research group on human capital working paper series · social mobility trends in canada: going up...
TRANSCRIPT
Research Group on Human Capital Working Paper Series
Social Mobility Trends in Canada: Going up the Great Gatsby Curve
Working Paper No. 19-03
Marie Connolly, Catherine Haeck, and David Lapierre
May 2019 (revised version)
https://grch.esg.uqam.ca/en/working-papers-series/
SocialMobilityTrendsinCanada:GoinguptheGreatGatsbyCurve
MarieConnolly*,CatherineHaeckandDavidLapierre
GroupederecherchesurlecapitalhumainUniversityofQuebecinMontreal
Thisversion:May27,2019
AbstractWhile cross-sectional increases in inequality are a cause for concern, the study of theintergenerational transmissionof socioeconomic status isperhapsmore relevant.How issocialstatusreproducedfromonegenerationtothenext?Recentworkhashighlightedtherelationship, if not causal then correlational, between inequality andmeasures of socialmobilityinacross-countrysetting.ThisrelationshipisdubbedtheGreatGatsbyCurve(Corak2013): places with higher inequality during one’s childhood are correlated with lowerintergenerationalincomemobilitybetweenthechildandhisorherparents.Inthispaper,newlydevelopedadministrativeCanadiantaxdataareexploitedtocomputemeasuresof intergenerational incomemobility at the national and provincial levels. Thisworkprovidesdetaileddescriptiveevidenceonthetrendsinsocialmobility.Resultsshowthatmobility has steadily declinedover time, and that there has been an increase in theinequalityof theparental incomedistribution, asmeasuredby theGini coefficient.HenceCanada, and all its provinces, have been “going up” the Great Gatsby Curve. The crosssectional,crosscountryrelationshipthusalsoholdswithinasamecountryovertime,leadingcredencetothemorecausalthancorrelationalnatureoftherelationship,thoughcausalityisnot formally tested here. The decrease in mobility, particularly for children born in thebottomquintileof the incomedistribution,shouldbeofconcernto federalandprovincialpolicymakersalikeandhighlightstheneedforadditionalresearchinordertoprovideequalopportunitiestoallchildren.JEL:J62,D63Keywords:socialmobility,intergenerationaltransmissions,incomeinequality,GreatGatsbycurve,Canada*Correspondingauthor,connolly.marie@uqam.caTheauthorswouldliketothanktheFondsderechercheduQuébec-Sociétéetculturefortheirfunding(grant2016-PU-195586) and the Social Analysis and Modelling Division at Statistics Canada, in particular YuriOstrovskyandGrantSchellenberg,formakingthisworkpossible.CristianStraticaprovidedsuperbresearchassistance.Theauthorsalso thankparticipantsat theCRDCNandSCSEconferencesandat seminarsat theMontreal Applied Micro Group, McGill University, Dalhousie University and the University of Ottawa forcomments.Allerrorsremaintheirown.TheanalysispresentedinthispaperwasinpartconductedatStatisticsCanada’sFederalResearchDataCentreandattheQuebecInteruniversityCentreforSocialStatistics,whichispartoftheCanadianResearchDataCentreNetwork(CRDCN).TheservicesandactivitiesprovidedbytheQICSSaremadepossiblebythefinancialorin-kindsupportoftheSocialSciencesandHumanitiesResearchCouncil(SSHRC), the Canadian Institutes of Health Research (CIHR), the Canada Foundation for Innovation (CFI),StatisticsCanada,theFondsderechercheduQuébec-Sociétéetculture(FRQSC),theFondsderechercheduQuébec-Santé(FRQS)andtheQuebecuniversities.Theviewsexpressedinthispaperarethoseoftheauthors,andnotnecessarilythoseoftheCRDCNoritspartners.
1
1. Introduction
InCanadaasinmostotherpartsoftheworld,incomeinequalityisontherise.Whilecross-
sectionalincreasesininequalityareacauseforconcern,thestudyoftheintergenerational
transmission of socioeconomic status is perhaps more relevant. How is social status
reproducedfromonegenerationtothenext?Recentworkhashighlightedtherelationship,
ifnotcausalthencorrelational,betweeninequalityandmeasuresofsocialmobilitysuchas
the intergenerational elasticity of income in a cross-country setting. This relationship is
dubbed theGreat Gatsby Curve (Corak 2013): placeswith higher inequality duringone’s
childhoodare correlatedwith lower intergenerational incomemobilitybetween the child
andhisorherparents.Thisisparticularlyworrisomeinthecontextofincreasinginequality:
are children from low-income backgrounds ever more likely to remain in a low-income
situationasadultsthaneverbefore?
ResearchintheUnitedStatesreachesdivergentconclusionsdependingonthedatasource
andmethods:Chettyetal.(2014a)usetaxdataandfindthattherank-rankrelationshiphas
notchangedbetweenthe1971and1982birthcohorts,whileDavisandMazumder(2017)
usetheNationalLongitudinalSurveysanddocumentdeclinesinintergenerationalmobility
between cohorts born 1942-53 and 1957-64 in both the rank-rank relationship and the
intergenerationalelasticityofincome.Upuntilrecently,thistypeofanalysiswasimpossible
inCanadaduetothelackofsuitabledata.Thishaschangedwiththerecentadditionofnew
birth cohorts to Statistics Canada’s Intergenerational Income Database (IID). The earlier
birthcohorts,bornfrom1963to1970,featuredinCorakandHeisz(1999)’sgroundbreaking
workonintergenerationaltransmissionofincome.Thelaterbirthcohortsnowstretchupto
1985.
In this paper, all available birth cohorts in the IID are exploited to computemeasuresof
intergenerationalincomemobilityandlookattrendsatthenationalandprovinciallevels.
Thesampleissplitintofivesuccessivebirthcohorts,spanningfrom1963to1985.Whilethe
focus is on rank mobility, shown to be more stable over the lifecycle than the
intergenerational elasticity, and on quintile transitions matrices, intergenerational
elasticities are also presented. Child income is measured over five years for three non-
2
overlappingagegroups:whenthechildis25to29,30to34,and35to39yearsold.Late
thirtieshavebeenfoundtobemorerepresentativeoflifetimeincome,buttheyoungerbirth
cohort, born from1982 to 1985,has not reached those ages yet. Parental income is also
measuredoverfiveyears,whenthechildisaged15to19yearsold.
Findings show thatmobility has steadily declined over time, nationally and provincially.
Relativemobility,asmeasuredbytherank-rankcorrelation,wentfrom0.189forchildren
born1963-66to0.234forthoseborn1982-85.Lookingattransitionmatrices,theprobability
thatachildfromthebottomquintileoftheparentalincomedistributionstaysinthebottom
quintileofhisorherincomedistributionwentfrom27.1%upto32.6%,withacorresponding
declineoftheprobabilityforthosekidstoreachmiddleclass.Theresultsarethusmorein
linewithDavisandMazumder’sfindingsofdecliningmobilityintheUnitedStatesthanwith
Chettyetal.’sfindingsofstablecopulabetweenparentalandchildincomedistributions.
