Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
WhatisnewinPWT9.0? RobertC.Feenstra,RobertInklaarandMarcelP.Timmer June2016 ThereleaseofthePennWorldTableversion9.0representsthefirstsubstantialchange to the ‘Next Generation of the Penn World Table’ of PWT versions 8.0 and 8.1, see Feenstra, Inklaar and Timmer (2015). If you are a first-time user of PWT, Section I of Feenstraetal.(2015)isstilltherecommendedstartingpoint,asthemainstructureofthe databaseanddefinitionofitsvariablesareunchangedinPWT9.0.Thatsaid,PWT9.0 containsimportantnewandreviseddata.Thisdocumentprovidesanoverviewofthe changes,withamoredetaileddiscussionofparticulartopicsinspecificdocuments. Thechangesfallinthreebroadcategories,namely,i)theincorporationofnewpurchasing powerparities(PPPs)datafromthe2011InternationalComparisonProgram(ICP)and other sources; ii) the incorporation of revised and extended National Accounts data, coveringtheperiodupto2014;andiii)revisedestimatesoffactorinputdataandlabor costshares. I. ICP2011andothernewPPPdata ThelatestroundofPPPsincludedinPWT8wasfortheyear2005.WithPWT9.0,weadd thePPPsfromICP2011tothesetofPPPbenchmarksandshiftourreferenceyearfrom 2005to2011.WorldBank(2014)markedthereleaseoftheresultsofICP2011,withdata on PPPs for consumption and investment for nearly 180 countries. Like ICP 2005, coverage was essentially global, though with 32 additional countries notably more extensive.FifteencountriesdidnotparticipateinanyICProundbefore2011,sothese were not covered in PWT8, but are included in PWT9. This increases the number of countriesfrom167to182andtheshareofworldpopulationcoveredbyPWTfrom96.9 to98.5percent.1 1Thelistofnewcountriesis:Algeria,Anguilla,Aruba,BritishVirginIslands,CaymanIslands,Curaçao,Haiti, Montserrat,Myanmar,Nicaragua,Seychelles,SintMaarten,StateofPalestine,TurksandCaicosIslandsand theUnitedArabEmirates.SomeofthesehadpreviouslybeencoveredinPWT7andearlierversionsbased onalternativepriceinformationthatwasnotcomparableinqualitytoICP. 1 Inaddition,importantmethodologicalissuesthatwereidentifiedafterthereleaseofICP 2005 (see World Bank, 2013), were solved for ICP 2011. These issues were primarily relatedtothepricecomparisonacrossmajorregionsoftheworld.InICP,pricesarefirst compared across countries within a region, such as Africa or Asia-Pacific. In that comparison, products that are particularly important for a region can be taken into account.Comparingpricesacrossregionsrequiresacommonglobalproductlistand,as turnedout,theICP2005globalproductlistincludedmanyproductsthatweretypicalin theconsumptionbasketsofhigh-incomecountries,buthigh-pricedluxuryitemsinlowincomecountries.Duetothisbias,theICP2005pricesofregionswithpredominantly low-incomecountrieswereoverestimatedrelativetohigh-incomeregionsandthereal GDPlevelofcountriessuchasChinawereseverelyunderestimated–ashadalsobeen separatelyestablishedbyFeenstra,Ma,NearyandRao(2013). Deaton and Aten (2016) and Inklaar and Rao (2016) demonstrate that this bias was presentandimportantinICP2005.Theyalsoconcludedthatthissourceofbiaswasnot present in ICP 2011.Moregenerally,the ICP2011havebeenbroadlyacceptedby the research community as the most sophisticated and reliable so far, without major methodologicalorpracticalflaws.InklaarandRao(2016)thereforeconstructedasetof relativepricesfortheyear2005,basedonICP2005data,butapplyingICP2011methods and correcting for the bias in ICP 2005. In PWT 8.1, we already relied on these biasadjustedrelativeprices.ThismeansthattheincorporationofICP2011resultsdoesnot