Discriminant Analysis Using Quadratic Discriminant Function

In SPSS, there is no subroutine for Quadratic Discriminant Analysis, I write this macro to compute quadratic score and posterior probability. If you have any problem, please let me know.

SET MXLOOP =99999.

MATRIX.

GET DATA/VARIABLES=X1 to Xp/MISSING=OMIT.

GET Group/VARIABLES=Y/MISSING=OMIT.

COMPUTE GMAX=CMAX(Group).

COMPUTE GMIN=CMIN(Group).

COMPUTE NR=NROW(DATA).

COMPUTE NC=NCOL(DATA).

COMPUTE RN=MAKE(GMAX-GMIN+1,1,1).

COMPUTE SSIGMA=MAKE(NC*(GMAX-GMIN+1),NC,0).

COMPUTE PI=4*ARTAN(1).

COMPUTE QDA=MAKE(NR, (GMAX-GMIN)+1,-999).

COMPUTE QDAPOST=MAKE(NR, 2*(GMAX-GMIN)+2,-999).

COMPUTE TEMP2=0.

LOOP I=GMIN TO GMAX.

COMPUTE TEMP1=0.

LOOP J=1 TO NR.

DO IF (Group(J)=I).

COMPUTE TEMP1=TEMP1+1.

END IF.

END LOOP.

COMPUTE TEMPDATA=MAKE(TEMP1,NC,999).

LOOP K=1 TO TEMP1.

COMPUTE TEMPDATA(K,:)=DATA(K+TEMP2,:).

END LOOP.

COMPUTE PG=NROW(TEMPDATA)/NR.

COMPUTE TEMP2=TEMP1+TEMP2.

COMPUTE MEAN=T(CSUM(TEMPDATA)/NROW(TEMPDATA)).

COMPUTE SIGMA=(SSCP(TEMPDATA)-NROW(TEMPDATA)*MEAN*T(MEAN))/(NROW(TEMPDATA)-1).

PRINT SIGMA.

LOOP U=1 TO NC.

COMPUTE SSIGMA(4*(I-1)+U,:)=SIGMA(U,:).

END LOOP.

LOOP V=1 TO NR.

COMPUTE XX=T(DATA(V,:)).

COMPUTE QD=TEMP1/NR*(2*PI)&**(-1*NC/2)*DET(SIGMA)&**(-0.5)*EXP(-0.5*T(XX-MEAN)*INV(SIGMA)*(XX-MEAN)).

COMPUTE QDA(V,I)=LN(QD)+(NC/2)*LN(2*PI).

COMPUTE QDAPOST(V,I)=LN(QD)+(NC/2)*LN(2*PI).

END LOOP.

END LOOP.

LOOP W= 1 TO NC.

LOOP X=1 TO NR.

COMPUTE QDAPOST(X,GMAX-GMIN+1+W)=QDA(X,W)/RSUM(QDA(X,:)).

END LOOP.

END LOOP.
LOOP X=1 TO NR.
LOOP GPRED= 1 TO GMAX-GMIN+1.
DO IF (QDAPOST(X,GPRED)= RMAX(QDA(X,:))).
COMPUTE QDAPOST(X,2*(GMAX-GMIN)+3)=GPRED.
END IF.
END LOOP.
END LOOP.

PRINT QDAPOST.

END MATRIX.