micro economics presentation and need guidance to help me learn.
The instructor normally asks about graph explanation, I have no idea which function or graph that I should focus for this topic
Requirements: 1-2 pages
DOESMANAGEMENTMATTER?EVIDENCEFROMINDIA*NicholasBloomBennEifertAprajitMahajanDavidMcKenzieJohnRobertsAlong-standingquestioniswhetherdifferencesinmanagementpracticesacrossÞrmscanexplaindifferencesinproductivity,especiallyindevelopingcountrieswherethesespreadsappearparticularlylarge.Toinvestigatethis,weranamanagementÞeldexperimentonlargeIndiantextileÞrms.Wepro-videdfreeconsultingonmanagementpracticestorandomlychosentreatmentplantsandcomparedtheirperformancetoasetofcontrolplants.WeÞndthatadoptingthesemanagementpracticesraisedproductivityby17%intheÞrstyearthroughimprovedqualityandefÞciencyandreducedinventory,andwithinthreeyearsledtotheopeningofmoreproductionplants.WhyhadtheÞrmsnotadoptedtheseproÞtablepracticespreviously?Ourresultssuggestthatinformationalbarriersweretheprimaryfactorexplainingthislackofadoption.Also,becausereallocationacrossÞrmsappearedtobeconstrainedbylimitsonmanagerialtime,competitionhadnotforcedbadlymanagedÞrmstoexit.JELCodes:L2,M2,O14,O32,O33.I.INTRODUCTIONEconomistshavelongpuzzledoverwhytherearesuchastoundingdifferencesinproductivityacrossbothÞrmsandcountries.Forexample,U.S.plantsinindustriesproducinghomogeneousgoodslikecement,blockice,andoakßooring*FinancialsupportwasprovidedbytheAlfredSloanFoundation,theFreemanSpogliInstitute,theInternationalInitiative,theGraduateSchoolofBusinessatStanford,theInternationalGrowthCentre,theInstituteforResearchintheSocialSciences,theKauffmanFoundation,theMurthyFamily,theKnowledgeforChangeTrustFund,theNationalScienceFoundation,theToulouseNetworkforInformationTechnology,andtheWorldBank.Thisre-searchwouldnothavebeenpossiblewithoutourpartnershipwithKayAdams,JamesBenton,andBreckMarshall;thededicatedworkoftheconsultingteamofAsifAbbas,SaurabhBhatnagar,ShaleenChavda,RohanDhote,KarlGheewalla,KushaGoyal,ManishMakhija,AbhishekMandvikar,ShrutiRangarajan,JitendraSatpute,ShreyanSarkar,AshutoshTyagi,andRavindraVasant;andtheresearchsupportofTroySmith.Wethanktheeditor,LarryKatz;sixanonym-ousreferees;ourformaldiscussantsSusantuBasu,FrancescoCaselli,RayFisman,NaushadForbes,CaseyIchniowski,VojislovMaksimovic,RamadaNada,PaulRomer,andSteveTadelis;aswellasalargenumberofseminaraudiences.!TheAuthor(s)2013.PublishedbyOxfordUniversityPress,onbehalfofPresidentandFellowsofHarvardCollege.Allrightsreserved.ForPermissions,pleaseemail:journals.permissions@oup.comTheQuarterlyJournalofEconomics(2013),1Ð51.doi:10.1093/qje/qjs044.1 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
display100%productivityspreadsbetweenthe10thand90thpercentile(Foster,Haltiwanger,andSyverson2008).Thisprod-uctivitydispersionappearsevenlargerindevelopingcountries,withHsiehandKlenow(2009)estimatingthattheratioofthe90thtothe10thpercentilesoftotalfactorproductivityis5.0inIndianand4.9inChineseÞrms.Onenaturalexplanationfortheseproductivitydifferencesliesinvariationsinmanagementpractices.Indeed,theideathatÔÔmanagerialtechnologyÕÕaffectstheproductivityofinputsgoesbackatleasttoWalker(1887),isemphasizedbyLeibenstein(1966),andiscentraltotheLucas(1978)modelofÞrmsize.Althoughmanagementhaslongbeenemphasizedbythemedia,businessschools,andpolicymakers,economistshavetypicallybeenskepticalaboutitsimportance.Onereasonforskepticismovertheimportanceofmanage-mentisthebeliefthatproÞtmaximizationwillleadÞrmstomin-imizecosts(e.g.,Stigler1976).Asaresult,anyresidualvariationsinmanagementpracticeswillreßectÞrmsÕoptimalresponsestodifferingmarketconditions.Forexample,Þrmsindevelopingcountriesmaynotadoptqualitycontrolsystemsbecausewagesaresolowthatrepairingdefectsischeap.Hence,theirmanage-mentpracticesarenotbad,buttheoptimalresponsetolowwages.Asecondreasonforthisskepticismisthecomplexityofthephenomenonofmanagement,makingithardtomeasure.Recentwork,however,hasfocusedonspeciÞcmanagementpractices,whichcanbemeasured,taughtinbusinessschools,andrecom-mendedbyconsultants.ExamplesofthesepracticesincludekeyprinciplesofToyotaÕsleanmanufacturing,includingqualitycon-trolprocedures,inventorymanagement,andcertainhumanre-sourcesmanagementpractices.AgrowingliteraturemeasuresmanysuchpracticesandÞndslargevariationsacrossestablish-mentsandastrongassociationbetweenthesepracticesandhigherproductivityandproÞtability.1However,suchcorrelations1.SeeforexampletheextensivesurveysinBloomandVanReenen(2011)andLazearandOyer(2012).Inrelatedworklookingatmanagers(ratherthanman-agementpractices),BertrandandSchoar(2003)useamanager-ÞrmmatchedpanelandÞndthatmanagerÞxedeffectsmatterforarangeofcorporatedecisions,whereasLocke,Qin,andBrause(2007)showbettermanagementpracticesareassociatedwithimprovedworkertreatment,andBloometal.(2010)showbettermanagementpracticesareassociatedwithmoreenergyefÞcientproduction.QUARTERLYJOURNALOFECONOMICS2 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
maybepotentiallymisleading.Forexample,proÞtableÞrmsmaysimplyÞnditeasiertoadoptbettermanagementpractices.ThisarticleprovidestheÞrstexperimentalevidenceontheimportanceofmanagementpracticesinlargeÞrms.Theexperi-menttooklarge,multiplantIndiantextileÞrmsandrandomlyallocatedtheirplantstotreatmentandcontrolgroups.TreatmentplantsreceivedÞvemonthsofextensivemanagementconsultingfromalargeinternationalconsultingÞrm.Thisconsultingdiag-nosedopportunitiesforimprovementinasetof38operationalmanagementpracticesduringtheÞrstmonth,followedbyfourmonthsofintensivesupportfortheimplementationoftheserec-ommendations.Thecontrolplantsreceivedonlytheonemonthofdiagnosticconsulting.ThetreatmentinterventionledtosigniÞcantimprovementsinquality,inventory,andoutput.WeestimatethatwithintheÞrstyearproductivityincreasedby17%;basedonthesechangesweimputethatannualproÞtabilityincreasedbyover$300,000.Thesebetter-managedÞrmsalsoappearedtogrowfaster,withsuggestiveevidencethatbettermanagementallowedthemtodelegatemoreandopenmoreproductionplantsinthethreeyearsfollowingthestartoftheexperiment.TheseÞrmsalsospreadthesemanagementimprovementsfromtheirtreatmentplantstootherplantstheyowned,providingrevealedpreferenceevidenceontheirbeneÞcialimpact.Giventhislargepositiveimpactofmodernmanagement,thenaturalquestioniswhyÞrmshadnotpreviouslyadoptedthesepractices.Ourevidence,thoughspeculative,suggeststhatinfor-mationalconstraintswerethemostimportantfactor.Formanysimple,alreadywidespreadpractices,likethemeasurementofqualitydefects,machinedowntime,andinventory,ÞrmsthatdidnotemploythemapparentlybelievedthatthepracticeswouldnotimproveproÞts.Theownersclaimedtheirqualitywasasgoodasthatofother(local)Þrmsand,becausetheywereproÞtable,theydidnotneedtointroduceaqualitycontrolprocess.Forlesscommonpractices,likedailyfactorymeetings,standar-dizedoperatingprocedures,orinventorycontrolnorms,Þrmstyp-icallyweresimplyunawareofthesepractices.AlthoughthesetypesofleanmanagementpracticesarecommoninJapanandtheUnitedStates,theyappeartoberareindevelopingcountries.WhydidcompetitionnotforcebadlyrunÞrmstoexit?Thereasonappearstobethatcompetitivepressureswereheavilyre-stricted:importsbyhightariffs,entrybythelackofexternalDOESMANAGEMENTMATTER?3 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
Þnance,andreallocationbylimitedmanagerialtime.Managerialtimewasconstrainedbythenumberofmalefamilymembers.NonÐfamilymemberswerenottrustedbyÞrmownerswithanydecision-makingpower,andasaresultÞrmsdidnotexpandbeyondthesizethatcouldbemanagedbyclose(almostalwaysmale)familymembers.Notsurprisingly,wefoundthatthenumberofmalefamilymembershadmorethanthreetimestheexplanatorypowerforÞrmsizeastheirmanagementpractices.Themajorchallengeofourexperimentwasitssmallcross-sectionalsamplesize.Wehavedataononly28plantsacross17Þrms.Toaddressconcernsoverstatisticalinferenceinsmallsam-ples,weimplementedpermutationtestswhosepropertiesarein-dependentofsamplesize.Wealsoexploitedourlargetimeseriesofaround100weeksofdataperplantbyusingestimatorsthatrelyonlargeT(ratherthanlargeN)asymptotics.WebelievetheseapproachesareusefulforaddressingsampleconcernsandalsopotentiallyforotherÞeldexperimentswherethedatahasasmallcross-sectionbutlongtime-seriesdimension.Thisarticlerelatestoseveralstrandsofliterature.First,thereisthelargebodyofliteratureshowinglargeproductivitydifferencesacrossplants,especiallyindevelopingcountries.Fromtheoutset,thisliteraturehasattributedmuchofthesespreadstodifferencesinmanagementpractices(Mundlak1961).ButproblemsinmeasurementandidentiÞcationhavemadethishardtoconÞrm.Forexample,SyversonÕs(2011)recentsurveyoftheproductivityliteratureconcludesthatÔÔnopotentialdrivingfactorofproductivityhasseenahigherratioofspeculationtoempiricalstudy.ÕÕDespitethis,therearestillfewexperimentsonproductivityinÞrms,andnone(untilnow)involvinglargemultiplantÞrms(McKenzie2010).Second,ourarticlebuildsontheliteratureonÞrmsÕmanage-mentpractices.TherehasbeenalongdebatebetweentheÔÔbestpracticeÕÕview,thatsomemanagementpracticesareuniversallygoodsothatallÞrmswouldbeneÞtfromadoptingthem(Taylor1911),andtheÔÔcontingencyview,ÕÕthatoptimalpracticesdifferacrossÞrmsandsoobserveddifferencesneednotreßectbadmanagement(Woodward1958).Muchoftheempiricalliteraturetryingtodistinguishbetweentheseviewshasbeenbasedoncasestudiesorsurveys,makingithardtodistinguishbetweendifferentexplanationsandresultinginlittleconsensusinthemanagementliterature.ThisarticleprovidesexperimentalevidencesuggestingthatthereisasetofpracticesthatatQUARTERLYJOURNALOFECONOMICS4 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
leastinoneindustrywouldbeproÞtable,onaverage,forÞrmstoadopt.Third,recentlyanumberofotherÞeldexperimentsinde-velopingcountries(e.g.,Drexler,Fischer,andSchoar2010;KarlanandValdivia2011;BruhnandZia2011;Bruhn,Karlan,andSchoar2012;Karlan,Knight,andUdry2012)havebeguntoestimatetheimpactofbasicbusinesstrainingandadviceonmicro-andsmallenterprises.2Thisresearchhassofardeliveredmixedresults.SomestudiesÞndsigniÞcanteffectsofbusinesstrainingonÞrmperformancealthoughotherstudiesÞndnoeffect.Theevidencesuggeststhatdifferencesinthequalityandintensityoftraining,andthesizeoftherecipiententerprisesareimportantfactorsdeterminingtheimpactofbusinesstraining.Ourresearchbuildsonthisliteraturebyprovidinghigh-qualitymanagementconsultingtolarge,multiplantorganizations.II.MANAGEMENTINTHEINDIANTEXTILEINDUSTRYII.A.WhyWorkwithFirmsintheIndianTextileIndustry?DespiteIndiaÕsrecentrapidgrowth,totalfactorproductivityinIndiaisabout40%ofthatoftheUnitedStates(Caselli2011),withalargevariationinproductivity,spanningafewhighlypro-ductiveÞrmsandmanylow-productivityÞrms(HsiehandKlenow2009).Incommonwithotherdevelopingcountriesforwhichdataareavailable,IndianÞrmsarealsotypicallypoorlymanaged.EvidenceofthisisseeninFigureI,whichplotsresultsfromtheBloomandVanReenen(2010)(henceforthBVR)surveysofman-ufacturingÞrmsintheUnitedStatesandIndia.TheBVRmeth-odologyscoresÞrmsfrom1(worstpractice)to5(bestpractice)onmanagementpracticesrelatedtomonitoring,targets,andincen-tives.AggregatingthesescoresyieldsabasicmeasureoftheuseofmodernmanagementpracticesthatisstronglycorrelatedwithawiderangeofÞrmperformancemeasures,includingproductiv-ity,proÞtability,andgrowth.ThetoppanelofFigureIplotsthesemanagementpracticescoresforasampleof695randomlychosenU.S.manufacturingÞrmswith100to5,000employeesandthesecondpanelfor620similarlysizedIndianones.Theresults2.SeeMcKenzieandWoodruff(2012)foranoverviewofbusinesstrainingevaluations.DOESMANAGEMENTMATTER?5 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
revealathicktailofbadlyrunIndianÞrms,leadingtoaloweraveragemanagementscore(2.69forIndiaversus3.33forU.S.Þrms).IndianÞrmstendnottocollectandanalyzedatasystem-aticallyintheirfactories,theytendnottosetandmonitorcleartargetsforperformance,andtheydonotexplicitlylinkpayorpromotionwithperformance.ThescoresforBrazilandChinainthethirdpanel,withanaverageof2.67,aresimilar,suggestingthatthemanagementofIndianÞrmsisbroadlyrepresentativeoflargeÞrmsinemergingeconomies.ToimplementacommonsetofmanagementpracticesacrossÞrmsinourÞeldexperimentandmeasureacommonsetofout-comes,wefocusedononeindustry.WechosetextileproductionbecauseitisthelargestmanufacturingindustryinIndia,ac-countingfor22%ofmanufacturingemployment.Thefourthpanelshowsthemanagementscoresforthe232textileÞrmsintheBVRIndiansample,whichlookverysimilartoIndianman-ufacturingingeneral.FIGUREIManagementPracticeScoresacrossCountriesHistogramsusingBloomandVanReenen(2007)methodology.Double-blindsurveysusedtoevaluateÞrmsÕmonitoring,targets,andoperations.Scoresfrom1(worstpractice)to5(bestpractice).Samplesare695U.S.Þrms,620IndianÞrms,1,083BrazilianandChineseÞrms,232IndiantextileÞrms,and17experimentalÞrms.Datafromhttp://www.worldmanagementsurvey.com.QUARTERLYJOURNALOFECONOMICS6 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