Paralleltothisdeclineinmobility,findingsalsodocumentanincreaseintheinequalityofthe
parentalincomedistribution,asmeasuredbytheGinicoefficient.HenceCanada,andallits
provinces,havebeen“goingup”theGreatGatsbyCurve.Thecrosssectional,crosscountry
relationshipthusalsoholdswithinacountryovertime,leadingcredencetothemorecausal
thancorrelationalnatureof the relationship, thoughcausality isnot formally testedhere.
ThisstudyprovidesdetaileddescriptiveevidenceonthetrendsinsocialmobilityinCanada,
exploitingnewlyavailableadministrativetaxdata.Thedecreaseinmobility,particularlyfor
children born in the bottom quintile of the income distribution, should be of concern to
federalandprovincialpolicymakersalike,andhighlightstheneedforadditionalresearchon
the mechanisms behind the decline in mobility in order to understand which policy or
programwouldstandthebestchanceatimprovingequalityofopportunitiesforallchildren.
Theremainderofthepaperisstructuredasfollows.Sections2and3containinformationon
themethodologyand thedata.The findingsarepresented inSection4. Section5offersa
discussionandconcludes.
3
2. Methodology
Thegoalistomeasuretheintergenerationaltransmissionofincome,orthedegreetowhich
an individual’s incomedependsonthatofhisorherparents.The focusofthestudy ison
incomebecausethedatacontaindetailed,reliable,andlongitudinalinformationonincome
forsuccessivecohortsofyouthandtheirparents,whichallowsthecomputationofstatistics
bothovertimeandacrossspace.Whileotherdimensionsofsocialmobilitymaybeimportant
or interesting, such as occupation, education, or cultural capital, no administrative data
source in Canada contains the necessary information to study those dimensions. The
intergenerational elasticity (IGE) of income has been widely used in the literature as a
measureofmobility(orlackthereof),includingstudiesbasedonearlyIIDcohortsinCanada
such as Corak and Heisz (1999) and Chen et al. (2017). The IGE of income is typically
computed by estimating a linear model where the natural logarithm of child income is
explainedbythenaturallogarithmofparentalincome.However,duetothepresenceoflogs,
asDahlandDeLeire(2008),Chettyetal.(2014b)andConnolly,CorakandHaeck(2019)have
noted,IGEestimatesaresensitivetothetreatmentofverysmallvaluesofincomeinaway
thatmeasuresofrankmobilityarenot.Rankmobilityisdefinedinmuchthesamefashionas
the IGE, except that child and parental income aremeasured as percentile ranks in their
respectiveincomedistribution.Sincerankmobilityprovestobemuchmorerobusttothe
treatmentoflowincomesandacrossmodelspecifications,thisanalysisoftrendsinsocial
mobilityisbasedonrankmobility,whichiscomputedbyestimatingthefollowingmodel:
𝑅",$% = 𝛼% + 𝛽%𝑅"*+,$% + 𝜀$%, (1)
where𝑅",$% is the income rank of child i—the t generation—in province p,𝑅"*+,$% is the
incomerankofhisorherparent(s)—thet−1generation—,and𝜀$%isarandomterm.The
slopefromEquation(1),𝛽%,canbeinterpretedasameasureofintergenerationalmobility:
thehigherthe𝛽%,themoreparentalincomerankexplainschildincomerank,andtheless
mobilitythereis.Equation(1)isestimatedbothatthenationalandprovinciallevel,where
geographical location is fixedduring teenageyears.The ranks canbe computed from the
nationaldistributionortheprovincialdistribution,andestimatesusingbothtypesofranks
willbepresented.MoredetailsonthedatafollowinSection3.
4
ModelsbasedonEquation(2)belowarealsoestimated:
𝑙𝑛/𝑌",$%1 = 𝑎% + 𝑏%𝑙𝑛/𝑌"*+,$%1 + 𝑒$%, (2)
where𝑌",$%ischildi’sfromprovinceptotalincome,𝑌"*+,$%ishisorherparents’income,and
𝑒$%isarandomterm.Theslopeofthismodel,b,istheintergenerationalelasticityofincome.
Another typeofmobilitymeasure that canbe computed is the transitionmatrix.Quintile
transitionmatricesareused,whereeachcolumnreferstoaquintileoftheparentalincome
distribution, each row to a quintile of the child income distribution, and each cell is a
conditionalprobability.Conditionalprobabilitiesaredenoted𝑃6,7:theprobabilityofmoving
fromorigino(parentalincomequintile)todestinationd(childincomequintile),orPr(child
income𝜖quintiled|parentalincome𝜖quintileo).AsinCorak(2017)andConnolly,Corak
andHaeck (2019), the focus isona fewkeypointsof thetransitionmatrix:𝑃+,+,𝑃+,9, and
𝑃+,:*;.Theintergenerationalcycleofpoverty,𝑃+,+,capturestheprobabilityforachildraised
inabottomquintileofthefamilyincomedistributiontoremaininthebottomquintilehimor
herselfasanadult.Therags-to-richesmovement,𝑃+,9,istheprobabilitytoescapethebottom
quintileandattainthetopincomequintile.Finally,𝑃+,:*;,measurestheprobabilitytoleave
the bottom incomequintile and reach themiddle three quintiles,which could be loosely
definedasthemiddleclass.
3. The Intergenerational Income Database
This study isbasedonall current cohortsof StatisticsCanada’s Intergenerational Income
Database. This database, first developed in the mid 1990s (see Corak and Heisz, 1999),
containslongitudinaladministrativetaxfilesofsuccessivecohortsofchildreninCanadaand
theirparents.TheoriginalIIDcohorts,1982,1984and1986,coveredchildrenbornbetween
1963and1970,inclusively.MorerecentcohortsarenowavailableintheIID,11991,1996
and 2001, coveringmost birth years between 1972 and 1985. In the IID, each cohort is
identifiedusingthefirstfiscalyearinwhichthelinkbetweentheparentsandthechildis
1TheauthorsthanktheFondsderechercheduQuébec-SociétéetcultureforfundingthatallowedthenewcohortstobeaddedtotheIID(grant2016-PU-195586).
5
attempted.Forexample, the1982cohort includeschildrenbornbetween1963and1966,
whowerematchedtotheirparentsasof1982,whentheywere16to19yearsold. If the
matchcouldnotbeestablishedin1982,thematchwasattemptedagainoverthenextthree
years. In the current study, toavoidoverlappingbirthyearsacrossdifferent cohorts, the
1984IIDcohortisdividedintwo,poolinghalfwiththe1982IIDcohortandtherestwiththe
1986IIDcohort,dependingonbirthyear,andthenremovingduplicates.Table1givesthe
birthyearsandnumberofobservationspercohortasdefinedforthispaper.Thereareover
onemillionchildrenperbirthcohort,foratotalofclosetosixmillionchild-parentspairs.
