Power Divergence Test Statistic

[marygc]

 I am looking for SAS syntax or SAS macro with Code that will perform Goodness of Fits on discrete categorical data. 

Specifically one that shows how to do the math described by N. Cressie, T.R.C. Read, in their paper entitled 

"Multinomial goodness-of-fit tests," J. Roy. Statist. Soc. B 46 (1984) 440–464.

I do have code now for 4 of the 5 test statistics I need.

1. proc freq gives the Pearson's Chi Square,
2. Proc Univarite produces the Komogorov Smirnov test statistic,
3. Proc Probit produces the Log Likelihood Ratio as does Proc GenMod,
4. Proc Multtest  calculates the Freeman-Tukey test statistic,

but I have not been able to find the Power divergence test show in the image attached here.   I found some R code so maybe I can run it inside SAS enterprise guide?  But I do NOT have a SAS/IML or SAS Enterprise Miner license so I can not run the R code inside a node that way.  Please advise.

[marygc]

Specifically one that shows how to do the math described by N. Cressie, T.R.C. Read, in their paper entitled 

"Multinomial goodness-of-fit tests," J. Roy. Statist. Soc. B 46 (1984) 440–464.

I do have code now for 4 of the 5 test statistics I need.

1. proc freq gives the Pearson's Chi Square,
2. Proc Univarite produces the Komogorov Smirnov test statistic,
3. Proc Probit produces the Log Likelihood Ratio as does Proc GenMod,
4. Proc Multtest  calculates the Freeman-Tukey test statistic,

but I have not been able to find the Power divergence test show in the image attached here.   I found some R code so maybe I can run it inside SAS enterprise guide?  But I do NOT have a SAS/IML or SAS Enterprise Miner license so I can not run the R code inside a node that way.  Please advise.

[PGStats · Accepted solution]

It can be computed with a SQL query from proc freq output. For example

proc freq data=sashelp.heart;
where weight_status is not missing;
table sex*weight_status / out=cellcounts outexpect;
run;

%let lambda=(2/3);

proc sql;
select
    &lambda as lambda, 
    2/(&lambda*(1+&lambda))*sum(COUNT*((COUNT/EXPECTED)**&lambda-1)) as powerDivergence
from cellCounts;
quit;