Both OLS and WLS/GLS (after setting the weights to zero for the inadmissible cases) give some predictions outside the range [0,1]. This demonstrates the uselessness of the linear probability model.READ (DATA12-1) Accept GPA Bio Chem Phy Red Prb Qnt Age Gender OLS Accept GPA Bio Chem Phy Red Prb Qnt Age Gender /PREDICT=Yhat1 PRINT Yhat1 GENR D=(Yhat1.GT.0)*(Yhat1.LT.1) GENR SigmaSq=Yhat1*(1-Yhat1) GENR Wt=1/SigmaSq*D OLS Accept GPA Bio Chem Phy Red Prb Qnt Age Gender /PREDICT=Yhat2 WEIGHT=Wt PRINT Yhat1 D SigmaSq Wt Yhat2 PROBIT Accept GPA Bio Chem Phy Red Prb Qnt Age Gender
Updated March 24, 2000
noelroy@morgan.ucs.mun.ca