Table1:TheIntergenerationalIncomeDatabasecohortsIIDcohort Birthyears IIDcount IIDweightedcount
1982-84 1963to1966 1,219,470 1,566,2401984-86 1967to1970 1,158,900 1,555,2801991 1972to1975 1,095,160 1,474,1401996 1977to1980 1,166,440 1,557,8002001 1982to1985 1,349,190 1,633,270
Source:Authors’calculationsbasedontheIID
IntheIID,taxfilesareavailableannuallystartingin1978forthe1982,1984and1986IID
cohortsandstartingin1981forthemorerecentcohorts.Thelastavailableyearoftaxdata
for all cohorts is currently 2014. From those tax files, a variety of incomemeasures are
availableforeachindividualofthefamily:boththemotherandthefather,thechild,aswell
asthespouseofthechildwhenreportedonthechild’staxreturn.Theanalysisisbasedon
total income from all sources, as defined by the Canada Revenue Agency. This includes
earnings, interest and investment income, self-employment net income, taxable capital
gains/lossesanddividends,andbenefits.Inthemainanalysispre-taxtotalincomeisused;
after-taxincomeisalsoconsideredinonepartoftheanalysis.
Parentalincomeisdefinedastheaverageannualparentalincomewhenthechildisaged15
to 19 years old, including bothmother and father if present. Taking a five-year average
reducesbiasesresulting fromtransitoryshocksto income.A five-yearwindowduringthe
lateteenageyearsisusedtoreflectresourcesavailabletothechildduringthosecrucialyears
ofhumancapitalaccumulationandtransitionsbetweensecondaryandpostsecondaryschool
orbetweenschoolingandthelabormarket.Theseagesarealsotargetedbecausetheyare
6
the closest to the year of the family link in the tax files, thus ensuring that the family
circumstancesthatapplytothechild’slifeareeffectivelycaptured.Becausebiologicallinks
arenotidentifiedintheIIDandbecauseonlytheadultsidentifiedasthechild’sparentsat
themomentofthelinkcanbetracked,theearliertheperiodconsideredinthechild’slife(e.g.
earlychildhood),thelowerthecertaintythattheactualfamilyofthechildatthatmomentof
hisorherlifeiscapturedaccurately.Anotherreasontouseparentalincomewhenthechild
is15to19yearsoldisthatsincethetaxfilesstartin1978,dataonparentalincomeduring
earlychildhoodforthefirstbirthcohortsarenotavailablesuchthatcomparabilityacross
cohortswouldbeachallenge.Futurestudiescouldtrytoinvestigatetheeffectofparentalage
atchildbirthonmeasuresoftotalincomerank.
Comparabilityacrosscohortsbecomesanimportantissuewhendecidinghowtocompute
the child incomemeasure. If five-year averages are used for child income, as isdone for
parents,thenthelatestthatcanbestudiedinthechild’slifeforthelastcohort,thoseborn
between1982and1985,isfromtheagesof25to29yearsold.Achildbornin1985willturn
29yearsoldin2014,thelastyearofdataavailableatthetimeofwriting.Theseages,25to
29,arecomparabletowhatChettyetal.(2014b)useintheirmainanalysis(ages29to32)
andinsensitivitychecks(ages26to29).Tostudytrendsoverthelongestperiodpossible
usingtheIID,five-yearaveragesfromage25to29areusedinthemainanalysis.Childincome
averagesarealsocomputedfortheagesof30to34and35to39yearsold.Thefirstfourof
thefivecohortsareusedwhenchildincomeiscapturedbetweentheagesof30and34years
old,andonlythefirstthreeofthefivecohortsareusedifagesisconstrainedto35to39years
old.Notethatwheneveraparentorachildisnotfoundinthetaxfilesforagivenyear,an
income of $0 is imputed for that year. The sample is then restricted to children that are
observedatleastonceoverthefive-yearperiod,andtoparentsandchildrenthathavean
averageincomeofatleast$500.Thistreatmentiscoherentwiththesampleselectiondone
byCorak(2017).
Alldollarfiguresareconvertedto2016CanadiandollarsusingtheAll-itemsConsumerPrice
Index(CANSIMTable326-0021).Oncethefive-yearaveragesarecomputedforbothchild
andparentalincome,percentileranksarecomputedusingthenationaldistributionformost
of the analysis, and the provincial distribution for some complementary analysis. The
7
provinceofresidenceisfixedatthetimeofthelinkbetweenthechildandtheparents,i.e.
whenthechildisaged16to19.Thismeansthatifachildmovesoutaprovincebythetime
hisorherincomeismeasured,thatchildisstillassignedtotheplacewhereheorshegrew
up,aswasdoneinChettyetal.(2014b),Corak(2017)andConnolly,CorakandHaeck(2019).
Studyingtheeffectofgeographicalmobilityonincomemobilityisleftforfurtherstudies.
Table2presentssomedescriptivestatisticsforeachbirthcohortinthesamplesstudied.The
toppanelsshowsstatisticsontheparents.Averageparentalincomeincreasedovertime,as
diditsstandarddeviationaswelltheGinicoefficientscomputedusingbothbefore-taxand
after-taxincome,areflectionofincreasingincomeinequality.Thenumberofparentslinked
toachilddependsonwhofiledataxreturnontheyearoftheparent-childmatch(i.e.when
thechildis16to19):ifonlyoneparentfiledataxreturn,thenonlythatparentisobserved
intheIID.Forthe1963birthcohort,thefractionofchildrenforwhomonlyoneparentfiled
taxeswas30.8%.Thisfractiondoesnotreallyreflectthemaritalarrangementofthechild’s
parents:single-earnerhouseholdsmaybeamarriedcouple,but ifonlyonespouse filesa
return,thentheyarereportedhereasa“one-parenttaxfiler,”despitebeingmarried.This
situationismorelikelytohappenformarriedcouplesinwhichoneofthetwoparentsdoes
not work, typically the mother. The rise in women’s labour force participation and the
increase indual-earner families (StatisticsCanada,2016)explains the substantialdrop in
one-parenttaxfilerfractionovertime,to21.8%,despitetheincreaseinlone-parentfamilies
documented elsewhere (Statistics Canada, 2015). That said, theway the parental income
variableisconstructedinthisanalysisisnotproblematic:parentalincomesimplyrepresents
householdincomearoundthetimeoftheparent-childmatch,thusreflectingtheresources
availabletothehouseholdthatthechildlivedin.Ifahouseholdhasoneparentearnerand
tax filer, then only that parent’s income counts towards household income,whether it is
becausetheparentisaloneparent,oramarriedparentinasingle-earnerhousehold.
