1

PAA2016AnnualMeeting,Washington

Moreinequality,moreviscosity?

IntergenerationalmobilityinEuropeandtheUS

LouisChauvel1 AnneHartung

Draftversion,March07,2016

AbstractThepopularity of the so-calledGreatGatsbyCurve (GGC,Blanden2013,Corak2013, Torche 2015) that describes a strong macro relation between incomeinequality(Gini)andintergenerationalearningselasticityisdemonstratedbytheincreasing number of studies investigating this relationship. Still, the exactmechanism underlying the association between income inequality andintergenerational immobility is not yetwell understood (McCall andPercheski2010:339).This studywill contribute (a) to filling this gapby complementingthe “economic” income-based regressions approachwith a “sociological” class-based categorical modelling approach (Xie 2003; Pisati’s unidiff model, PisatiandSchizzerotto2004)and(b)totesttherobustnessoftheGreatGatsbyCurveby proposing an innovativemethodological approach. Applying this approachcan be useful for the current methodological debate on imputation methods(typically two sample two stages least squares TSTSLS) that could produceupward-biased intergenerational elasticities, leading to an overestimation GGCrelation.Toavoidthisbias,weuseanewmeasureofelasticitynetofGini(logitrankofincome)toreproducetheGGCwithamorerobustapproach.Besides, concerns about international comparability of themethods used havebeenraised;ourapproachisimputedinthecontextofamultilevelrandomslopemodel where the country-specific slopes (BLUPS Best Linear UnbiasedPredictor) are identified as ameasure of socioeconomic reproduction. A third,empirical contribution of this paper lies using a harmonised methods acrosscountriesandinwideningtheevidencebasebyincludingalargesetofcountiesnotincludedinpreviouspublications:InadditionaltotheUS(PSID),weincluded26 European countries (EU Survey on Income and Living Conditions, EU-SILC,module 2005 and 2011 on Intergenerational transmission ofpoverty/disadvantage).The results confirmour expectations. First, the resultsbasedonaGini-neutralmeasure are more modest in terms of explained variance of the GGC but arestrongandsignificant:theeconomictraditionalthoughtheassociationbetween1UniversityofLuxembourg.Correspondingauthor:Louis.Chauvel@uni.lu

2

Gini and intergenerational reproduction remain rather solid. Secondly, in linewiththeliterature(Blanden2013,Torche2015),wefindthatthe“sociological”approach differs considerably from the “economic” one in the sense that thelatter has a strong result in terms of impact of inequality on (im)mobility.Wescrutinizethediscrepancyfromananalyticalperspectiveandconcludewithnewchallengesforsociologicalresearch.

Acknowledgement

Wewould like to thankMaurizioPisati forprovidingamodifiedversionofhisStataunidiff.adoprogram.Introduction

The linkage of static and dynamic income inequalities has experienced anupsurgeintheeconomicliteratureoverthelastdecade.Mostnotably,thefamed“GreatGatsbyCurve”(Corak,2013,Figure2)mapping22countriesonagraphwith income inequality(Gini)onthexaxisandgenerationalearningselasticityon the y axis has provokedmanydiscussions,which spilled over to the policydomain. Inbrief, the “GreatGatsbyCurve”withaR-squaredof .75anda slopecloseto1suggeststhatmorestatic inequalityisassociatedwithmoredynamicinequality (intergenerational reproduction) or in otherwords, themore equalsocieties in thepast, themoreequalityofopportunity today.Thenatureof therecent controversial academic debate around the “Great Gatsby Curve” and itsreliability is in essence methodological discussing the different approaches ofestimating intergenerational mobility (panel methods versus imputationstrategies),whosegenerallogicswewillbrieflydiscussbelow.AlthoughmanyauthorsconfirmtheGGCconcluding“Moreinequality,lesssocialmobility”(Andrews and Leigh 2009; Blanden 2013, Corak 2013, Torche 2015),relatively little is known about the link between income inequality andintergenerational mobility (Jäntti and Jenkins 2013) and especially about themechanisms behind (Jerrim and Macmillan 2015). On top of this, the relativelevel of intergenerational mobility differs depending on the measure used:income, educationor social class (Blanden2013,Torche2015). Critiqueshavebeen developed on some of these approaches2and these results are often

2 The widely used two samples two stages least squares (TSTSLS) approach seems tooverestimate the slope (intergenerational elasticity) and lacks robustness (Jerrim et al 2014).Onekey issue is that the father’s income ismostlynotavailableandthereforeestimatedonanauxiliarydataset.Theflawinherenttothismethodisthatthelog(income)isdependentonGini:ThehighertheGini,themorestretchedthedistribution-eveniftheintrinsicregimeofmobilityisnotaffected.ThischangeinGiniisastructuraltransformation.However,whenGinirises,alsothe elasticity increases. A second line of criticism concerns of uncertainty about thecomparabilityof intergenerationalearningsmobilityacrosscountrieshavebeenraised. Jerrim,Choi, and Rodríguez (2014) for instance point out that in the production (supposedly) cross-nationally comparable findings different empirical methodologies have been applied and

3

debatedamongeconomists,whilesociologistsremainlessactiveinthisrespect.This calls for further research, in particular including a larger set of differentcountries.Our aim is to contribute to these debates with this analysis of the wellharmonizedEU-SILCdatausingacombinedmeasureofparentalsocio-economicorigin,whichwetransformintoarankscaleandlinkthisranktotheindividual’ssocio-economicpositioninasociety–firstintermsofincomeandthenintermsofsocialclass.Inthisway,weareabletoassessthesensitivityofpriorresearchand gain additional information on the interrelation of intergenerationalmobilityandinequality.

