Google
 

Trailing-Edge - PDP-10 Archives - decuslib20-02 - decus/20-0026/multr.doc
There are 2 other files named multr.doc in the archive. Click here to see a list.
SUBROUTINE MULTR

PURPOSE
   PERFORM A MULTIPLE LINEAR REGRESSION ANALYSIS FOR A
   DEPENDENT VARIABLE AND A SET OF INDEPENDENT VARIABLES.  THIS
   SUBROUTINE IS NORMALLY USED IN THE PERFORMANCE OF MULTIPLE
   AND POLYNOMIAL REGRESSION ANALYSES.

USAGE
   CALL MULTR (N,K,XBAR,STD,D,RX,RY,ISAVE,B,SB,T,ANS)

DESCRIPTION OF PARAMETERS
   N	 - NUMBER OF OBSERVATIONS.
   K	 - NUMBER OF INDEPENDENT VARIABLES IN THIS REGRESSION.
   XBAR  - INPUT VECTOR OF LENGTH M CONTAINING MEANS OF ALL
	   VARIABLES. M IS NUMBER OF VARIABLES IN OBSERVATIONS.
   STD	 - INPUT VECTOR OF LENGTH M CONTAINING STANDARD DEVI-
	   ATIONS OF ALL VARIABLES.
   D	 - INPUT VECTOR OF LENGTH M CONTAINING THE DIAGONAL OF
	   THE MATRIX OF SUMS OF CROSS-PRODUCTS OF DEVIATIONS
	   FROM MEANS FOR ALL VARIABLES.
   RX	 - INPUT MATRIX (K X K) CONTAINING THE INVERSE OF
	   INTERCORRELATIONS AMONG INDEPENDENT VARIABLES.
   RY	 - INPUT VECTOR OF LENGTH K CONTAINING INTERCORRELA-
	   TIONS OF INDEPENDENT VARIABLES WITH DEPENDENT
	   VARIABLE.
   ISAVE - INPUT VECTOR OF LENGTH K+1 CONTAINING SUBSCRIPTS OF
	   INDEPENDENT VARIABLES IN ASCENDING ORDER.  THE
	   SUBSCRIPT OF THE DEPENDENT VARIABLE IS STORED IN
	   THE LAST, K+1, POSITION.
   B	 - OUTPUT VECTOR OF LENGTH K CONTAINING REGRESSION
	   COEFFICIENTS.
   SB	 - OUTPUT VECTOR OF LENGTH K CONTAINING STANDARD
	   DEVIATIONS OF REGRESSION COEFFICIENTS.
   T	 - OUTPUT VECTOR OF LENGTH K CONTAINING T-VALUES.
   ANS	 - OUTPUT VECTOR OF LENGTH 10 CONTAINING THE FOLLOWING
	   INFORMATION..
	   ANS(1)  INTERCEPT
	   ANS(2)  MULTIPLE CORRELATION COEFFICIENT
	   ANS(3)  STANDARD ERROR OF ESTIMATE
	   ANS(4)  SUM OF SQUARES ATTRIBUTABLE TO REGRES-
		   SION (SSAR)
	   ANS(5)  DEGREES OF FREEDOM ASSOCIATED WITH SSAR
	   ANS(6)  MEAN SQUARE OF SSAR
	   ANS(7)  SUM OF SQUARES OF DEVIATIONS FROM REGRES-
		   SION (SSDR)
	   ANS(8)  DEGREES OF FREEDOM ASSOCIATED WITH SSDR
	   ANS(9)  MEAN SQUARE OF SSDR
	   ANS(10) F-VALUE

REMARKS
   N MUST BE GREATER THAN K+1.

SUBROUTINES AND FUNCTION SUBPROGRAMS REQUIRED
   NONE

METHOD
   THE GAUSS-JORDAN METHOD IS USED IN THE SOLUTION OF THE
   NORMAL EQUATIONS.  REFER TO W. W. COOLEY AND P. R. LOHNES,
   'MULTIVARIATE PROCEDURES FOR THE BEHAVIORAL SCIENCES',
   JOHN WILEY AND SONS, 1962, CHAPTER 3, AND B. OSTLE,
   'STATISTICS IN RESEARCH', THE IOWA STATE COLLEGE PRESS,
   1954, CHAPTER 8.