搜索
克隆巴赫（信度）系数（Cronbach'salpha），是心理或教育测验中最常用的信度评估信度工具。其依据一定公式估量测验内部的一致性，作为信度的指标。克服部分折半法的缺点，是目前社会研究最常使用的信度指标，它是测量一组同义或平行测“总和”的信度。克隆巴赫系数公式α﹦（n/n-1）(1-∑Si2/St2)α为信度系数，n为测验题目数，S2i为每题各被试得分的方差，S2t为所有被试所得总分的方差。一般来说，该系数愈高，即工具的信度愈高。在基础研究中，信度至少应达到0.80才可接受，在探索性研究中，信度只要达到0.70就可接受，介于0.70－0.98均属高信度,而低于0.35则为低信度，必须予以拒绝。SPSSFAQWhatdoesCronbach'salphamean?Cronbach'salphaisameasureofinternalconsistency,thatis,howcloselyrelatedasetofitemsareasagroup.Ahighvalueofalphaisoftenused(alongwithsubstantiveargumentsandpossiblyotherstatisticalmeasures)asevidencethattheitemsmeasureanunderlying(orlatent)construct.However,ahighalphadoesnotimplythatthemeasureisunidimensional.If,inadditiontomeasuringinternalconsistency,youwishtoprovideevidencethatthescaleinquestionisunidimensional,additionalanalysescanbeperformed.Exploratoryfactoranalysisisonemethodofcheckingdimensionality.Technicallyspeaking,Cronbach'salphaisnotastatisticaltest-itisacoefficientofreliability(orconsistency).Cronbach'salphacanbewrittenasafunctionofthenumberoftestitemsandtheaverageinter-correlationamongtheitems.Below,forconceptualpurposes,weshowtheformulaforthestandardizedCronbach'salpha:HereNisequaltothenumberofitems,c-baristheaverageinter-itemcovarianceamongtheitemsandv-barequalstheaveragevariance.Onecanseefromthisformulathatifyouincreasethenumberofitems,youincreaseCronbach'salpha.Additionally,iftheaverageinter-itemcorrelationislow,alphawillbelow.Astheaverageinter-itemcorrelationincreases,Cronbach'salphaincreasesaswell(holdingthenumberofitemsconstant).AnexampleLet'sworkthroughanexampleofhowtocomputeCronbach'salphausingSPSS,andhowtocheckthedimensionalityofthescaleusingfactoranalysis.Forthisexample,wewilluseadatasetthatcontainsfourtestitems-q1,q2,q3andq4.Youcandownloadthedatasetbyclickingonalpha.sav.TocomputeCronbach'salphaforallfouritems-q1,q2,q3,q4-usethereliabilitycommand:RELIABILITY/VARIABLES=q1q2q3q4.Hereistheresultingoutputfromtheabovesyntax:Thealphacoefficientforthefouritemsis.839,suggestingthattheitemshaverelativelyhighinternalconsistency.(Notethatareliabilitycoefficientof.70orhigherisconsideredacceptableinmostsocialscienceresearchsituations.)Inadditiontocomputingthealphacoefficientofreliability,wemightalsowanttoinvestigatethedimensionalityofthescale.Wecanusethefactorcommandtodothis:FACTOR/VARIABLESq1q2q3q4/FORMATSORTBLANK(.35).Hereistheresultingoutputfromtheabovesyntax:LookingatthetablelabeledTotalVarianceExplained,weseethattheeigenvalueforthefirstfactorisquiteabitlargerthantheeiganvalueforthenextfactor(2.7vs.0.54).Additionally,thefirstfactoraccountsfor67%ofthetotalvariance.Thissuggeststhatthescaleitemsareunidimensional.