Economists’viewontheassociationbetweenstaticanddynamicinequality

Intergenerational elasticities (IGE) have been estimatedwith two strategies inthe pertinent literature. The first “ideal” strategy relies on administrative orpaneldata comprising– if sufficiently long– actual informationon income (orearnings)ofparentsinthepastandchildrentodayrespectively(Jäntti&Jenkins,2013) – but is rarely feasible. This first approach uses ordinary least squares(OLS) techniques to estimate the son’s income based on the father’s as apredictor(Nicoletti&Ermisch,2008).Wherepaneldataandthusactualinformationontheparents’incomeinthepastis not available, two-sample strategies have been applied (Björklund & Jäntti,1997; Lefranc, 2011; Lefranc & Trannoy, 2005) (Andrews and Leigh 2007)matchinginformationonchildren’s incomeandparents’characteristicssuchaseducation,experienceandoccupation fromtwodifferentsources (Jerrimetal.,2014).Thesestudiesestimatefirsttheearningsequationoftheparentsbasedonauxiliary data representing the parent’s income distribution obtainingcoefficientsofthese incomedeterminants. Inasecondstep, thecoefficientsareusedtopredicttheincomeoftheparentbasedonthemaindata(representativefor the adult children), using the socio-economic characteristics of theparentsreportedby thechildren(Torche,2013).Someauthorsprefer torefer to theseapproachesasimputationmethods(Jerrimetal.,2014).Inherent to the traditional GGC approach are three main problems that havebeen emphasized by different scholars: (a) the cross-national comparabilitywhen applying two different strategies, (b) the overestimation of estimatedintergenerational elasticity and thus the GGC, and (c), whichwill be ourmainargumenthere,theinabilitytoaccountforastructuraltransformationssuchasanincreasingGiniasforinstancewitnessedintheUS.Wewillnextelaborateonargument(b)and(c)inmoredetail.In a nutshell, the argument (b) adheres to the risk that estimates ofintergenerationalelasticitydifferaccordingtothemethodapplied.Containinginadditiontotheassociationbetweenoriginanddestinationalsothenetimpactof

emphasizetheneedtoproduce“morerobustandreliableestimatesofearningsmobilitythatcanbelegitimatelycomparedacrosscountries”(Jerrimetal.2014:22).

4

the (instrumental) variable used to predict parental income, the two-samplemethodsoverestimate the trueslopeandproduceanupwardbiasedestimatedelasticity(Jerrimetal.,2014).Theestimatorshouldthereforebeinterpretedasanupperbound for the intergenerationalelasticity (Nicoletti&Ermisch,2008;Torche, 2013). 3 When applying two-sample strategies, the variance in theparents’ distribution decreases inflating estimated intergenerational elasticity(Andrews&Leigh,2009).4SincetheGGCmixesresultsbasedonthefirstandthesecond estimator (11 out of 21 included countries have applied the TSTSLSmethodology),criticsclaimthat thekey findingsdono longerholdwhenusingthesameapproachforallcountries(Jerrimetal.,2014).5Argument(c)islinkedtothisfeature.StructuraltransformationssuchasarisingGinimay(upward)biastherelationdisplayedintheGGCifitisbasedonlogofincomeasit isusuallythecase.WhenGinirises,alsotheelasticity increasesasour simulation based on the PSID (in which the parental income is available)shows(Table1).Table1:Twosamplesleastsquaresestimationofcurrencyelasticityafter

rescalingoffathers’andsons’Gini(US)

Note:WesimulatechangesintheGinisforfathersandsonsbasedwiththeformulaaln(p/1-p)=ln(medianisedincome)wherea=Gini(Fisk-Champernowne-Dagumdistributions).Source:PSID.

Table 1 displays intergenerational measures based on currency elasticity (indollars)andtheirdependencyonchangesintheGini:whentheGiniofthesonsincome distribution is larger than the Gini among the fathers’ incomedistribution,theintergenerationalelasticityincreases.6The central problem TS2SLS methodology is thus that (pseudo)-fathersestimates (or averages) are imputed and create a non-realistic distribution offathers where Gini(fathers)<<Gini(sons). This intergenerational Gini gap islarger in countrieswithhigherGini of the sons’ incomedistributionand could

3The larger R-squared in the first-stage regression, the smaller this bias (Nicoletti & Ermisch,2008).4On the contrary, the intergenerational correlation (IGC) is a measure of intergenerationalpersistence,whichadjustsfordifferencesinincomevariancebetweentheparentsandchildrenandisthusamorerobustmeasureofintergenerationalmobility(BjorklundandJantti2009).Asincomesofbotharerequired,whichisoftennotavailable,thecorrelationisonlyrarelyincluded(Andrews & Leigh, 2009 as an exception). Moreover, it cannot reflect nonlinearities in theintergenerational economic association (Corak, Lindquist, & Mazumber, 2014; Torche,forthcoming). Yet, informationonboth the elasticityand the correlation is desirable (Blanden,2013).5In addition, Jerrim et al. (2014) stipulate that theTSTSLS approach should not be applied incross-nationalcomparisonssince thepoorTSTSLSproxiesofparents’earningsproducebiasedcoefficientsofintergenerationalassociations.6WhenbothGinisareequal(diagonal),thisissuedoesnotoccur.

