多元回归分析假设检验_第1页
多元回归分析假设检验_第2页
多元回归分析假设检验_第3页
多元回归分析假设检验_第4页
多元回归分析假设检验_第5页
已阅读5页,还剩37页未读 继续免费阅读

下载本文档

版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领

文档简介

多元回归分析假设检验1第1页,课件共42页,创作于2023年2月AssumptionsoftheClassicalLinearModel(CLM)Sofar,weknowthatgiventheGauss-Markovassumptions,OLSisBLUE,Inordertodoclassicalhypothesistesting,weneedtoaddanotherassumption(beyondtheGauss-Markovassumptions)Assumethatuisindependentofx1,x2,…,xkanduisnormallydistributedwithzeromeanandvariance

s2:u

~

Normal(0,s2)2第2页,课件共42页,创作于2023年2月CLMAssumptions(cont.)UnderCLM,OLSisnotonlyBLUE,butistheminimumvarianceunbiasedestimator,whichmeansthatOLShasthesmallestvarianceamongunbiasedestimators;wenolongerhavetorestrictourcomparisontoestimatorsthatarelinearinyi.WecansummarizethepopulationassumptionsofCLMasfollowsy|x~Normal(b0+b1x1+…+bkxk,s2)Whilefornowwejustassumenormality,clearthatsometimesnotthecaseLargesampleswillletusdropnormality3第3页,课件共42页,创作于2023年2月..x1x2ThehomoskedasticnormaldistributionwithasingleexplanatoryvariableE(y|x)=b0+b1xyf(y|x)Normaldistributions4第4页,课件共42页,创作于2023年2月NormalSamplingDistributions5第5页,课件共42页,创作于2023年2月ThetTest6第6页,课件共42页,创作于2023年2月ThetTest(cont)KnowingthesamplingdistributionforthestandardizedestimatorallowsustocarryouthypothesistestsStartwithanullhypothesisForexample,H0:bj=0Ifacceptnull,thenacceptthatxjhasnoeffectony,controllingforotherx’s7第7页,课件共42页,创作于2023年2月ThetTest(cont)8第8页,课件共42页,创作于2023年2月tTest:One-SidedAlternatives

Besidesournull,H0,weneedanalternativehypothesis,H1,andasignificancelevelH1maybeone-sided,ortwo-sidedH1:bj>0andH1:bj<0areone-sidedH1:bj

0isatwo-sidedalternativeIfwewanttohaveonlya5%probabilityofrejectingH0ifitisreallytrue,thenwesayoursignificancelevelis5%9第9页,课件共42页,创作于2023年2月One-SidedAlternatives(cont)Havingpickedasignificancelevel,a,welookupthe(1–

a)thpercentileinatdistributionwithn–k–1dfandcallthisc,thecriticalvalueWecanrejectthenullhypothesisifthetstatisticisgreaterthanthecriticalvalueIfthetstatisticislessthanthecriticalvaluethenwefailtorejectthenull10第10页,课件共42页,创作于2023年2月yi=b0+b1xi1+…

+bkxik+uiH0:bj=0H1:bj>0One-SidedAlternatives(cont)0ca(1-a)Failtorejectreject11第11页,课件共42页,创作于2023年2月AnExample:HourlyWageEquationWagedetermination:(wooldridge,p123)log(wâge)=0.284+0.092educ+0.0041exper+0.022tenure

(0.104)(0.007)(0.0017)(0.003)

n=526R2=0.316Whetherthereturntoexper,controllingforeducandtenure,iszerointhepopulation,againstthealternativethatitispositive.H0:bexper=0vs.H1:bexper>0Thetstatisticist=0.0041/0.0017≈2.41Thedegreeoffreedom:df=n-k-1=526-3-1=522Thecriticalvalueof5%is1.645Andthetstatisticislargerthanthecriticalvalue,ie.,2.41>1.645Thatis,wewillrejectthenullhypothesisandbexperisreallypositive.01.645(1-a)Failtoreject5%reject12第12页,课件共42页,创作于2023年2月Anotherexample:StudentPerformanceandSchoolSizeWhethertheschoolsizehaseffectonstudentperformance?math10,mathtestscores,revealthestudentperformancetotcomp,averageannualteachercompensationstaff,thenumberofstaffperonethousandstudentsenroll,studentenrollment,revealtheschoolsize.TheModelEquationmath10=b0+b1totcomp+b2staff+b3enrollH0:benroll=0,H1:benroll<0TheEstimatedEquationmath10=2.274+0.00046totcomp+0.048staff-0.00020enroll