Withintextiles,ourexperimentwascarriedoutin28plantsoperatedby17Þrmsinthewovencottonfabricindustry.Theseplantsweavecottonyarnintocottonfabricforsuits,shirtsandhomefurnishings.TheypurchaseyarnfromupstreamspinningÞrmsandsendtheirfabrictodownstreamdyeingandprocessingÞrms.AsshowninthebottompanelofFigureI,the17ÞrmsinvolvedhadanaverageBVRmanagementscoreof2.60,verysimilartotherestofIndianmanufacturing.Hence,ourparticularsampleof17IndianÞrmsalsoappearsbroadlysimilarintermsofmanagementpracticestomanufacturingÞrmsinmajordevelop-ingcountriesmoregenerally.II.B.TheSelectionofFirmsfortheFieldExperimentThesampleÞrmswererandomlychosenfromthepopulationofallpubliclyandprivatelyownedtextileÞrmsaroundMumbai,basedonlistsprovidedbytheMinistryofCorporateAffairs(MCA).3WerestrictedattentiontoÞrmswithbetween100to1,000employeestofocusonlargerÞrmsbutavoidmultinationals.Geographically,wefocusedonÞrmsinthetownsofTarapurandUmbergaon(thelargesttwotextiletownsinthearea)becausethisreducedthetraveltimefortheconsultants.Thisyieldedasampleof66potentialsubjectÞrms.AlloftheseÞrmswerethencontactedbytelephonebyourpartneringinternationalconsultingÞrm.Theyofferedfreecon-sulting,fundedbyStanfordUniversityandtheWorldBank,aspartofamanagementresearchproject.Wepaidfortheconsult-ingservicestoensurethatwecontrolledtheinterventionandcouldprovideahomogeneousmanagementtreatmenttoallÞrms.WewereconcernedthatiftheÞrmsmadeanycopayments,theymighthavetriedtodirecttheconsulting,forexample,askingforhelponmarketingorÞnance.OfthisgroupofÞrms,34expressedaninterestintheprojectandweregivenafollow-upvisitandsentapersonallysignedletterfromStanford.Ofthese34Þrms,17agreedtocommitseniormanagementtimetothe3.TheMCAlistcomesfromtheRegistrarofBusiness,withwhomallpublicandprivateÞrmsarelegallyrequiredtoregisterannually.OfcoursemanyÞrmsdonotregisterinIndia,butthisisgenerallyaproblemwithsmallerÞrms,notwithmanufacturingÞrmsofmorethan100employees,whicharetoolargeandperman-enttoavoidgovernmentdetection.DOESMANAGEMENTMATTER?7 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
consultingprogram.4WerefertotheseÞrmsinthesubsequentdiscussionasprojectÞrms.ThisofcoursegeneratesaselectionbiasinthatourresultsarevalidonlyforthesampleofÞrmsthatselectedintotheexperiment(Heckman1992).Wetooktwostepstoassesstheextentofthebias.First,wecomparedtheprojectÞrmswiththe49nonprojectÞrmsandfoundnosigniÞcantdifferences,atleastinobservables.5Second,inlate2011weranadetailedground-basedsurveyofeverytextileÞrmaroundMumbaiwith100to1,000employees(seeOnlineAppendixA2fordetails).WeidentiÞed172suchÞrmsandmanagedtointerview113ofthem(17projectÞrmsand96nonprojectÞrms).TheinterviewstookplaceattheÞrmsÕplantsorheadquartersandfocusedonowner-ship,size,managementpractices,andorganizationaldatafrom2008to2011.Wefoundthe17projectÞrmswerenotsigniÞcantlydifferentintermsofpreinterventionobservablesfromthe96non-projectÞrmsthatrespondedtothissurvey.6Althoughthepreviousresultsarecomfortinginthatourtreatmentandcontrolplantsappearedsimilartotheindustrypreinterventionalongobservables,thereisstillthepotentialissuethatselectionintotheexperimentalsamplewasdrivenbyunobservables.Wecannotruleoutthispossibility,thoughwenotethatthesignofthebiasisambiguousÑtheexperimentaleffectmaybelargerthantheeffectinthegeneralpopulationifÞrmswithmoretogainaremorelikelytoparticipate,oritmaybesmallerifÞrmswiththemosttogainfromimprovementarealso4.ThemainreasonsweweregivenforrefusingfreeconsultingwerethattheÞrmsdidnotbelievetheyneededmanagementassistanceorthatitrequiredtoomuchtimefromtheirseniormanagement(onedayaweek).ItisalsopossibletheseÞrmsweresuspiciousoftheoffer,givenmanyÞrmsinIndiahavetaxandregulatoryirregularities.5.TheseobservablesforprojectandnonprojectÞrmsaretotalassets,em-ployeenumbers,totalborrowings,andtheBVRmanagementscore,withvalues(p-valuesofthedifference)of$12.8mversus$13.9m(.841),204versus221(.552),$4.9mversus$5.5m(.756),and2.52versus2.55(.859),respectively.6.TheseobservablesforprojectandnonprojectÞrmsincludedage,largestplantsizein2008(inloomnumbers),largestplantsizein2008(inemployees),andadoptionofbasictextilemanagementpracticesin2008(seeOnlineAppendixTableAI)withvalues(p-valuesofthedifference)of22versus22.6years(.796),38versus42looms(.512),93versus112employees(.333),and0.381versus0.324practiceadoptionrates(.130),respectively.Wecomparedthesevaluesacrossthe17projectÞrmsandthe96nonprojectÞrmsusing2008datatoavoidanyeffectoftheexperiment.QUARTERLYJOURNALOFECONOMICS8 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
themostskepticalofwhatconsultantscando.7Nevertheless,becausetypicalpolicyeffortstooffermanagementtrainingtoÞrmswillalsorelyonÞrmsvolunteeringtoparticipate,webe-lieveourestimateoftheeffectofimprovingmanagementispolicyrelevantforthetypesofÞrmsthattakeadvantageofhelpwhenitisoffered.II.C.TheCharacteristicsoftheExperimentalFirmsTheexperimentalÞrmshadtypicallybeeninoperationfor20yearsandallwerefamily-owned.8Theyallproducedfabricforthedomesticmarket(althoughsomealsoexported).TableIreportssummarystatisticsforthetextilemanufacturingpartsoftheseÞrms(manyoftheÞrmshaveotherbusinessesintextileprocess-ing,retail,andevenrealestate).OnaveragetheseÞrmshadabout270employees,assetsof$13million,andsalesof$7.5mil-lionayear.ComparedtoU.S.manufacturingÞrms,theseÞrmswouldbeinthetop2%byemploymentandthetop4%bysales,andcomparedtoIndiamanufacturingtheyareinthetop1%bybothemploymentandsales(HsiehandKlenow2010).Hence,thesearelargemanufacturingÞrmsbymoststandards.9TheseÞrmsarealsocomplexorganizations,withamedianoftwoplantsperÞrm(plusaheadofÞceinMumbai)andfourreportinglevelsfromtheshopßoortothemanagingdirector.InalltheÞrms,themanagingdirectorwasthelargestshareholder,andalldirectorswerefamilymembers.TwoÞrmswerepubliclyquotedontheMumbaiStockExchange,althoughmorethan50%oftheequityineachwasheldbythemanagingfamily.InFiguresII,III,andIVandExhibitsO1toO4intheOnlineAppendix,weincludeasetofphotographsoftheplants.Theseareincludedtoprovidesomebackgroundinformationontheirsize,productionprocesses,andinitialstateofmanagement.Eachplantsiteinvolvedseveralmultistorybuildingsandoperated7.Thereisnowsomeevidenceontheimportanceofself-selectioninlaboratoryexperiments.Harrison,Lau,andRustrom(2009)Þndthattheseeffectsarerela-tivelysmallintheclassofexperimentstheyexamined,whereasLazear,Malmendier,andWeber(2012)Þndstrongerevidenceofself-selectionintoexperi-mentsonsocialpreferences.8.Interestingly,everysingleÞrminour113industrysamplewasalsofamily-ownedandmanaged.9.NotethatmostinternationalagenciesdeÞnelargeÞrmsasthosewithmorethan250employees.DOESMANAGEMENTMATTER?9 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
continuouslyÑ24hoursaday(intwo12-hourshifts),365daysayear.Thefactoryßoorswereoftendirty(FigureII)anddisorga-nized(FigureIII),andtheiryarnandsparepartsinventorystoresfrequentlylackedanyformalizedstoragesystems(FigureIV).Thisdisorganizedproductionledtofrequentqualitydefects(oilstains,brokenthreads,wrongcolors,etc.)necessitatinganextensivecheckingandmendingprocessthatemployed19%ofthefactorymanpower,onaverage.TABLEITHEFIELDEXPERIMENTSAMPLEAllTreatmentControlDiffMeanMedianMinMaxMeanMeanp-valueNumberofplants28n/an/an/a199n/aNumberofexperimentalplants20n/an/an/a146n/aNumberofÞrms17n/an/an/a116n/aPlantsperÞrm1.652141.731.50.393EmployeesperÞrm273250705002912360.454Employees,experimentalplants134132602501441140.161Hierarchicallevels4.44374.44.40.935Annualsales($m)perÞrm7.4561.415.67.068.370.598Currentassets($m)perÞrm8.505.211.8929.338.837.960.837Dailymtrs,experimentalplants5,5605,1302,26013,0005,7575,0910.602BVRmanagementscore2.602.611.893.282.502.750.203Managementadoptionrates0.2620.2570.0790.5530.2550.2880.575Age,experimentalplant(years)19.416.524620.516.80.662Qualitydefectsindex5.243.890.6116.44.477.020.395Inventory(1,000kilograms)61.172.87.4117.061.460.20.945Output(picks,million)23.325.46.932.122.125.80.271Productivity(inlogs)2.902.902.123.592.912.860.869Notes.Dataprovidedattheplantand/orÞrmleveldependingonavailability.NumberofplantsisthetotalnumberoftextileplantsperÞrmincludingthenonexperimentalplants.Numberofexperimentalplantsisthetotalnumberoftreatmentandcontrolplants.NumberofÞrmsisthenumberoftreatmentandcontrolÞrms.PlantsperÞrmreportsthetotalnumberoftextileplantsperÞrm.SeveraloftheseÞrmshaveotherbusinessesÑforexample,retailunitsandrealestatearmsÑwhicharenotincludedinanyoftheÞgureshere.EmployeesperÞrmreportsthenumberofemployeesacrossallthetextileproductionplants,thecorporateheadquarters,andsalesofÞce.Employees,experimentalplantsreportsthenumberofemployeesintheexperimentplants.HierarchicallevelsdisplaysthenumberofreportinglevelsintheexperimentalplantsÑforexample,aÞrmwithworkersreportingtoforeman,foremantooperationsman-ager,operationsmanagertothegeneralmanager,andgeneralmanagertothemanagingdirectorwouldhaveÞvehierarchicallevels.Annualsales($m)andCurrentassets($m)arebothin2009US$millionvalues,exchangedat50rupees=US$1.Dailymtrs,experimentalplantsreportsthedailymetersoffabricwovenintheexperimentplants.Notethatabout3.5metersisrequiredforafullsuitwithjacketandtrousers,sothemeanplantproducesenoughforabout1,600suitsdaily.BVRmanagementscoreistheBloomandVanReenen(2007)managementscorefortheexperimentalplants.ManagementadoptionratesaretheadoptionratesofthemanagementpracticeslistedinAppendixTableA.Iintheexperimentalplants.Ageofexperimentalplant(years)reportstheageoftheplantfortheexperimentalplants.Qualitydefectindexisaseverityweightedmeasureofproductionqualitydefects.Inventoryisthestockofyarnperintervention.Outputistheproductionoffabricinpicks(onepickisasinglerotationoftheweavingshuttle),andProductivitywhichislog(value-added)Ð0.42*log(capital)Ð0.58*log(totalhours).Allper-formancemeasuresarepooledacrosspreÐdiagnosticphasedata.QUARTERLYJOURNALOFECONOMICS10 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
FIGUREIIManyPartsofTheseFactoriesWereDirtyandUnsafeFIGUREIIITheFactoryFloorsWereFrequentlyDisorganizedDOESMANAGEMENTMATTER?11 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
III.THEMANAGEMENTINTERVENTIONIII.A.WhyUseManagementConsultingasanIntervention?TheÞeldexperimentaimedtoimprovemanagementpracticesinthetreatmentplants(whilekeepingcapitalandlaborinputsconstant)andmeasuretheimpactofdoingsoonÞrmperformance.Toachievethis,wehiredamanagementconsultancyÞrmtoworkwiththeplantsastheeasiestwaytochangeplant-levelmanage-mentrapidly.WeselectedtheconsultingÞrmusinganopentender.Thewinnerwasalargeinternationalmanagementcon-sultancythatisheadquarteredintheUnitedStatesandhasabout40,000employeesinIndia.Thefull-timeteamof(upto)sixconsultantsworkingontheprojectatanytimeallcamefromtheMumbaiofÞce.TheseconsultantswereeducatedatleadingIndianbusinessandengineeringschools,andmostofthemhadpriorexperienceworkingwithU.S.andEuropeanmultinationals.Selectingahigh-qualityinternationalconsultingÞrmsub-stantiallyincreasedthecostoftheproject.10However,itmeantFIGUREIVMostPlantsHadMonthsofExcessYarn,UsuallySpreadacrossMultipleLocations,OftenwithoutAnyRigorousStorageSystem10.AtthebottomoftheconsultingqualitydistributioninIndiaconsultantsarecheap,buttheirqualityispoor.Atthetopend,ratesaresimilartothoseintheQUARTERLYJOURNALOFECONOMICS12 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
thatourexperimentalÞrmsweremorepreparedtotrustthecon-sultants,whichwasimportantforgettingarepresentativesamplegroup.ItalsoofferedthelargestpotentialtoimprovethemanagementpracticesoftheÞrmsinourstudy.TheÞrst(andmain)waveoftheprojectranfromAugust2008toAugust2010,withatotalconsultingcostof$1.3million,approximately$75,000pertreatmentplantand$20,000percon-trolplant.ThisisdifferentfromwhattheÞrmsthemselveswouldhavetopayforthisconsulting,whichtheconsultantsindicatedwouldbeabout$250,000.ThereasonsforourlowercostsperplantarethattheconsultancyÞrmchargedusprobonorates(50%ofcommercialrates)asaresearchproject,providedfreepartnertime,andenjoyedconsiderableeconomiesofscalework-ingacrossmultipleplants.ThesecondwaveranfromAugust2011toNovember2011,withatotalconsultingcostof$0.4mil-lion,andfocusedoncollectinglonger-runperformanceandman-agementdata.TheintentiontoundertakethiswavewasnotmentionedinAugust2010(whentheÞrstwaveÞnished)toavoidanticipationeffects.Althoughtheinterventionofferedhigh-qualitymanagementconsulting,thepurposeofourstudywastousetheimprovementsinmanagementgeneratedbythisinterventiontounderstandif(andhow)modernmanagementpracticesaffectÞrmperform-ance.LikemanyrecentdevelopmentÞeldexperiments,thisinterventionwasprovidedasamechanismofconvenienceÑtochangemanagementpracticesÑandnottoevaluatethemanage-mentconsultantsthemselves.III.B.TheManagementConsultingInterventionTheinterventionaimedtointroduceasetofstandardman-agementpractices.Basedontheirpriorindustryexperience,theconsultantsidentiÞed38keypracticesonwhichtofocus.ThesepracticesencompassarangeofbasicmanufacturingprinciplesthatarestandardinU.S.,European,andJapaneseÞrms,andcanbegroupedintofollowingÞveareas..Factoryoperations:RegularmaintenanceofmachinesandrecordingthereasonsforbreakdownstolearnfromUnitedStatesbecauseinternationalconsultingcompaniestargetmultinationalsandemployconsultantswhoareoftenU.S.-orEuropean-educatedandhaveaccesstointernationallabormarkets.DOESMANAGEMENTMATTER?13 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