5

thereforepotentiallyaccountforpartsofthelog-incomeGGC.Inotherwords,wehave to find a measure of exchange mobility, net of Gini change (structuraltransformation) to test this empirically. The main risk of the conventionalelasticityapproach is thentogeneratetrivialresults: incomegapsarestronger(elasticity)whereincomegapsaredeeper(Gini).Asociologicalperspective

Sociologists investigating inequality of opportunity through a social class lenshave long been providing ambiguous statements on the link between theintensityofeconomic inequalityandintergenerational immobility(EriksonandGoldthorpe 1992, Goldthorpe 2013). In a European comparisons of socialmobility,theteamofBreen(2004,Figure3)findsnorelationbetweenfluidityofnationsandtheirrankingsintermsofGiniindices.Recentresearch(Mitniketal2016) making use of occupational based information and unidiff log-linearmodels (Xie 1992) show the increase in inequality in the US seems to haveenabledtheprofessional-managerialclasstobetterprotecttheirinterestinclassreproduction and maintain their positions in the upper social hierarchy overgenerations. But this kind of approach is not very common in sociology,comparedtothemassiveuseofbigdatabasedlong-termresearchineconomics(Chettyetal2014).A general issue is thus the contradictory evidence on the association of socialreproductionandinequalitydependingontheconceptused.Findingsbasedonregressions of continuous measures such as earnings and income do notnecessarilycoincidewiththosebasedoncategoricalmeasuressuchasclassandoccupation due to their conceptual difference (Neckermann and Torche 2007,Torche 2015). The class approach groups the set of occupations into a fewdiscrete strata, which are not only defined by income but rather otherdifferentiatingcharacteristicssuchasindustrialsectororauthority.Within-classdispersionofincomemaythusvaryandthusalsothedegreeofintergenerationalcorrelation(Blanden2013).However,anapproachtoeconomichierarchyshouldberedevelopedinsociology(Pareto 1896; Nielsen 2007), in particular in the domain of intergenerationalmobility (Girod 1986). For sociologists, the GGC hypothesis is interesting intermsofsocialclasstheory.IfhigherGiniindicesindeedgowithstrongerincomeintergenerational elasticity (socio-economic reproduction), this has importantimplications for social class structure: stronger inequality means a higherpredictability of children’s socioeconomic position when parents’ are known.Otherwise,itmeansthatthepercentageofpredictedvarianceofincomesofkids(predictedbyparents’position)mightincreasewhenGiniindicesarehigher.Inadiscretemodel,theoverlapsbetweensocioeconomicoriginsshoulddeclinewithinequality;conversely,equalitariancountriesarethenexpectedtoshowmassiveoverlapsbetweensocialorigins.

6

Figure1: Possible scenarios of changes in the class structure as result of

increasinginequality(wideningofincomedistribution)

Source:ownillustrationFigure 1 illustrates possible outcomes of changes in the class structure whenincomeinequalityisrising:Comparedtothebaselinesituation,scenario1refersto a stretching of the whole structure, similar to an elastic loom, so that theclassescanincreasetheirincome(moveupward)butalsoproportionallyspreadtheirvariance.Still,asitisthecaseinthebaselinesituation,personsfromlowerclassesmayhavehigherincomesthansomepersonsfromthenexthigherclass.Scenario 2 is a “telescopic” change, in the sense that the spread (variance) ofeachclassremainsthesamebutthedifferenceorgapbetweenthemincreases;the ratio of the predicted variance (by origins) by total variance is increasingcompared to scenario 1, and the relative overlaps decline. In scenario 2,intergenerationalelasticityincreasesdramatically.In a nutshell, we will answer the following questions in view of the differenteconomic and sociological concepts: Canwe confirm the link between incomeinequality(Giniindex)andintergenerationalmobilitynetofstructuralchangesinthe shape of distribution? Even with Gini independent measures of mobility?What is the difference between “economic” approaches (income based) and“sociological”ones(occupationalclass)andhowcanitbeexplained?Methods

Incomeversusincomeranks

Many authors have underlined the importance of using rank positions whenanalyzing the income distribution (Chauvel, 2015; Ebert, 1999; Jenkins & VanKerm, 2006; Mujcic & Frijters, 2013; Van Kerm, 2004; Chetty and al 2014).

0

25

50

75

100

125

150

Lower class

Medium class

Higher class

Lower class

Medium class

Higher class

Lower class

Medium class

Higher class

Inc

om

e (

cu

rre

nc

y)

Baseline Scenario 1 Scenario 2

7

Comparing the income distribution with an “elastic band” where a merestretching signifies an increasing inequality (Gini) in a given population, nopositional changes should occur. When calculated in terms of incomes, anincreasingGinireferstoastretcheddistribution,even if the intrinsicregimeofmobility is not affected (Jäntti & Jenkins, 2013). The change in Gini indiciesmeasure thus structural transformation of inequality. This structural mobilitydiffers from exchange mobility, the re-ranking within a population, throughwhichinequalityisnotaffected.Using ranks avoids the issue with the conventional currency elasticitymeasurement, i.e. that intergenerationalelasticity isvery reactive to inequalitytrends:Wheninequality(Gini)rises,elasticityincreases.“Changesintheextentof mobility mostly reflects the evolution of cross-section earnings inequality,ratherthanvariationsinpositionalmobility.”(Lefranc,2011:1)Thismechanismmayaccount fora largeextentofassociationshown in theGreatGatsbycurve.Lefranc shows this for the evolution of intergenerational mobility but this isproblematic for international comparisons as well. We will therefore apply arankbasedelasticity.Thepercentilerankbasedelasticity