Table2:Descriptivestatistics
BirthcohortVariable 1963 1967 1972 1977 1982Averageparentalincome(beforetax) $78,800 $77,700 $82,100 $81,200 $89,200 (84,300) (82,800) (91,800) (104,200) (167,300)
8
Ginicoefficient,parentalbefore-taxincome 36.360 37.773 39.267 41.177 44.374Ginicoefficient,parentalafter-taxincome 34.081 34.699 36.039 37.461 40.526Fractioninone-parenttaxfilerhousehold 0.308 0.296 0.248 0.231 0.218Averagechildincome Ages25-29 $34,100 $32,000 $34,100 $35,300 $36,700
(23,100) (24,200) (27,800) (27,000) (30,100)Ages30-34 $42,100 $43,600 $46,100 $47,700 —
(44,300) (130,600) (44,800) (41,600) Ages35-39 $51,500 $53,600 $56,500 — —
(73,400) (77,600) (63,500) (73,400) Ginicoefficient, Ages25-29 38.774 41.337 41.497 41.333 43.323childbefore-taxincome Ages30-34 45.150 46.844 46.085 45.723 —
Ages35-39 48.860 50.459 49.273 — —Ginicoefficient, Ages25-29 31.179 32.733 33.113 33.503 35.244childafter-taxincome Ages30-34 35.310 36.779 36.237 36.143 —
Ages35-39 38.886 39.796 38.686 — —Fractionfemale(ages25-29) 0.490 0.487 0.489 0.489 0.491Fractionsingle,children Ages25-29 0.511 0.491 0.525 0.558 0.579
Ages30-34 0.308 0.311 0.320 0.350 —Ages35-39 0.246 0.252 0.264 — —
N(excludingparentsandchildren Ages25-29 1,469,010 1,437,190 1,357,900 1,442,150 1,501,800withincomeunder$500and Ages30-34 1,421,820 1,386,020 1,321,720 1,408,010 —childrenmissingallyears) Ages35-39 1,393,430 1,364,920 1,301,780 — —Sampleselection-parents Fractionmissingatleastonce 0.075 0.081 0.109 0.093 0.110Fractionwithincomeunder$500 0.018 0.022 0.018 0.017 0.022Sampleselection-children Fractionmissingallyears Ages25-29 0.029 0.038 0.045 0.043 0.044
Ages30-34 0.050 0.064 0.069 0.065 —Ages35-39 0.068 0.080 0.083 — —
Fractionmissingatleastonce Ages25-29 0.199 0.240 0.245 0.245 0.247Ages30-34 0.225 0.248 0.249 0.241 —Ages35-39 0.230 0.252 0.247 — —
Fractionwithincomeunder Ages25-29 0.015 0.018 0.019 0.017 0.018$500 Ages30-34 0.026 0.027 0.021 0.018 —
Ages35-39 0.027 0.025 0.021 — —Note:Standarddeviationsinparentheses.AllfiguresareweightedusingIIDweights.Source:Authors’calculationsbasedontheIID
ThesecondpanelofTable2presentsstatisticsthatrelatetothechild.Similarlytothetrends
observedfortheparents,incomeinequalityisontherise,asseenthroughtheincreaseinthe
Ginicoefficients.Thebottomtwopanelsinvestigatethesampleselection.Atmost2.2%of
parentsand2.7%of childrenareexcludedbecause their total income isunder$500.The
9
fractionofchildrenthatarecompletelymissingfromthetaxdataovertherelevantfive-year
intervalincreaseswiththeageofthechild,goingfrom3-4%atages25to29,to5-7%atages
30to34,to7-8.3%when35to39yearsold.Thedeathrateisstillverylowatthoseages,so
the attrition most likely reflects outmigration. A nontrivial fraction of both parents and
childrenhaveatleastoneyearoftaxfilesmissing:upto11%fortheparents,andupto25.2%
for the children. In robustness analyses, the sample is further restricted to parents and
children that are observed all years over the five-yearwindow, and similar findings are
obtained.
4. Findings
Thissectionpresentsthefindingsfromthisstudy.Foreaseofpresentation,birthcohortsare
referred to by the first birth year of the cohort (e.g. 1963 birth cohort designates the
1963−1966 birth cohort). The first subsection presents the main findings regarding
intergenerational mobility at the national level, including results on rank mobility,
intergenerational elasticities, anddifferencesby gender of the child.Nonlinearity in rank
mobilityisexplorednext.Then,rankmobilityattheprovinciallevelisinvestigated,followed
by trends in the transitionmatrices. The next subsection presents a version of theGreat
GatsbyCurve,linkinginequality(theGinicoefficient)torankmobility.Robustnessanalyses
areshownlast.
4.1. Mobility at the national level
Mobility estimates at the national level are presented first. Table 3 shows the estimated
slopesfromtherankmobilityEquation(1),bybirthcohortandageatwhichchildincomeis
measured,forsonsanddaughterscombinedandseparately,aswellastheintergenerational
elasticityestimatesfromEquation(2).Ahigherrank-rankslopeorahigherIGEmeansthat
parentalincomehasahigherexplanatorypoweronchildincome,andthusmobilityislower.
Table3:Mobilityestimates
Birthcohort 1963 1967 1972 1977 1982
Rank-rank
10
All ages25-29 0.189 0.188 0.216 0.215 0.234ages30-34 0.201 0.214 0.232 0.235 —ages35-39 0.201 0.214 0.230 — —
Sons ages25-29 0.188 0.187 0.214 0.219 0.233ages30-34 0.227 0.237 0.252 0.253 —ages35-39 0.238 0.247 0.257 — —
Daughters ages25-29 0.204 0.200 0.228 0.216 0.238ages30-34 0.193 0.205 0.223 0.224 —ages35-39 0.182 0.194 0.214 — —
Intergenerationalelasticity(IGE) All ages25-29 0.153 0.154 0.199 0.210 0.224
ages30-34 0.171 0.184 0.223 0.239 —ages35-39 0.180 0.192 0.232 — —
Sons ages25-29 0.154 0.157 0.200 0.215 0.223ages30-34 0.192 0.202 0.238 0.251 —ages35-39 0.214 0.221 0.255 — —
Daughters ages25-29 0.165 0.160 0.206 0.210 0.227ages30-34 0.166 0.178 0.219 0.234 —ages35-39 0.161 0.174 0.219 — —
Note:Standarderrorsareall0.001orlowerfortherankmobilityestimatesandfortheIGEforsonsanddaughterscombined.FortheIGEforsonsanddaughtersseparately,standardserrorsareallbelow0.002.Source:Authors’calculationsbasedontheIID
Aclearpatterntowardsdecreasingintergenerationalmobilitycanbeobservedwhenmoving
fromtheearlierbirthcohortstothelaterbirthcohorts:therank-rankslopehasbeensteadily
goingup,birthcohortafterbirthcohort.Whenmeasuredatages25to29,therank-rankslope
wentfrom0.189to0.234betweenthe1963cohortandthe1982cohort,a24%increase.This
trendispresentforbothsons(0.188to0.233)anddaughters(0.204to0.238),albeitfora
somewhatsmallerextentforthelatter.Intergenerationalelasticitiesalsoexhibitanupward
patternacrossbirthcohorts,hencetowardslowermobility:theIGEwentfrom0.153to0.224
forsonsanddaughterscombined,withsimilartrendswhengendersaretakenseparatelyand
againlowermobilityforsonsrelativetodaughters.Thesameupwardtrendcanbeseenwhen
childincomeismeasuredatolderages,evenifthepointestimatestendtobeslightlylarger
theolderthechildis.
AlthoughCanada ismovingtowards lessequalopportunities, therank-rankcorrelationof
the1982birthcohortinCanadaat0.234isstilllowerthanthe0.344estimatereportedby
11
Chettyetal.(2014b)fortheUnitedStates.
Foreaseofexposition,Figures1through3presenttheestimatesfromTable3ingraphical
form.Figure1presentstheresultingestimatesoftherank-rankslope,ameasureofrelative
mobility, while Figure 2 shows the estimates of the IGE, and Figure 3 looks at sons and
daughters,separately.
[InsertFigure1abouthere.]
[InsertFigure2abouthere.]
[InsertFigure3abouthere.]
