Issue with Little's MCAR macro

Arbormazoo — Sat, 20 Jul 2019 20:05:17 GMT

I am not a proficient macro user and have not used PROC IML prior to now, so forgive me if the answer is obvious. I'm working with SAS 9.4.

When I run the code, the only error I receive is:

"ERROR: (execution) Matrix has not been set to a value."

I am familiar with the fact that this error generally means that something has not been defined (ex. y=x, but x doesn't exist). I have scoured the previous boards, and read through SAS documents, and have not been able to identify why the final %mcartest (line 4329) would result in this error. Previous problems, using the same macro, were related to data input and perhaps errors within the code, but I have not come across any examples where the source of error is the %mcartest.

If anyone can provide some feedback/suggestions, I would greatly appreciate it. I have attached the original code as a word document and the log (with error statement) as a PDF for convenience, as well.

Thank you, in advance.

**********************************************************************************************************************************************

%macro mcartest
(indata= ems.recodeptsd /* Input DATASET name */
,testvars= sc1 ptsd1_1 ptsd1_2 ptsd1_3 ptsd1_4 ptsd1_5 ptsd1_6 ptsd1_7 ptsd1_8 ptsd1_9 ptsd1_10 ptsd1_11 ptsd1_12 ptsd1_13
ptsd1_14 ptsd1_15 ptsd1_16 ptsd1_17 /* SPECIFY VARIABLE SET FOR THE MCAR TEST */
,misscode= . /* SPECIFY THE MISSING VALUE CODE */
);
/******************************************************************************************************************
* This SAS macro implements the chi-square test for a missing completely at random (MCAR) mechanism, as *
* outlined in Little's (1998) JASA article. Note that the macro requires SAS version 8.2 (or higher) because *
* PROC MI is used to obtain ML estimates of the covariance matrix and mean vector. * *
******************************************************************************************************************/

%local numvars ;
%let numvars = %sysfunc(countw(&testvars));

data one;
set &indata (keep=&testvars);
array m[&numvars] &testvars ;
array r[&numvars] r1 - r&numvars ;

do i = 1 to &numvars;
if m[i] = &misscode then m[i] = .;
r[i] = not missing(m[i]);
end;
drop i;
run;

proc sort;
by r1-r&numvars;
run;

proc mi data = one nimpute = 0 noprint;
var &testvars;
em outem = emcov;
run;

proc iml;

use one;
read all var {&testvars} into y;
read all var {%do i = 1 %to &numvars; r&i %end;} into r;
use emcov;
read all var {&testvars} into em;

mu = em[1,];
sigma = em[2:nrow(em),];

/* ASSIGN AN INDEX VARIABLE DENOTING EACH CASE'S PATTERN */

jcol = j(nrow(y), 1 , 1);

do i = 2 to nrow(y);
rdiff = r[i,] - r[i - 1,];
if max(rdiff) = 0 & min(rdiff) = 0 then jcol[i,] = jcol[i - 1,];
else jcol[i,] = jcol[i - 1,] + 1;
end;

/* NUMBER OF DISTINCT MISSING DATA PATTERNS */

j = max(jcol);

/* PUT THE NUMBER OF CASES IN EACH PATTERN IN A COL VECTOR M */
/* PUT THE MISSING DATA INDICATORS FOR EACH PATTERN IN A MATRIX RJ */

m = j(j, 1, 0);
rj = j(j, ncol(r), 0);

do i = 1 to j;
count = 0;
do k = 1 to nrow(y);
if jcol[k,] = i then do;
count = count + 1;
end;
if jcol[k,] = i & count = 1 then rj[i,] = r[k,];
m[i,] = count;
end;
end;

/* COMPUTE D^2 STATISTIC FOR EACH J PATTERN */

d2j = j(j, 1, 0);

do i = 1 to j;

/* OBSERVED VALUES FOR PATTERN J */
yj = y[loc(jcol = i),loc(rj[i,] = 1)];

/* VARIABLE MEANS FOR PATTERN J */
ybarobsj = yj[+,]/nrow(yj);

/* D = P X Pj MATRIX OF INDICATORS (SEE P. 1199) */
Dj = j(ncol(y), rj[i,+], 0);

count = 1;
do k = 1 to ncol(rj);
if rj[i,k] = 1 then do;
Dj[k, count] = 1;
count = count + 1;
end;
end;

/* REDUCE EM ESTIMATES TO CONTAIN OBSERVED ELEMENTS */
muobsj = mu * Dj;
sigmaobsj = t(Dj) * sigma * Dj;

/* THE CONTRIBUTION TO THE D^2 STATISTIC FOR EACH OF THE J PATTERNS */
d2j[i,] = m[i,] * (ybarobsj - muobsj) * inv(sigmaobsj) * t(ybarobsj - muobsj);

end;

/* THE D^2 STATISTIC */
d2 = d2j[+,];

/* DF FOR D^2 */
df = rj[+,+] - ncol(rj);
p = 1 - probchi(d2,df);

/* PRINT ANALYSIS RESULTS */
file print;
put "Number of Observed Variables = " (ncol(rj)) 3.0;
put "Number of Missing Data Patterns = " (j) 3.0; put;
put "Summary of Missing Data Patterns (0 = Missing, 1 = Observed)"; put;
put "Frequency | Pattern | d2j"; put;
do i = 1 to nrow(rj);
put (m[i,]) 6.0 " | " @;
do j = 1 to ncol(rj);
put (rj[i,j]) 2.0 @;
end;
put " | " (d2j[i,]) 8.6;
end;
put;
put "Sum of the Number of Observed Variables Across Patterns (Sigma psubj) = " (rj[+,+]) 5.0; put;
put "Little's (1988) Chi-Square Test of MCAR"; put;
put "Chi-Square (d2) = " (d2) 10.3;
put "df (Sigma psubj - p) = " (df) 7.0;
put "p-value = " (p) 10.3;

%mend mcartest;

%mcartest
quit;
run;

Re: Issue with Little's MCAR macro

Reeza — Sat, 20 Jul 2019 23:04:45 GMT

you need to also show how you're calling the macro and the log. It's possible the issue is with your data.

topic Re: Issue with Little's MCAR macro in SAS/IML Software and Matrix Computations

Issue with Little's MCAR macro

Re: Issue with Little's MCAR macro