Thanks for your reply LVM. I did sort the IDs prior to execution.  The dataset is large and SAS claims that sorting IDs and taking the variable out of the CLASS statement speeds up the execution.  Hence, no CLASS statement. What you say about high inter-correlations makes sense and the non-existence of the convergence point does as well.  However, wouldn't you expect that the mathematically artificial intercept will be somewhere close to the sample-wide event rate?  If you center all of your covariates at the mean, that is. I don't understand why you would expect the intercept to get smaller?  Why not larger?  I don't understand the reason for deviation from the raw rate, much less the direction. Let me know what output you would like to see.  LOGISTIC regression produces an intercept of -2.9888+-.2406.  GLIMMIX regression with identical set of co-variates PLUS random effect of Hospital gets me -4.1852+-.2491.  That corresponds to the predicted sample event rate of 4.8% versus 1.15%.  Huge difference!  WHY!? In this particular example, the variables are not centered around their means so I would not expect the rate to be near the sample-wide unadjusted rate of 10.8%.  I did hope, however, that the overall rate will be similar. Here's the syntax for LOGISTIC: proc logistic data=analyze desc;   model Mortality = <same long list of covariates>; run; ... View more

Here's the same question rephrased: what is the interpretation of the fixed intercept in the mixed effects DIST=binary LINK=logit GLIMMIX model? Normally, it is the logit of event rate at the average/zero level of the covariates' vector.  Does this still hold in the case of GLIMMIX?  Does not seem to be the case.  As soon as I start using co-variates, the intercept value drops significantly even if I use all continuous predictors re-scaled to mean=0.  If the interpretation of the standard LOGISTIC model applies, why is the intercept so much lower?  How does the presence of random effects affect the value of the fixed intercept?  Stumped here ... View more

Hmm, I think I used Quadrature and Double-Dog but this definitely gives me something to look into.  THANKS!  I'll check it out and report back to the group. Follow-up: I tested Restricted PL, Quad and LaPlace and all of them produce virtually identical results.  Back to the drawing board. Message was edited by: Haris Subacius ... View more

And thought I was as clear as it gets Here's the code for full table as well as a PROC REPORT I am currently using--it's a standard table 1 from any medical journal: data have;   input Variable $;   N1 = ranuni(0); N2 = ranuni(0);   Group1 = put(ranuni(0),8.2); Group2 = put(ranuni(0),8.2);   PVal= ranuni(0);   cards; Var1 Var2 Var3 Var4 AgeGr1 AgeGr2 AgeGr3 Race1 Race2 Race3 Region1 Region2 Region3 Region4 ; run; proc report data=have;   column ('Predictor' Variable)   ('Number of Patients' N1 N2)   ('Mean/STD' Group1 Group2)   ('p-value' PVal);   define Variable / display '';   define N1 / display center;   define N2 / display center;   define Group1 / display center;   define Group2 / display center;   define Pval / display center ''; run; Output is the table below (for some reason the (headers are cut off but they don't matter).  At the moment I am typing the bold category headings manually and would like to automate it.  Tried COMPUTE BEFORE but got lost.          Other Pt Variables A 0.9945648 0.4985552   0.32   0.12 0.6276928 B 0.0805072 0.6616233   0.05   0.23 0.4454616 C 0.9524591 0.6282662   0.03   0.38 0.304287 D 0.5737679 0.1142044   0.57   0.49 0.7099973 Age Group AgeGr1 0.9518248 0.2908635   0.35   0.89 0.6725817 AgeGr2 0.507329 0.5758729   0.84   0.09 0.6611321 AgeGr3 0.5157527 0.4067989   0.98   0.40 0.4416961 Race Race1 0.1075504 0.9987347   0.96   0.25 0.5914531 Race2 0.9669518 0.6722821   0.88   0.48 0.4796646 Race3 0.8172502 0.8970209   0.42   0.07 0.6066189 Geographical Region Region1 0.6131752 0.7856588   0.40   0.52 0.0752222 Region2 0.9814383 0.9802798   0.48   0.52 0.6117008 Region3 0.176767 0.2434878   0.82   0.66 0.6990772 Region4 0.8758262 0.9445898   0.72   0.54 0.8402694 ... View more

Tables are all generated outside of PROC REPORT using SQL, FREQ, MEANS, etc.  It's for the demographics table of a manuscripts with predictors as rows, groups as columns, Ns, p-values, etc.  At the moment, I am only interested in making the output prettier and removing the manual step of adding the section headings.  I don't think what I am looking for can be made entirely in PROC REPORT or any one procedure for that matter. Thanks, Haris ... View more

Thank you Cynthia, This code works for what I need; however, I have ten columns in my table.  The equivalent of adding weight as well as seven additional numeric columns to the example you created.  I can write 9 COMPUTE statements but is there a way to apply the same formats to the entire row rather than just one cell?  I tried replacing _COL_ with _ROW_ but that does not seem to work.  Listing Weight next to Height on the COMPUTE line also does not work. Appreciate your help! ... View more

I ended up calculating a rank variable for the x-axis and creating a format for the ranks that contain the values of the third variables for the ranks. ... View more

Thanks for the suggestion, Ballardw.  Nothing there that looks like what I want but a good resource. Also, have done very limited amount of programming in TEMPLATE.  Useful skill but a steep learning curve. I ended up transposing the Tall file into a wide one and using the following code: proc sgplot data=df noautolegend;   highlow x=Outcome low=D1 high=D11 /type=bar;   highlow x=Outcome low=D2 high=D10 /type=bar;   highlow x=Outcome low=D3 high=D9  /type=bar;   highlow x=Outcome low=D4 high=D8  /type=bar;   highlow x=Outcome low=D5 high=D7  /type=bar;   highlow x=Outcome low=D6 high=D6  /type=bar;   scatter x=Outcome y=OR / markerattrs=(color=black symbol=DiamondFilled size=15)   yerrorlower=Lower yerrorupper=Upper errorbarattrs=(color=black thickness=2);   xaxis labelattrs=(weight=bold size=14);   yaxis label='Odds Ratio (95% CI)' labelattrs=(weight=bold size=14); run; Any feedback or suggestions for improvement is greatly appreciated. ... View more

Here's another SGPLOT routine that gives me something similar to what I am looking for but with estimated rather than literal percentiles, 50th, 25th, and 75th percentiles and whiskers at 0 and 100. ... View more