Apparent Discrepancy Between Proc Tabulate and a Manual Calculation (SAS EG, v 7.13 HF4)

[davis]

Good afternoon,

I have a quick question, and was wondering if someone in the community could help.

I am running proc tabulate on some data.  To be sure the code ran as intended, I manually computed a few cells.

Attached is a snapshot from proc tabulate, which reports a 3.69 row percent, a numerator of 642,583 and denominator of 16,780,658.

As a hedge, I manually computed 642,583 / 16,780,658 and came up with 3.83%.  I ASSUME the following code is incorrect. 

Thoughts on the apparent discrepancy?  Thanks.

-----------

proc tabulate

     data=DSN format=comma20.2;

     CLASS STATE        /     ORDER=UNFORMATTED MISSING;

     CLASS Flag1           /     ORDER=UNFORMATTED MISSING;

     CLASS D120_flag   /     ORDER=UNFORMATTED MISSING;

     TABLE /* Row Dimension */

           STATE*(

                D120_flag*(

                N

                RowPctN)

                ALL={LABEL="Total (ALL)"}*N)

                ALL={LABEL="Total (ALL)"}*(

                N

                RowPctN

                D120_flag*(

                N

                ColPctN)),

                /* Column Dimension */

           Flag1;

           ;

* freq cur_upb_amt;

run;

;

[Astounding]

One feature that you have to appreciate is that PROC TABULATE automatically removes all observations that have a missing value for any CLASS variable. So any time FLAG1 or STATE or D120_FLAG is missing, PROC TABULATE ignores the observation. That probably accounts for the difference you see.

If you want to change the result, take a look at the MISSING option.

[davis]

Thanks.

The coding error has been identified.

Row values were not totaled.  So, the denominator used in the calculation was not the 'total'.

The calculation makes sense now that this adjustment was made.

Manual checks.  What would we do without them.