failures.Keepingthefactoryßoortidytoreduceacci-dentsandfacilitatethemovementofmaterials.Establishingstandardproceduresforoperations..Qualitycontrol:Recordingqualityproblemsbytype,ana-lyzingtheserecordsdaily,andformalizingprocedurestoaddressdefectstopreventtheirrecurrence..Inventory:Recordingyarnstocksonadailybasis,withoptimalinventorylevelsdeÞnedandstockmonitoredagainstthese.Yarnsorted,labeled,andstoredinthewarehousebytypeandcolor,andthisinformationloggedontoacomputer..Humanresourcesmanagement:Performance-basedin-centivesystemsforworkersandmanagers.Jobdescrip-tionsdeÞnedforallworkersandmanagers..Salesandordermanagement:Trackingproductiononanorder-wisebasistoprioritizecustomerordersbydeliverydeadline.Usingdesign-wiseefÞciencyanalysissopricingcanbebasedonactual(ratherthanaverage)productioncosts.Thesepractices(listedinAppendixTableA.I)formasetofpreciselydeÞnedbinaryindicatorsthatwecanusetomeasurechangesinmanagementpracticesasaresultoftheconsultingintervention.11Ageneralpatternatbaselinewasthatplantsre-cordedavarietyofinformation(ofteninpapersheets),buthadnosystemsinplacetomonitortheserecordsorroutinelyusethemindecisions.Thus,although93%ofthetreatmentplantsrecordedqualitydefectsbeforetheintervention,only29%monitoredthemonadailybasisorbytheparticularsortofdefect,andnonehadanystandardizedsystemtoanalyzeandactonthisdata.Theconsultingtreatmenthadthreephases.TheÞrstphase,calledthediagnosticphase,tookonemonthandwasgiventoalltreatmentandcontrolplants.Itinvolvedevaluatingthecurrentmanagementpracticesofeachplantandconstructingaperform-ancedatabase.Constructionofthisdatabaseinvolvedsettingupprocessesformeasuringarangeofplant-levelmetricsÑsuchas11.WeprefertheseindicatorstotheBVRmanagementscoreforourworkhere,becausetheyareallbinaryindicatorsofspeciÞcpracticesthataredirectlylinkedtotheintervention.Incontrast,theBVRindicatormeasurespracticesatamoregen-erallevelonaÞve-pointordinalscale.Nonetheless,thesumofour38preinterven-tionmanagementpracticescoresiscorrelatedwiththeBVRscoreat0.404(p-valueof.077)acrossthe17projectÞrms.QUARTERLYJOURNALOFECONOMICS14 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
output,efÞciency,quality,inventory,andenergyuseÑonanon-goingbasis,plusextractinghistoricaldatafromexistingrecords.Forexample,tofacilitatequalitymonitoringonadailybasis,asinglemetric,calledtheQualityDefectsIndex(QDI),wascon-structedasaseverity-weightedaverageofthemajortypesofde-fects.Attheendofthediagnosticphase,theconsultingÞrmprovidedeachplantwithadetailedanalysisofitscurrentman-agementpracticesandperformanceandrecommendationsforchange.Thisphaseinvolvedabout15daysofconsultingtimeperplantoverthecourseofamonth.Thesecondphasewasafour-monthimplementationphasegivenonlytothetreatmentplants.Inthisphase,theconsultingÞrmfolloweduponthediagnosticreporttohelpintroduceasmanyofthekeymanagementpracticesastheÞrmscouldbepersuadedtoadopt.Theconsultantassignedtoeachplantworkedwiththeplantmanagementtoputtheproceduresintoplace,Þne-tunethem,andstabilizethemsothatemployeescouldreadilycarrythemout.Forexample,oneofthepracticeswasholdingdailymeetingsformanagementtoreviewproductionandqualitydata.TheconsultantattendedthesemeetingsfortheÞrstfewweekstohelpthemanagersrunthem,providedfeedbackonhowtorunfuturemeetings,andadjustedtheirdesign.Thisphasealsoinvolvedabout15daysamonthofcon-sultingtimeperplant.Thethirdphasewasameasurementphase,whichlastedintheÞrstwaveuntilAugust2010,andtheninthesecond(follow-up)wavefromAugust2011toNovember2011.Thisinvolvedcollectionofperformanceandmanagementdatafromalltreatmentandcontrolplants.Inreturnforthiscontinuingdata,theconsultantsprovidedlightconsultingadvicetothetreatmentandcontrolplants.Thisphaseinvolvedabout1.5daysamonthofconsultingperplant.Insummary,thecontrolplantswereprovidedwiththediag-nosticphaseandthenthemeasurementphases(totaling273con-sultanthoursonaverage),andthetreatmentplantswereprovidedwiththediagnostic,implementation,andthenmeasure-mentphases(totaling781consultanthoursonaverage).III.C.TheExperimentalDesignWewantedtoworkwithlargeÞrmsbecausetheircomplexitymeanssystematicmanagementpracticeswerelikelytobeDOESMANAGEMENTMATTER?15 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
important.However,providingconsultingtolargeÞrmsisexpen-sive,whichnecessitatedanumberoftrade-offsdetailedhere.1.Cross-sectionalsamplesize.Weworkedwith17Þrms.Weconsideredhiringcheaperlocalconsultantsandprovidingmorelimitedconsultingtoasampleofseveralhundredplantsinmorelocations.Twofactorspushedagainstthis.First,manylargeÞrmsinIndiaarereluctanttoletoutsidersintotheirplants,probablybecauseofcomplianceissueswithvariousregulations.Tominimizeselectionbias,weofferedahigh-qualityintensiveconsultinginterventionthatÞrmswouldvalueenoughtoriskallowingoutsidersintotheirplants.Thishelpedmaximizeinitialtake-up(26%asnotedinSectionII.B)andretention(100%,asnoÞrmsdroppedout).Second,theconsensusfromdiscussionswithmembersoftheIndianbusinesscommunitywasthatachievingameasurableimpactinlargeÞrmswouldrequireanextendeden-gagementwithhigh-qualityconsultants.Obviously,thetrade-offwasthatthisledtoasmallsamplesize.WediscusstheestimationissuesthisgeneratesinSectionIII.D.2.Treatmentandcontrolplants.The17Þrmsthatagreedtoparticipateintheprojecthad28plantsbetweenthem.Ofthese,25wereeligibletobetreatmentorcontrolplantsbecausewecouldobtainhistoricperformancedatafromthem.Randomiza-tionoccurredattheÞrmlevelandwasconductedbycomputer.WeÞrstrandomlychosesixÞrmstobethecontrolÞrms,andoneeligibleplantfromeachofthemtobethecontrolplants.Theremaining11ÞrmswerethenthetreatmentÞrms.OurinitialfundingandcapacityconstraintsoftheconsultingteammeantthatwecouldstartwithfourplantsasaÞrstround,whichstartedinSeptember2008.Wethereforerandomlychose4ofthe11treatmentÞrmstobeinround1,randomlyselectingoneplantfromeachÞrm.InApril2009,westartedasecondroundoftreatment.ThiscomprisedselectingarandomplantfromeachoftheremainingseventreatmentÞrms,and,becausefundingallowedforit,threemoreplantsselectedatrandomfromthetreatmentÞrmswithmultipleplants.Purerandomization,ratherthanstratiÞcationorrerandomizing,wasusedineachstep.ThiswasdonebothbecauseofinitialuncertaintyastohowmanyplantswewouldhavefundingtotreatandbecauseofconcernsaboutvarianceestimationandpowerwhenstratiÞedQUARTERLYJOURNALOFECONOMICS16 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
randomizationisusedinverysmallsamples(BruhnandMcKen-zie2009).Theresultisthatwehave11treatmentÞrms,with14treat-mentplantsamongthem,and6controlÞrms,eachwithacontrolplant.Wepickedmoretreatmentthancontrolplantsbecausethestaggeredinitiationoftheinterventionsmeantthedifferenttreatmentgroupsprovidedsomecross-identiÞcationforeachotherandbecausewebelievedthetreatmentplantswouldbemoreusefulforunderstandingwhyÞrmshadnotadoptedman-agementpracticesbefore.TableIshowsthatthetreatmentandcontrolÞrmswerenotstatisticallydifferentacrossanyofthecharacteristicswecouldobserve.12TheremainingeightplantswerethenclassiÞedasÔÔnonexperimentalplantsÕÕ:threeincontrolÞrmsandÞveintreatmentÞrms.Thesenonexperimentalplantsdidnotdirectlyreceiveanyconsultingservices,butdataontheirmanagementpracticeswerecollectedinbimonthlyvisits.3.Timing.Theconsultinginterventionwasexecutedinthreeroundsbecauseofthecapacityconstraintofthesix-personcon-sultingteam.TheÞrstroundstartedinSeptember2008withfourtreatmentplants.InApril2009asecondroundwith10treatmentplantswasinitiated,andinJuly2009theÞnalroundwith6controlplantswascarriedout.FirmrecordsusuallyallowedustocollectdatagoingbacktoacommonstartingpointofApril2008.WestartedwithasmallÞrstroundbecauseweexpectedtheinterventionprocesstogeteasierovertimeduetoaccumulatedexperience.Thesecondroundincludedalltheremainingtreat-mentÞrmsbecause(1)theconsultinginterventionstaketimetoaffectperformanceandwewantedthelongesttimewindowtoobservethetreatmentÞrms,and(2)wecouldnotmixthetreat-mentandcontrolÞrmsacrossimplementationrounds.13ThethirdroundcontainedthecontrolÞrms.12.Wetestfordifferencesinmeansacrosstreatmentandcontrolplants,clus-teringattheÞrmlevel.13.Eachroundhadaone-daykick-offmeetinginvolvingpresentationsfromseniorpartnersfromtheconsultingÞrm.ThishelpedimpresstheÞrmswiththeexpertiseoftheconsultingÞrmandhighlightedthepotentialforperformanceim-provements.Becausethismeetinginvolvedaprojectoutline,andwedidnottellÞrmsaboutthedifferenttreatmentandcontrolprograms,wecouldnotmixthegroupsinthemeetings.DOESMANAGEMENTMATTER?17 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
III.D.SmallSampleSizeThefocusonlargeÞrmsmeantwehadtoworkwithasmallnumberofÞrms.Thisraisesthreebroadissues.AÞrstpotentialconcerniswhetherthesamplesizeistoosmalltoidentifysigniÞ-cantimpacts.Asecondiswhattypeofstatisticalinferenceisap-propriategiventhesamplesize.ThethirdpotentialconcerniswhetherthesampleistoosmalltoberepresentativeoflargeÞrmsindevelopingcountries.Wediscusseachconcerninturnandthestepswetooktoaddressthem.1.SigniÞcanceofresults.Eventhoughwehadonly20experi-mentalplantsacrossthe17projectÞrms,weobtainedstatistic-allysigniÞcantresults.ThereareÞvereasonsforthis.First,thesearelargeplantswithabout80loomsandabout130employeeseach,sothatidiosyncraticshocksÑlikemachinebreakdownsorworkerillnessÑtendedtoaverageout.Second,thedatawerecollecteddirectlyfromthemachinelogs,andhadverylittle(ifany)measurementerror.Third,theÞrmswerehomogeneousintermsofsize,product,region,andtechnology,sotimedummiescontrolledformostexternalshocks.Fourth,wecol-lectedweeklydata,whichprovidedhigh-frequencyobservationsoverthecourseofthetreatment.Theuseoftheserepeatedmeas-urescanreducethesamplesizeneededtodetectagiventreat-menteffect,althoughthisistemperedbythedegreeofserialcorrelationintheoutputdata(McKenzie2012),whichwasaround0.8forourperformancemetrics.Finally,theinterventionwasintensive,leadingtolargetreatmenteffectsÑforexample,thepointestimateforthereductioninqualitydefectswasalmost50%.2.Statisticalinference.Asecondconcernisoverusingstatis-ticalteststhatrelyonasymptoticargumentsinthecross-sectionaldimension(here,thenumberofÞrms)tojustifythenormalapproximation.Weusethreealternativestoaddressthisconcern.First,weuseÞrm-clusteredbootstrapstandarderrors(Cameron,Gelbach,andMiller2008).Second,weimple-mentpermutationprocedures(forboththeintentiontotreatandinstrumentalvariablesestimators)thatdonotrelyonasymptoticapproximations.Third,weexploitourlargeTsampletoQUARTERLYJOURNALOFECONOMICS18 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
implementproceduresthatrelyonasymptoticapproximationsalongthetimedimension(withaÞxedcross-sectionaldimension).143.Representativenessofthesample.AthirdconcernwithoursmallsampleishowrepresentativeitisoflargeÞrmsindevelop-ingcountries.Inpart,thisconcernrepresentsageneralissueforÞeldexperiments,whichareoftenrunonindividuals,villages,orÞrmsinparticularregionsorindustries.Inoursituation,wefocusedononeregionandoneindustry,albeitIndiaÕscommercialhub(Mumbai)anditslargestindustry(textiles).Comparingoursampletothepopulationoflarge(100to5,000employee)ÞrmsinIndia,bothoverallandintextiles,suggeststhatoursmallsampleisatleastbroadlyrepresentativeintermsofmanagementprac-tices(seeFigureI).InOnlineAppendixFigureOI,wealsoplotresultsonaplant-by-plantbasistofurtherdemonstratethattheresultsarenotdrivenbyanyparticularplantoutlier.Althoughwehaveasmallsample,theresultsarerelativelystableacrosstheindividualsampleplants.III.E.PotentialConßictofInterestAÞnaldesignchallengewasthepotentialforaconßictofinterestinhavingourconsultingÞrmmeasuringtheperform-anceoftheexperimentalplants.Toaddressthis,weÞrsthadtwograduatestudentscollectivelyspendsixmonthswiththeconsultingteaminIndiaoverseeingthedailydatacollection.Second,abouteveryothermonthonememberoftheresearchteamvisitedtheÞrms,metwiththedirectors,andpresentedthequality,inventory,andoutputdatatheconsultantshadsentus.Thiswaspositionedasawaytoinitiatediscussionsontheimpactoftheexperimentwiththedirectors,butitalsoservedtocheckthatthedatawewerereceivingreßectedreality.WewouldlikelyhavereceivedsomepushbackiftheresultshadbeenatvariancewiththeownersÕownjudgment.Finally,someofthelong-rundata,likethenumberofplants,isdirectlyobserv-able,soitwouldbehardfortheconsultingÞrmtofabricatethis.14.ThesepermutationandlargeTproceduresaresummarizedintheAppendixanddetailedinOnlineAppendixB.DOESMANAGEMENTMATTER?19 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