leadtomajorshiftsinincomelevelsoflower-incomerelativetohigher-incomecountries, thoughasshownbelow,individualcountrydifferencescanbelarge. Theshiftinreferenceyearfrom2005to2011meansthatallvariablesthatweredenoted in2005USdollarsinPWT8.0and8.1arenowdenotedin2011USdollars.Fortheperiod 1950to2005,thisshiftinreferenceyearhasnoeffectonrealGDP,otherthantoincrease allvaluesby12percent–theincreaseintheUSGDPdeflatorbetween2005and2011.2 For the years from 2006 onwards, however, PPPs are revised as consumption and investment PPPs are now based on interpolation between the 2005 and 2011 ICP benchmarkresults,whiletheywereextrapolatedfromICP2005inPWT8. 2RevisionstoNationalAccountswillalsoleadtochanges,seeSectionII. 2 There are three sets of countries for which PPPs and real GDP change in a different fashion: 1. NineteencountriesfromCentralAmericaandtheCaribbeandidnotparticipateinICP 2005,butdidparticipateinICP2011andatleastoneearlierICPround.InPWT8,the PPPsandrealGDPnumbersforthesecountrieswereextrapolatedfromthatearlier ICPround(1996inmanycases)to2011,thelatestyearinPWT8.InPWT9thePPPs forthatperiodareinterpolatedusingdatafromtheirearlierICPbenchmarkandthe new2011round.3 2. ForcountriesintheEuropeanUnion(EU)and/orOECD,PWTreliesnotonlyonthe ICP benchmark data, but also includes their more frequent PPP benchmark comparisons;i.e.annualdataforEUcountriesandtriennialdatafornon-EUOECD countries.InPWT8,EUPPPdatauntil2010wasused;fornon-EUOECDcountriesthe mostrecentPPPdatawerefor2008.InPWT9.0,forEUcountriesweincorporatePPP dataupto2014.,Fornon-EUOECDcountriesthe2011ICPPPPsarethemostrecent available. 3. TurkmenistanandUzbekistanhaveonlyparticipatedinICP1996,sotheirPPPdata continuestobeextrapolated. InadditiontothenewconsumptionandinvestmentPPPsfromICP2011,wealsoextend thePPPinformationforexportsandimports.AsinPWT8,thesetradePPPsarebasedon theframeworkintroducedbyFeenstraandRomalis(2014),butwhiletradePPPswere previouslyavailablefortheperiodfrom1984to2007,thetradePPPdatainPWT9.0 extendthrough2014. Toillustratetheimpactofthesechanges,Figure1plotstheratiooftheGDPopricelevel inPWT9.0relativetothepricelevelinPWT8.1fortheyear2011forthe142countries that participated in ICP 2005 and in ICP 2011. In PWT 8.1, the price levels for these countrieswerebasedonextrapolationsfrom2005basedonrelativeinflation,whilein PWT9.0,weusetheICP2011PPPsincombinationwithnewlyestimatedtradePPPs.As thefigureshows,thedifferencescanbesubstantial,withpricelevels(relativetotheUS) 3ThelistofcountriesnotinICP2005butinICP2011andinanearlierICProundis:AntiguaandBarbuda, Bahamas, Barbados, Belize, Bermuda, Costa Rica, Dominica, Dominican Republic, El Salvador, Grenada, Guatemala, Honduras, Jamaica, Panama, Saint Kitts and Nevis, Saint Lucia, Saint Vincent and the Grenadines, Suriname and Trinidad and Tobago. Zimbabwe did participate in ICP 2005, but the results werenotincorporatedinPWT8duetothedistortingimpactofhyperinflationonpricesandexchangerates. 3 lowerby,onaverage,6percentand16countriesforwhichthedifferenceislargerthan 25percent(upordown). Asdiscussedabove,thereareno(remaining)systematicdifferencesintheunderlying measurementmethodologyofICP2005andICP2011andthepricelevelsarealsonot systematicallydifferent–i.e.thedifferencesarenotrelatedtoincomelevel.Yetindividual countrydifferencesarelarge.ThishaslongbeenafeatureofconsecutiveICPbenchmark rounds and, as the current results demonstrate, even better-funded and more closely harmonizedICProundssufferfromthis. DifferenceinPL_GDP&,PWT9.0/PWT8.1−1 -0.4 -0.2 0.0 0.2 0.4 