4.2. Nonlinearity in rank mobility
Figure4presentsfivebinnedscatterplotsofrankmobility:oneforeachbirthcohort.TheX-
axisrepresentsthepercentileranksoftheparentalincomedistributionandtheY-axisshows
themeanchildpercentilerankusingincomemeasuredwhenthechildisaged25to29to
allowforacomparisonacrossallcohorts.Eachpointthereforecapturestheaveragerankof
childrenforeachoftheparent’spercentilerank.Theequationshownatthebottomrightof
each subfigure shows the corresponding estimates of Equation (1). These binned scatter
plots are a way to visualize the estimation of Equation (1). Note that the rank-rank
relationshipdoesnotappearaslinearaswhatChettyetal.(2014b)hadfoundfortheUnited
States, especially at the bottom of the parental income distribution (below the 20th
percentilesforparents).
[InsertFigure4abouthere.]
To further investigate the nonlinearity, locally-weightednonparametric smoothedscatter
plotsarepresented inFigure5, separately foreachbirth cohort.Darker linesare for the
earlierbirthcohortsandlighteronesforthemorerecent.Thecurvesappearnottobelinear:
theslopeismorepronouncedatthebottomoftheparentalincomedistribution.Moreover,
the curves for thevariousbirth cohortsare roughly stackedaccording tobirthyears: the
earlierbirth cohortshavehighermeanchild ranks than the laterones for lowerparental
12
ranks, and lower mean child rank for higher parental ranks. This is coherent with the
decliningrankmobilitythatwasalreadynotedusingalinearmodelintheprevioussection,
but indicates that the changes inmobility observed over time come from changes in the
bottomandtopquintileoftheparentalincomedistribution.Aninterestingresulttolookat
itiswherethecurvesmeeteachotherastheparentalincomerankincreases.Thecurvefor
the1967cohortmeetsthe1963curveatparentalincomepercentilerankof16,whereasthe
1982curveonly reaches the1963cohort atpercentile36, a full20 ranks furtherup the
incomedistribution.Thismeansthatthechildrenfromthebottom36%ofthe1982parental
incomedistributionhaveonaveragelowerincomeranksthansimilarlypositionedchildren
fromtheearliestbirthcohort. Inasense, ithasgrownmoredifficult toreachtherelative
positionofthechildrenofthe1963cohortforchildrenfromlow-incomebackgroundsinlater
birthcohorts.Thereverseistrueatthetopendoftheparentalincomedistribution.
[InsertFigure5abouthere.]
LookingatFigures4and5,itseemsreasonablethatamodelallowingtwoslopes,onebelow
andoneabovethe20thpercentileoftheparentalincomedistribution,wouldprovideagood
fit for the data. Consequently, a version of Equation (1), the rank mobility equation, is
estimatedusingasplinemodelallowing for twoslopes.Theresultingslopeestimatesare
plottedinFigure6,wherethepanelontheleftisfortheslopeforparentalranks1to20,and
theoneontherightisforranksfrom21to100.NoteherethatthescalesoftheY-axesare
notthesameforthetwopanels.Lookingateitherofthetwoslopes,asimilarpatterncanbe
seen,withestimatedslopesshowinganincreasingtrendfromthe1963cohorttothe1982
cohort(rightpanel),albeitwithaslightlydifferentintermediatetrajectoryfortheslopesfor
lowerparentalranks(leftpanel).ThepanelontheleftofFigure6showsthattheslopefor
thebottomquintileofparentspeakedforthe1972birthcohortatover0.7beforegoingback
downtoreachcloseto0.55forthe1982cohort.Acrossallyears,theslopesaremuchhigher,
somobilitymuchlower,forthebottomquintileoftheparentalincomedistributionthanthey
are for the top fourquintiles, and the rateof increasehasbeen stronger for thatbottom
quintile, too. Incontrast, for therestof theparental incomedistribution,mobility, though
decreasing,isrelativelyhigh.Althoughtheslopesinthebottomquintilearehigherin1963
and1972comparedto1977and1982,theaveragerankofachildgivenhisorherparents’
13
rankisalmostalwayslowerin1977and1982thanthatoftheearliercohort(seeFigure5).
Astimepasses,itisgettingmuchharderforachildfromthebottomincomequintiletoreach
higherincomeranks.Whatmattersformobilitymaynotsimplybetheslopeoftherank-rank
relationship,butalsoitsaveragelevel,orinthepresentcasewherethecurvesmeet.
[InsertFigure6abouthere.]
4.3. Rank mobility at the provincial and territorial level
Uptonow,theanalysisfocusedonthenationallevel.Thissectionnowturnstorankmobility
attheprovinciallevel.Threesetsofresultsarepresentedfrommodelsestimatedseparately
for each province or territory: onewhere parental and child ranks are computed at the
nationallevel;onewhereranksarecomputedwithineachprovinceorterritory;andfinally
onewhereranksarecomputednationallybuttheincomemeasuredusedisnetoftaxes.Using
localranksisonewaytotakeintoaccountthedifferentrealitiesofthedifferentprovinces
andterritories,andinasensetocontrolforcostoflivingdifferences.Ifasalaryof$60,000
doesnotmeanthesamethinginAlbertaandinNovaScotia,thenusingwithin-provinceranks
should help compare the relative positions of each. The next three figures present the
resultingslopeestimates,usingnationalranks(Figure7)thenprovincialorterritorialranks
(Figure8),andafter-taxincome(Figure9).StartingwithFigure7,onecanseethatthetrend
ofdecreasingmobilityobservedatthenationallevelisalsopresentinallprovinces,albeitto
differentdegrees.OnlytheYukonshowsanoveralldecreaseintherank-rankslope,butit
shouldbenotedthatthepopulationsizefortheYukonisquitesmallcomparedtotheother
regions.Therearesubstantialdifferencesacrossprovinces,withsomeprovincesshowing
lowslopes/highmobility,suchasPrinceEdwardIslandorNewfoundlandandLabrador,with
slopes around 0.15-0.23, and other provinces with high slopes/low mobility, like
Saskatchewan and Manitoba, both with slopes above 0.3 for the latest birth cohort.
Saskatchewan is also theprovince that saw the greatest increase in the slope coefficient,
goingfrom0.17forthe1963birthcohorttoalmostdoublethatforthe1982birthcohort.By
and large, using ranks computed at the provincial or territorial does not change the
conclusionsdrawnfromlookingatFigure6;theestimatedslopesarepracticallyidentical.
The samecanbe saidaboutusingafter-tax incomerather thangross income to compute
14
incomeranks:Figure9displayspatternsthatareverysimilartothoseofFigures7and8.
[InsertFigure7abouthere.]
[InsertFigure8abouthere.]
[InsertFigure9abouthere.]