IV.THEIMPACTONMANAGEMENTPRACTICESInFigureV,weplottheaveragemanagementpracticeadop-tionofthe38practicesforthe14treatmentplants,the6controlplants,the8nonexperimentalplants,andour96nonprojectÞrmssurveyedin2011.Thismanagementpracticescoreisthepropor-tionofthe38practicesaplanthadadopted.ThisdatafortheprojectÞrmsisshownat2-monthintervalsstarting10monthsbeforethediagnosticphaseandextendingtoatleast24monthsafter.ThenonprojectÞrmdatawascollectedatayearlyfre-quencyusingretrospectiveinformation.FortheprojectÞrms,datafromthediagnosticphaseonwardwerecompiledfromdirectobservationatthefactory,anddatafrombeforethediag-nosticphasewerecollectedfromdetailedretrospectiveinterviewsoftheplantmanagementteam.ForthenonprojectÞrms,datawerecollectedduringtheinterviewfromdirectfactoryobserva-tionanddetaileddiscussionwiththemanagers(detailsinOnlineAppendixAII).FigureVshowssixresults.First,allplantsstartedoffwithlowbaselineadoptionratesofthesetofmanagementpractices.15Amongthe28individualplantsintheprojectÞrms,theinitialadoptionratesvariedfromalowof7.9%toahighof55.3%,sothateventhebest-managedplantinthegrouphadjustoverhalfofthekeytextilemanufacturingpracticesinplace.ThisisconsistentwiththeresultsonpoorgeneralmanagementpracticesinIndianÞrmsshowninFigureI.16Forexample,manyoftheplantsdidnothaveanyformalizedsystemforrecordingorimprovingproductionquality,whichmeantthatthesamequalitydefectcouldariserepeatedly.Mostalsohadnotorganizedtheiryarninventories,soyarnstoreswerefrequentlymixedbycolorandtype,withoutlabelingorcomputerizedentry.Theproductionßoorwasoftenblockedbywaste,tools,andmachinery,impedingtheßowofworkersandmaterialsaroundthefactory.Second,theinterventiondidsucceedinchangingmanage-mentpractices.Thetreatmentplantsincreasedtheiruseofthe38practicesby37.8percentagepointsonaveragebyAugust2010,whenthemainwaveended(anincreasefrom25.6%to15.Thepretreatmentdifferencebetweenthetreatment,control,andotherplantgroupsisnotstatisticallysigniÞcant,withap-valueonthedifferenceof.550(seeAppendixTableA.I).16.Interestingly,Clark(1987)suggestsIndiantextileplantsmayhaveevenbeenbadlymanagedintheearly1900s.QUARTERLYJOURNALOFECONOMICS20 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
63.4%).Theseimprovementsinmanagementpracticeswerealsopersistent.Themanagementpracticeadoptionratesdroppedbyonly3percentagepoints,onaverage,betweentheendoftheÞrstwaveinAugust2010(whentheconsultantsleft)andthestartofthesecondwaveinAugust2011.Third,notallpracticeswereadopted.TheÞrmsappearedtoadoptthepracticesthatweretheeasiesttoimplementand/orhadthelargestperceivedshort-runpay-offs,likethedailyquality,inventory,andefÞciencyreviewmeetings.Ifso,thechoiceofpracticeswasendogenousanditpresumablyvariedwiththecost-beneÞtcalculationforeachpractice.17FIGUREVTheAdoptionofKeyTextileManagementPracticesoverTimeAverageadoptionratesofthe38keytextilemanufacturingmanagementpracticeslistedinTableA.I.Shownforthe14treatmentplants(diamond),6controlplants(plussign),the5nonexperimentalplantsinthetreatmentÞrmstowhichtheconsultantsdidnotprovideanydirectconsultingassistance(smallcircle),the3nonexperimentalplantsinthecontrolÞrms(largecircle),and96plantsfromtherestoftheindustryaroundMumbai(square).Scoresrangefrom0(ifnoneofthegroupofplantshaveadoptedanyofthe38managementpractices)to1(ifallofthegroupofplantshaveadoptedallofthe38manage-mentpractices).InitialdifferencesacrossallthegroupsarenotstatisticallysigniÞcant.The96plantsfromtherestoftheindustryweregiventhesamediagnosticphasestartdateasthecontrolplants(July2009).17.See,forexample,Suri(2011)forarelatedÞndingonheterogeneousagri-culturaltechnologyadoptioninKenya.DOESMANAGEMENTMATTER?21 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
Fourth,thetreatmentplantsÕadoptionofmanagementprac-ticesoccurredgraduallyandnonuniformly.Inlargepart,thisreßectsthetimetakenfortheconsultingÞrmtogaintheconÞ-denceofthedirectors.Initiallymanydirectorswereskepticalaboutthesuggestedmanagementchanges,andtheyoftenstartedbypilotingtheeasiestchangesaroundqualityandinventoryinonepartofthefactory.Oncethesestartedtogenerateimprove-ments,thesechangeswererolledoutandtheÞrmsthenbeganintroducingthemorecompleximprovementsaroundoperationsandhumanresources.Fifth,thecontrolplants,whichweregivenonlytheone-monthdiagnostic,increasedtheiradoptionofthemanagementpractices,butbyonly12percentagepoints,onaverage.Thisissubstantiallylessthantheincreaseinadoptioninthetreat-mentÞrms,indicatingthatthefourmonthsoftheimplemen-tationphasewereimportantinchangingmanagementpractices.However,itisanincreaserelativetotherestoftheindustryaroundMumbai(thenonprojectplants),whichdidnotchangetheirmanagementpracticesonaveragebetween2008and2011.Finally,thenonexperimentalplantsinthetreatmentÞrmsalsosawasubstantialincreaseintheadoptionofmanagementpractices.IntheseÞveplants,theadoptionratesincreasedby17.5percentagepointsbyAugust2010.ThisincreaseoccurredbecausethedirectorsofthetreatmentÞrmscopiedthenewprac-ticesfromtheirexperimentalplantsovertotheirotherplants.Interestingly,thisincreaseinadoptionratesissimilartothecon-trolÞrmsÕ12percentagepointincrease,suggestingthatcopyingbestpracticesacrossplantswithinÞrmscanbeasleastaseffect-iveatimprovingmanagementpracticesasshort(one-month)burstsofexternalconsulting.V.THEIMPACTOFMANAGEMENTONPERFORMANCEV.A.IntentiontoTreatEstimatesWeestimatetheimpactoftheconsultingservicesonman-agementpracticesviathefollowingintentiontotreat(ITT)equation:OUTCOMEi,t¼aTREATi,tþbDURINGi,tþctþdiþei,t,ð1ÞQUARTERLYJOURNALOFECONOMICS22 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
whereOUTCOMEisoneofthekeyperformancemetricsofquality,inventory,output,andtotalfactorproductivity(TFP).18TFPisdeÞnedaslog(valueadded)Ð0.42*log(capital)Ð0.58*log(labor),wherethefactorweightsarethecostsharesforcottonweavingintheIndianAnnualSurveyofIndustry(2004Ð5),capitalincludesallphysicalcapital(land,buildings,equip-ment,andinventory),andlaborisproductionhours(seeOnlineAppendixAIfordetails).TREATi,ttakesthevalueofoneforthetreatmentplantsstartingonemonthaftertheendoftheinter-ventionperiodanduntiltheendofthestudyandiszeroother-wise,whileDURINGi,ttakesthevalueofoneforthetreatmentplantsforthesix-monthwindowfromthestartofthediagnosticperiod.Thectareafullsetofweeklytimedummiestocontrolforseasonality,andthediareafullsetofplantdummiesthatwereincludedtocontrolfordifferencesbetweenplantssuchasthescalingofQDI(perpiece,perroll,orpermeteroffabric)ortheloomwidth(apickÑonepassoftheshuttleÑonadouble-widthloomproducestwiceasmuchfabricasapickonsingle-widthloom).TheparameteragivestheITT,whichistheaverageimpactoftheimplementationinthetreatedplants,andbshowstheshort-termimpactduringtheimplementation.19InadditiontothisspeciÞcation,inAppendixTableA.IIweestimatetheimpactonoutcomesofourindexofmanagementpractices,usingtheconsultingservicesasaninstrument,andcomparetheinstrumentalvariables(IV)resultstotheÞxed-effectsestimates.WeÞndthattheÞxed-effectsestimatestendtounderstatethegaininperformancefrombettermanage-ment,whichisconsistentwithchangesinmanagementbeingmorelikelytobeimplementedwhenoutcomesaredeclining.WeuseperformancedatauptothestartofSeptember2010,sincethedataarenotcomparableafterthisdatebecauseofin-vestmentinnewloomsinsometreatmentplants.TheÞrm18.Westudyquality,inventory,andoutputbecausethesearerelativelyeasytomeasurekeyproductionmetricsformanufacturing.TheyalsodirectlyinßuenceTFPbecausepoorqualityleadstomoremendingmanpower(increasinglabor)andwastesmorematerials(loweringvalueadded),highinventoryincreasescapital,andloweroutputreducesvalueadded.19.Inthecasethatavariesacrossplants,ourestimateofawillbeaconsistentestimateoftheaveragevalueofai.Notethatbecausethediagnosticwasreceivedbybothtreatmentandcontrolplants,theITTestimatestheeffectoftheimplementa-tionontreatmentplantsinasituationwherebothtreatmentandcontrolplantsreceivethediagnostic.DOESMANAGEMENTMATTER?23 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
directorsbeganreplacingolderSulzerandRapierloomswithJacquardlooms,whichproducehighermark-upfabricbutre-quiremoreadvancedqualitycontrolandmaintenancepractices.ThisstartedinSeptember2010aftertheend-of-summerproduc-tionsurgefortheweddingseasonandtheDiwaliholiday.InTableIIcolumn(1)weseethattheITTestimateforqual-itydefectsshowsasigniÞcantdropof25%occurringjustduringtheimplementationperiod,eventuallyfallingfurthertoa43%drop.20ThisisshownovertimeinFigureVI,whichplotstheQDIscoreforthetreatmentandcontrolplantsrelativetothestartofthetreatmentperiod:September2008forround1treat-mentandApril2009forround2treatmentandcontrolplants.21Thescoreisnormalizedto100forbothgroupsofplantsusingpretreatmentdata.Togeneratepoint-wiseconÞdenceintervalsweblock-bootstrappedovertheÞrms.ThetreatmentplantsstartedtoreducetheirQDIscores(i.e.,improvequality)signiÞcantlyandrapidlyfromaboutweek5onward,whichwasthebeginningoftheimplementationphasefollowingtheinitialone-monthdiagnosticphase.ThecontrolÞrmsalsoshowedamildanddelayeddownwardtrendintheirQDIscores,consistentwiththeirslowertake-upoftheseprac-ticesintheabsenceofaformalimplementationphase.Thelikelyreasonforthishugereductionindefectsisthatmeasuring,classifying,andtrackingdefectsallowsÞrmstoad-dressqualityproblemsrapidly.Forexample,afaultyloomthatcreatesweavingerrorswouldbepickedupinthedailyQDIscoreanddealtwithinthenextdayÕsqualitymeeting.Withoutthis,theproblemwouldoftenpersistforseveralweeks,sincethecheckingandmendingteamhadnomechanism(orincentive)toreducedefects.Inthelongerterm,theQDIalsoallowedmanagerstoidentifythelargestsourcesofqualitydefectsbytype,design,yarn,loom,andweaverandstarttoaddressthesesystematically.Forexample,designswithcomplexstitchingthatgeneratelargenumbersofqualitydefectscouldbedroppedfromthesalescata-log.Thisabilitytoimprovequalitydramaticallythroughsystem-aticdatacollectionandevaluationisakeyelementofthelean20.Notethatqualityisestimatedinlogs,sothatthepercentagereductionisÐ43.1=exp(Ð0.564)Ð1.21.Becausethecontrolplantshavenotreatmentperiod,wesettheirtimingtozerotocoincidewiththe10round2treatmentplants.Thismaximizestheoverlapofthedata.QUARTERLYJOURNALOFECONOMICS24 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
TABLEIITHEIMPACTOFMODERNMANAGEMENTPRACTICESONPLANTPERFORMANCE(1)(2)(3)(4)(5)(6)(7)(8)DependentvariableQualitydefectsInventoryOutputTFPQualitydefectsInventoryOutputTFPSpeciÞcationITTITTITTITTWeeksoftreatmentWeeksoftreatmentWeeksoftreatmentWeeksoftreatmentInterventioni,t0.564**0.245**0.090**0.154*(0.235)(0.117)(0.037)(0.084)Duringimplementationi,t0.293**0.0700.0150.048(0.137)(0.093)(0.031)(0.056)Cumulativetreatmenti,t0.032**0.015**0.006***0.009**(0.013)(0.005)(0.002)(0.004)SmallsamplerobustnessIbragimov-Mueller(95%CI)[1.65,0.44][0.83,-0.02][0.05,0.38][0.014,0.79]Permutationtest(p-value).001.060.026.061TimeFEs127127127127127127127127PlantFEs2018201820182018Observations1,8072,0522,3931,8311,8072,0522,3931,831Notes.Allregressionsuseafullsetofplantandcalendarweekdummies.StandarderrorsarebootstrapclusteredattheÞrmlevel.Interventionisaplantleveldummyequalto1aftertheimplementationphaseattreatmentplantsand0otherwise.Duringimplementationisadummyvariableequalto1fromthebeginningofthediagnosticphasetotheendoftheimplementationphaseforalltreatmentplants.Cumulativetreatmentisthecumulativeweeksoftreatmentsincethebeginningoftheimplementationphaseineachplant(0inboththecontrolgroupandpriortotheimplementationphaseinthetreatmentgroup).Qualitydefectsisthelogofthequalitydefectsindex(QDI),whichisaweightedaveragescoreofqualitydefects,sohighernumbersimplyworsequalityproducts(morequalitydefects).Inventoryisthelogofthetonsofyarninventoryintheplant.Outputisthelogoftheweavingproductionpicks.TFPisplant-leveltotalfactorproductivitydeÞnedaslog(output)measuredinproductionpickslesslog(capital)timescapitalshareof0.42lesslog(labor)timeslaborcostsshareof0.58.ITTreportstheintentiontotreatresultsfromregressingthedependentvariabledirectlyontheinterventiondummy.TimeFEsreportthenumberofcalendarweektimeÞxedeffects.PlantFEsreportsthenumberofplant-levelÞxedeffects.Twoplantsdonothaveanyinventoryonsite,sonoinventorydataareavailable.Smallsamplerobustnessimplementstwodifferentprocedures(describedingreaterdetailintheAppendixandOnlineAppendixB)toaddressissuesofplanthetero-geneity,withinplant(andÞrm)correlation,andsmallsampleconcerns.Ibragimov-Mueller(95%CI)report95%conÞdenceintervalestimatesfromÞrm-by-Þrmparameterestimatestreatingtheestimatesasdrawsfromindependent(butnotidenticallydistributed)normaldistributionsandconductsatwo-samplet-test.Permutationtestreportsthep-valuesfortestingthenullhypothesisthatthetreatmenthasnoeffectfortheITTparameterbyconstructingapermutationdistributionoftheITTestimateusingthe12,376possiblepermutationsoftreatmentassignment.ThesetestshaveexactÞnitesamplesize.***denotes1%,**denotes5%,*denotes10%signiÞcance.DOESMANAGEMENTMATTER?25 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
manufacturingsystemofproduction,andinfactmanyU.S.auto-motiveplantssawreductionsindefectsofover90%followingtheadoptionofleanproductionsystems(see,forexample,Womack,Jones,andRoos1990).AtthebottomofTableIIwealsopresentresultsfromourrobustnesschecks:theIbramigov-Mueller(IM)andper-mutationtests.Theresultsareconsistentwithareductioninqualitydefects.First,lookingatthepermutationteststhathaveexactsize,weseethattheITTissigniÞcantatthe5%level(thep-valueis.01).TheIMapproachthatexploitsasymp-toticsinTratherthanNÞndsthattheITTresultsareconsistentwithlargeimprovementsinqualitythoughtheconÞdenceinter-valsarewide.Column(2)reportstheresultsforinventorywitha21.7%(=exp(Ð0.245)Ð1)post-treatmentreduction,andnosigniÞcantchangeduringtheimplementationphase.FigureVIIshowstheFIGUREVIQualityDefectsIndexfortheTreatmentandControlPlantsDisplaystheaverageweeklyqualitydefectsindex,whichisaweightedindexofqualitydefects,soahigherscoremeanslowerquality.Thisisplottedforthe14treatmentplants(plussigns)andthe6controlplants(diamonds).Valuesnormalizedsobothserieshaveanaverageof100priortothestartoftheintervention.ToobtainconÞdenceintervalswebootstrappedtheÞrmswithreplacement250times.NotethatseasonalityduetoDiwaliandtheweddingseasonaffectsbothgroupsofplants.QUARTERLYJOURNALOFECONOMICS26 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