Figure1,Differenceinthe2011GDPopricelevel,PWT9.0vs.8.1,forcountriesin ICP2005andICP2011 MWI TCD DJI LAO AZE YEM IDN IRQ EGY VEN OMN GNQ MAC JOR 6 8 10 LogofCGDP&percapita(PWT9.0) 12 Note:CGDP&percapitafromPWT9.0isin2011USdollars.PL_GDP&fromPWT8.1wasinUlatedbyafactor 1.123toaccountforUSinUlation.Includedarethe142countriesthatwereinICP2005andICP2011. Figure2showsthatthedifferencesarenotablylargerforthesetofcountriesthatwere partofICP2011andanearlierICProundbutdidnotparticipateinICP2005.Forinstance, thepricelevelforBarbados(BRB)is80percenthigherinPWT9.0thaninPWT8.1,while thepricelevelforElSalvador(SLV)is80percentlower.ThePPPforZimbabweis920 percenthigherandnotshowninthefigure.Thesenewerfiguresshouldbeconsideredan improvementoverthepreviousestimates.Thisis,inpart,becausewewouldexpectmore recentpricecomparisonstomoreaccuratelyreflecttherelativepricelevelofacountry comparedwithextrapolationsfrom(much)earliercomparisons,butalsobecausetheICP 4 2011 round relied on more extensive data collection and improved methodologies, especiallycomparedtotheICProundsbeforeICP2005.Thisisalsoillustratedbythefact that in PWT8, the observations for El Salvador in recent decades had been flagged as outliersduetoimplausiblyhighrelativepricelevels. DifferenceinPL_GDP&,PWT9.0/PWT8.1−1 -1.0 -0.5 0.0 0.5 1.0 Figure2,Differenceinthe2011GDPopricelevel,PWT9.0vs.8.1,forcountriesin ICP2011butnotinICP2005 BRB DMA LCA BLZ HND DOM CRI VCT GTM BHS PAN GRD JAM SUR KNA TTO ATG BMU SLV 8.5 9 9.5 10 LogofCGDP&percapita(PWT9.0) 10.5 11 Note:CGDP&percapitafromPWT9.0isin2011USdollars.PL_GDP&fromPWT8.1wasinTlatedbyafactor 1.123toaccountforUSinTlation.Includedarethe19countriesthatwerenotinICP2005butwere inICP2011andinPWT8.1;Zimbabwe,withadifferenceof9.2,isomitted. ThesefiguresillustratethataPPPestimateforaspecificgivencountryinaspecificyear issubjecttoasizeablelevelofuncertainty,especiallyiftheestimateisnotbasedona recentICPbenchmarkbutextrapolatedoverlongerperiods.Aswealsoremarkedinour ‘UserGuidetoPWT8’(Feenstra,InklaarandTimmer,2013),thisimpliesthatcautionis inorderwhenrelyingonthepointestimateofrelativeincomeforaparticularcountryin aparticularyearandthat‘true’incomelevelsmaybe10–20percenthigherorlower.4 Thebroadercross-countrypatternofpricesismuchlessaffectedbythisuncertainty.This is most easily demonstrated using the Balassa-Samuelson/Penn-effect relationship betweenthelogpricelevelandlogincomelevel–measuredasexchange-rate-converted GDPpercapita.Estimatingthisrelationshipforthe142countriesfromFigure1forthe 4SeealsoRaoandHajarghast(2016),whoestimatestandarderrorsofrelativeprices,whichimplyasimilar confidenceinterval. 5 logpricelevelfromPWT9.0yieldsanearlyidenticalcoefficientaswhenusingthelog pricelevelfromPWT8.1,namely0.22. II. GDPdatafromtheNationalAccounts Revisions in the PWT are due to incorporation of new PPP data from the ICP, which mainlyaffectspricelevels,aswellasnewNationalAccounts(NA)dataofcountrieswhich mainlyaffectsnominalGDPlevelsandrealgrowthrates.Whiletypicallyreceivingless attentioninthediscussions,thissourceofrevisionscanbeasleastasimportantthanPPP revisionsformakingrealGDPcomparisonsacrosscountries.NationalAccountsdatain PWT8coveredtheperiodupto2011andwerefromtheversionoftheUnitedNations National Accounts Main Aggregates Database compiled in 2012. Since then, nearly all countries have revised their National Accounts data, in part because more complete sourcematerialhasbecomeavailableforthemostrecentyears,butmanycountrieshave alsomadecomprehensiverevisions.Agrowingnumberofcountrieshasshiftedfromthe