(6.113)(0.00010)(0.040)(0.00022)

n=408,R2=0.0541df=408-3-1=404,t=-0.00020/0.00022≈-0.91,c=-1.645-0.91>-1.645,sowecan’trejectthenullhypothesis.-1.645reject-09113第13页,课件共42页,创作于2023年2月One-sidedvsTwo-sidedBecausethetdistributionissymmetric,testingH1:bj<0isstraightforward.ThecriticalvalueisjustthenegativeofbeforeWecanrejectthenullifthetstatistic<–c,andifthetstatistic>than–cthenwefailtorejectthenullForatwo-sidedtest,wesetthecriticalvaluebasedona/2andrejectH1:bj

0iftheabsolutevalueofthetstatistic>c14第14页,课件共42页,创作于2023年2月yi=b0+b1Xi1+…

+bkXik+uiH0:bj=0H1:bj>0c0a/2(1-a)-ca/2Two-SidedAlternativesrejectrejectfailtoreject15第15页,课件共42页,创作于2023年2月SummaryforH0:bj=0Unlessotherwisestated,thealternativeisassumedtobetwo-sidedIfwerejectthenull,wetypicallysay“xjisstatisticallysignificantatthea%level”Ifwefailtorejectthenull,wetypicallysay“xjisstatisticallyinsignificantatthea%level”16第16页,课件共42页,创作于2023年2月AnExample:DeterminantsofCollegeGPA(wooldridge,p128)Variables:colGPA,collegeGPAskipped,theaveragenumberoflecturesmissedperweekACT,achievementtestscorehsGPA,highschoolGPATheestimatedmodelĉolGPA=1.39+0.412hsGPA+0.015ACT–0.083skipped

(0.33)(0.094)(0.011)(0.026)

n=141,R2=0.234H0:bskipped=0,H1:bskipped≠

0fd:n-k-1=137,thecriticalvaluet137=1.96Thetstatisticis|-0.083/0.026|=3.19>

t137=1.96,sowewillrejectthenullhypothesisandthebskippedissignanificantlybeyondzero.-1.96reject-3.191.96reject17第17页,课件共42页,创作于2023年2月TestingotherhypothesesAmoregeneralformofthetstatisticrecognizesthatwemaywanttotestsomethinglikeH0:bj=aj

Inthiscase,theappropriatetstatisticis18第18页,课件共42页,创作于2023年2月AnExample:CampusCrimeandEnrollment(wooldridge,p129)Variablescrime,theannualnumberofcrimesoncollegecampusesenroll,studentenrollment,revealthesizeofcollege.Theregressionmodellog(crime)=b0+b1log(enroll)+uWhetherb1=1,thatisH0:b1=1,H1:b1>1log(crime)=-6.63+1.27log(enroll)(1.03)(0.11)n=97R2=0.585df:n-k-1=95,thecriticalvalueat5%ist95=1.645Thet-statisticis(1.27-1)/0.11≈2.45>t95=1.645Sowerejectthenullhypothesisandtheevidenceprovethatb1>1.19第19页,课件共42页,创作于2023年2月AnExample:HousePricesandAirPollution(wooldridge,p131)Variablesprice,medianhousingprice;nox,theamountofnitrogenoxideintheair,inpartspermillion;dist,aweighteddistanceofthecommunityfromfiveemploymentcenters,inmiles;rooms,theaveragenumberofroomsinhousesinthecommunityStratio,theaveragestudent-teacherratioofschoolsinthecommunity.Theestimatedmodellog(priĉe)=11.08-0.954log(nox)-0.134log(dis)+0.255rooms-0.052stratio(0.32)(0.117)(0.043)(0.019)(0.006)

n=506R2=581ThenullhypothesisH0:blog(nox)=-1,H1:blog(nox)≠-1Thetstatisticis(-0.954-(-1))/0.117=0.393,andthecriticalvaluet=1.96.0.393<1.96,sowecan’trejectthenullhypothesis.20第20页,课件共42页,创作于2023年2月ConfidenceIntervals