plotofinventoryovertime.Thefallininventoryintreatmentplantsoccurredbecausetheywerecarryingaboutfourmonthsofrawmaterialsinventoryonaveragebeforetheintervention,includingalargeamountofdeadstock.Becauseofpoorrecordsandstoragepractices,theplantmanagerstypicallydidnotevenknowtheyhadthesestocks.Aftercataloging,theÞrmssoldorusedupthesurplusyarn(byincorporationintonewdesigns),andthenintroducedrestockingnormsforfuturepurchasesandmoni-toredagainsttheseweekly.Thisprocesstooksometime,andhenceinventoriesdonotrespondasquicklyasqualitytotheintervention.Incolumn(3)ofTableII,welookatoutputandseea9.4%(=exp(+0.09)Ð1)increaseinoutputfromtheintervention,withnosigniÞcantchangeduringtheimplementationperiod.Severalchangesdrovethisincrease.First,thehalvinginqualitydefectsmeanttheamountofoutputbeingscrapped(5%beforethebegin-ningoftheexperiment)fellconsiderably(weestimatebyabout50%).Second,undertakingroutinemaintenanceoftheloomsandcollectingandmonitoringbreakdowndatapresumablyalsohelpedreducedmachinedowntime.VisualdisplaysaroundthefactoryßoortogetherwiththeincentiveschemesalsoencouragedworkerstoimproveoperatingefÞciencyandattendancelevels.Finally,keepingthefactoryßoorcleanandtidyreducedthenumberofuntowardincidentsliketoolsfallingintomachinesorfactoryÞres.Incolumn(4),weshowtheresultsforlogTFPreportinga16.6%(exp(0.154)Ð1)increaseinthetreatmentÞrmsposttreat-mentcomparedtothecontrolÞrms.Productivityincreasedbe-causeoutputwentup(asshownincolumn(3)),capitaldropped(becauseoflowerinventorylevels,asshownincolumn(2)),andmendinglabordropped(asthenumberofqualitydefectsfellasshownincolumn(1)).FigureVIIIshowsthetimeproÞleofprod-uctivity,showingittooktimetoseeimpacts,whichiswhythereisnosigniÞcanteffectduringimplementation.Incolumns(5)to(8)ofTableII,weestimateresultsusingatime-varyingtreatmentindicator,whichisweeksofcumulativeimplementation.WeincludedthesefourcolumnssincethechangesinmanagementpracticesandoutcomesoccurredslowlyoverthetreatmentperiodasFiguresVtoVIIIhighlight.Wealsoplottedthedifferenceinquality,inventory,andoutputaftertreatmentonaplant-by-plantbasis(OnlineAppen-dixFigureOI).AlthoughthereissomeevidencethattheresultsDOESMANAGEMENTMATTER?27 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
varyacrossplants,wefoundnooutliersdrivingthesecoefÞcientdifferences.22Asanothercheck,weestimatedtheresultswithtimedummiesforweeksrelativetothestartofthediagnosticphase(ratherthancalendartime)andfoundslightlylargeref-fects(seeOnlineAppendixTableOI).Finally,weplottedthemainperformancemetricsforeachwavebycalendartime,againshowinggraduallyrisingeffectsoftheinterventiononim-provingquality,reducinginventory,andraisingTFP(seeOnlineAppendixFiguresOIItoOIV).FIGUREVIIYarnInventoryfortheTreatmentandControlPlantsDisplaystheweeklyaverageyarninventoryplottedfor12treatmentplants(plussigns)andthe6controlplants(diamonds).Valuesnormalizedsobothserieshaveanaverageof100priortothestartoftheintervention.ToobtainconÞdenceintervalswebootstrappedtheÞrmswithreplacement250times.Twotreatmentplantsmaintainnoon-siteyarninventory.NotethatseasonalityduetoDiwaliandtheweddingseasonaffectsbothgroupsofplants.22.NotethattheIMprocedurealsoallowsforÞrm-levelheterogeneity.Inpar-ticular,theprocedureestimatesÞrm-by-ÞrmcoefÞcients,whichformthebasisforinference.QUARTERLYJOURNALOFECONOMICS28 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
UsingtheresultsfromTableII,weestimateatotalincreaseinproÞtsofaround$325,000perplantperyear(detailedinOnlineAppendixB).WecouldnotobtainaccountingdataontheseÞrmsÕproÞtsandlosses.PublicaccountsdataareavailableonlywithalagoftwotothreeyearsattheÞrmlevel(ratherthanplant,whichiswhatwewouldwant)andmaynotbecompletelyreliable(accordingtoourinterviews).Firmswerealsoreluctanttoprovideinternalaccounts,thoughtheyindicatedthatproÞtswereoftenintherangeof$0.5mto$2mperyear.23SoweinferthechangesinproÞtsfromtheimprovementsinquality,inventory,andefÞciency.Theestimatesaremedium-runbasedonthechangesovertheperiodoftheexperiment.Inthelongerrun,theimpactmightbegreaterifothercomplementarychangesFIGUREVIIITotalFactorProductivityfortheTreatmentandControlPlantsDisplaystheweeklyaverageTFPforthe14treatmentplants(plussigns)andthe6controlplants(diamonds).Valuesnormalizedsobothserieshaveanaverageof100priortothestartoftheintervention.ToobtainconÞdenceinter-valswebootstrappedtheÞrmswithreplacement250times.Notethatseason-alityduetoDiwaliandtheweddingseasonaffectsbothgroupsofplants.23.ItisnotevenclearifÞrmsactuallykeepcorrectrecordsoftheirproÞts,giventheriskstheyentail.Forexample,anyemployeethatdiscoveredsuchrecordscouldusethemtoblackmailtheowners.DOESMANAGEMENTMATTER?29 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
happen(likeÞrmsupgradingtheirdesignportfolio)orsmalleriftheÞrmsbackslideonthesemanagementchanges.ToestimatethenetincreaseinproÞt,wealsoneedtocalculatethedirectcostsofimplementingthesechanges(ignoringfornowanycostsofconsulting).Thesecostsweresmall,averaginglessthan$3,000perÞrm.24Sogiventhe$250,000thattheconsultancyreporteditwouldhavechargedanindividualÞrmforcomparableservicesifitpaiddirectly,thisimpliesabouta130%one-yearrateofreturn.AfurtherquestionwhichariseswhenviewingFiguresVthroughVIIIiswhetherthecontrolÞrmsbeneÞtedfromtheprac-ticestheyintroducedasaresultofthediagnosticphase,andifnot,why?WeÞrstnotethatwehavenocounterfactualforthecontrolgroup,soareunabletosaywhatwouldhavehappenediftheyhadnotimplementedthefewpracticesthattheydidimple-ment.ThefactthatweseegraphicallyatmostasmallgradualdeclineinqualitydefectratesforthecontrolgroupinFigureVImaythusrepresentanimprovementrelativetowhatwouldhavehappenedabsentthesenewmanagementpractices,orelseshowlittlereturnfromthespeciÞcpracticesthecontrolÞrmsdidim-plement.Second,AppendixTableA.IshowsthatthebiggestchangesinpracticesforthecontrolÞrmswerelargelytheresultsofourmeasurementÑrecordingqualitydefect-wise,monitoringmachinedowntimedaily,andmonitoringclosingstockweekly.Implementingthesepracticestomonitorandelectronicallyrecorddatawithoutimplementingthecomplementarypracticestouseandactonthesedataisintuitivelylikelytohavemorelimitedeffects.However,becausewedonothaveexperimentalvariationinwhichpracticeswereimplemented,wecannotinves-tigatesuchcomplementaritiesformally.V.B.Long-runEffectsoftheManagementInterventionToevaluatethelong-runeffectsofthemanagementinterven-tion,wecollecteddataontheevolutionofthenumberofplantsforeachÞrm.Thenumberofplantsisagoodlong-runperformanceindicatorasitis(1)easytomeasureandrecollectsocouldbeaccuratelycollectedovertimefortheentiresetofÞrmsinour2011survey,and(2)isnotinßuencedbychangesinloomtech-nologiesthatmadecomparingplant-leveloutputovertime24.About$35ofextralabortohelporganizethestockroomsandfactoryßoor,$200onplasticdisplayboards,$200forextrayarnracking,$1,000onrewards,and$1,000forcomputerequipment.QUARTERLYJOURNALOFECONOMICS30 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
difÞcult.Wealsocollecteddataontwoothersizemeasures:thenumberofloomsperplantandthenumberofemployeesperplant.InTableIII,weseeincolumn(1)thatinthe2011cross-sectionthenumberofplantsperÞrmishigherforbettermana-gedÞrmsandforÞrmswithmoremaleadultfamilymembers.Themanagementvariableistheshareofthe16managementpracticesmeasuredinthe2011survey,whichtheÞrmsadoptedeachyear.25ThenumberofadultmalefamilymembersistheresponsetoquestionÔÔhowmanyfamilymemberscouldcurrentlyworkasdirectorsintheÞrm?,ÕÕwhichaimstorecordthesupplyoffamilymanagersbyincludingfamilymemberscurrentlyworkingtheÞrmplusinanyotherÞrms(butexcludingthoseinfull-timeeducationorinretirement).26Intermsofmagnitudes,althoughbothindicatorsareatleastweaklysigniÞcant,wefoundthatthenumberoffamilymemberswasmuchmoreimportantinexplain-ingÞrmsizethanthemanagementscore.ThereductioninR-squaredfromdroppingthenumberoffamilymemberswas10.1%(15.9%versus5.8%),morethanthreetimesthereductionof3.3%(15.9%versus12.6%)fromdroppingthemanagementterm.Thisreßectsthefactthatthenumberofmalefamilymem-bersappearedtobethedominantfactordeterminingÞrmsize,andmanagementpracticesappearedtoplayonlyasecondaryrole.Incolumn(2)weregressthetimeseriesofÞrmsizeintermsofthenumberofplantsinour17projectÞrmsona(0/1)post-treatmentindicator,includingasetoftimeandÞrmÞxedeffects,usingyearlydatafrom2008to2011andclusteringbyÞrm.WeÞndapositivecoefÞcient,suggestingthatimprovingmanage-mentpracticesintheseÞrmshelpedthemexpand,althoughthisisonlysigniÞcantatthe10%level.TreatmentÞrmstoldusthathavingbettermanagementpracticesenabledeachfamilydirectortooverseemoreproductionactivity,astheadditionaldataallowedthemtodelegatemoredecisionstotheirplantmanagerswhilecloselymonitoringthem.Expansionoccurredthroughincreasingthenumberofplants(asopposedto25.These16practicesareasubsetofthe38practiceslistedinAppendixTableA.Iandwerechosenbecausetheycouldbemostaccuratelymeasuredduringasinglevisittoaplant.26.Werefertothisasmalefamilymembersbecauseitwasextremelyraretohavefemaledirectors.Wecameacrossonlyonefemaledirectorinthe113surveyedÞrms.ShewasonlyrunningtheÞrmbecauseofherhusbandÕsheartattack.DOESMANAGEMENTMATTER?31 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
TABLEIIILONG-RUNIMPACTOFTHEEXPERIMENTONFIRMSIZEANDDECENTRALIZATIONFirmsizeDelegationtoplantmanagement(1)(2)(3)(4)(5)(6)DependentvariableNo.ofplantsNo.ofplantsNo.ofplantsz-scorez-scorez-scoreSampleIndustryExperimentIndustryIndustryExperimentIndustryTimeperiod20112008Ð20112008Ð201120112008Ð20112008Ð2011Managementi,t1.040*0.597y(0.563)(0.370)Malefamilymembersi,t0.210***0.010(0.065)(0.042)Posttreatmenti,t0.217*0.259**0.103**0.171***(0.122)(0.110)(0.049)(0.035)Plantmanagerrelatedi0.423***(0.150)Plantmanagertenurei0.014**(0.007)SmallsamplerobustnessPermutationtests(p-value)n/a0.210.02n/a0.120.001TimeFEsn/a33n/a33Plant/FirmFEsn/a17121n/a28128Observations10768468120108499Notes.Thesizedependentvariableincolumns(1)Ð(3)isthenumberofplantsintheÞrm.Thedecentralizationdependentvariableincolumns(4)Ð(6)isthez-scoreindexofplantdecentralization,whichisthesumofthefourz-scored(normalizedtoameanof0andstandarddeviationof1)individualresponsesoverplantmanagerautonomyoverweaverhiring,juniormanagerhiring,sparepartspurchasingauthority,anddaysthedirectordoesnotvisitthefactory(seeOnlineAppendixA.Ifordetails).Columns(1)Ð(3)arerunattheÞrmlevel(becauseÞrm-sizeisaÞrm-levelvariable)andcolumns(4)Ð(6)arerunattheplantlevel(becausedecentralizationisaplant-levelvariable).Managementistheadoptionshareofthe16managementpracticesstarredinAppendixTableA.IanddiscussedinOnlineAppendixA.I,averagedacrossallplantswithinthesameÞrmincolumns(1)Ð(3).Malefamilymembersisthenumberofadultsonsandbrothersoftheintervieweddirector,whichincludesallmalefamilymemberscurrentlyworking(evenworkinginanotherÞrm)butexcludesthoseinschoolofuniversity.ThisisdesignedtomeasurethesupplyofmalefamilymembersthatcouldworkintheÞrm.Posttreatmenttakesthevalue1foratreatmentÞrm/plantaftertheimplementationphaseand0otherwise.Plantmanagerrelatedreportsiftheplantmanagerisrelatedtothedirector,includingcousins,uncles,andotherindirectfamilymembers.PlantmanagertenuremeasuresthenumberofyearstheplantmanagerhasbeenworkingattheÞrm.TimeFEsreportthenumberofcalendarweektimeÞxedeffects.Firm/PlantFEsreportsthenumberofÞrm-levelÞxedeffects(columns(1)Ð(3))orplant-levelÞxedeffects(columns(4)Ð(6)).StandarderrorsclusteredattheÞrmlevelinallcolumns.Permutationtestreportsthep-valuesfortestingthenullhypothesisthatthetreatmenthasnoeffectfortheITTparameterbyconstructingapermutationdistributionoftheITTestimateusingthe12,376possiblepermutationsoftreatmentassignment.***denotes1%,**denotes5%,*denotes10%,ydenotes15%signiÞcance.QUARTERLYJOURNALOFECONOMICS32 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
expandingcurrentplants)asthisapparentlyreducedtheriskofunionizationandregulatoryproblems.27Interestingly,wefoundnoimpactofmanagementpracticesonthenumberofemployeesperplantorthenumberofloomsperplant.28Thisisbecausealthoughoutputwasrisingateachplant,employmentwasusu-allyfalling(duetoareductionintheamountofmendinglabor)andloomnumberscouldmoveineitherdirectiondependingonwhethertheplantupgradedtoJacquardlooms.OnepossiblereasonfortheweaksigniÞcanceisthesmallsamplesize.Incolumn(3)ofTableIIIweÞndthatourtreatmentÞrmsareincreasingplantnumbersincomparisontothewholeindustryaswellasjustthecontrolÞrms.WeÞndacoefÞcientof0.259plantsonthetreatmentdummy,suggestingthatoverthe3+yearsfollowingthestartoftheinterventionourtreatmentÞrmsopened0.259moreplantsonaveragethantheindustry(whoseaveragenumberofplantopeningswas0.120).Incolumn(4)weinvestigatethereasonforthispossiblein-creaseinnumberofplantsperÞrmintermsofthedegreeofdelegationfromthedirectorstomanagersintheplants.ThisismeasuredastheprincipalfactorcomponentoffourquestionsÑtwoonthedegreeofdelegationtoplantmanagersoverhiringweaversandhiringmanagers,oneontherupeeinvestmentspendinglimitofaplantmanager,andoneonthenumberofdaysperweekthedirectorvisitedtheplant.WeÞndincolumn(4),lookingÞrstatthecross-section,thatmoredecentralizedplantstendtobebettermanagedandtohaveaplantmanagerrelatedtothedirector(i.e.,acousinoruncle)andalong-tenuredplantmanager.Incolumn(5)weregressdelegationonposttreat-mentandÞndasigniÞcantpositivecoefÞcient,suggestingbettermanagementleadstomoredelegation.Finally,incolumn(6)wecomparethechangeintreatmentÞrmstothewholeindustryandalsoÞndasigniÞcantincreaseindelegationwhencomparedtothislargersample.Thisisconsistentwiththestoryweheard27.Theownerstoldusthatlargerplantsattractmoreregulatoryandunionattention.Forexample,boththeFactoriesAct(1947)andDisputesActs(1947)regulateplantsratherthanÞrms,andunionizationdrivestendtobeplant-ratherthanÞrm-based.28.InanidenticalspeciÞcationtocolumn(2)inTableIII,exceptwithemploy-mentandloomnumberasthedependentvariable,weÞndpointestimates(stand-arderrors)ofÐ1.28(6.19)and2.38(3.24),respectively.InterestinglyBruhn,Karlan,andSchoar(2012)alsoÞndnoeffectofconsultingonemploymentdespiteÞndinglargepositiveeffectsonsalesandproÞts.DOESMANAGEMENTMATTER?33 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