We could standardize the variables as simple ranks or as fractional ranks(between 0 and 1), which would, however, contract the Pareto-tails. Analternativewouldbe toreshape thedistributionsof incomeasanormalcurve,but once again we lose the properties of Pareto tails specific to income andwealth distributions. The logitrank approach offers a standardization strategyconsistent with the Pareto characteristics of income distributions that haveimportantproperties.Weproceedas follows.Letp∈[0;1]bethepercentilerankof individual i in theincome distribution, so that the logged odds of the percentileln (�!/(1− �!)measures the relative social power of individual i (“Logit rank” (Copas, 1999),O’Brien, 1978; compare also to the log of Positional Status Index used byRotman, Shavit and Shalev 2015). Based on the so-created rank positions, wederivetheGiniα,ameasureofthedegreetowhichthesepositionsarestretchedbetween the top and the bottom, from the following equation based on theChampernowne-I-Fisk quantile distribution (Champernowne, 1953; Chauvel,2015;Dagum,1977;Fisk,1961): ln �! = α ln (�!/(1− �!))or�! = � �! where the medianized income�! = �!/������,�! = ln(�!) = ln (�!/������),andthelogitrank�! = logit �! = ln (�!/(1− �!)).(Chauvel,2015:3)Let us introduce percentile rank based elasticity (RE).While the Great Gatsbycurve is basedonCurrency-Elasticity (CE)7,which isGini-dependent, i.e.whentheGinirises,CEincreases,weavoidthisissuebyreplacingtheyearlycountry-

7TheCEisbasedonloggedincomessothatinflationandgrowthisabsorbedbytheconstant.

8

logged incomes by logit rank�! = logit �! = ln (�!/(1− �!)). This rank basedelasticityisindependentoftheGiniinparents’orchildren’sgenerationandthusa measure of mobility net of structural changes in both distributions and isthereforeamoresuitablestrategytocomparedifferentcountries.8Multilevelmodel

Inordertoobtaincountry-specificgradientsofsocioeconomicoriginlogitranksonchildrenincomelogitrank,weapplyamultilevelrandominterceptsrandomslopesmodel(Rabe-Hesketh&Skrondal2012)withthefollowinggeneralform:

withrandomcountry-specificinterceptζ1jandrandomcountry-specificslopeforparentalbackgroundβ2+ζ2jforthe52subsamplesincluded(26countriesattwotimepoints).Underthisspecification,β2+ζ2jcanbeunderstoodasthestrengthofthe impactofparents’ relativeposition in thesocioeconomichierarchyof theirnationon their childrenachievements.Thismodelhasalreadybeenapplied tosocioeconomic gradient (logitrank of both socioeconomic origin and ego’sincome)ofhealthonthesamedatasets(Chauvel&Leist2015)todemonstratethat higher Gini indices increases social origin’s role on inequalities of health.The GGC hypothesis goes with the existence of a strong correlation betweennationalβ2+ζ2jandtheGiniindices.The52(countryxyear)β2+ζ2jandtherelatedstandarderrorsareestimated ina singlemultilevelmodelwith thebest linearunbiasedpredictions(BLUPs),thencorrelatedtothe52Giniindices.The dependent variable is here the “logit rank” (at the country-year level),loggedoddofthepercentileln(p/1-p)intheincomedistribution(post-tax,post-transfer, disposable household income). The main explanatory variable is thelogit rank of the scores in the Multiple Correspondence Analysis (MCA, Burtmethod)ofeducationandoccupationofthefatherandofthemother(6classesEGPscheme)againatthecountry-yearlevel(seebelow).Loglinearassociationmodel

For analyzing social mobility in terms of social classes, we apply log-multiplicativelayermodel(Xie1992,2003),whichisamoreparsimoniouslog-linearmodelcompared to thesaturatedmodel thatsuitswell researchdesignsthatareinterestedincountry-orlayer-specificmobility(Xie1992).

with parental versus offspring’s social class across 52 layers. Note, that weassumethusthatthequalitativepatternofmobilityissimilaracrosscountriesastherow-column-layereffectisconstrainedhere.

8If the income inequality in all countries does not change between the two generations,log(income) based strategies are equally suitable. Using PPPwill also not solve the issue thatinequitymeasuredbytheGinidiffersamongparentsandtheirchildren.

!!" = !! + !!! + !! + !!! !!" + !!" !

9

Dataandvariables

OuranalysisdrawsontheUSPanelStudyofIncomeDynamics(PSID)andtheEUSurvey on Income and Living Conditions (EU-SILC). While the PSID containsgenerally information on the parent’s socio-economic origin and the familycontext,fortheEU-SILCcomparableinformationexistsonlyforthewaves2005and 2011, which includes modules on the “intergenerational transmission ofpoverty/disadvantages” that comprises a set of questions on the parents’educationandoccupation.Weincludethefollowing27countriesinouranalysis:Austria, Belgium, Czech Republic, Cyprus, Germany, Denmark, Estonia, Spain,Finland, France, Greece, Hungary, Ireland, Iceland, Italy, Lithuania, Latvia,Luxembourg, the Netherlands, Norway, Poland, Portugal, Slovenia, Slovakia,Sweden,theUKandtheUS.AmajoradvantageofusingtheEU-SILC is theavailabilityofa largenumberofharmonized variables. On the other hand, the EU-SILC has the limitation ofobtaining its data with different sampling strategies across the countries. Inmany Nordic countries, which are characterized by low inequality and highmobility, administrative data is complementedwith additional forms filled outby citizens (e.g. NL, SE), leading to high rates of non-response and a potentialbiasthatisnotcorrectedbytheavailableweights.Todatethereisnoevidenceoftheexistenceofsuchapotentialdesignartifactandmanyotherresearchersusethe EU-SILC for intergenerational mobility analyses (some excluding certaincountries).In order to obtain our parental origin variable, we combine the maximumamountofinformationavailableontheparents’socio-economicbackgroundandconstructascalewithMultipleCorrespondenceAnalysis(MCA;Burtmethod)ofeducation and occupation of the father (6 classes EGP scheme) and mother’seducation and occupation. For robustness checks we tried differentcombinations of these variables, and the results donot diverge iswe focus onmotheror fatheralone.The firstdimensionof theMCAweuse forour furtheranalysisexplains88.6%inertia(predictfactorcoordinates),thismeansthatthedifferent dimensions of hierarchy (occupations, education of parents) arestronglyrelated.Scoresarelogit-rankedbycountry-year.OnedataissuewiththeEU-SILC, however, is that the high percentage of missing information on theparents’background,inparticularinSweden(morethan50%ofmissingvalues).Forconsistencywithotherstudies,weexcludethosemissingcases.The harmonization of the PSID andEU-SILC variables across generations doesunfortunatelynotallowforgreatdetail.Education(ISCED)offatherandsonhasbeenharmonized into three levels: (1) low:pre-primary,primaryeducationorlower secondary education, (2)medium:upper secondary education andpost-secondarynon-tertiaryeducation,and(3)high: firststageoftertiaryeducationand second stage of tertiary education. However, as the comparability ofeducational qualifications is limited across generations and countries, thesevariablesmustbeinterpretedwithcaution.