4.5. Trends in transition matrices
Inthissubsection,pointsfrom5x5transitionmatricesarepresented,showingconditional
probabilitiesforthevariousincomequintiles.Again,findingsarepresentedgraphicallyfor
easeofexposition.Figure10shows,conditionalonparentalincomequintile,theprobability
thatachildhasanincomeinthebottomquintileofhisorherincomedistribution.Formost
parentalincomequintiles,thisprobabilityisrelativelystableacrossthefivebirthcohorts,
andperhapsslightlydecreasingfortheverytopincomequintile.Thelargestmovementis
seenforchildrenwhogrewupinfamiliesatthebottomoftheincomedistribution.𝑃+,+,the
probabilitytoremaininthebottomquintile,hasgrownfrom0.27to0.33,a22%increase
betweentheoldestcohortandtheyoungestone.Ithasthusbecomemoredifficultforachild
fromadisadvantageousbackgroundtogetoutofalowincomesituation.
[InsertFigure10abouthere.]
Figure11isconstructedthesamewayasFigure10,butthistimeshowingtheprobabilityfor
achildtohaveanincomeinthetopincomequintileofhisorherdistribution.Figure11shows
thatall theprobabilitiesremain constantacross the fivebirth cohorts.The flipsideof the
increasein𝑃+,+observedinFigure10isthusnotthatchildrenfromlow-incomefamilieshave
a lower probability to reach to top income quintile, but rather that they have a lower
probabilitytobeinoneofthethreemiddleincomequintiles(quintiles2,3or4).Thiscanbe
seen in Figure 12,which presents the probability for a child to be in thosemiddle three
quintiles,conditionalonparental incomequintile.Thedrop inthesetransitionstomiddle
class,sotospeak,canclearlybeseenforchildrenwhogrewupinbottom-quintilefamilies.
Forthosechildren,theprobabilitytoreachoneofthethreemiddlequintileswentfrom0.60
forthe1963birthcohortto0.56forthe1982cohort,anearly8%decrease.
15
[InsertFigure11abouthere.]
[InsertFigure12abouthere.]
4.6. The Canadian Great Gatsby Curve
NowthatmeasuresofrankmobilityforfivesuccessivebirthcohortsofCanadianshavebeen
computed,theycanbeputinrelationwithameasureofincomeinequalityfortheparental
incomedistribution,totraceoutawithin-countryversionoftheGreatGatsbyCurve.Figure
13doesjustthisfortheslopecoefficient,ameasureofrelativemobility,atthenationallevel.
Eachdotrepresentsabirthcohort,anditisclearthatCanadahasbeen“goingup”thecurve,
bymoving towards the top right. In doing so, the successive cohorts are tracing out the
upward sloping curve, too, showing that the classic inequality-mobility relationship
describedintheliteratureacrosscountriesforagiventimeperiod(Corak2013)alsoholds
withinacountryacrossbirthcohorts.Figure14issimilartoFigure13,thistimeusingafter-
taxincome.TheGinicoefficientsfortheparentalincomedistributionarelowerwhenusing
after-taxincome,aresultoftheincomeredistributionthroughthetaxsystem,buttheslope
coefficientsremainverysimilar.TheresultingGreatGatsbycurveisthussimplyshiftedto
theleft.
[InsertFigure13abouthere.]
[InsertFigure14abouthere.]
Anotherwaytovisualizetherelationshipbetweeninequalityandmobilityistolookwithin
Canada by plotting a provincial version of Figure 13, that is one where each point
correspondstoaprovinceandbirthcohort.Again,theinequality-mobilityrelationshipholds
within a given country for a given point in time (looking at a given birth cohort across
provinces),andwithinagivencountrywhenallcohortsarepooledtogether(seethebest
linearfitlineonFigure15).Figure16presentsthesamepoints,connectedbyprovincethis
time.Giventhateveryprovincehasseenanincreaseininequalityovertheperiodstudied,so
increasing Gini coefficients, each provincial series is labeled by adding the province
abbreviationtotherightoftheseries,i.e.nexttothepointcorrespondingtothe1982birth
cohort.Themovementof“goingup”theGreatGatsbyCurveisnotjusttrueforCanadaoverall
16
(asseeninFigure13);itistrueforeverysingleprovinceofCanada.Someprovinceshavea
relatively steep mobility-inequality profile, meaning that their decrease in mobility is
relatively more important than their increase in inequality (for example Saskatchewan),
while others have a flatter profile, with relatively large increases in inequality but not
necessarilylargeincreasesintherank-rankslopecoefficient(forexampleOntario).
[InsertFigure15abouthere.]
[InsertFigure16abouthere.]
4.7. Robustness analyses
Inthissectionrobustnessanalysesarepresented.Onepossiblesourceofbiascouldcome
fromthefactthatsomeindividualscannotbefoundinthetaxfilesinsomeyears.Table2
showedthatwhenlookingatages25to29,3to4%ofchildrencannotbelinkedtoanytax
file.Thosepercentagesare5to7%whenconsideringages30to34and7to9%forages35
to39.Moreover,somechildrencanonlybefoundinsomeyears.Thequestionofwhattodo
whenchildrenaremissing fromthetax filescouldbe important. In themainanalysis, the
sampleexcludesparentsandchildrenwithaverage incomeunder$500andchildrenthat
cannotbefoundinanytaxfile.Inarobustnessanalysis,thesampleisfurtherrestrictedto
includeonlyparentsandchildrenthatareobservedforthefull5-yearwindowduringwhich
incomeismeasured.TheresultingslopeestimatesfromEquation(1)areinTable4.
17
Table4:Mobilityestimatesfromalternativespecifications
Birthcohort 1963 1967 1972 1977 1982
Rank-rankallquintilesoftheparentalincomedistributionAll ages25-29 0.189 0.188 0.216 0.215 0.234
ages30-34 0.201 0.214 0.232 0.235 —ages35-39 0.201 0.214 0.230 — —
Restrictedset ages25-29 0.186 0.182 0.209 0.206 0.226ages30-34 0.201 0.214 0.231 0.233 —ages35-39 0.202 0.215 0.229 — —
Rank-rankbottomquintileoftheparentalincomedistributionAll ages25-29 0.483 0.616 0.722 0.460 0.544
ages30-34 0.432 0.585 0.657 0.392 —ages35-39 0.439 0.563 0.635 — —
Restrictedset ages25-29 0.513 0.655 0.810 0.566 0.700ages30-34 0.460 0.624 0.758 0.540 —ages35-39 0.465 0.594 0.735 — —
Rank-ranktopfourquintilesoftheparentalincomedistributionAll ages25-29 0.172 0.160 0.181 0.178 0.193
ages30-34 0.197 0.203 0.214 0.217 —ages35-39 0.201 0.208 0.218 — —
Restrictedset ages25-29 0.166 0.155 0.171 0.166 0.181ages30-34 0.192 0.200 0.209 0.210 —ages35-39 0.197 0.206 0.214 — —
Note:Standarderrorsareall0.001fortherankmobilityestimatesforallquintilesandforthetopfourquintiles.Forbottomquintile,standardserrorsareallaround0.01.Source:Authors’calculationsbasedontheIID
Tables4presentstherank-rankslopeestimatesforthewholedistribution(toppanel)aswell
asseparatelyforthebottomquintileoftheparentalincomedistribution(middlepanel)and
for the other four quintilesof the parental incomedistribution (bottompanel). Although
therearesomedifferenceswhencomparingtheresultsfromthemainsampletothoseofthe
restrictedset,thepatternsmostlyhold.Forexample,thetoppanelshowsthattheestimated
slopewentfrom0.189to0.234acrossthebirthcohortsinthemainanalysis,andfrom0.186
to 0.226 in the restricted set. Looking at the bottom quintile of the parental income
distribution in the middle panel, the increase in the slope coefficient for the later birth
cohortsissomewhatlargerintherestrictedsetthanitisinthemainsample:0.566and0.700
forthe1977and1982birthcohorts,respectively,fortherestrictedset,comparedto0.460
18
and0.544fortheanalysissample.Overall,thepotentialbiascomingfromtheunavailability
oftaxfilesforsomeindividualsdoesnotappeartobemuchofaconcern.