accountingrulesofthe1993editionoftheSystemofNationalAccounts(SNA)tothe2008 edition,whichrequirescapitalizationofresearchanddevelopment(R&D)expenditure (amongstmanyotherchanges). By itself, these new accounting rules have a fairly modest effect, increasing GDP by around2–3percentinadvancedeconomies(Eurostat,2014).Achangeintheaccounting systemcanalsobeanoccasionforotherchanges,suchasshiftstonewsourcesorarebenchmarking.5Forexample,thetransitiontoSNA2008intheEUledtoGDPrevisions dueto‘statisticalimprovements’of1.4percentfortheEU-28asawhole,butlargerfor individualcountries:e.g.amountingto2.6percentofGDPintheUKand5.9percentin theNetherlands(Eurostat2014). Butwhilethesechangesarecertainlynoteworthy,theyaresmallcomparedtorevisions in several African countries. In recent years, statistical systems for measuring GDP in countries like Nigeria and Ghana have been overhauled and revamped with major consequencesforlevelsandgrowthratesofGDP.InGhana,thelevelofGDPwasrevised upwards by 60 percent in 2010 (Jerven, 2013), while in Nigeria the GDP level was 5InNationalAccountspractice,itiscommonformanyaggregatestobeextrapolatedfromabenchmark yearusing,forinstance,moretimelybutlesscomprehensivesourcematerial. 6 increased by 89 percent.6 These revisions are welcome, since they provide a more comprehensiveviewoftheseeconomies,butalsoalarming,astheysuggestsubstantial uncertaintyaboutthetruesizeofAfricaneconomies.Suchconcernsaboutthereliability ofNationalAccountsestimatesarenotnew.Previousresearchhasaimedtoprovidean alternative perspective based on detailed Demographic and Health Surveys (Young, 2012) and on nighttime light intensity (Henderson, Storeygard and Weil, 2012). On a morepositivenote,areportbytheAfricanDevelopmentBank(2013)showsthatthesize of the revisions in Ghana and Nigeria have been exceptions rather than the rule for Africancountriesupdatingtheiraccountingmethodologies. Figure3,RevisionstothelevelGDPinlocalcurrencyunitsfortheyear2011 0.6 NGA COD GDPrevisions,PWT9.0/PWT8.1−1 -0.2 0.0 0.2 0.4 SDN TZA LBRMWI IRQ ZWE KEN ZMB GNB BGD MRT SWZ MDV MNG ECU ARG SYR GMB -0.4 GAB 4 6 8 10 LogofCGDP%percapita(PWT9.0) 12 Moregenerally,theyarguethatAfricancountriesaremakinggreatereffortstoproduce timelyandreliableNationalAccountsstatistics.Yetotherevidenceinthereportsuggests there could be future surprises comparable to Ghana or Nigeria: of the 44 countries surveyed,27reliedonaNationalAccountsbenchmarkthatwas10oreven20yearsout of date. Given that the once-a-decade re-benchmarking in some European countries 6 http://www.economist.com/news/finance-and-economics/21600734-revised-figures-show-nigeriaafricas-largest-economy-step-change. 7 already lead to sizeable revisions, it would not be surprising if there were very large revisionsinsomeAfricancountriesinthefuture. ToillustraterecentrevisionstoPWTsourcedata,Figure3showsthechangeinthelevel ofGDPinlocalcurrencyunitsfortheyear2011betweentheNationalAccountsdatathat wereusedforPWT8andthedatausedforPWT9.Nigeria’s(NGA)revisionisthelargest buttherearelargeupward(andalsosomedownward)adjustmentsinothercountries, suchasGabon(GAB),Liberia(LBR)andBangladesh(BGD). Figure4,RevisionstogrowthofGDP,annualandfive-year .1 Five-year(2006–2011) .1 Annual(2010–2011) BDI BDI .05 GDPgrowthrevisions,PWT9.0/PWT8.1−1 -.05 0 .05 ZWE -.05 0 MDV MWI TCD SDN OMN 6 8 10 LogofCGDP%percapita(PWT9.0) -.1 -.1 GMB 12 6 8 10 LogofCGDP%percapita(PWT9.0) 12 Dataoneconomicgrowthisalsosubjecttochangingmethodsandrevisions.Aprominent exampleistherevisionofthemethodsforestimatingIndia’seconomicgrowthinJanuary 2015,whichimpliedmuchfastergrowththanhadpreviouslybeenreported:the20132014 GDP growth rate was revised upwards from 4.7 to 6.9 percent.7 These revisions continuetoexercisepolicymakersandanalysts,raisingquestionsaboutthe‘true’Indian 7Seehttp://in.reuters.com/article/india-gdp-idINKBN0L319Z20150130. 