10

Results

The aim of this section is to replicate the “Great Gatsby Curve” using theassociation between social origin rank and children’s income rank rather thantheintergenerationalelasticityusedintheeconomicliterature.Beforedoingso,we investigate the link between the father’s origin and his children’s outcomewiththelogit-rankapproachacrossthedifferenttypesofwelfarestatesbasedonEbbinghaus’(2006)typology(Figure2).Figure2:Theorigingradientindifferenttypesofwelfarestates

Source:EU-SILC2005/2011,PSID2005.Consistentwith the sociological literature the degree of social reproduction islowest intheNordiccountries.CentralEuropeanwelfarestatesandevenmoreso theother types exhibit amuchhigherdegreeof origindependence.TheUSseparatedfromtheEuropeanwelfarestatesexhibitsthesteepestorigingradientandthusgreatestintergenerationalpersistenceofallwelfarestatesinvestigated.

In a next step, we run a multilevel regression with random intercepts andrandom slopes to obtain the country-specific origin gradients. The estimatedbest linear unbiased predictions (BLUPs) of the random effects showing thevariation for both the intercept and the estimated beta coefficient(s) aredisplayedinFigure3.

-2-1

01

2D

estin

atio

n: L

og

itra

nk o

f in

co

me

-5 -4 -3 -2 -1 0 1 2 3 4 5Origin

Nordic Centre

Anglo Latin

Postsocialist US

11

Figure3:TheorigingradientacrossEUcountries

Notes:Best linearunbiasedpredictor (BLUP)basedonrandomcoefficientnullmodel.Labels:lowercases-2005(e.g.uk),uppercases-2011(e.g.UK).Source:EU-SILC2005/2011.Figure3allowsdetectingmoredetailsinthecountrydifferencesandexceptionsfrom theprevious figure.TheNordic countriesexhibitagainamuch less steepslope suggesting a lower degree of social reproduction than the rest of thecountries investigated here. The Eastern European and Southern countries,whichhaveoftenbeenexcludedfromtheGGC(JerrimandMacmilan2015),fallmostly on the other side of the spectrum with rather large degrees of origindependence. However, Slovakia (2005) and Lithuania (2011) as compared tootherEasterncountries,hasamuchhighersocialmobility,whichiscomparabletothoseofAustria,Belgium(2005)andtheUK(2005)forinstance.ThelargestorigindependencecanbefoundinHungaryandLuxembourg(both2011),whichisevenhigherthanintheUS.Ifweplotthesecountry-specificgradientsagainsttheGini(Figure4),weobtainourkeyresult,thelogit-rankbasedGreatGatsbyCurve,thatcorrectsforchangesin the income distributions among fathers and sons. Two observations can bemade: first, the conclusion compared to the log-income based GGC (Figure 4)remains relatively stable: themore inequality, themore viscosity. Second, theexplained variance (see Annex A.1) is lower using our method (Rsquared(M1)=.343) compared to the log-income based approach (Rsquared(M3)=.661) indicating that the changes in Gini do indeed explain anenormouspartoftheassociationbetweeninequalityandsocialreproductiononthemacrolevel.

12

Figure4: Logit-rankbasedGGC: (a) Europe and theUS and (b) excluding

post-socialistcountries

R2=.343 R2=.570

Source:EU-SILC2005/2011Figure5:Log-incomebasedGGC:(a)EuropeandtheUSand(b)excluding

post-socialistcountries

R2=.661 R2=.773

Source:EU-SILC2005/2011Ifweparalleltheresultsbasedonsocialclass/unidiff“sociological”andlogitrankincome based/elasticity “economic” concepts (Figure 6), not surprisingly,obvious differences appear. The link to economic inequality is not givenwhenlooking at social reproduction in terms of EGP classes instead of income. Rsquaredclosetozeroconfirmsthis.Inotherwords,therelationdescribedintheGreatGatsbycurvedoesnotholdforintergenerationalclassmobility.Thisisinlinewithfindingsofotherstudiescomparingincomeandsocialclassapproachestointergenerationalmobility(Blanden2013,Torche2015).Inthiscomparison,anaïveconclusionwouldbethatthemuchstrongerGGCcurveofthe“economic”approach compared to the “sociological” one suggests it is more powerful toconsider incomes than social class. The gradient of social origin on today’sincome hierarchy is stronger in more unequal countries. This goes with theresult that the intensity of social class reproduction is not correlated witheconomic inequality. The rigidity of the economic structure is correlated toeconomicinequality.Inequalitariancountries,thedifferentsocioeconomiclevelsoforiginlooselypredictchild’sincomeandtheyoverlapmassivelyinthisrespect(see Figure 1) and conversely in unequal countries parents’ relative position