Anotherwaytouseclassicalstatisticaltestingistoconstructaconfidenceintervalusingthesamecriticalvalueaswasusedforatwo-sidedtestA(1-a)%confidenceintervalisdefinedas21第21页,课件共42页,创作于2023年2月Computingp-valuesforttests

Analternativetotheclassicalapproachistoask,“whatisthesmallestsignificancelevelatwhichthenullwouldberejected?”So,computethetstatistic,andthenlookupwhatpercentileitisintheappropriatetdistribution–thisisthep-value

p-valueistheprobabilitywewouldobservethetstatisticwedid,ifthenullweretrue22第22页,课件共42页,创作于2023年2月Stataandp-values,ttests,etc.Mostcomputerpackageswillcomputethep-valueforyou,assumingatwo-sidedtestIfyoureallywantaone-sidedalternative,justdividethetwo-sidedp-valueby2Stataprovidesthetstatistic,p-value,and95%confidenceintervalforH0:bj=0foryou,incolumnslabeled“t”,“P>|t|”and“[95%Conf.Interval]”,respectively23第23页,课件共42页,创作于2023年2月TestingaLinearCombination

Supposeinsteadoftestingwhetherb1isequaltoaconstant,youwanttotestifitisequaltoanotherparameter,thatisH0:b1=b2,orb1-b2=0Usesamebasicprocedureforformingatstatistic24第24页,课件共42页,创作于2023年2月TestingLinearCombination(cont)25第25页,课件共42页,创作于2023年2月TestingaLinearCombo(cont)So,touseformula,needs12,whichstandardoutputdoesnothaveManypackageswillhaveanoptiontogetit,orwilljustperformthetestforyouInStata,afterregyx1x2…xkyouwouldtypetestx1=x2togetap-valueforthetestMoregenerally,youcanalwaysrestatetheproblemtogetthetestyouwant26第26页,课件共42页,创作于2023年2月Example:SupposeyouareinterestedintheeffectofcampaignexpendituresonoutcomesModelisvoteA=b0+b1log(expendA)+b2log(expendB)+b3prtystrA+uH0:b1=-b2,orH0:q1=b1+b2=0b1=q1

b2,sosubstituteinandrearrange

voteA=b0+q1log(expendA)+b2log(expendB-expendA)+b3prtystrA+u27第27页,课件共42页,创作于2023年2月Example(cont):Thisisthesamemodelasoriginally,butnowyougetastandarderrorforb1

b2=q1directlyfromthebasicregressionAnylinearcombinationofparameterscouldbetestedinasimilarmannerOtherexamplesofhypothesesaboutasinglelinearcombinationofparameters:b1=1+b2;b1=5b2;b1=-1/2b2;etc28第28页,课件共42页,创作于2023年2月MultipleLinearRestrictionsEverythingwe’vedonesofarhasinvolvedtestingasinglelinearrestriction,(e.g.b1=0orb1=b2)However,wemaywanttojointlytestmultiplehypothesesaboutourparametersAtypicalexampleistesting“exclusionrestrictions”

–wewanttoknowifagroupofparametersareallequaltozero29第29页,课件共42页,创作于2023年2月TestingExclusionRestrictionsNowthenullhypothesismightbesomethinglikeH0:bk-q+1=0,...,bk=0ThealternativeisjustH1:H0isnottrueCan’tjustcheckeachtstatisticseparately,becausewewanttoknowiftheqparametersarejointlysignificantatagivenlevel–itispossiblefornonetobeindividuallysignificantatthatlevel30第30页,课件共42页,创作于2023年2月ExclusionRestrictions(cont)Todothetestweneedtoestimatethe“restrictedmodel”withoutxk-q+1,,…,xkincluded,aswellasthe“unrestrictedmodel”withallx’sincludedIntuitively,wewanttoknowifthechangeinSSRisbigenoughtowarrantinclusionofxk-q+1,,…,xk