8 rateofgrowth.8TheissueofNationalAccountsrevisionsandtheirimpactonresearch was also raised more generally by Johnson, Larson, Papageorgiou and Subramanian (2013), who show that cross-country growth regression studies relying on annual growth rates of GDP can be severely affected by moving from one vintage of National Accountsdatatothenext.Figure4illustratesthispoint,showingrevisionstotheannual growthofthevolumeofGDP(inlocalcurrencyunits)between2010and2011andthe revisiontotheaverageannualgrowthratebetween2006and2011.Theaveragegrowth overthefive-yearperiodshowsnotablysmallerrevisionsthantheannualgrowthrate, confirmingtheJohnsonetal.(2013)finding. III. Capital,laborandTFP InadditiontorevisionsinPPPsandGDP,PWT9alsoincludesimprovementsinthesource materialanddatacompilationforthelaborandcapitalinputdata.Adiscussionofthese changesisgivenbelowforeachtopic.Thebasicmethodusedtomeasurecapitalandlabor hasnotbeenchanged,forafullexposition,seeFeenstra,InklaarandTimmer(2015), specifically (online) Appendix C. In conjunction with these new estimates, we also providenewdetailedsourcematerial,tobemoreusefulandmoretransparent. Specifically,wenowprovideadetailedlaborfile,thatdetailsthesourcesandmethods forthedataonemployment,yearsofschoolingandthelaborshare,aswellasthevarious alternativelaborsharemeasuresthatcanbeusedtoassessthesensitivityofourchoices. Wealsoprovideadetailedcapitalfile,thatprovidesabreakdownofinvestment,capital stocksanddepreciationbyfourassets–structures,machinery,transportequipmentand otherassets,whichincludesoftwareandotherintellectualpropertyproducts. • Investmentdata.OneofthemaininnovationsinPWT8wasthereintroductionof capitalstocksseriesbasedonestimatesofinvestmentbyasset.Thoseestimateswere based on National Accounts data, detailed expenditure data from ICP benchmarks, andestimatesbasedonoutputintheconstructionindustryandsupply(production+ imports – exports) of machinery and equipment, the so-called Commodity Flow Method(CFM). 8Seee.g.http://www.economist.com/news/finance-and-economics/21696546-few-economists- wholeheartedly-believe-indias-stellar-growth-rate-elephant 9 In PWT 9.0, this basic approach is unchanged, but its implementation has greatly improved.First,weincludesubstantiallymoreinvestmentdata,directlytakenfrom national accounts sources, reducing our reliance on the indirect CFM estimates. Second,weincorporatedatacompiledunderthenewSystemofNationalAccounts, which includes investment in R&D. This improved dataset is the result of a collaborationwithTheConferenceBoardandwillunderlienotjustPWT9.0,butalso upcomingversionsoftheTheConferenceBoard’sTotalEconomyDatabase.9Ajoint paperprovidingamoredetaileddiscussionoftheconstructionandfeaturesofthese dataisplannedforreleaseinJune2016. • Laborshare.InPWT8weintroducedthevariableLABSHthatgivesestimatesofthe shareoflaborincomeinnominalGDP.Itisrelativelystraightforwardtodetermine theshareoflaborincomeofemployeesinGDP,asthisinformationisaregularpartof the National Accounts of countries. Estimating the labor income of self-employed workersismorechallenging.Ifacountryreportsthetotalincomeofself-employed, knownasmixedincome,thereisaclearupperboundtooveralllaborincome,leading