at2005AT2011

be2005

BE2011 cy2005

CY2011cz2005

CZ2011

de2005

DE2011

dk2005

DK2011

ee2005EE2011es2005 ES2011

fi2005FI2011

fr2005FR2011

gr2005GR2011

hu2005

HU2011

ie2005IE2011

is2005

IS2011

it2005

IT2011lt2005

LT2011

lu2005

LU2011

lv2005LV2011

nl2005

NL2011

no2005

NO2011

pl2005

PL2011

pt2005

PT2011

se2005

SE2011

si2005SI2011

sk2005

SK2011

uk2005

UK2011

US2005

-.2

-.1

0.1

.2

.2 .25 .3 .35 .4gini

at2005AT2011

be2005

BE2011 cy2005

CY2011

de2005

DE2011

dk2005

DK2011

es2005 ES2011

fi2005FI2011

fr2005FR2011

gr2005GR2011

ie2005IE2011

is2005

IS2011

it2005

IT2011

lu2005

LU2011

nl2005

NL2011

no2005

NO2011

pt2005

PT2011

se2005

SE2011

uk2005

UK2011

US2005

-.2

-.1

0.1

.2

.2 .25 .3 .35 .4gini

at2005AT2011be2005

BE2011 cy2005

CY2011cz2005

CZ2011

de2005

DE2011

dk2005

DK2011

ee2005EE2011

es2005ES2011

fi2005FI2011

fr2005 FR2011

gr2005GR2011hu2005

HU2011

ie2005IE2011

is2005

IS2011

it2005IT2011

lt2005

LT2011

lu2005

LU2011

lv2005LV2011

nl2005

NL2011

no2005

NO2011

pl2005PL2011

pt2005

PT2011

se2005SE2011

si2005SI2011

sk2005

SK2011uk2005

UK2011

US2005

0.05

.1.15

.2 .25 .3 .35 .4gini

at2005AT2011be2005

BE2011 cy2005

CY2011

de2005

DE2011

dk2005

DK2011

es2005ES2011

fi2005FI2011

fr2005 FR2011

gr2005GR2011

ie2005IE2011

is2005

IS2011

it2005IT2011lu2005

LU2011

nl2005

NL2011

no2005

NO2011

pt2005

PT2011

se2005SE2011

uk2005

UK2011

US2005

0.05

.1.15

.2 .25 .3 .35 .4gini

13

predictsbetterchild’sincome,andsubgroupsofsocialoriginsoverlaplessonthechildrenlogitrankincomescale.Theoccupationalsocialclasssocialmobilitytableanalyzedbytheunidiffmodelsuppresses this correlation. This may seem paradoxical at first sight, but at acloser look the explanation is that within each social class of destination, thericherarethekidsofricherparents:higherlevelincomemedicaldoctorsarethechildrenofprofessionals, for instance,whilepublicsectorgeneralpractitionerscouldbemoreoftenthekidscomingfromincome-modestsocialmilieus.Figure 6: Income-based/elasticity (logitrank) GGC (top) vs. social class-

based/unidiff GGC (bottom), Europe with and without post-socialist

countriesR2=.343 R2=.570

R2=.012 R2=.079

Source:EU-SILC2005/2011,USandpost-socialistcountriesexcluded.Interesting, however, is the link between the both concepts: the correlation ofthe residuals (in the simple regression of elasticity on Ginis) of both is ratherstrong(r=.645whenpost-socialistscountriesareincluded),indicatingthatbothconceptsarestronglylinked.Investigationtheresiduals,i.e.whichcountriesfallabove and below the fit line in Figure 6, gives an idea ofwhich countries aremore or less prone towards viscosity as expected given their Gini. The strongcorrelation of the residuals of elasticity versus unidiff approachesmeans thatrelatively to a Gini level, fluid and viscous countries of both approaches aresimilar.Relativelyhigherviscositywith respect to income ranks canespeciallybe found in Hungary, Luxembourg and many Eastern European countries forinstance,whilerelativelylessorigindependenceprevailsintheNordiccountriesgiventheir(loworhigh)levelofinequality.Withrespecttosocialclass,thereare

