INCONSISTENCY BETWEEN ACTUAL AND SPECIFIC TAPE LABELS 



When the population of ohservations are on tape or in a cataloged data set, if 

 the leader label on input unit AT APE is not identical to the label on control card 7, 

 the above diagnostic message is generated and program execution is terminated. This 

 provides some assurance that you are reading the appropriate input file. 



MATRIX INVERSION FAILURE, NCO = ---,1 = , SINGULARITY AT J = --, WITH A(.I,J) = 



This diagnostic message is generated when subroutine INVERT is unable to success- 

 fully invert a matrix. The variable NCO in this message indicates at what point in the 

 program the inversion failed. When NGO equals 1, the inversion failure has occurred at 

 statement RISK2610. When NCO equals 999, the inversion failure has occurred at state- 

 ment RISK5360. In all other cases, NGO will be a two or three digit number. The final 

 digit indicates where in the program inversion has failed; the initial one or two 

 digits indicate the iteration on which inversion failed. The various values of NCO are: 



where X is the iteration number on which matrix inversion fails. The variable I in 

 this message is the number of the last observation added to the solution before inver- 

 sion failed. J is the location in the matrix to be inverted that caused inversion 

 failure, and A(J,J) is the actual value of the matrix element that caused inversion 

 failure . 



COEFFICIENTS HAVE DIVERGED TO POINT THAT P EQUALS 



Occasionally for a sorted or weighted set of observations, the regression coeffi- 

 cients may be driven to the point where the estimates of P approach D or 1 . This 

 situation results in a division by 0, which would terminate program execution on the 

 IBM 360/67; therefore, the value of P is checked and the above diagnostic message is 

 generated any time P is equal to or 1 . 



When this message occurs, the user has two courses of action that may allow 

 completion of the analysis: (a) Rearranging the data to eliminate the effects of 

 sorting (unfortunately, this is frequently not feasible) , and (b) increasing the 

 input parameter ELIM (the limit for the sum of weights) , which reduces the effects 

 both of large variability in the weights and of sorted data. 



EXECUTION TERMINATED, SBX = 



This diagnostic message is generated by circumstances similar to those that 

 generate the previous message. However, in this case, the regression coefficients 

 have been driven to the point that P cannot be evaluated by the computer. 'Alien this 

 occurs, the user has the two alternatives discussed for the previous diagnostic 

 message . 



NGO 



Location of Inversion Failure 



X2 

 X3 

 X0 



RISK3460 

 RISK3830 

 RISK4520 



L7 



