanswer is ‘withinayear’.This isbecause if,
say,thegapbetweenthebirthandinterviewdateisof10months,ananswer‘withinayear’
shouldbecodedasslowreturn(asopposedto fastreturn),giventhat the totalamountof
timetherespondentwouldhavebeenonparentalleavewouldbetwoyears.Toaccountfor
this,Itakeinformationaboutthebirthandinterviewmonth,substractthemandforanswers
‘withinayear’ I lookatthedifferencebetweenbirthandinterviewrate.Ifthedifferenceis
32
elevenortwelvemonths,Icodetheanswerasslowreturn.Ifthedifferenceamountstoone
tosixmonths,Icodeitasfastreturn.Ifthedifferenceamountstoseventotenmonths,itis
quiteambiguous,soIdroptheobservation.
Codingofcountryoforiginsubjecttomigrationbackground
For those observations which have ‘direct migration background’, I look at the variable
‘countryoforigin’.Fortheobservationswith‘indirectmigrationbackground’theprocessto
tracebackthecountryoforiginismorecomplex.Firstly,Ilookatthevariable‘motherand
father countryoforigin’. If thisone isexistent, I checkwhether they come from the same
country.Whenthatisthecase,Icodetheobservationwiththepertinentcountryoforigin.
Whenthecountryoforiginforthefatherisdifferentfromthemother,Idroptheobservation
(furtheranalysiscouldbepursuedtotryto includetheseobservations). If themotherand
fathercountryoforiginisnotavailable,Ilookatthemotherandfatherpersonalnumber,I
look for themin thedatasetand I lookat thevariable ‘countryoforigin’or ‘nationality’, if
this one is not Germany. From 60 observations with ‘indirect migration background’, 43
haveaknowncountryoforigin.Fromthose,32havethesamecountryoforiginfrommother
andfather.
33
Bibliography
Aghion,P.,Algan,Y.,&Cahuc,P.(2011).CivilSocietyandtheState:TheInterplay
betweenCooperationandMinimumWageRegulation.JournaloftheEuropean
EconomicAssociation,9(1),3‐42.
Aghion,P.,Algan,Y.,Cahuc,P.,&Shleifer,A.(2010).RegulationandDistrust.Quarterly
JournalofEconomics,125(3),1015‐1049.
Alesina,A.F.,Algan,Y.,Cahuc,P.,&Giuliano,P.(2010).FamilyValuesandtheRegulation
ofLabor.NationalBureauofEconomicResearchWorkingPaperSeries,No.15747.
Algan,Y.,&Cahuc,P.(2007).TherootsoflowEuropeanemployment:Familyculture?
NBERInternationalSeminaronMacroeconomics2005,65‐+.
Banfield,E.C.(1958).Themoralbasisofabackwardsociety.NewYork:FreePress.
Barro,R.J.,&McCleary,R.M.(2003).Religionandeconomicgrowthacrosscountries.
AmericanSociologicalReview,68(5),760‐781.
Bergemann,A.,&Riphahn,R.T.(2011).Femalelaboursupplyandparentalleave
benefits‐thecausaleffectofpayinghighertransfersforashorterperiodoftime.
AppliedEconomicsLetters,18(1),17‐20.
Blum,S.(2012).GermanyCountryNote.InD.Erler(Ed.).
Borck,R.(2011).AdieuRabenmutter‐TheEffectofCultureonFertility,FemaleLabour
Supply,theGenderWageGapandChildcare.CESifoWorkingPaperSeriesNo.
3337.
Esping‐Andersen,G.(1990).Thethreeworldsofwelfarecapitalism.Cambridge:Polity.
34
Esping‐Andersen,G.(1999).SocialfoundationsofpostindustrialeconomiesAvailable
from
https://shibboleth2sp.sams.oup.com/Shibboleth.sso/Login?entityID=https://lse
.ac.uk/idp&target=https://shibboleth2sp.sams.oup.com/shib?dest=http://www.
oxfordscholarship.com/SHIBBOLETH?dest=http://dx.doi.org/10.1093/0198742
002.001.0001
Featherstone,K.,&Radaelli,C.M.(2003).ThepoliticsofEuropeanization.Oxford:Oxford
UniversityPress.
Ferge,Z.,&Kolberg,J.E.(1992).SocialpolicyinachangingEurope.FrankfurtamMain
Boulder,Colo:CampusVerlag;
WestviewPress.
Fernandez,R.(2007).AlfredMarshallLecture‐Women,work,and,culture.Journalof
theEuropeanEconomicAssociation,5(2‐3),305‐332.
Fernandez,R.,Fogli,A.,&Olivetti,C.(2004).Mothersandsons:Preferenceformation
andfemalelaborforcedynamics.QuarterlyJournalofEconomics,119(4),1249‐
1299.
Ferrera,M.(1996).The'SouthernModel'ofWelfareinSocialEurope.Journalof
EuropeanSocialPolicy,6(1),17‐37.
Fleckenstein,T.(2011).ThePoliticsofIdeasinWelfareStateTransformation:Christian
DemocracyandtheReformofFamilyPolicyinGermany.SocialPolitics,18(4),
543‐571.
Green,D.P.,Leong,T.Y.,Kern,H.L.,Gerber,A.S.,&Larimer,C.W.(2009).Testingthe
AccuracyofRegressionDiscontinuityAnalysisUsingExperimentalBenchmarks.
PoliticalAnalysis,17(4),400‐417.
Guiso,L.,Sapienza,P.,&Zingales,L.(2004).Theroleofsocialcapitalinfinancial
development.AmericanEconomicReview,94(3),526‐556.
35
Guiso,L.,Sapienza,P.,&Zingales,L.(2009).CulturalBiasesinEconomicExchange?
QuarterlyJournalofEconomics,124(3),1095‐1131.
Haisken‐DeNew,J.P.a.M.H.(2010).PanelWhiz:EfficientDataExtractionofComplex
PanelDataSets,anExampleUsingtheGermanSOEP.JournalofAppliedSocial
ScienceStudies,130(4),643‐654.
Ichino,A.,&Maggi,G.(2000).Workenvironmentandindividualbackground:Explaining
regionalshirkingdifferentialsinalargeItalianfirm.QuarterlyJournalof
Economics,115(3),1057‐1090.
Kluve,J.T.,M.(2009).NowDaddy'sChangingDiapersandMommy'sMakingHerCareer.
RuhrEconomicPapers(145).
Leitner,S.(2010).GermanyoutpacesAustriainchildcarepolicy:thehistorical
contingenciesof'conservative'childcarepolicy.JournalofEuropeanSocialPolicy,
20(5),456‐467.
Naldini,M.(2003).ThefamilyintheMediterraneanwelfarestate.London,Portland,Ore.:
FrankCass.
Reher,D.S.(1998).FamilytiesinwesternEurope:Persistentcontrasts.Populationand
DevelopmentReview,24(2),203‐+.
Roland,G.(2004).Understandinginstitutionalchange:Fast‐movingandslow‐moving
institutions.StudiesinComparativeInternationalDevelopment,38(4),109‐131.
Tabellini,G.(2010).CultureandInstitutions:EconomicDevelopmentintheRegionsof
Europe.JournaloftheEuropeanEconomicAssociation,8(4),677‐716.