at2005AT2011

be2005

BE2011 cy2005

CY2011cz2005

CZ2011

de2005

DE2011

dk2005

DK2011

ee2005EE2011es2005 ES2011

fi2005FI2011

fr2005FR2011

gr2005GR2011

hu2005

HU2011

ie2005IE2011

is2005

IS2011

it2005

IT2011lt2005

LT2011

lu2005

LU2011

lv2005LV2011

nl2005

NL2011

no2005

NO2011

pl2005

PL2011

pt2005

PT2011

se2005

SE2011

si2005SI2011

sk2005

SK2011

uk2005

UK2011

US2005

-.2

-.1

0.1

.2

.2 .25 .3 .35 .4gini

at2005AT2011

be2005

BE2011 cy2005

CY2011

de2005

DE2011

dk2005

DK2011

es2005 ES2011

fi2005FI2011

fr2005FR2011

gr2005GR2011

ie2005IE2011

is2005

IS2011

it2005

IT2011

lu2005

LU2011

nl2005

NL2011

no2005

NO2011

pt2005

PT2011

se2005

SE2011

uk2005

UK2011

US2005

-.2

-.1

0.1

.2

.2 .25 .3 .35 .4gini

at2005

AT2011be2005

BE2011

cy2005

CY2011cz2005

CZ2011

de2005

DE2011dk2005

DK2011

ee2005

EE2011

es2005

ES2011

fi2005

FI2011

fr2005

FR2011 gr2005

GR2011

hu2005HU2011

ie2005

IE2011

is2005

IS2011

it2005

IT2011

lt2005

LT2011

lu2005

LU2011

lv2005LV2011

nl2005NL2011

no2005

NO2011

pl2005PL2011

pt2005

PT2011

se2005

SE2011si2005SI2011

sk2005

SK2011

uk2005UK2011

.2.3

.4.5

.6

.2 .25 .3 .35 .4gini

at2005

AT2011be2005

BE2011

cy2005

CY2011

de2005

DE2011

dk2005

DK2011

es2005

ES2011

fi2005

FI2011

fr2005

FR2011gr2005

GR2011

ie2005

IE2011

is2005

IS2011

it2005

IT2011

lu2005

LU2011

nl2005

NL2011

no2005

NO2011

pt2005

PT2011

se2005

SE2011

uk2005UK2011

.3.35

.4.45

.5.55

.2 .25 .3 .35 .4gini

14

afewdifferences.LuxembourgagaindisplaysarelativelyhigherviscositygiventhelevelofGinibutnowalsomanyCentralEuropeancountriescanbefoundonthis side of the spectrum. Relatively more open with respect to the classstructuregiventheGinilevelareIcelandandtheAnglo-Saxoncountries.Thesefindingssuggestinanutshellthatsocialclassmobilitydepictsadifferentlinktoinequalitythanincome-basedmobilityduetointergenerationalearningspersistence within social classes. In other words, in countries with highinequality,inagivenclassofdestiny,descendantsofhigher-classoriginsobtainhigherincomesthanthosefromlower-classorigins.Conclusions

In this paper, our aimwas first to reproduce theGGCwith amore robust andincome-distributionneutralmeasurethatisindependentofeconomicinequality,namely Gini. The risk of the conventional approach is to obtain trivial results:incomepredictsstrongergapswheregapsbetweenincomesarestronger.Therethe logitrank approach offers a standardization strategy consistent with thePareto characteristics of incomedistributions. Second,we tested if the provenrelationbetweenGiniand intergenerationalmobilityalsoholds forsociologicalconceptsofreproduction.Wecontributetotheexistingliteratureinseveralways.First,weincludealargesetofnewcountriesvis-à-visCorak(2013)makinguseofharmormisedinsteadof country-specific data comprising the US (PSID) and Europe (EU-SILC).Second, the use of Gini-independent measure of social mobility (vis-à-visincome) allows us scrutinize the macro-relation between economic inequalityand intergenerational mobility and feed the methodological debate with newinsights.OurresultsconfirmandrefinetheGreatGatsbyCurve:WefindamuchmoremodestassociationwiththeGini-independentlogitrankmeasureweuse.Theassociation is thuspartlydueto theGini itself,or inmoretechnical terms,theupwardbiasoftheintergenerationalelasticitybasedonlogofincome.This study, third, contributes to the interdisciplinary literature by providingadditional insights based on both economic and sociological concepts. In linewiththefewpreviousstudies(Blanden2013,Torche2015),socialclassmobilityshows a different link to inequality. Our explanation is the intergenerationalearningspersistencewithinsocialclassesisthatincountrieswithhigheconomicinequality, inagivenclassofdestiny,kidsofhigher-classoriginsobtainhigherincomes than those from lower-class origins. Reflecting different processes,sociological and economic concepts need to be seen in conjunction to betterreflectonthelinkofsocio-economicinequalityandsocialviscosity.Sociologicaland economic concepts reflect in other words utterly different processes andfuture research will profit much from investigating both traditions inconjunction.

15

References

Aaberge,Rolf,Björklund,Anders,Jäntti,Markus,Palme,Mårten,Pedersen,Peder

J, Smith,Nina,&Wennemo,Tom. (2002). Income inequalityand incomemobility in the Scandinavian countries compared to the United States.ReviewofIncomeandWealth,48(4),443-469.

Andrews, Dan, & Leigh, Andrew. (2009). More inequality, less social mobility.AppliedEconomicsLetters,16,1489-1492.

Björklund,Anders,& Jäntti,Markus. (1997). Intergenerational IncomeMobilityin Sweden Compared to the United States. The American EconomicReview,87(5),1009-1018.doi:10.2307/2951338

Blanden, Jo. (2013). Cross-national rankings of intergenerational mobility: acomparison of approaches from economics and sociology. Journal ofEconomicSurveys,27(1),38-73.

Champernowne, D. G. (1953). A Model of Income Distribution. EconomicJournal(63),318–351.

Chauvel, L. (2015). The Intensity and Shape of Inequality: TheABGMethod ofDistributional Analysis. Review of Income and Wealth, n/a-n/a. doi:10.1111/roiw.12161

Chauvel L, AK Leist, 2015, Socioeconomic hierarchy and health gradient inEurope: theroleof incomeinequalityandofsocialorigins, Internationaljournalforequityinhealth14(1),132.

Chetty, Raj, Nathaniel Hendren, Patrick Kline, Emmanuel Saez, and NicholasTurner. 2014. Is the United States still a land of opportunity? Recenttrends in intergenerational mobility. NBER Working Paper 19844,Cambridge,MA.

Copas, J. (1999). The Effectiveness of Risk Scores: The Logit Rank Plot QuickView. Journal of the Royal Statistical Society, Series C (AppliedStatistics)(48),165–183.

Corak, Miles. (2013). Inequality from Generation to Generation. In R. Rycroft(Ed.), The Economics of Inequality, Poverty, and Discrimination in the21stCentury(pp.107-123).SantaBarbara,CA:ABC-CLIO.

Dagum,C. (1977).ANewModel ofPersonal IncomeDistribution: SpecificationandEstimation.EconomieAppliquée(30),413–436.

Ebbinghaus,Bernhard(2006)ReformingEarlyRetirementinEurope,JapanandtheUSA.Oxford:OUP.