fromthedirectorsandtheconsultantsthatbettermanagementpracticesenabledthedirectorstodecentralizemoredecisionmakingtotheplantmanagers,increasingÞrmsizebyrelaxingtheconstraintonmalefamilymembersÕmanagementtime.V.C.AretheImprovementsinPerformanceduetoHawthorneEffects?Hawthorneeffectsinvolvethepossibilitythatjustrunningexperimentsandcollectingdatacanimproveperformance,rais-ingconcernsthatourresultscouldbespurious.However,wethinktheseareunlikelytobeamajorfactorinourstudy,foranumberofreasons.First,ourcontrolplantsalsohadtheconsult-antsonsiteoverasimilarperiodoftimeasthetreatmentÞrms.Bothsetsofplantsgottheinitialdiagnosticperiodandthefollow-upmeasurementperiod,withtheonlydifferencebeingthetreatmentplantsalsogotanintensiveconsultingduringtheintermediateimplementationstagewhilethecontrolplantshadbrieferbutneverthelessfrequentvisitsfromtheconsultantscollectingdata.Neitherthetreatmentnorcontrolplantsweretoldthattwogroupsexisted,sotheywerenotawareofbeingÔÔtreatmentÕÕorÔÔcontrolÕÕplants.29Hence,itcannotbesimplythepresenceoftheconsultantsorthemeasurementofperform-ancethatgeneratedtherelativeimprovementinperformanceoftreatmentÞrms.Second,theimprovementsinperformancetooktimetoariseandtheyaroseinquality,inventory,andefÞciency,wherethemajorityofthemanagementchangestookplace,andinthelongerrunledtotreatmentÞrmsopeningadditionalplants.Third,theseimprovementspersistedaftertheimplementationperiod,30includingthelong-runchangeinthenumberofplants,soarenotsometemporaryphenomenaduetoincreasedattention.Finally,theÞrmsthemselvesalsobelievedtheseim-provementsarosefrombettermanagementpractices,whichwasthemotivationforthemextensivelycopyingthesepracticesovertotheirothernonexperimentalplants(seeFigureV)andopeningnewplants.29.WetoldtheÞrmstheWorldBankwantedtoinvestigatetheeffectofman-agementpracticesonperformance,sowewereprovidingmanagementadvicetotheÞrmsandcollectingperformancedata.Thus,theÞrmsbelievedtheywereinvolvedinaÔÔdifferenceÕÕratherthanaÔÔdifference-of-differencesÕÕexperiment.30.NotethatsincewedidnotinformÞrmsinAugust2010thatwewouldrevisittheminAugust2011,persistenceofmanagementpracticesduetoanticipationeffectsisunlikely.QUARTERLYJOURNALOFECONOMICS34 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
Arelatedbutharderissueiswhethertheimprovementsarethelastingimpactoftemporaryincreasesinhumancapital.Forexample,thehighlyeducatedconsultantscouldhavesimplyinspiredtheemployeestoworkharder.Ourviewisthatthisispossible,althoughitwasnotthemainchannelofeffectsfromconsulting.First,anecdotally,weheardnoevidenceforthishap-peningfromthedirectors,managers,orconsultants.Second,thefactthatalmost50%ofthemanagementchangeswerecopiedacrossplantssuggeststhatatleastpartofthiswasnotlocalizedplant-levelhumancapitaleffects.Third,thefactthatthetreat-mentledtotheopeningofnewplantssuggeststheeffectswereseenassufÞcientlypersistentbythedirectorstojustifymakinglong-runinvestmentsinnewplants,whichseemsinconsistentwithmotivationeffectswhichseemmoretemporary.V.D.ManagementSpilloversGiventheevidenceontheimpactofthesepracticesonÞrmperformance,anotherkeyquestionishowmuchthesepracticesspilledovertootherÞrms.Toaddressthisinourground-basedsurvey(runfromNovember2011toJanuary2012)weaskedeveryoneofthe96nonprojectÞrmsiftheyhadheardoftheStanfordÐWorldBankproject,andifso,whattheyknew(seeOnlineAppendixAIIfordetails).Wefoundtworesults.First,thelevelofspilloverswasextremelylow,withonly16%ofÞrmshavingheardoftheprojectandonly2%havingheardanydetailsofactualpractices.Thereasonsforthelimitedspilloverwerepartlythatownerswerereluctanttodiscussthedetailsoftheirbusinesswithoutsiders,andpartlybecausetheydidnotwanttogivetheircompetitorsinformationorriskthemtryingtohireawaytheirplantmanagers.Second,totheextentthatanyspilloversoccurred,thesewereentirelylocalÑoutoftheÞvemaintextiletownsaroundMumbai,spilloversonlyaroseinTarapurandUmbergaon,whereourexperimentalÞrmswerelocated.Asaresult,althoughspill-oversacrossplantswithinÞrmsoccurredrapidlyasshowninFigureV,spilloversbetweenÞrmswereverylimited.VI.WHYDOBADLYMANAGEDFIRMSEXIST?GiventheevidenceinthepriorsectionofthelargeimpactofmodernmanagementpracticesonproductivityandproÞtability,DOESMANAGEMENTMATTER?35 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
theobviousquestioniswhythesemanagementchangeswerenotintroducedbefore.VI.A.WhyAreFirmsBadlyManaged?Ourexperimentdoesnotdirectlyanswerthisquestion,butwecanuseinformationgeneratedbytheexperimentandadd-itionalinformationgatheredintheÞeldtodrawsomeprelimin-aryconclusions.Inparticular,weaskedtheconsultantstodocument(everyothermonth)thereasonforthenonadoptionofanyofthe38practicesineachplant.Todothisconsistently,wedevelopedaßowchart(OnlineAppendixExhibitOV),whichrunsthroughaseriesofquestionstounderstandtherootcauseforthenonadoptionofeachpractice.Theconsultantscollectedthisdatafromdiscussionswithowners,managers,andworkers,plustheirownobservations.Asanexampleofhowthisßowchartworks,imagineaplantthatdoesnotrecordqualitydefects.TheconsultantwouldÞrstaskiftherewassomeexternalconstraint,likelaborregulations,preventingthis,whichwefoundnevertobethecase.31Theywouldthenaskiftheplantwasawareofthispractice,whichintheexampleofrecordingqualitytypicallywasthecase.Thecon-sultantswouldthencheckiftheplantcouldadoptthepracticewiththecurrentstaffandequipment,whichagainforqualityrecordingsystemswasalwaystrue.Then,theywouldaskiftheownerbelieveditwouldbeproÞtabletorecordqualitydefects,whichwasoftentheconstraintonadoptingthispractice.Theownerfrequentlyarguedthatqualitywassogoodtheydidnotneedtorecordqualitydefects.Thisviewwasmistaken,however,because,althoughtheseplantsÕqualitywasoftenrelativelygoodcomparedtootherlow-qualityIndiantextileplants,itwaspoorbyinternationalstandards.So,inthiscase,thereasonfornon-adoptionwouldbeÔÔincorrectinformationÕÕbecausetheownerap-pearedtohaveincorrectinformationonthecost-beneÞtcalculation.Theoverallresultsfornonadoptionofmanagementpracticesaretabulatedattwo-monthintervalsstartingthemonthbeforetheinterventioninTableIV.Therowsreportthedifferentrea-sonsfornonadoptionasapercentageofallpractices.Theseare31.ThisdoesnotmeanlaborregulationsdonotmatterforsomepracticesÑforexample,ÞringunderperformingemployeesÑbuttheydidnotdirectlyimpingeontheimmediateadoptionofthe38practices.QUARTERLYJOURNALOFECONOMICS36 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
splitintononadoptionreasonsforcommonpractices(thosethat50%ormoreoftheplantswereusingbeforetheexperiment,likequalityandinventoryrecordingorworkerbonuses)anduncom-monpractices(thosethatlessthan5%oftheplantswereusinginadvance,likequalityandinventoryreviewmeetingsormanagerbonuses).FromTableIV,severalresultsareapparent.First,forthecommonpractices,themajorinitialbarriertoadoptionwasthatÞrmshadheardofthepracticesbutthoughttheywouldnotbeproÞtabletoadopt.Forexample,manyoftheÞrmswereawareofpreventivemaintenance,butfewofthemthoughtitwasworthdoing.Theypreferredtokeeptheirma-chinesinoperationuntiltheybrokedown,andthenrepairthem.Thisaccountedforslightlyover45%oftheinitialnonadop-tionofpractices.Second,fortheuncommonpractices,themajorinitialbarriertotheadoptionwasalackofinformationabouttheirexistence.Firmsweresimplynotawareofthesepractices.Thesepracticesincludeddailyquality,efÞciencyandinventoryreviewmeetings,postingstandardoperatingprocedures,andhavingvisualaidsaroundthefactory.ManyofthesearederivedfromtheJapanese-inspiredleanmanufacturingrevolutionandarenowstandardacrossNorthAmerica,Japan,andnorthernEuropebutnotindevelopingcountries.32Third,astheinterventionprogressed,thelackofinformationconstraintontheuncommonpracticeswasrapidlyovercomeinbothtreatmentandcontrolÞrms.Itwaseasytoexplaintheex-istenceoftheuncommonmanagementpractices,sothenonadop-tionratesofthesepracticesfellrelativelyrapidly:from98.5%inthetreatmentgroupsonemonthbeforetheexperimentto63.2%atninemonths(adropof35.3percentagepoints).Fourth,theincorrectinformationconstraintswerehardertoaddressbecausetheownersoftenhadstrongpriorbeliefsabouttheefÞcacyofapractice,andittooktimetochangethese.Thiswasoftendoneusingpilotchangesonafewmachinesintheplantorwithevidencefromotherplantsintheexperiment.Forex-ample,theconsultantstypicallystartedbypersuadingtheman-agerstoundertakepreventivemaintenanceonasetoftrialmachines,andonceitwasprovensuccessful,itwasrolledout32.Thisignoranceofbestpracticesseemstobecommoninmanydevelopingcountrycontexts,forexample,inpineapplefarminginGhana(ConleyandUdry2010).DOESMANAGEMENTMATTER?37 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
TABLEIVREASONSFORTHENONADOPTIONOFTHE38MANAGEMENTPRACTICES(%OFALLPRACTICES),BEFOREANDAFTERTREATMENTNonadoptionreasonGroupManagementpracticetypeTimingrelativetotreatment1monthbefore1monthafter3monthsafter5monthsafter7monthsafter9monthsafterLackofinformation(plantsneverheardofthepracticebefore)TreatmentCommon3.33.20.5000TreatmentUncommon64.019.12.91.500ControlCommon1.900000ControlUncommon67.823.722.022.022.022.0Incorrectinformation(heardofthepracticebeforebutthinkitisnotworthdoing)TreatmentCommon3022.415.415.214.414.4TreatmentUncommon30.950.750.749.349.347.1ControlCommon18.518.518.518.518.518.5ControlUncommon27.152.550.950.949.249.2Ownertime,ability,orpro-crastination(theowneristhereasonfornonadoption)TreatmentCommon1.10.80.50.81.60.8TreatmentUncommon3.713.213.213.213.214.0ControlCommon3.73.73.73.73.73.7ControlUncommon3.420.318.618.618.618.6Other(varietyofotherreasons)TreatmentCommon000000TreatmentUncommon2.11.51.52.22.22.2ControlCommon000000ControlUncommon000000TotalnonadoptionTreatmentCommon34.626.416.316.016.015.2TreatmentUncommon98.584.678.266.265.163.2ControlCommon25.122.222.222.222.222.2ControlUncommon98.396.691.591.589.889.8Notes.Percentages(%)ofpracticesnotadoptedbyreason.Commonpracticesaretheeightpracticeswithmorethan50%initialadoption,mainlyqualityanddowntimerecording,andworkerbonuses(seeAppendixTableA.Ifordetails).Uncommonpracticesarethe10practiceswithlessthan5%initialadoption,mainlystandardoperatingprocedures,inventorynorms,reviewmeetings,andmanagerincentiveschemes.Timingisrelativetothestartofdiagnosticphase.Covers532practicesinthetreatmentplants(38practicesin14plants),and228practicesinthecontrolplants(38practicesin6plants).NonadoptionwasmonitoredeveryothermonthusingthetoolshowninOnlineAppendixExhibitOV,basedonregulardiscussionswiththeÞrmsÕdirectors,managers,andworkers.QUARTERLYJOURNALOFECONOMICS38 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
totherestofthefactory.Astheconsultantsdemonstratedthepositiveeffectoftheseinitialpracticechanges,theownersin-creasinglytrustedthemandadoptedmoreoftherecommenda-tions,likeperformanceincentivesformanagers.Thus,thecommonpracticenonadoptionratesstartedatamuchlowerlevelbutwereslowertofall:droppingfrom34.6%onemonthbeforetheexperimentforthetreatmentplantsto15.2%afterninemonths(adropof19.4percentagepoints).Fifth,oncetheinformationalconstraintswereaddressed,otherconstraintsarose.Forexample,eveniftheownersbecameconvincedoftheneedtoadoptapractice,theywouldoftentakeseveralmonthstodoso.Amajorreasonisthattheownerswereseverelytimeconstrained,workinganaverageof68hoursperweekalready.33AlthoughinitiallyownersÕtimeac-countedforonly3.7%ofnonadoptionintreatmentplants,byninemonthsitaccountedfor14.0%asabacklogofmanagementchangesbuiltupthattheownersstruggledtoimplement.Finally,wedidnotÞndevidenceforthedirecteffectofcapitalconstraints,whichareasigniÞcantobstacletotheexpansionofmicro-enterprises(e.g.,DeMel,McKenzie,andWoodruff2008).OurevidencesuggestedthattheselargeÞrmswerenotcash-constrained,atleastfortangibleinvestments.Wecollecteddataonalltheinvestmentsmadebyour17projectÞrmsduringthetwoyearsofdatacollection.Themean(median)investmentwas$880,000($140,000).Soinvestmentsonthescaleof$3,000(theÞrst-yearcostsofthesemanagementchangesexcludingtheconsultantsÕfees)areunlikelytobedirectlyimpededbyÞnancialconstraints.Ofcourse,Þnancialconstraintscouldimpedehiringinternationalconsultants.Theestimatedmarketcostofourfreeconsultingwouldbe$250,000,and,asanintangibleinvestment,itwouldbedifÞculttocollateralize.Hence,althoughÞnancialconstraintsdonotappeartodirectlyblocktheimplementationofbettermanagementpractices,theymayhinderÞrmsÕabilitytoimprovetheirmanagementusingexternalconsultants.Nevertheless,inconversationswithfactorymanagers,inabilitytoborrowtoÞnanceconsultingnevercameupasareasonfornotusingthem,suggestingthisisnotthe(main)bindingconstraint.33.Therewasalsoevidencesuggestiveofprocrastinationinthatsomeownerswoulddefertakingquickdecisionsfornoapparentreason.Thismatchesupwiththeevidenceonprocrastinationinothercontexts,forexampleKenyanfarmersinvestinginfertilizer(Dußo,Kremer,andRobinson,2011).DOESMANAGEMENTMATTER?39 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
