How to solve system worning in Proc Npar1way?

[Diels_O]

Dear professors, my sas software warning when computing the shift of a 7-level ordinal outcome using Hodges-Lehmann
 Estimation, what should I do to deal with this problem?

WARNING: Computing exact confidence limits for this problem may require much time and memory. Press the system interrupt
key to terminate exact computations.

Here's my code.

proc npar1way hl alpha=.05 data=work.stroke;
      class group;
      var D90_mRS_score;
      exact hl;
      ods select WilcoxonScores HodgesLehmann;
   run;

[Rick_SAS]

The warning alerts you to the fact that you have requested a computation that is very complex and potentially time-consuming. If your program runs in a reasonable amount of time, you can ignore the warning.  Here is an example for which the WARNING is printed, but the computation finishes in a fraction of a second:

proc npar1way hl alpha=.05 data=sashelp.class;
      class sex;
      var height;
      exact hl;
      ods select WilcoxonScores HodgesLehmann;
   run;

The details of the EXACT statement are in the doc. For this problem, you are requesting exact confidence limits. 

If the HL computation is taking too long, you'll have to remove the EXACT statement and use the asymptotic estimates. For most exact computations, you can use the MC option to produce a Monte-Carlo estimate of the exact statistic, but the HL exact statistic does not support the MC option.

[Diels_O]

Thanks, how can I perform MC for a confidence intervel, can you give an example?

[Rick_SAS]

Can you provide sample data?

[Diels_O]

Here I upload my sample data, and I want to get the shift of the ordinal variables(mrs90d and NIHSS_24H, NIHSS_7D)， thanks very much!

[Rick_SAS · Accepted solution]

It appears that you have uploaded a "sav" file, which is a SPSS file. You can use PROC IMPORT to convert that file to SAS.

PROC IMPORT
  DATAFILE='C:\Downloads\test.sav'
  OUT=work.stroke
  DBMS=SAV replace;
RUN;

It looks like the data has 300 observations. The split between Group=0 and Group=1 is N0=169 and N1=131.  For samples that large, you are not going to be able to run an exact test for the LH confidence intervals. However, the asymptotic standard error and the traditional HL confidence limits should provide good estimates for your data.

[Diels_O]

Thanks, wish you a nice day!

[Diels_O]

Dear Rick, I have posted my sample data, and can you show me how to calculate the difference of ordinal outcomes and confidence intervel?