资源描述：
《多分类logistic回归》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。
1、多分类logistic回归Logistic回归要求应变量Y是0/1两分类变量。如果Y是多分类变量，如疾病结局可有恶化、好转、痊愈三种结局，就应该使用该模块多分类(multinomiallogistic)逻辑回归。该模块不仅输出Multinomiallogistic回归结果，而且给出按序合并生成0/1两分类变量后的ordinarylogistics回归结果。如Y为0、1、2三分组变量，将有两种合并方法：1和2合并然后与0比，或合并0和1然后与2比。该模块同时给出每个自变量对Y的分布的预测值，并用图形显示。例：DEMO数据分析对SBP的三分组进行回归分析，输
2、入界面如下图所示输出结果：MultinomialLogisticRegressionOutcome:SystolicBP,mmhg三分组SEX:MaleMultinomialLogisticRegression:Odd.ratio(95%CI)P.valueSBP.T3=SBP.T3=1SBP.T3=201.017.3557(0.8369,359.9301)0.3025(0.0156,5.8780)(Intercept)(ref.)0.06510.42961.00.9100(0.7993,1.0360)0.9973(0.8808,1.1292)BMI(r
3、ef.)0.15390.96581.00.9837(0.9613,1.0065)1.0381(1.0164,1.0603)AGE(ref.)0.15930.00051.01.0226(0.5852,1.7870)0.9584(0.5525,1.6624)factor(OCCU.NEW)2(ref.)0.93750.87981.00.6797(0.2983,1.5489)1.1612(0.5524,2.4411)factor(EDU.NEW)2(ref.)0.35820.69331.01.2542(0.5313,2.9607)1.1300(0.4953,2
4、.5781)factor(EDU.NEW)3(ref.)0.60520.7714Predictedprobabilityforeachsub-group(withotherXs=meanorthefirstlevel)Sub-groupP(SBP.T3=0)P(SBP.T3=1)P(SBP.T3=2)OCCU.NEW=10.287140.367140.34572OCCU.NEW=20.288900.377730.33337EDU.NEW=10.287140.367140.34572EDU.NEW=20.306080.265990.42793EDU.NEW
5、=30.252260.404530.34321OrdinarylogisticregressionOR95%CILow95%CIUppP-valueBMI0.98790.90051.08380.7962AGE1.03431.01821.0507<0.0001factor(OCCU.NEW)20.94760.63491.41420.7921factor(EDU.NEW)21.19550.67062.13120.5449factor(EDU.NEW)31.16890.63102.16520.61970
6、11.06020.12419.05740.95741
7、2
8、5.16340.601144.35440.1346SEX:FemaleMultinomialLogisticRegression:Odd.ratio(95%CI)P.valueSBP.T3=SBP.T3=1SBP.T3=201.00.0294(0.0021,0.4057)3e-040(0.0000,0.0060)(Intercept)(ref.)0.0085<0.00011.01.0944(0.9858,1.2150)1.2288(1.0915,1.3833)BMI(ref.)0.09080.00071.01.0419(1.0142,1.0703)1.1
9、138(1.0810,1.1475)AGE(ref.)0.0028<0.00011.00.9164(0.5460,1.5381)0.3478(0.1865,0.6487)factor(OCCU.NEW)2(ref.)0.74110.00091.00.9895(0.5246,1.8666)0.7482(0.3356,1.6680)factor(EDU.NEW)2(ref.)0.97410.47821.02.4945(1.1156,5.5776)0.7209(0.1957,2.6558)factor(EDU.NEW)3(ref.)0.02600.6228Pr
10、edictedprobabilityforeachsub-group(witho