VI.B.HowDoBadlyManagedFirmsSurvive?Wehaveshownthatmanagementmatters,withimprove-mentsinmanagementpracticesimprovingplant-leveloutcomes.Oneresponsefromeconomistsmightbetoarguethatpoorman-agementcanatmostbeashort-runproblem,becauseinthelongrunbetter-managedÞrmsshouldtakeoverthemarket.YetmostofourÞrmshavebeeninbusinessformorethan20years.OnereasonbetterrunÞrmsdonotdominatethemarketap-pearstobeconstraintsongrowthderivedfromlimitedmanager-ialspanofcontrol.IneveryÞrminoursample,beforethetreatment,onlymembersoftheowningfamilyhadpositionswithanyrealdecision-makingpoweroverÞnance,purchasing,operations,oremployment.Non-familymembersweregivenonlylower-levelmanagerialpositionswithauthorityonlyoverbasicday-to-dayactivities.Theprincipalreasonseemstobethatfamilymembersdidnottrustnon-familymembers.Forex-ample,theywereconcernediftheylettheirplantmanagerspro-cureyarntheymaydosoatinßatedratesfromfriendsandreceivekickbacks.34AkeyreasonforthisinabilitytodecentralizeappearstobetheweakruleoflawinIndia.Evenifdirectorsfoundmanagersstealing,theirabilitytosuccessfullyprosecutethemandrecovertheassetsislikelyminimalbecauseoftheinefÞciencyofIndiancourts.AcompoundingreasonfortheinabilitytodecentralizeinIndianÞrmsseemstobetheprevalenceofbadmanagementprac-tices,asthismeanttheownerscouldnotkeepgoodtrackofma-terialsandÞnance,andsomightnotevenbeabletoidentifymismanagementortheftwithintheirÞrms.35ThisisconsistentwiththegeneralÞndingofBloom,Sadun,andvanReenen(2012)acrossarangeofcountriesthatÞrmsheadquarteredinhigh-trustregionsaremorelikelytodecentralize,asareÞrmsincountrieswithbetterruleoflaw.34.Thisalsolinkstowhyplantmanagers(versusdirector-owners)didnotdir-ectlyadoptthese38practicesthemselves.Theyhadbothlimitedcontroloverfac-torymanagementandalsolimitedincentivestoimproveperformancebecausepromotionisnotpossible(onlyfamilymemberscanbecomedirectors)andtherewerenobonussystems(theÞrmsdidnotcollectenoughperformancedata).35.AcompoundingfactorisnoneoftheseÞrmshadaformalizeddevelopmentortrainingplanfortheirmanagers,sotheylackedcareermotivation.Incontrast,IndiansoftwareandÞnanceÞrmsthathavegrownmanagementbeyondthefound-ingfamiliesplaceahugeemphasisondevelopmentandtraining(seealsoBanerjeeandDußo2005).QUARTERLYJOURNALOFECONOMICS40 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
Asaresultofthisinabilitytodelegate,Þrmscouldlikelyexpandbeyondthesizethatcouldbemanagedbyasingledirectoronlyifotherfamilymemberswereavailabletoshareexecutiveresponsibilities.Thus,aswesawinTableIII,animportantpre-dictorofÞrmsizewasthenumberofmalefamilymembersoftheowners.ThismatchestheideasoftheLucas(1978)spanofcontrolmodel,thattherearediminishingreturnstohowmuchadditionalproductivitybettermanagementtechnologycangeneratefromasinglemanager.IntheLucasmodel,thelimitstoÞrmgrowthrestricttheabilityofhighlyproductiveÞrmstodrivelower-productivityonesfromthemarket.InourIndianÞrms,thisspanofcontrolrestrictionseemstobebinding,sounproduct-iveÞrmsarelikelyabletosurvivebecausemoreproductiveÞrmscannotexpand.OurÞndingthatimprovedmanagementpracticewasconnectedwithincreaseddelegationtoplantmanagersandinvestmentinnewplantssupportsthisview.36EntryofnewÞrmsintotheindustryalsoappearslimitedbythedifÞcultyofseparatingownershipfromcontrol.ThesupplyofnewÞrmsisconstrainedbythenumberoffamilieswithÞnanceandmalefamilymembersavailabletobuildandruntextileplants.SinceotherindustriesinIndiaÑsoftware,construction,andrealestateÑaregrowingrapidly,theattractivenessofnewinvestmentintextilemanufacturingisrelativelylimited.Finally,a35%tariffoncottonfabricimportsinsulatesIndiantextileÞrmsagainstforeigncompetition.Hence,theequilibriumappearstobethatwithIndianwageratesbeingextremelylow,Þrmscansurvivewithpoormanage-mentpractices.Becausespansofcontrolareconstrained,pro-ductiveÞrmsarelimitedfromexpanding,soreallocationdoesnotdriveoutbadlyrunÞrms.Becauseentryislimited,newÞrmsdonotenterrapidly.ThesituationapproximatesaMelitz(2003)ÐstylemodelwithÞrmsexperiencinghighdecreasingre-turnstoscaleduetoLucas(1978)spanofcontrolconstraints,highentrycosts,andlowinitialproductivitydraws(becausegoodmanagementpracticesarenotwidespread).Theresultant36.Powell(2012)buildstheintuitionthattheinabilityofownersindevelopingcountriestotrustnonfamilymanagerslimitsÞrmsizeintoacontractingmodelandshowsthisexplainsawidenumberofÞrmandmacrofacts.DOESMANAGEMENTMATTER?41 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
equilibriumhaslowaverageproductivity,lowwages,lowaverageÞrmsize,andalargedispersionofproductivity.37TheideathatÞrmsmightberuninahighlyinefÞcientmannertracesbackatleasttoLeibenstein(1966).Themanage-mentliteratureonthereasonsforsuchinefÞciencyhastendedtofocusonthreemaindrivers(e.g.,Rivkin2000;GibbonsandHenderson2011).TheÞrstisthemotivationproblem(peopleknowwhatwouldimproveperformancebutlacktheincentives)andisrelatedtoourlimitedcompetitionandlackofdelegationstories.Thesecondandthirdmatchourinformationalstory:theinspirationproblem(thedecisionmakersknowtheyarenotefÞ-cientbutdonotknowhowtoÞxthis)andtheperceptionproblem(theÞrmÕsdecisionmakersdonotrealizetheyareinefÞcient).Ofcourse,onequestionaroundinformationalstoriesiswhyÞrmsdonotlearn,inparticularfromrandomvariationsintheiroperatingconditions.Forexample,ifthefactoryßoorwaslessclutteredatsometimesthanothers,thiswouldgeneratedatathatwouldsuggestthevalueofneatness.RecentworkbySch-wartzstein(2012)andHanna,Mullainathan,andSchwartzstein(2012)highlights,however,thatcognitivelimitationscouldpre-ventdirectorsandmanagersfromnotingtheimpactofsuchvari-ablesthattheyhadnotthoughtimportant,sotheinformationwouldbeignored.Certainly,ourÞrmsweresufÞcientlylargeandcomplexthatthedirectorshadtofocusoncertainareas,andweseethischoiceoverwhattofocusonasanimportantareaforfuturemanagementresearch.VI.C.WhyDoFirmsNotUseMoreManagementConsulting?Finally,whydotheseÞrmsnothireconsultantsthemselves,giventhelargegainsfrombettermanagement?AprimaryreasonisthattheseÞrmsarenotawaretheyarebadlymanaged,asillustratedinTableIV.Ofcourse,consultingÞrmscouldap-proachÞrmsforbusiness,pointingoutthattheirpracticeswerebadandoffertoÞxthem.ButIndianÞrmsarebombardedwithsolicitationsfrombusinessesofferingtosavethemmoneyoneverythingfromtelephonebillstoyarnsupplies,andthusareunlikelytobereceptive.MaybeconsultingÞrmscouldgofurtherandoffertoprovidefreeadviceinreturnforanexpost37.CaselliandGennaioli(2011)calibrateaneconomywithfamilyÞrmsthatareunabletogrowduetodelegationconstraintsandÞndareductioninTFPof35%,suggestingthesekindsofdistortionscanbequantitativelyimportant.QUARTERLYJOURNALOFECONOMICS42 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
proÞt-sharingdeal.Butmonitoringthiswouldbeextremelyhard,giventheÞrmsÕreluctancetorevealproÞts.Moreover,theclientÞrminsuchanarrangementmightworrythattheconsultancywouldtwistitseffortstoincreaseshort-termproÞtsinwhichitwouldshareattheexpenseoflong-termproÞts.VII.CONCLUSIONSWeimplementedarandomizedexperimentthatprovidedmanagerialconsultingservicestotextileplantsinIndia.Thisexperimentledtoimprovementsinbasicmanagementpractices,withplantsadoptingleanmanufacturingtechniquesthathavebeenstandardfordecadesinthedevelopedworld.Theseimprove-mentsinmanagementpracticesledtoimprovementsinproduct-ivityof17%withintheÞrstyearfromimprovedqualityandefÞciencyandreducedinventoryandappeartohavebeenfol-lowedbyalonger-runincreaseinÞrmsize.Itappearsthatcompetitiondidnotdrivethesebadlymana-gedÞrmsoutofthemarketbecausetheinabilitytodelegatede-cisionsawayfromtheownersoftheÞrmimpededthegrowthofmoreefÞcientÞrmsand,thereby,interÞrmreallocation.Firmshadnotadoptedthesemanagementpracticesbeforebecauseofinformationalconstraints.Inparticular,formanyofthemorewidespreadpractices,althoughtheyhadheardofthesebefore,theywereskepticaloftheirimpact.Forlesscommonmanage-mentpractices,theysimplyhadnotheardofthem.Intermsoffutureresearch,wewouldliketoinvestigatetheextensionoftheseresultstootherindustries,countries,andÞrmcharacteristics.Inparticular,theÞrmsinourexperimentarelarge,multiplantÞrmsoperating24hoursadayacrossmultiplelocations,soarecomplextomanage.Othersimilarlysized(orlarger)ÞrmswouldpresumablyalsobeneÞtfromadoptingformalizedmanagementpracticesthatcontinuouslymonitortheproductionprocess.ButmuchsmallerÞrmsÑsuchasthetypicallysingle-personÞrmsstudiedinDeMel,McKenzie,andWoodruff(2008)Ñmaybesimpleenoughthattheownercandirectlyobservethefullproductionprocessandhencedoesnotneedformalmonitoringsystems.Otherinterestingexten-sionsinvolveexamininginmoredetailthespilloverofbettermanagementpracticesacrossÞrmswithinthesameindustryorDOESMANAGEMENTMATTER?43 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
regionandthecomplementarityofdifferentbundlesofmanage-mentpractices.Finally,whataretheimplicationsofthisforpublicpolicy?Certainly,wedonotwanttoadvocatefreeconsulting,givenitsextremelyhighcost.However,ourresearchdoessupportsomeofthecommonrecommendationstoimproveproductivity,likeincreasingcompetition(bothfromdomesticÞrmsandmultina-tionals)andimprovingtheruleoflaw.OurresultsalsosuggestthatÞrmswerenotimplementingbestpracticesontheirownbe-causeoflackofinformationandknowledge.Thissuggeststhattrainingprogramsforbasicoperationsmanagement,likeinven-toryandqualitycontrol,couldbehelpful,aswoulddemonstrationprojects,somethingwealsohopetoexploreinsubsequentwork.SupplementaryMaterialAnOnlineAppendixforthisarticlecanbefoundatQJEonline(qje.oxfordjournals.org).APPENDIX:SMALLSAMPLETESTSSUMMARY1.PermutationTestsPermutationproceduresusethefactthatorderstatisticsaresufÞcientandcompletetoproposeandderivecriticalvaluesforteststatistics.WeÞrstimplementedthisforthenullhypothesisofnotreatmenteffectfortheITTparameter.ThiscalculatestheITTcoefÞcientforeverypossiblecombinationof11treatmentÞrmsoutofour17projectÞrms(werunthisattheÞrmleveltoallowforÞrm-levelcorrelationsinerrors).Oncethisiscalculatedforthe12,376possibletreatmentassignments(17choose11),the2.5%and97.5%conÞdenceintervalsarecalculatedasthe2.5thand97.5thpercentilesofthetreatmentimpact.AtreatmenteffectoutsidetheseboundscanbesaidtobesigniÞcantatthe5%level.PermutationtestsfortheIVestimatoraremorecomplex,involvingimplementingaprocedurebasedonGreevyetal.(2004)andAndrewsandMarmer(2008)(seeOnlineAppendixB).2.T-AsymptoticClusteredStandardErrorsAnalternativeapproachistouseasymptoticestimatorsthatexploitthelargetimedimensionforeachÞrm.Todothis,weuseQUARTERLYJOURNALOFECONOMICS44 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
APPENDIXTABLEA.ITHETEXTILEMANAGEMENTPRACTICESADOPTIONRATESAreaSpeciÞcpracticePreinterventionlevelPostinterventionchangeTreatmentControlTreatmentControlFactoryoperationsPreventivemaintenanceiscarriedoutforthemachines*0.4290.6670.2860PreventivemaintenanceiscarriedoutpermanufacturerÕsrecommendations*0.07100.0710.167Theshopßoorismarkedclearlyforwhereeachmachineshouldbe0.0710.3330.2140.167Theshopßoorisclearofwasteandobstacles00.1670.2140.167Machinedowntimeisrecorded*0.5710.6670.3570Machinedowntimereasonsaremonitoreddaily*0.4290.1670.50.5Machinedowntimeanalyzedatleastfortnightly&actionplansim-plementedtotrytoreducethis*00.1670.7140DailymeetingstakeplacethatdiscussefÞciencywiththeproductionteam*00.1670.7860.5Writtenproceduresforwarping,drawing,weaving,&beamgaitingaredisplayed0.0710.1670.50VisualaidsdisplaydailyefÞciencyloomwiseandweaverwise0.2140.1670.6430.167Thesevisualaidsareupdatedonadailybasis0.14300.6430.167Sparesstoredinasystematicbasis(labelinganddemarkedlocations)0.14300.1430.167Sparespurchasesandconsumptionarerecordedandmonitored0.5710.6670.0710.167ScientiÞcmethodsareusedtodeÞneinventorynormsforspares000.0710DOESMANAGEMENTMATTER?45 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
APPENDIXTABLEA.I(CONTINUED)AreaSpeciÞcpracticePreinterventionlevelPostinterventionchangeTreatmentControlTreatmentControlQualitycontrolQualitydefectsarerecorded*0.92910.0710Qualitydefectsarerecordeddefectwise0.2860.1670.6430.833Qualitydefectsaremonitoredonadailybasis*0.2860.1670.7140.333Thereisananalysisandactionplanbasedondefectsdata*000.7140.167Thereisafabricgradationsystem0.5710.6670.3570ThegradationsystemiswelldeÞned0.5000.50.4290Dailymeetingstakeplacethatdiscussdefectsandgradation*0.0710.1670.7860.167Standardoperatingproceduresaredisplayedforqualitysupervisors&checkers000.7140InventorycontrolYarntransactions(receipt,issues,returns)arerecordeddaily*0.92910.0710Theclosingstockismonitoredatleastweekly*0.2140.1670.5710.5ScientiÞcmethodsareusedtodeÞneinventorynormsforyarn000.0830Thereisaprocessformonitoringtheagingofyarnstock0.23100.5380Thereisasystemforusinganddisposingofoldstock*000.6150.6Thereislocationwiseentrymaintainedforyarnstorage*0.35700.3570LoomplanningAdvanceloomplanningisundertaken0.4290.8330.2140Thereisaregularmeetingbetweensalesandoperationalmanagement0.4290.5000.1430QUARTERLYJOURNALOFECONOMICS46 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
