Chapter19CarryingOutanEmpiricalProjectWooldridge:IntroductoryEconometrics:AModernApproach,5eGoalofthischapterLearnhowtocompleteatermproject/writeatermpaperPosingaquestionKnowingpreciselywhatquestionyouwanttoanswerisessentialYoucanonlycollectyourdataifyouexactlyknowyourquestionYoucanonlyknowwhetheryoucancompleteyourprojectintheallottedtimeifyouknowwhetherthenecessarydataisavailableYoucanonlyknowifyourresearchquestionisofinteresttosomeoneifyoucanpreciselystateitanddiscussitwithyourclassmates/instructorCarryingoutanEmpiricalProjectFindinginterestingresearchquestionsChoosetheareaofeconomics/socialsciencesyouareinterestedinExamplesfortypicalresearchquestionsLaborEconomics:ExplainingwagedifferentialsPublicEconomics:EffectoftaxesoneconomicactivityEducationEconomics:EffectofspendingonschoolperformanceMacroeconomics:EffectofinvestmentonGNPgrowthLookforpublishedpapersonthechosentopicusingtoolssuchasEconLit,GoogleScholar,theJournalofEconomicLiterature(JEL)etc.CarryingoutanEmpiricalProjectYourresearchprojectshouldaddsomethingnewAddanewvariablewhoseinfluencehasnotbeenstudiedbeforeExpandeconomicquestionstoincludefactorsfromothersciencesStudyanexistingquestionformorerecentdata(maybeboring)UseanewdatasetorstudyaquestionforadifferentcountryTryoutnew/alternativemethodstostudyanoldquestionFindacompletelynewquestion(hardbutpossible)IthelpsifyourresearchquestionispolicyrelevantoroflocalinterestCarryingoutanEmpiricalProjectLiteraturereviewAliteraturereviewisimportanttoplaceyourpaperintocontextUseonlinesearchservicestosystematicallysearchforliteratureWhensearching,thinkofrelatedtopicsthatmayalsoberelevantAliteraturereviewcanbepartofintroductionoraseparatesectionDatacollectionMostquestionscanbeaddressedusingalternativetypesofdata(purecross-sections,repeatedcross-sections,timeseries,panels)CarryingoutanEmpiricalProjectDecidingontheappropriatedatasetManyquestionscaninprinciplebestudiedusingasinglecross-sectionButforareasonableceterisparibusanalysisoneneedsenoughcontrolsPaneldataprovidesmorepossibilitiesforconvincingceterisparibusanalysesasonecancontrolfortime-invariantunobservedeffectsExamplesforpaneldatasets:PSID(individuals),Compustat(firms)Paneldataforcities,counties,statesetc.areoftenpubliclyavailableDatasetsareoftenavailableonline,injournalarchives,orfromauthorsCarryingoutanEmpiricalProjectEnteringandstoringyourdataDataformats:1)printed,2)ASCII,3)spreadsheet,4)softwarespecificImportantidentifiers:1)observationalunit,2)timeperiodTimeseriesmustbeorderedaccordingtotimeperiodPaneldataareconvenientlyorderedasblocksofindividualdataItisalwaysimportanttocorrectlyidentifyandhandlemissingvaluesNonnummericaldataalsohavetobehandledwithgreatcareSoftwarespecificformatsoftenprovidegoodwaysofdocumentationCarryingoutanEmpiricalProjectInspecting,cleaning,andsummarizingyourdataItisextremelyimportanttobecomefamiliarwithyourdatasetEvendatasetsthatwereusedbeforemaycontainproblems/errorsLookatindividualentries/trytounderstandthestructureofyourdataUnderstandhowmissingvaluesarecoded;iftheyarecodedas“999“or“-1“,thiscanbeextremelydangerousforyouranalysisItisbettertousenonnummericalvaluesformissingvaluesUnderstandtheunitsofmeasurementofyourvariablesCarryingoutanEmpiricalProjectInspecting,cleaning,andsummarizingyourdataKnowwhetheryourdataisreal/nominal,seasonallyadjusted/unadjustedCheckifmeans,std.dev.,mins,andmaxsofyourdataareplausibleCleanyourdataofimplausiblevaluesandobviouscodingerrorsWhenmakingdatatransformations(differencing,growthrates)makesureyourdataiscorrectlyorderedandnowrongoperationsresultForexample,inapaneldataset,beawarethatthefirstobservationofeachcross-sectionalunithasnopredecessorCarryingoutanEmpiricalProjectEconometricAnalysisGivenyourresearchquestionandthedataavailable,youhavetodecideontheappropriateeconometricmethodstouseSomegeneralguidelinesOLSisstillthemostwidelyusedmethodandoftenappropriateMakesurethekeyassumptionsaresatisfiedinyourmodelAlwayscheckforpossibleproblemsofomittedvariables,self-selection,measurementerror,andsimultaneityCarryingoutanEmpiricalProjectSomegeneralguidelinesCarefullychoosefunctionalformspecifications(logs,squaresetc.)Beginnersmistake:donotincludevariablesthatarelistedasnumericalvaluesbuthavenoquantitativemeaning(e.g.,3-digitoccupations)TransformsuchvariablestodummyvariablesrepresentingcategoriesHandleordinalregressorsinasimilarway(e.g.,jobsatisfaction)Forordinaldependentvariables,thereareorderedlogit/probitmodelsOnecanalsoreduceorderedvariablestobinaryvariablesCarryingoutanEmpiricalProjectSomegeneralguidelinesThinkofsecondarycomplicationssuchasheteroscedasticitySpecificproblemsintimeseriesregressions:1)levelsvs.differences,2)trendsandseasonality,3)unitrootsandcointegrationCarryoutmisspecificationtestsandthinkaboutpossiblebiasesSensitivityanalysis:lookatvariationsofyourspecification/methodHopefully,resultsdonotchangeinasubstantialwayArethereproblemswithoutliers/influentialobservations?CarryingoutanEmpiricalProjectSpecificaspectstothinkofwhenusingpaneldataKeyassumptionsRandomeffects:regressorsunrelatedtoindividualspecificeffectsFixedeffects:regressorsrelatedtoindividualspecificeffectsThefixedeffectsassumptionisoftenmoreconvincingContemporaneousexogeneity:idiosyncraticerrorsareuncorrelatedwiththeexplanatoryvariablesofthesametimeperiodStrictexogeneity:idiosyncraticerrorsareuncorrelatedwiththeexplanatoryvariablesofalltimeperiods(oftenproblematic)CarryingoutanEmpiricalProjectSpecificaspectstothinkofwhenusingpaneldataMethodsforpaneldataPooledOLS:randomeffectsassumption,serialcorrelationoferrorterms,needsonlycontemporaneousexogeneityRandomeffectsestimation:randomeffectsassumption,moreefficientthanpooledOLS,needsstrictexogeneityFixedeffectsestimation:fixedeffectsassumption,problemwithtimeinvariantregressors,needsstrictexogeneityFirstdifferencing:similartofixedeffects,goodforlongertimeseriesCarryingoutanEmpiricalProjectDatamining/specificationsearchesTheprocessoflookingforthebestmodeliscalledspecificationsearchOften,onestartswithageneralmodelanddropsinsignificantvariablesIfthespecificationsearchentailsmanysteps,thisisproblematicOurassumptionsactuallyrequirethatthemodelisonlyestimatedonceIfonesequentiallyestimatesanumberofmodelsonthesamedata,theresultingteststatisticsandp-valuescannotbeinterpretedanymoreThis(difficult)problemisoftenignoredinpracticeOneshouldkeepthenumberofspecificationstepstoaminimumCarryingoutanEmpiricalProjectWritinganempiricalpaperAsuccesfulempiricalpapercombinesacareful,convincingdataanalysiswithgoodexplanationsandaclearexpositionIntroductionStatebasicobjectivesandexplainwhythetopicisimportantLiteraturereview:Whathasbeendone?Howdoyouaddtothis?Grabthereader‘sattentionbypresentingsimplestatistics,paradoxicalevidence,topicalexamples,orchallengestocommonwisdomOnemaygiveashortsummaryofresultsintheintroductionCarryingoutanEmpiricalProjectConceptual(ortheoretical)frameworkDescriptionofgeneralapproachtoansweringyourresearchquestionYoumaydelevop/useaformaleconomicmodelforthisForexample,settingupautilitymaximizationmodelofcriminalactivityclarifiesthefactorsthatmatterforexplainingcriminalactivityHowever,oftencommoneconomicsensesufficestodiscussthemainmechanismsandcontrolvariablesthathavetobetakenintoaccountAsoneisinmostcasesinterestedinansweringacausalquestion,aconvincingdiscussionofwhatvariablestocontrolforisessentialCarryingoutanEmpiricalProjectEconometricmodelsandestimationmethodsSpecifythepopulationmodelyouhaveinmindExample:EffectsofalcoholconsumptiononcollegeGPAExample:Timeseriesmodelofcity-levelcartheftsExplainyourfunctionalformchoicesCarryingoutanEmpiricalProjectEconometricmodelsandestimationmethodsAfterspecifyingapopulationmodel,discussestimationmethodsDescribehowyoumeasurethevariablesinyourpopulationmodelWhenusingOLS:Discusswhyexogeneityassumptionshold,andhowyoudealwithheteroscedasticity,serialcorrelationandthelikeWhenusingIV/2SLS:Explainwhyyourinstrumentalvariablesfulfilltheassumptions:1)exclusion,2)exogeneity,3)partialcorrelationWhenusingpanelmethods:Explainwhattheunobservedindividualspecificeffectsstandfor,andhowtheyareremoved/accountedforCarryingoutanEmpiricalProjectDataCarefullydescribethedatausedinyourempiricalanalysisNamethesourcesofyourdataandhowtheycanbeobtainedTimeseriesdataandshortdatasetsmaybelistedintheappendixIfyourdataisself-collected,includeacopyofthequestionnaireDiscusstheunitsofmeasurementofthevariablesofinterestPresentsummarystatisticsforthevariablesusedintheanalysisFortrendingvariables,growthratesorgraphsaremoreappropriateAlwaysstatehowmanyobservationsyouusefordifferentestimationsCarryingoutanEmpiricalProjectResultsPresentestimatedequations,or,iftherearetoomany,presenttablesAlwaysincludethingslikeR-squaredandthenumberofobservationsAreyourestimatedcoefficientsstatisticallysignificant?Aretheyeconomicallysignificant?Whatistheirmagnitude?Ifcoefficientsdonothavetheexpectedsigns,thismayindicatethereisaspecificationproblem,forexample,omittedvariablesRelatedifferencesbetween