toareasonableestimate.Whensuchinformationisnotavailable,PWT8assumedselfemployed earn the same average wage as employees or alternatively that selfemployed labor income equaled value added in agriculture, depending on which methodleadstoalowerlaborshare.10Thisapproachwasmotivatedtominimizethe risk of overestimation, which is high when using the former. However, this conservativeprocedurecouldleadtounderestimationoflaborshares. InPWT9.0,weattempttoidentifysuchcases,byconsideringtwocriteria:(1)does thechosenmethod(sameaveragewageorvalueaddedinagriculture)leadtoalabor shareoflessthan40percent,onaverage?And(2)istheshareofGDPgoingtofixed assets larger than 50 percent, on average? The first criterion is motivated by the observationthat,whenevermixedincomedataisavailable,thelaborshareisonly rarely smaller than 40 percent (i.e. in less than 10 percent of cases). The second criterionreliesonanestimateoftheincomeflowingtoownersoffixedassets.Tothat 9 See https://www.conference-board.org/data/economydatabase/. While the Total Economy Database alsoestimatestotalfactorproductivitygrowth,onlyPWTprovidesestimatesofcomparativecapitalinput levels. 10Inaddition,somecountriesreportanemployeecompensationshareofmorethan70percent.Sincesuch highsharesoccurveryrarelyifmixedincomedataisavailable,thissuggeststheemployeecompensation sharesalreadyincludesanestimateofself-employedlabourincome. 10 end,wesubtracttheWorldBank’sestimateofnaturalresourcerentsfromGDP.Take SaudiArabia:itsestimatedlaborshare(basedonmixedincomedata)is25percentof GDP, natural resource rents account for 40 percent of GDP, leaving 35 percent for owners of fixed assets, such as buildings and machinery. Whenever data on mixed incomeisavailable,theshareofincomegoingtoownersoffixedcapitalisonlyrarely larger than 50 percent. For both criteria, we consider the average across the full periodtoidentifycountrieswherethelaborshareisclearlyunderestimated.Inthose countries,weusethelarger,ratherthanthesmaller,ofthetwoalternativemethods.11 While we view this is a worthwhile refinement, the labor share estimates remain uncertain,especiallyincountrieswherenoinformationisavailableonmixedincome. For this reason, we provide additional detailed data underlying our compilations, namelythevariouslaborsharealternativesandinformationonwhichmethodisused forwhichspecificobservations. • AveragehoursworkedandTFP.PWT8andPWT9reporttheaveragehoursworked by persons engaged (variable AVH), sourced from The Conference Board’s Total EconomyDatabase.Thesedatacover65countriesandsincethisislessthanthe111 countriesforwhichTFPcouldbecomputedbasedonlaborandcapitalestimates,the data on average hours worked were not taken into account when computing TFP levels(CTFP)orgrowth(RTFPNA)inPWT8. InPWT9,wechangedthecomputationmethodtotakeintoaccounttheavailabledata onaveragehoursworkedforcountriesandyearsforwhichtheseareavailable,by imputing missing values. For countries without any information, average hour workedissetequaltoaveragehoursworkedintheUS.ThisensuresthatCTFPisnot affectedbythischoice,thoughaconsequenceisthatTFPgrowthishigher,onaverage, sinceUSaveragehoursworkedhavedeclinedatanaverageannualrateof0.2percent since 1950. This approach is clearly no substitute for careful measurement (or econometricmodeling)ofaveragehoursworked,as,forexample,increasesinincome levels tend to lead to decreases in average hours worked. The current approach 11Therearetwoexceptions,SudanandSierraLeone.InSudan,thisapproachleadstoalaboursharethat isexceptionallyhighin1996-1998,sowesetthelabourshareat0.9inthoseyears.InSierraLeone,new vintageNationalAccountsdataleadtoveryhighestimatesofthelabourshareafter2001,soweusethe employeecompensationshareforthoseyears. 