31第31页,课件共42页,创作于2023年2月TheFstatisticTheFstatisticisalwayspositive,sincetheSSRfromtherestrictedmodelcan’tbelessthantheSSRfromtheunrestrictedEssentiallytheFstatisticismeasuringtherelativeincreaseinSSRwhenmovingfromtheunrestrictedtorestrictedmodel

q=numberofrestrictions,ordfr

dfur

n–k

–1=dfur32第32页,课件共42页,创作于2023年2月TheFstatistic(cont)TodecideiftheincreaseinSSRwhenwemovetoarestrictedmodelis“bigenough”torejecttheexclusions,weneedtoknowaboutthesamplingdistributionofourFstatNotsurprisingly,F~Fq,n-k-1,whereqisreferredtoasthenumeratordegreesoffreedomandn–k–1asthedenominatordegreesoffreedom33第33页,课件共42页,创作于2023年2月0c(1-a)f(F)FTheFstatistic(cont)rejectfailtorejectRejectH0atasignificancelevelifF>ca34第34页,课件共42页,创作于2023年2月Example:thedeterminationsofleaguebaseballplayer’ssalaries(wooldridge,p143)Theregressionmodellog(salary)=b0+b1year+b2gamesyr+b3bavg+b4hrunsyr+b5rbisyr+usalary,the1993totalsalaryyears,yearsintheleaguegamesyr,averagegamesplayedperyearbavg,careerbattingaveragehrunsyr,homerunsperyearrbisyr,runsbattledinperyearThenullhypothesisisH0:b3=0,b4=0,b5=0,whichiscalledmultiplehypothesestestorjointhypothesestest.ThealternativehypothesisisH1:H0isnottrue.Theunrestrictedmodel:log(salary)=11.19+0.0689year+0.0126gamesyr+0.00098bavg+0.0144hrunsyr+0.0108rbisyr

(0.29)(0.0689)(0.0026)(0.00110)(0.0161)(0.0072)

n=353,SSR=183.186,R2=0.6278Therestrictedmodellog(salary)=11.22+0.0713year+0.0202gamesyr

(0.11)(0.0125)(0.0013)

n=353,SSR=198.311,R2=0.597135第35页,课件共42页,创作于2023年2月Example:thedeterminationsofleaguebaseballplayer’ssalaries,cont.Therestrictednumberandthedegreeofthefreedomofrestrictedmodelisq=3;Thedegreeoffreedomofunrestrictedmodelis353-5-1=347;ThentheFstatisticis36第36页,课件共42页,创作于2023年2月TheR2formoftheFstatisticBecausetheSSR’smaybelargeandunwieldy,analternativeformoftheformulaisusefulWeusethefactthatSSR=SST(1–R2)foranyregression,socansubstituteinforSSRuandSSRur37第37页,课件共42页,创作于2023年2月Example:Parent’sEducationinaBirthWeightEquation(wooldridge,p150)Variablesbwght,birthweightinpounds;cigs,averagenumberofcigarettesthemothersmokedperdayduringpregnancy;parity,thebirthorderofthischild;faminc,annualfamilyincome;motheduc,yearsofschoolingforthemother;fatheduc,yearsofschoolingforthefather.Model:bwght=b0+b1cigs+b2parity+b3faminc+b4motheduc+b5fatheduc+uWhethertheparents’educationhasanyeffectonbirthweight?ThisisstatedasH0:b4=0,b5=0,soq=2.bwght=114.524-0.596cigs+1.788parity+0.056faminc-0.370motheduc+0.472fatheduc

(3.728)(0.110)(0.659)(0.037)(0.320)(0.283)

n=1191R2=0.0387bwght=115.470-0.598cigs+1.832parity+0.067faminc

(1.656)(0.109)(0.658)(0.

温馨提示

  • 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
  • 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
  • 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
  • 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
  • 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
  • 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
  • 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。

评论

0/150

提交评论