5. Discussion and conclusion
Inthisstudy,decreasesinsocialmobilityacrossfiveconsecutivebirthcohortsinCanadaare
documentedusing a previously unavailable very large setof administrative tax files. The
focusisonrankmobility,andfindingsshowthattheslopebetweenachild’sincomerank,as
anadult,andhisorherparents’incomerankhasbeenonanincreasingtrend,goingfrom
0.189to0.234betweenthe1963-66birthcohortandthe1982-85cohort.Therealsoappears
tobesubstantialnonlinearityintherankmobilityequation,motivatingasplitoftherank-
rankslopeintwosegments:oneforparentalincomerankuptothe20thpercentile,theother
forranks21to100.Thebottomsegmenthasmuchhigherslopes,thuslowermobility,than
thetopsegment.Moreover,theaveragechildranksforthemorerecentcohortsarelower
than thoseof the first cohort for a substantial partof the bottomof theparental income
distribution: even childrenwhose parents are at the 30th to 35th percentiles have lower
average ranks for the1982-85cohort than for the1963-66cohort.Rankmobility is also
described at the provincial and territorial levels.While therehas been a deterioration of
mobilityineverysingleprovince,therearestilllargevariationsinrank-rankslopesacross
regions in most recent cohort, and the within-province variations over time have had
differentmagnitudes.Someplaces,notablytheMaritimes,haveamongthelowestrank-rank
slopes,meaningthemostequalopportunitiesforchildren,alongwiththesmallestincreases
inslopes.Ontheotherhand,Manitobagenerallydisplaystheleastintergenerationalmobility
amongst the10provinces, and it is inSaskatchewan that thedeteriorationof equalityof
opportunitieshasbeen theworst.A lookatquintile transitionmatrices revealsmuch the
samestoryofdecliningmobility:childrenborn inthebottom20%of families in termsof
incomehave become less likely to exit the bottomquintile themselves, and less likely to
transitionintothemiddleclass.
These results can seem to contradict Ostrovsky’s (2017) finding of stable rates of
intergenerational incomemobility between 1970 and 1984, using the same data source.
19
Ostrovsky’smeasurehoweverwasquitedifferent:heusedasanabsolutemobilitymeasure
thefractionofchildrenearningmorethantheirparentsatage30.Thischoicewasmotivated
bythedesiretocomparetheCanadiancasewithChettyetal.’s(2017)similarestimatesfor
theUnitedStates.Therestrictiontouse incomemeasuredatage30 forbothparentsand
childrenconstrainedthecohorts thatcouldbeusedfromtheIID.Duetodataconstraints,
Ostrovsky’s analysis starts with birth year 1970, but the present analysis documents
decreases in rankmobilitystartingwith the1963birth cohort. It is thuspossible thatby
havingtostartwiththe1970birthcohort,asubstantialpartofthedeclineobservedherein
was missed. Moreover, the absolute mobility measure used in Ostrovsky states which
fractionofchildrentipspastacertainincomethresholdandinsodoingdoesnotdescribethe
rank-rankrelationshipinthesamelevelofdetailsasdonehere.
Togetherwiththeincreasesinincomeinequalityobservedoverthesametimeperiod,the
increasinglystrongassociationuncoveredbetweenparentalincomerankandchildincome
rankmeansthatCanada,andeverysingleoneofitsprovinces,havebeen“goingup”theGreat
GatsbyCurve.Moreinequalityhasgonehandinhandwithlowermobility.Thisassociationis
purelycorrelationalatthisstage,andthecurrentdescriptiveanalysismakesnoattemptto
uncovercausalrelationships.Butthefactthattherelationshipholdsacrosscountriesfora
giventimeperiod,aswellasforagivencountryovertime,andwithinagivencountry(across
itsregions)bothatagivenpoint in timeandacrosstime,points tosomethingmorethan
simplyaspuriouscorrelation.Whentherungsofthesocioeconomicladderarefurtherapart,
itbecomesmoredifficulttoclimbit.Theoreticalmodelsexplainingthisrelationshipwould
beanimportantstepforwardinourunderstandingofsocialmobility.
Armedwithabatteryofdescriptivestatistics,amuchclearerportraitofthesituationofsocial
mobilityinCanadaanditsrecenttrendsemerges.Alotofquestionsremain.Whatmakesa
regionmoremobilethananother,andwhydosomeexperiencemuchstrongerdeclines?The
literaturesuggestssomefactorsthatcorrelatewithmoreorlessintergenerationalincome
transmission (see Chetty et al., 2014b, or Connolly, Corak and Haeck, 2019). But more
researchisneededtouncovercausalmechanismsandidentifypublicpoliciesthatmayplay
infavourofmoreequalopportunitiesforCanadianchildren.
20
References Becker,G.S.,&Tomes,N.(1979).AnEquilibriumTheoryoftheDistributionofIncomeand
IntergenerationalMobility.JournalofPoliticalEconomy,87(6),1153-1189.Becker,G.S.,&Tomes,N.(1986).HumanCapitalandtheRiseandFallofFamilies.Journalof
LaborEconomics,4(3,Part2),S1-S39.
Black,S.E.,&Devereux,P.J.(2011).RecentDevelopmentsinIntergenerationalMobility.InO.C.Ashenfelter,&D.E.Card(Eds.),HandbookofLaborEconomics,4B(16), (p.1487-1541).Amsterdam:North-Holland.
Chen,W.H.,Ostrovsky,Y.,&Piraino,P.(2017).LifecycleVariation,Errors-in-VariablesBiasand Nonlinearities in Intergenerational Income Transmission: New Evidence fromCanada.LabourEconomics,44(January),1-12.
Chetty,R.,Hendren,N.,Kline,P.,Saez,E.,&Turner,N.(2014a).IstheUnitedStatesStillaLandofOpportunity?Recent Trends in IntergenerationalMobility.TheAmerican EconomicReview,104(5),141-147.
Chetty,R.,Hendren,N.,Kline,P.,&Saez,E.(2014b).WhereistheLandofOpportunity?TheGeographyof IntergenerationalMobility in theUnitedStates.TheQuarterly JournalofEconomics,129(4),1553-1623.
Chetty, R., Grusky,D., Hell,M.,Hendren,N.,Manduca, R., &Narang, J. (2017). The FadingAmericanDream:TrendsinAbsoluteIncomeMobilitySince1940.Science,356(6336),398-406.
Connolly,M.,Corak,M.,&Haeck,C.(2019).IntergenerationalMobilitywithinandbetweenCanadaandtheUnitedStates.Forthcoming,JournalofLaborEconomics.