11 shouldthusbeseenmoreasa‘weakprior’ratherthanatruecorrectionforhours worked.Forcountrieswithdataforpartoftheirtimeseries,weassumednochange in average hours worked before the first observation. This ensure CTFP reflects higherorloweraveragehoursworked,whileavoidingstringentassumptionsabout theevolutionovertime. NotethatPWT9continuestoshowdataonaveragehoursworkedonlyforcountries andyearsforwhichtheTotalEconomyDatabasealsoreportsdata,astheimputed figures are removed after computing CTFP and RTFPNA. See the PWT program package(specificallygen_pwt.do)forthepreciseimplementationoftheimputation. • Human capital. In PWT8, a human capital index was estimated using data on the average years of schooling from Barro and Lee (2013) and rates of return on education from Psacharopoulos (1994). As we describe in more detail in our document‘HumancapitalinPWT9.0’,theBarroandLee(2013)dataforarangeof countriesishardtosquarewithseveralalternativesources.InPWT9,wetherefore drawinpartuponthedatafromBarroandLee(2013)andinpartondatabyCohen and Leker (2014), which updated the work of Cohen and Soto (2007). The precise implementationofthiscombinationofsourcesisalsodiscussedinthe‘Humancapital in PWT9.0’ document on the PWT website. This change in source material has a relatively small effect on cross-country comparisons, but more notable effects on growthratesforseveralcountries. References African Development Bank (2013), Situational Analysis of the Reliability of Economic StatisticsinAfrica:SpecialFocusonGDPMeasurement. Barro,RobertJ.andJong-WhaLee(2013),“Anewdatasetofeducationalattainmentin theworld,1950-2010”JournalofDevelopmentEconomics104:184–198. Cohen, Daniel and Marcelo Soto (2007), “Growth and human capital: good data, good results”JournalofEconomicGrowth12(1):51–76. Cohen, Daniel and Laura Leker (2014), “Health and Education: Another Look with the ProperData",mimeoParisSchoolofEconomics. 12 Deaton,AngusandBetinaAten(2016),“TryingtoUnderstandthePPPsinICP2011: WhyaretheResultssoDifferent?”forthcomingAmericanEconomicJournal: Macroeconomics. Eurostat(2014),“ESA2010shiftslevelofEUandeuroareaGDPupward,growthrates almostunaffected”EurostatNewsRelease, http://ec.europa.eu/eurostat/documents/42577/761078/2-17102014-BP-EN. Feenstra,RobertC.,HongMa,J.PeterNearyandD.S.PrasadaRao(2013),“WhoShrunk China?PuzzlesintheMeasurementofRealGDP”EconomicJournal123(573):1100– 1129. Henderson,J,Vernon,AdamStoreygardandDavidN.Weil(2012),“Measuringeconomic growthfromouterspace”AmericanEconomicReview102(2):994–1028. Inklaar,RobertandD.S.PrasadaRao(2016),“Cross-countryincomelevelsovertime:did the developing world suddenly become much richer?” forthcoming American EconomicJournal:Macroeconomics. Jerven,Morten(2013),“ComparabilityofGDPestimatesinSub-SaharanAfrica:theeffect ofrevisionsinsourcesandmethodssincestructuraladjustment”ReviewofIncome andWealth59(S1):S16–S36. Johnson,Simon,WilliamLarson,ChrisPapgeorgiouandArvindSubramanian(2013),“Is newerbetter?PennWorldTablerevisionsandtheirimpactongrowthestimates” JournalofMonetaryEconomics60:255–274. Rao, D.S. Prasada and Gholamreza Hajargasht (2016), “ Stochastic Approach to Computation of Purchasing Power Parities in the International Comparison Program(ICP)”JournalofEconometrics191(2):414–425. Young,Alwyn(2012),“TheAfricangrowthmiracle”JournalofPoliticalEconomy120(4): 696–739. 13