Dear Sir/Madam,
I have currently fitted an ordinal logistic model in SAS (using proc logistic) on a 7 level of response varible and I wanted to create the predicted probability plots in order to present and explain my results (Odds ratio and estimates) to non-statisticians. The following SAS code was used below to fit the probability plots for each of the explanatory variables versus the cumulative predicted probabilities.
My questions are:
1) Is there any problem with the code below that I do get a slope of the line on the opposite direction from this estimated Odds Ratio?
2) Is it the aplha which is =0.33 in the output line instead of 0.20 that is used when selecting the explanatory variables? Is there any explanation why my figures can give a line showing an association on the opposite direction from what is the OR shows? In addition most of my figures shows a stronger association than what has been estimated by the Proportional Odds model.
I was wondering if this caused due to the fact that the estimated lines are an interpolation of the scatter plots produced and hence it takes into account all extreme values of all explanatory variables.
I would really appreciate if anyone can advice.
Kind Regards,
Pantelia
proc logistic data = prim descending;
class iBestResponse ;
model idBestResponse = iSwitchTime iTimeFromDiagnosis iBestResponse / link = logit;
where iIStopResist=1 and iPOMpopulation=1;
output out=procLogResults p=predict l=lower u=upper xbeta=logit stdxbeta=selogit / alpha=.33;
run;
proc gplot data=procLogResults;
plot predict*iSwitchTime=_level_;
symbol1 v=none i=rxl c=black;
symbol2 v=none i=rxl c=red;
symbol3 v=none i=rxl c=green;
symbol4 v=none i=rxl c=blue;
symbol5 v=none i=rxl c=brown;
symbol6 v=none i=rxl c=magenta;
run;
proc gplot data=procLogResults;
plot predict*ibestresponse=_level_;
symbol1 v=none i=rxl c=black;
symbol2 v=none i=rxl c=red;
symbol3 v=none i=rxl c=green;
symbol4 v=none i=rxl c=blue;
symbol5 v=none i=rxl c=brown;
symbol6 v=none i=rxl c=magenta;
run;
proc gplot data=procLogResults;
plot predict*iTimeFromDiagnosis=_level_;
symbol1 v=none i=rxl c=black;
symbol2 v=none i=rxl c=red;
symbol3 v=none i=rxl c=green;
symbol4 v=none i=rxl c=blue;
symbol5 v=none i=rxl c=brown;
symbol6 v=none i=rxl c=magenta;
run;
You may leave the list at any time by sending the command
SIGNOFF allstat
to [log in to unmask], leaving the subject line blank.
|