Ebert,U. (1999). "Usingequivalent incomeofequivalentadults torank incomedistributions."SocialChoiceandWelfare(16):233-258.

Erikson,Robert,andJohnH.Goldthorpe.1992.Theconstantflux:Astudyofclassmobilityinindustrialsocieties.Oxford:ClarendonPress.

Fisk, P. R. (1961). The Graduation of Income Distributions. Econometrica(29),171–185.

GirodR.,1986,Evolutiondesrevenusetmobilitésociale(àGenève1950-1980),Genève-Paris,Droz.

Goldthorpe, John H. (2013). Understanding–and misunderstanding–socialmobility in Britain: The entry of the economists, the confusion ofpoliticians,andthelimitsofeducationalpolicy.JournalofSocialPolicy42(3):431–50.

16

Hout, M. (2015). "A Summary of What We Know about Social Mobility." TheANNALSoftheAmericanAcademyofPoliticalandSocialScience657(1):27-36.

Jäntti,Markus,&Jenkins,StephenP.(2013).Incomemobility.Jenkins, S. P. and P. VanKerm (2006). "Trends in income inequality, pro-poor

income growth and income mobility." Oxford Economic Papers 58( 3):531–548.

Jerrim,J.andL.Macmilan(2015)."IncomeInequality,IntergenerationalMobility,andtheGreatGatsbyCurve:IsEducationtheKey?"SocialForces.

Jerrim,John;ÁlvaroChoiandRosaSimancasRodríguez.2014.‘Two-sampleTwo-Stage Least Squares (TSTSLS) estimates of earnings mobility: howconsistentarethey?’WorkingPaper.

Lefranc,Arnaud,&Trannoy,Alain. (2005). IntergenerationalEarningsMobilityinFrance:IsFranceMoreMobilethantheUS?Annalesd'ÉconomieetdeStatistique(78),57-77.doi:10.2307/20079128

Lefranc, Arnaud. (2011). Educational expansion, earnings compression andchanges in intergenerational economicmobility : Evidence from Frenchcohorts,1931-1976.ThemaWorkingPaper(n°2011-11).

McCall, L. and C. Percheski (2010). "Income Inequality: New Trends andResearchDirections."AnnualReviewofSociology36(1):329-347.

Mitnik, P. A., et al. (2016). "Social Mobility in a High-Inequality Regime." TheANNALSoftheAmericanAcademyofPoliticalandSocialScience663(1):140-184.

Mujcic, R. and P. Frijters (2013). "Economic choices and status: measuringpreferencesforincomerank."OxfordEconomicPapers(65):47-73.

Neckerman,K.M.andF.Torche(2007)."Inequality:CausesandConsequences."AnnualReviewofSociology33(1):335-357.

Nicoletti,C.,&Ermisch, J. (2008). Intergenerationalearningsmobility:ChangesacrosscohortsinBritain.B.E.JournalofEconomicAnalysis&Policy,7(2),1-36.

Nielsen,F., "EconomicInequality,Pareto,andSociology:TheRouteNotTaken",AmericanBehavioralScientist,50,619–38,2007.

Pareto,V.,"Lacourbedesrevenus",LeMondeeconomique,juillet,99–100,1896.Pisati,M.,&Schizzerotto,A. (2004).The Italianmobility regime:1985-1997 In

Breen, Richard. (2004). Social mobility in Europe. Oxford: OxfordUniversityPress.

Rabe-Hesketh, S. andA. Skrondal (2012).Multilevel and longitudinalmodelingusingStata,StataPress,CollegeStation,TX.

Rotman,A.,etal.(inpress)"NominalandpositionalperspectivesoneducationalstratificationinIsrael."ResearchinSocialStratificationandMobility.

Torche, Florencia. (2013). How do we characteristically measure and analyzeintergenerationalmobility?Working Paper, Stanford Center on PovertyandInequality.

Torche, Florencia. (2015). Analyses of Intergenerational Mobility: AnInterdisciplinary Review. Annals of the American Political and SocialSciences,657:37-62.

Torche, Florencia. 2015. "Intergenerational Mobility and Equality ofOpportunity,"EuropeanJournalofSociology56(3):343-371.

17

Van Kerm, P. (2004). "What lies behind income mobility? Reranking anddistributional change in Belgium, Western Germany and the USA."Economica(71):223–239.

Xie, Yu. (1992). "The Log-Multiplicative Layer Effect Model for ComparingMobilityTables."AmericanSociologicalReview57:380-395.

Xie,Yu.(2003)."AssociationModel."Pp.30-33inEncyclopediaofSocialScienceResearchMethods,editedbyMichaelLewis-Beck,AlanBrymanandTimFutingLiao.ThousandOaks,Ca:Sage.

Xie, Y., et al. (2016). "Between- andWithin-Occupation Inequality: TheCase ofHigh-Status Professions." The ANNALS of the American Academy ofPoliticalandSocialScience663(1):53-79.

Annex

A.1Explainedcountry-levelvarianceofthethreeapproachesapplied

Logitrankapproach

(Figure3)Logincomeapproach(Figure4)

Socialclassapproach(Figure5)

M1-allcountries

M2-excludingEasternEU

M3-allcountries

M4-excludingEasternEU

M5-allcountries

M6-excludingEasternEU

Gini 1.006*** 1.375*** 0.546*** 0.640*** -0.168 0.447 (0.178) (0.158) (0.055) (0.059) (0.240) (0.246)Intercept -0.289*** -0.408*** -0.099*** -0.129*** 0.439*** 0.271*** (0.057) (0.048) (0.016) (0.017) (0.067) (0.067)r2 0.343 0.570 0.661 0.773 0.012 0.079N 53 37 53 37 52 36