subtle sources of bias and error

《subtle sources of bias and error》由会员上传分享，免费在线阅读，更多相关内容在工程资料-天天文库。

1、CHAPTER10SubtleSourcesofBiasandErrorJonathanA.SchwabishPleasenote:TheviewsexpressedinthischapterarethoseoftheauthorandshouldnotbeinterpretedasthoseoftheCongressionalBudgetOffice.Beforewegetstarted,letmebeclear:Igettoworkwithsomeofthebestsocioeconomicdataintheworld.Ihaveaccesstodataprovidedb

2、ytheU.S.SocialSecurityAdminis‐tration(SSA),whichprovidesinformationonearnings,governmentbenefits(specifi‐cally,SocialSecuritybenefits),andearningsforahugenumberofpeopleoveralargenumberofyears.Thedataisprovidedtothegovernmentthroughworkers’W-2taxformsorothergovernmentrecords.Bycomparison,sur

3、veydataisoftencollectedfrominterviewsbetweenaninterviewerandrespondent,butmayalsobecollectedonlineorthroughcomputerinterfacesinwhichthereisnointeractionbetweentheinterviewerandinterviewee.Administrativedataisbecomingmorewidelyavailableinmanysocialsciencefields,andwhilethatavailabilityisenab

4、lingresearcherstoasknewandinter‐estingquestions,thatdatahasalsoledtonewquestionsaboutvarioussourcesofbiasanderrorinsurveydata.Administrativedatahasbothadvantagesanddisadvantagesoverpubliclyavailablesur‐veydata.Themajoradvantageisthattheadministrativedatatendstobemoreaccuratethansurveydatabe

5、causeitisnotsubjecttothetypicalerrorsfoundinsurveydata.Sucherrorsinclude:•Nonresponse(therespondentfailstoansweraquestion)•Recallerror(therespondentincorrectlyanswersaquestionorcannotrecallinfor‐mationandfailstoanswer)•Proxyreporting(onepersonrespondstothesurveyforanother)•Sampleselection(t

6、hesurveysampledoesnotrepresentthepopulation)129Furthermore,administrativedataistypicallynotimputed(thatis,thefirmorinstitutionconductingthesurveyhasnotinsertedvaluesformissingobservations)ortop-orbot‐tomcoded(thatis,variablesatthetailsofthedistributionarenotchangedtoprotecttheidentityofresp

7、ondents).Finally,administrativedatamayincludealargernumberofobservationsforlongertimeperiodsthanmaybeavailableinsurveydata.Administrativedataalsohaveseveraldisadvantages,however.First,informationistyp‐icallyavailableononlyalimitedsetofdemographicvariable