模型开发
注意:如果使用全模型法,所有自变量必须是连续型
data imputed;
set imputed; 离散型的转化
resr = (res='R');
resu = (res='U');
run;
proc logistic data = imputed des; des表示1对应发生响应事件
model ins = $screened resr resu /
屏幕上所有变量
selection=score best=1; 全模型法 score
run;
将res离散型变量转化成哑变量
计算SBC统计量
ods listing close; ods设置输出格式 ods output Nobs=Nobs bestsubsets =score; 输出这两个数据集 proc logistic data = imputed des ; model ins = resr resu brclus2 checks ccbal mmbal income ilsbal posamt nsfamt cd irabal age sav dda invbal crscore brclus3 cc brclus1 cashbk miacctag micrscor moved acctage dirdep savbal ddabal sdb ccpurc inarea atmamt phone mmcred inv depamt brclus4 atm lores / selection=score best=1; run; ods listing;
data _NULL_;
data subset;
set score;
sbc = -scorechisq + log(&obs)*(numberofvariables +1); 变量个数越少越好 score卡方越大越好
run;