Chapter 24 - Logistic regression

《Chapter 24 - Logistic regression》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、Chapter24LogisticRegressionContentlistPurposeoflogisticregression568Assumptionsoflogisticregression569Thelogisticregressionequation570Interpretinglogoddsandoddsratio573ModelÞtandlikelihoodfunction574SPSSactivityÐalogisticregressionanalysis575Bytheendofthischapteryo

2、uwillunderstand:1Thepurposesoflogisticregression.2HowtouseSPSStoperformlogisticregression.3HowtointerprettheSPSSprintoutoflogisticregression.IntroductionThischapterextendsourabilitytoconductregression,inthiscasewheretheDVisanominalvariable.Ourpreviousstudiesonregre

3、ssionhavebeenlimitedtoscaledataDVs.ThepurposeoflogisticregressionThecruciallimitationoflinearregressionisthatitcannotdealwithDV’sthataredichoto-mousandcategorical.Manyinterestingvariablesinthebusinessworldaredichotomous:forexample,consumersmakeadecisiontobuyornotbu

4、y,aproductmaypassorfailqualitycontrol,therearegoodorpoorcreditrisks,anemployeemaybepromotedornot.Arangeofregressiontechniqueshavebeendevelopedforanalysingdatawithcategoricaldependentvariables,includinglogisticregressionanddiscriminantanalysis(DA)(Chapter25).Logisti

5、calregressionisregularlyusedratherthanDAwhenthereareonlytwocategoriesofthedependentvariable.LogisticregressionisalsoeasiertousewithSPSSthanDAwhenthereisamixtureofnumericalandcategoricalIV’s,becauseitincludesproceduresforgeneratingthenecessarydummyvariablesautomatic

6、ally,requiresfewerassumptions,andismorestatisticallyrobust.DAstrictlyrequiresthecontinuousindependentvariablesLOGISTICREGRESSION569(thoughdummyvariablescanbeusedasinmultipleregression).Thus,ininstanceswheretheindependentvariablesarecategorical,oramixofcontinuousand

7、categorical,andtheDViscategorical,logisticregressionisnecessary.Logisticregression.Determinestheimpactofmultipleindependentvariablespresentedsimultaneouslytopredictmembershipofoneorotherofthetwodependentvariablecategories.Sincethedependentvariableisdichotomouswecan

8、notpredictanumericalvalueforitusinglogisticregression,sotheusualregressionleastsquaresdeviationscriteriaforbestfitapproachofminimizingerroraroundt