Corak,M.(2013).IncomeInequality,EqualityofOpportunity,andIntergenerationalMobility.TheJournalofEconomicPerspectives,27(3),79-102.
Corak,M.(2017).DividedLandscapesofEconomicOpportunity:TheCanadianGeographyofIntergenerationalIncomeMobility.UniversityofChicago,HumanCapitalandEconomicOpportunityWorkingPaperNumber2017-043.
Corak, M., & Heisz, A. (1999). The Intergenerational Earnings and Income Mobility ofCanadian Men: Evidence from Longitudinal Income Tax Data. Journal of HumanResources,34(3),504-533.
Dahl,M.W.,&DeLeire,T.(2008).TheAssociationbetweenChildren'sEarningsandFathers'LifetimeEarnings:EstimatesUsingAdministrativeData.DiscussionPaperNo.1342-08,UniversityofWisconsin-Madison,InstituteforResearchonPoverty.
Davis, J., & Mazumder, B. (2017). The Decline in Intergenerational Mobility After 1980.FederalReserveBankofChicago,WP2017-05,March2017.
Ostrovsky,Y.(2017).DoingasWellasOne'sParents?:TrackingRecentChangesinAbsoluteIncome Mobility in Canada. Economic Insights, Catalogue no. 11-626-X — No. 073,StatisticsCanada.
21
Statistics Canada (2015). Lone-parent families: The new face of an old phenomenon.CanadianMegatrends,Catalogueno.11-630-X,2015002,ISBN978-0-660-25887-4.
Statistics Canada (2016). The rise of the dual-earner family with children. CanadianMegatrends,Catalogueno.11-630-X,2016005,ISBN978-0-660-25887-4.
22
Figures
Figure1:Rankmobilitybyagegroupandbirthcohort
Note:This figureshowstheestimatedslopecoefficientofEquation(1),byagegroupandbirthcohort.Source:Authors’calculationsbasedontheIID
23
Figure2:Intergenerationalelasticity,byagegroupandbirthcohort
Note:This figureshowstheestimatedslopecoefficientofEquation(2),byagegroupandbirthcohort.Source:Authors’calculationsbasedontheIID
24
Figure3:Rankmobilitybyagegroupandbirthcohort,sonsanddaughtersseparately
Note:This figureshowstheestimatedslopecoefficientofEquation(1),byagegroupandbirthcohort,forsonsanddaughtersseparately.Source:Authors’calculationsbasedontheIID
25
Figure4:Rankmobilitybybirthcohort
Note: This figure shows binned scatter plots of rank mobility by birth cohort. Each dotcorrespondstoonepercentilerankoftheparentalincomedistribution.TheY-axisgivesthemeanvalueofchildrankbyparentalincomerank.Childindividualincomeismeasuredatages25to29.TheestimateofEquation(1)isshownatthebottomrightofeachsubfigure.Source:Authors’calculationsbasedontheIID
26
Figure5:Nonlinearityinrankmobilitybybirthcohort
Note:ThisfigureshowsthelowessestimationsofEquation(1),bybirthcohort.Source:Authors’calculationsbasedontheIID
27
Figure6:Rankmobilitybelowandabovethe20thpercentileofparentalincome
Note:Thisfigureshowstheestimatedslopecoefficientofrankmobility,byagegroupandbirthcohort,whereEquation(1)ismodifiedtoallowtwoslopes:oneforparentalranks1to20,andtheotherforparentalranks21to100.Source:Authors’calculationsbasedontheIID
28
Figure7:Rankmobilitybyprovinceorterritoryandbirthcohortusingnationalranks
Note: This figure shows the estimated slope coefficient of Equation (1), by province orterritoryandbirthcohort,withrankscomputedatthenationallevel.Source:Authors’calculationsbasedontheIID
29
Figure8:Rankmobilitybyprovinceorterritoryandbirthcohortusingprovincialor
territorialranks
Note: This figure shows the estimated slope coefficient of Equation (1), by province orterritoryandbirthcohort,withrankscomputedattheprovincialorterritoriallevel.Source:Authors’calculationsbasedontheIID
30
Figure9:Relativerankmobilitybyprovinceorterritoryandbirthcohortusingafter-
taxincomeandranksatthenationallevel
Note:ThisfigureshowstheestimatedinterceptcoefficientofEquation(1),byprovinceorterritoryandbirthcohort,withrankscomputedatthenationallevel.Source:Authors’calculationsbasedontheIID
31
Figure10:Probabilityforachildtobeinbottomincomequintilebyparentalincome
quintileandbirthcohort
Note:Thisfigureshowstheprobabilityforachildtobeinthebottomincomequintileofhisorherincomedistribution,byparentalincomequintileandbirthcohort.Source:Authors’calculationsbasedontheIID
32
Figure 11: Probability for a child to be in top income quintile by parental income
quintileandbirthcohort
Note:Thisfigureshowstheprobabilityforachildtobeinthetopincomequintileofhisorherincomedistribution,byparentalincomequintileandbirthcohort.Source:Authors’calculationsbasedontheIID
33
Figure12:Probabilityforachildtobeinmiddlethreeincomequintilesbyparental
incomequintileandbirthcohort
Note:Thisfigureshowstheprobabilityforachildtobeinthemiddlethreeincomequintilesofhisorherincomedistribution,byparentalincomequintileandbirthcohort.Source:Authors’calculationsbasedontheIID
34
Figure13:GreatGatsbyCurveforrelativerankmobility
Note:ThisfigureshowstherelationshipbetweentheGinicoefficientoftheparentalincomedistribution(ontheX-axis)andtheestimatedslopeofEquation(1)(ontheY-axis).Thelineisthebestlinearfit.Eachpointcorrespondstoabirthcohort.Source:Authors’calculationsbasedontheIID
35
Figure14:GreatGatsbyCurveforrelativerankmobility,usingafter-taxincome
Note:ThisfigureshowstherelationshipbetweentheGinicoefficientoftheparentalafter-taxincomedistribution(ontheX-axis)andtheestimatedslopeofEquation(1)(ontheY-axis).Thelineisthebestlinearfit.Eachpointcorrespondstoabirthcohort.Source:Authors’calculationsbasedontheIID
36
Figure15:GreatGatsbyCurveforrelativerankmobilityattheprovinciallevel
Note:ThisfigureshowstherelationshipbetweentheGinicoefficientoftheparentalincomedistribution(ontheX-axis)andtheestimatedslopeofEquation(1)(ontheY-axis,wheretheranks are computed at the national level). The line is the best linear fit. Each pointcorrespondstoaprovinceorterritoryandbirthcohort.Source:Authors’calculationsbasedontheIID
37
Figure16:GreatGatsbyCurveforrelativerankmobilitybyprovince
Note:ThisfigureshowstherelationshipbetweentheGinicoefficientoftheparentalincomedistribution(ontheX-axis)andtheestimatedslopeofEquation(1)(ontheY-axis,wheretheranksarecomputedatthenationallevel).Eachpointcorrespondstoaprovinceandbirthcohort.Thepointsforagivenprovinceareconnectedfromtheoldestbirthcohorttothemorerecent,withtheprovinceabbreviationnexttothepointforthemorerecentbirthcohort.Source:Authors’calculationsbasedontheIID