APPENDIXTABLEA.I(CONTINUED)AreaSpeciÞcpracticePreinterventionlevelPostinterventionchangeTreatmentControlTreatmentControlHumanresourcesThereisarewardsystemfornon-managerialstaffbasedonperformance*0.5710.6670.0710Thereisarewardsystemformanagerialstaffbasedonperformance*0.2140.1670.2860Thereisarewardsystemfornonmanagerialstaffbasedonattendance0.2140.3330.3570TopperformersamongfactorystaffarepubliclyidentiÞedeachmonth0.07100.3570Roles&responsibilitiesaredisplayedformanagersandsupervisors000.6430SalesandordersCustomersaresegmentedfororderprioritization0000.167Orderwiseproductionplanningisundertaken0.69210.2310HistoricalefÞciencydataisanalyzedforbusinessdecisionsregardingdesigns000.0710AllAverageofallpractices0.2560.2880.3780.120p-valueforthedifferencebetweentheaverageofallpractices0.5100.000Notes.Reportsthe38individualmanagementpracticesmeasuredbefore,duringandafter(August2010)themanagementintervention.Thepracticeswithanasteriskwerethosealsocollectedinthe2011survey(seeOnlineAppendixAI).ThecolumnsunderPreinterventionlevelofadoptionreportthepreinterventionshareofplantsadoptingthispracticeforthe14treatmentand6controlplants.ThecolumnsunderPostinterventionchangeinadoptionreportthechangesinadoptionratesbetweenthepreinterventionperiodandfourmonthsaftertheendofthediagnosticphase(rightaftertheendoftheimplementationphaseforthetreatmentplants)forthetreatmentandcontrolplants.Thep-valueforthedifferencebetweentheaverageofallpracticesreportsthesigniÞcanceofthedifferenceintheaveragelevelofadoptionandtheincreaseinadoptionbetweenthetreatmentandcontrolgroups.DOESMANAGEMENTMATTER?47 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
APPENDIXTABLEA.IIOLSANDIVESTIMATIONSOFTHEEFFECTOFMANAGEMENTPRACTICESONPLANTPERFORMANCE(1)(2)(3)(4)(5)(6)(7)(8)SpeciÞcationOLSIV2ndstageOLSIV2ndstageOLSIV2ndstageOLSIV2ndstageDependentvariableQualitydefectsQualitydefectsInventoryInventoryOutputOutputTFPTFPManagementi,t-0.558-1.694**-0.404-0.833**0.1190.310**0.1670.477**(0.440)(0.850)(0.259)(0.347)(0.099)(0.128)(0.176)(0.242)SpeciÞcationIV1ststageIV1ststageIV1ststageIV1ststageDependentvariableManagementManagementManagementManagementCumulativetreatmenti,t0.018***0.017***0.019***0.019***(0.002)(0.002)(0.002)(0.002)SmallsamplerobustnessIbragimov-Mueller(95%CI)[19.00,5.22][5.28,-1.18][4.77,0.11][1.50,-0.57][0.27,1.38][0.15,1.21][3.59,11.97][0.27,1.90]IVPermutationTests(95%CI)[28.05,-0.18][6.85,0.37][0.09,0.78][0.44,3.91]FirststageF-test73.4172.88107.5586.00TimeFEs127127127127127127127127PlantFEs2020181820202020Observations1,8071,8072,0522,0522,3932,3931,8311,831Notes.Allregressionsuseafullsetofplantandcalendarweekdummies.StandarderrorsbootstrapclusteredattheÞrmlevel.Qualitydefectsisalogofthequalitydefectsindex(QDI).Inventoryisthelogofthetonsofyarninventoryintheplant.Outputisthelogoftheweavingproductionpicks.Managementistheadoptionshareofthe38managementpracticeslistedinOnlineAppendixTableAI.Cumulativetreatmentisthecumulativeweeksofsincebeginningtheimplementationphaseineachplant(0inthecontrolgroupsandpriortotheimplementationphase).OLSreportsresultswithplantestimations.IVreportstheresultswherethemanagementvariablehasbeeninstrumentedwithweeksofcumulativetreatment.TimeFEsreportthenumberofcalendarweektimeÞxedeffects.PlantFEsreportsthenumberofplant-levelÞxedeffects.Twoplantsdonothaveanyinventoryonsite,sonoinventorydataareavailable.Smallsamplerobustnessimplementsthreedifferentprocedures(describedingreaterdetailinOnlineAppendixB)toaddressissuesofplantheterogeneity,withinplant(andÞrm)correlation,andsmallsampleconcerns,where95%CIreports95%conÞdenceintervals.Ibragimov-MuellerestimatesparametersÞrmbyÞrmandthentreatstheestimatesasadrawfromindependent(butnotidenticallydistributed)normaldistributions.PermutationTestIreportsthep-valuesfortestingthenullhypothesisthatthetreatmenthasnoeffectfortheITTparameterbyconstructingapermutationdistributionoftheITTestimateusing1,000possiblepermutations(outof12,376)oftreatmentassignment.IV-PermutationtestsimplementsapermutationtestfortheIVparameterusing1,000possiblepermutations(outof12,376)oftreatmentassignment.ThesetestshaveexactÞnitesamplesize.***denotes1%,**denotes5%,*denotes10%signiÞcance.QUARTERLYJOURNALOFECONOMICS48 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
IbramigovandMueller(2010)toimplementat-statisticbasedestimatorthatisrobusttosubstantialheterogeneityacrossÞrmsaswellastoautocorrelationacrossobservationswithinaÞrm.ThisapproachrequiresestimatingtheparameterofinterestseparatelyforeachtreatmentandcontrolÞrmandthencomparingtheaverageofthe11treatmentÞrmestimatestothoseofthe6controlÞrmsusingastandardt-testforgroupedmeans(allowingforunequalvariances)with5degreesoffreedom(seeOnlineAppendixB).Suchaprocedureisvalidinthesenseofhavingthecorrectsize(foraÞxedsmallnumberofÞrms)aslongasthetimedimensionislargeenoughthattheestimateforeachÞrmcanbetreatedasadrawfromanormaldistribution.StanfordUniversity,CEP,CEPR,NBER,andSIEPROverlandAdvisorsStanfordUniversityandSCIDWorldBank,BREAD,CEPR,andIZAStanfordUniversityandSIEPRReferencesAndrews,DonaldW.K.,andVadimMarmer,ÔÔExactlyDistribution-freeInferenceinInstrumentalVariablesRegressionwithPossiblyWeakInstruments,ÕÕJournalofEconometrics,142,no.1(2008),183Ð200.Banerjee,Abhijit,andEstherDußo,ÔÔGrowththroughtheLensofDevelopmentEconomics,ÕÕInHandbookofEconomicGrowth,Volume1,ed.PhilippeAghionandStephenDurlauf(Amsterdam:ElsevierNorth-Holland,2005).Bertrand,Marianne,andAntoinetteSchoar,ÔÔManagingwithStyle:TheEffectsofManagersonCorporatePolicy,ÕÕQuarterlyJournalofEconomics,118,no.4(2003),1169Ð1208.Black,Sandra,andLisaLynch,ÔÔWhatÕsDrivingtheNewEconomy?TheBeneÞtsofWorkplaceInnovation,ÕÕEconomicJournal,114,no.493(2004),97Ð116.Bloom,N.,G.Genakos,R.Martin,andR.Sadun,ÔÔModernManagement:GoodfortheEnvironmentorJustHotAir?ÕÕEconomicJournal,120,no.544(2010),551Ð572.Bloom,Nicholas,RaffaellaSadun,andJohnVanReenen,ÔÔTheOrganizationofFirmsacrossCountries,ÕÕQuarterlyJournalofEconomics,127,no.4(2012),1663Ð1705.Bloom,Nicholas,andJohnVanReenen,ÔÔMeasuringandExplainingManagementPracticesacrossFirmsandCountries,ÕÕQuarterlyJournalofEconomics,122,no.4(2007),1341Ð1408.ÑÑÑ,ÔÔWhyDoManagementPracticesDifferAcrossFirmsandCountries?ÕÕJournalofEconomicPerspectives(2010),203Ð224.ÑÑÑÔÔHumanResourceManagementandProductivity,ÕÕInHandbookofLaborEconomics,ed.OrleyAshenfelterandDavidCard(Amsterdam:ElsevierNorth-Holland,2011).Bruhn,Miriam,DeanKarlan,andAntoinetteSchoar,ÔÔTheImpactofOfferingConsultingServicestoSmallandMediumEnterprisesÕÕ(Mimeo,WorldBank,2012).DOESMANAGEMENTMATTER?49 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
Bruhn,Miriam,andDavidMcKenzie,ÔÔInPursuitofBalance:RandomizationinPracticeinDevelopmentFieldExperiments,ÕÕAmericanEconomicJournal:AppliedEconomics,1,no.4(2009),200Ð232.Bruhn,Miriam,andBilalZia,ÔÔStimulatingManagerialCapitalinEmergingMarkets:TheImpactofBusinessandFinancialLiteracyforYoungEntre-preneurs,ÕÕWorldBankWorkingPaper5642,2011.Cameron,A.Colin,JonahGelbach,andDouglasMiller,ÔÔBootstrap-BasedImprovementsforInferencewithClusteredErrors,ÕÕReviewofEconomicsandStatistics,90,no.3(2008),414Ð427.Cappelli,Peter,andDavidNeumark,ÔÔDoÔHighPerformanceÕWorkPracticesImproveEstablishment-LevelOutcomes?ÕÕIndustrialandLaborRelationsReview,54,no.4(2001),737Ð775.Caselli,Francesco,ÔÔCREILecturesinGrowthÕÕ(Mimeo,LSE,2011).Caselli,Francesco,andNicolasGennaioli,ÔÔDynasticManagement,ÕÕ(Mimeo,LSE,2011).Clark,Greg,ÔÔWhyIsnÕttheWholeWorldDeveloped?LessonsfromtheCottonMills,ÕÕJournalofEconomicHistory,47(1987),141Ð173.Conley,Tim,andChristopherUdry,ÔÔLearningaboutaNewTechnology:PineappleinGhana,ÕÕAmericanEconomicReview,100,no.1(2010),35Ð69.DeMel,Suresh,DavidMcKenzie,andChristopherWoodruff,ÔÔReturnstoCapitalinMicroenterprises:EvidencefromaFieldExperiment,ÕÕQuarterlyJournalofEconomics,113,no.4(2008),1329Ð1372.Drexler,Alejandro,GregFischer,andAntoinetteSchoar,ÔÔFinancialLiteracyTrainingandRuleofThumbs:EvidencefromaFieldExperiment,ÕÕ(Mimeo,MIT,2010).Dußo,Esther,MichaelKremer,andJonathanRobinson,ÔÔNudgingFarmerstoUseFertilizer:TheoryandExperimentalEvidencefromKenya,ÕÕAmericanEconomicReview,101,no.6(2011),2350Ð2390.Foster,Lucia,HaltiwangerJohn,andChadSyverson,ÔÔReallocationFirmTurnover,andEfÞciency:SelectiononProductivityorProÞtability?ÕÕAmericanEconomicReview,98,no.1(2008),394Ð425.Gibbons,Robert,andRebeccaHenderson,ÔÔRelationalContractsandOrganiza-tionalCapabilities,ÕÕOrganizationScience,23,no.5(2011),1350Ð1364.Greevy,Robert,JeffreyH.Silber,AvitalCnaan,andPaulR.Rosenbaum,ÔÔRan-domizationInferencewithImperfectComplianceintheACE-InhibitorÕÕJournaloftheAmericanStatisticalAssociation,99,no.465(2004),7Ð15.Hanna,Rema,SendhilMullainathan,andJoshSchwartzstein,ÔÔLearningthroughNoticing:TheoryandExperimentalEvidenceinFarmingÕÕ(Harvard:Mimeo,2012).Harrison,Glenn,MortenLau,andElisabethRustrom,ÔÔRiskAttitudes,Randomi-zationtoTreatment,andSelf-SelectionintoExperiments,ÕÕJournalofEco-nomicBehaviorandOrganization,70,no.3(2009),498Ð507.Heckman,James,ÔÔRandomizationinSocialPrograms,ÕÕInEvaluatingWelfareandTrainingPrograms,ed.C.ManskiandI.GarÞnkle(CambridgeMA:HarvardUniversityPress,1992).Hsieh,Chiang-Tai,andPeteKlenow,ÔÔMisallocationandManufacturingTFPinChinaandIndia,ÕÕQuarterlyJournalofEconomics,124,no.4(2009),1403Ð1448.ÑÑÑ,ÔÔDevelopmentAccounting,ÕÕAmericanEconomicJournal:Macroeco-nomics,2,no.1(2010),207Ð223.Ibragimov,Rustam,andUlrichMuller,ÔÔt-statisticBasedCorrelationandHeterogeneityRobustInference,ÕÕJournalofBusinessandEconomicStatistics,28,no.4(2010),453Ð468.Ichniowski,Casey,KathrynShaw,andGiovannaPrennushi,ÔÔTheEffectsofHumanResourceManagement:AStudyofSteelFinishingLines,ÕÕAmericanEconomicReview,87,no.3(1997),291Ð313.Karlan,Dean,RyanKnight,andChrisUdry,ÔÔHopingtoWin,ExpectedtoLose:TheoryandLessonsonMicroenterpriseDevelopment,ÕÕNBERworkingpaper18325,2012.Karlan,Dean,andMartinValdivia,ÔÔTeachingEntrepreneurship:ImpactofBusinessTrainingonMicroÞnanceClientsandInstitutions,ÕÕReviewofEconomicsandStatistics,93,no.2(2011),510Ð527.QUARTERLYJOURNALOFECONOMICS50 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from
Lazear,Edward,UlrikeMalmendier,andRobertoWeber,ÔÔSortinginExperimentswithApplicationtoSocialPreferences,ÕÕAmericanEconomicJournal:AppliedEconomics,4,no.1(2012),136Ð163.Lazear,Edward,andPaulOyerÔÔPersonnelEconomics,ÕÕInHandbookofOrganizationalEconomics,ed.RobertGibbonsandJohnRoberts(Princeton:PrincetonUniversityPress,2012).Leibenstein,H.,ÔÔAllocativeEfÞciencyvs.X-efÞciency,ÕÕAmericanEconomicReview,56,no.3(1966),392Ð415.Locke,Richard,FeiQin,andAlbertoBrause,ÔÔDoesMonitoringImproveLaborStandards?LessonsfromNike,ÕÕIndustrialandLaborRelationsReview,61,no.1(2007),3Ð31.Lucas,Robert,ÔÔOntheSizeDistributionofBusinessFirms,ÕÕBellJournalofEconomics(1978),508Ð523.McKenzie,David,ÔÔImpactAssessmentsinFinanceandPrivateSectorDevelopment:WhatHaveWeLearnedandWhatShouldWeLearn?ÕÕWorldBankResearchObserver,25,no.2(2010),209Ð233.ÑÑÑ,ÔÔBeyondBaselineandFollow-up:TheCaseforMoreTinExperiments,ÕÕJournalofDevelopmentEconomics(2012),forthcoming.McKenzie,David,andChristopherWoodruff,ÔÔWhatAreWeLearningfromBusinessTrainingandEntrepreneurshipEvaluationsaroundtheDevelopingWorld?ÕÕ(Mimeo,WorldBank,2012).Melitz,Marc,ÔÔTheImpactofTradeonIntra-industryReallocationsandIntra-industryProductivity,ÕÕEconometrica,71,no.6(2003),1695Ð1725.Milgrom,Paul,andJohnRoberts,ÔÔTheEconomicsofModernManufacturing:Technology,StrategyandOrganization,ÕÕAmericanEconomicReview,80,no.3(1990),511Ð528.Mundlak,Yair,ÔÔEmpiricalProductionFunctionFreeofManagementBias,ÕÕJournalofFarmEconomics,43,no.1(1961),44Ð56.Powell,Michael,ÔÔProductivityandCredulityinIndustryEquilibriumÕÕ(Mimeo,MIT,2012).Rivkin,Jan,ÔÔImitationofComplexStrategies,ÕÕManagementScience,46,no.6(2000),824Ð844.Schwartzstein,Josh,ÔÔSelectiveAttentionandLearning,ÕÕWorkingPaper,Dartmouth,2012.Stigler,George,ÔÔTheXistenceofX-efÞciency,ÕÕAmericanEconomicReview,66,no.1(1976),213Ð216.Syverson,Chad,ÔÔMarketStructureandProductivity:AConcreteExample,ÕÕJournalofPoliticalEconomy,112,no.6(2004),1181Ð1222.ÑÑÑ,ÔÔWhatDeterminesProductivityattheMicroLevel?ÕÕJournalofEconomicLiterature,49,no.2(2011),326Ð365.Suri,Tavneet,ÔÔSelectionandComparativeAdvantageinTechnologyAdoption,ÕÕEconometrica,79,no.1(2011),159Ð209.Taylor,Frederick.PrinciplesofScientiÞcManagement(NewYork:HarperandBrothers,1911).Walker,Francis,ÔÔTheSourceofBusinessProÞts,ÕÕQuarterlyJournalofEconomics,1,no.3(1887),265Ð288.Womack,James,DanielJones,andDanielRoos,TheMachinethatChangedtheWorld(NewYork:SimonandSchuster,1990).Woodward,John,ManagementandTechnology(Cambridge:CambridgeUniver-sityPress,1958)DOESMANAGEMENTMATTER?51 at University of California, Los Angeles on January 15, 2013http://qje.oxfordjournals.org/Downloaded from