Calculating Delta P or Standardized Co-efficeints in PROC LOGISTIC

[deleted_user]

I've read (Menard, 2002) that standardized co-efficients produced in SAS PROC LOGISTIC are only partially standardized- does anyone know if this is still true, or if there has been an update since this was published.

Secondly, Menard recommends instead to calculate the Delta-P statistic

Delta P = exp([L.sub.1])/1 + exp([L.sub.1]) - exp([L.sub.0])/1 + exp([L.sub.0])

Is anyone aware of an option in PROC LOGISTIC that would give this statistic?

-thanks

[statsplank]

Hi slutsky_fan,

Can you please post a full reference for Menard 2002? I am curious why coefficients produced by PROC LOGISTIC are called standardized/partially standardized - to my knowledge they are not standardized at all...

Regarding your second statement, could you clarify instead of what Menard recommends calculating this Delta-P statistic?

Also, I'm not sure what are the L.sub.1 and L.sub.0 in your formula - could you put some detail about these?

Thank you!

[deleted_user]

sorry, I've gotten involved in other projects ( end of semester chaos) since my post- I'm trying to figure out where I left off. To partially answer your question, with regard to PROC LOGISTIC I'm referring to using the 'stb' option to get standardized co-efficients. ( which I', confused about if they are 'fully standardized.'


In Forest Science 52(6) 2006 'Best Predictors for Postfire Mortality of Ponderosa Pine Trees in the Intermountain West' the authors ranked the significant variables using the stb option. I think this would only work for quantitative varaibles vs. categorical.

I've got to get back to the library to get the specifics of the Menard citation.

I got the formula for delta p here:

http://findarticles.com/p/articles/mi_m0HCZ/is_3_29/ai_83315687/pg_5/

and more discussion here:

http://www.springerlink.com/content/f183527308p220n1/fulltext.pdf

( I'm not sure if anyone can access this or if I only have access via my institution)

I'll repost when I obtain the Menard pub.

[Alfredo]

For delta_p calculation, try the following code:

proc logistic data=&data ;

   model &depvar (event="1") = ... ;

   ods output parameterestimates=_parms (where=(variable ne "Intercept"));

run;

proc sql noprint; select sum(&depvar)/count(*) into :prior from &data; quit;

data delta_p;

   set _parms; retain P0 &prior;

   L0=log(P0/(1-P0));

   L1=L0+estimate;

   P1=exp(L1)/(1+exp(L1));

   Delta_P=P1-P0;  drop L0 L1;

run;