Tuesday, March 9, 2010

A macro for Jarque–Bera test

Jarque–Bera test is an important test for normality. This macro works with multivariates.

Note: PROC UNIVARIATE actually computes excess Kurtosis.

%macro jbtest(data = , var = );
   ods listing close;
   proc univariate data = &data ;
       var &var;
      ods output moments = _1;
   run;
  
   data _2;
      set _1;
      label = label1; value = nValue1; output;
      label = label2; value = nValue2; output;
      drop cvalue: label1 label2 nvalue:;
   run;

   proc transpose data = _2 out = _3;
      by varname notsorted;
      id label;
      var value;
   run;

   data _4;
      set _3;
      jb = (skewness**2 * n)/6 + ((kurtosis)**2 *n)/24;
      p = 1 - probchi(jb, 2);
      lable jb = 'JB Statistic' p = 'P-value'
         kurtosis = 'Excess Kurtosis';
   run;
   ods listing;
   proc print data = _4 label;
      title "Jarque-Bera test for variable &var from data &data";
      var varname n skewness kurtosis jb p;
   run;

   proc datasets nolist;
      delete _:;
   quit;
%mend;

/* Run a test using a data set from SAS's HELP library */
%jbtest(data = sashelp.class, var = age height weight); 

Good math, bad engineering

As a formal statistician and a current engineer, I feel that a successful engineering project may require both the mathematician’s abilit...