If you only wanted to impute then, yes, using the FCS DISCRIM or LOGISTIC method should work. ... View more

Yes it makes sense.  To impute ordinal data like you have you would need to use either the DISCRIM or LOGISTIC method, with the fomer being preferred if you have 7 levels.  You can use either the FCS or MONOTONE statement, depending on the missing data pattern.   That will take care of the imputation step, but you cannot use multiple imputation and cluster analysis because the theory does not lend itself to calculating combined estimates after running the cluster step due to the absence of point estimates and standard errors.   So I would not use multiple imputation in this case.  You could try some single imputation method or use FASTCLUS rather than CLUSTER which is more forgiving with missing values.    ... View more

Typically you would only use multiple imputation when you have a specific analysis in mind.  Once you impute the data, what do you plan to do with it? ... View more

After thinking about this a little more, I am curious about your comment which I initially missed regarding the odds ratio only being reasonable when you report it in standard deviation units.  I am wondering about the distribution of that particular variable.  Are the values really large or really small and how big exactly is the standard deviation? Take a look at the summary statistics for that variable after the imputation (maybe a Proc MEANS with a BY statement) and make sure they look correct.  Again I would check the convergence of your Proc MI code (you can post the LOG if you have any questions).     You could also try standardizing that variable, especially if it has extreme values or extreme variation and see if you get more meaningful results. ... View more

Since you have the OUTEST= data set, you would use the DATA= option in MIANALYZE.   proc mianalyze data=outcoxreg1; modeleffects age_cat0 age_cat1 p02sex educ_cat0 educ_cat1 active_cat0 active_cat1 bmi_cat0 bmi_cat1 p_cvd p_diab depr_cat0 depr_cat1 apoe_cat0 apoe_cat1 apoe_cat2 apoe_cat4 apoe_cat5 &biom_trans.; run;   If you are still interested in figuring out why the estimates are so large then I would suggest you check the imputation models to make sure nothing strange was going on with them. ... View more

It would be helpful to see the LOG from both the Proc MI and Proc LOGISTIC steps as well.  I suspect that there may be an issue with separation related to the biomarker variable.  This usage note will help to explain separation if you are not sure what it is and what to do about it. https://support.sas.com/kb/22/599.html ... View more

There is something odd about the output that you attached.  It appears to be for a different table, namely, ANIMALCAT2*EAR_F_PROTEIN.  Can you double-check to make sure you are looking at the right output.  The other thing I would check is that you have properly subsetted.  Here is an example using that table you sent. data test; do animalcat='A','B','C','D'; do sex='M','F'; input count; do i=1 to count; output; end; end; end; datalines; 67 51 424 166 1612 155 920 30 ; data subset; set test; if animalcat in ('A','B') then animalcat2=animalcat; run; proc surveyfreq data=subset; tables animalcat*sex; run; proc surveyfreq data=subset nomcar; tables animalcat2*sex; run; ... View more

You will need to subset the data to get the 2x2 comparisons and then run SURVEYFREQ independently for each comparison.  The trick however is that in subsetting, you will need to make sure you treat the groups as true subgroups (i.e. domains) in order to make sure the standard errors are correct.  To do this, you will need to set all the groups not in the comparison to missing and then use the NOMCAR option on the SURVEYFREQ statement.   Roughly speaking it would look like this, creating a new variable for each comparison.   data new; set old; if category in ('A', 'B') then newcatab=category; else newcatab=.;   proc surveyfreq data=new nomcar; tables newcatab*sex ....; .... run; ... View more

The ordering of the variables on the VAR statement effects the way in which the imputation models are configured because of the montonicity.  It assumes that the variables are listed such that they exhibit a monotone missing pattern from left to right.  In other words, the left-most variable has no missing values, the second has fewer then the third and so on.  This means that the imputation models are built such that only a variable that is to the left of a given variable may be used in its imputation model   Specifically there is a problem with this MONOTONE statement: monotone discrim(Z_1= y1 X_1   y2 X_2   y3 X_3   y4 X_4           y5 X_5   y6 X_6   y7 X_7  /classeffects=include);   Since y2 (and several others) comes in the VAR list after Z_1 then it cannot be used as a covariate.  Only those variables listed to the left of Z_1 on the VAR statement could be used.  If you want to use models that include other variables then you will need to relax the monotone assumption and use the FCS approach instead. ... View more

You'll need the SOLUTION option on the RANDOM statement ... View more

You can also us MIANALYZE to combine the random effects as well.  You would have to do something like the example below.   data hlm3; seedval = 9273448; do _imputation_=1 to 3; do school = 1 to 5; call rannor(seedval,usi); call rannor(seedval,ust); do class = 1 to 3; call rannor(seedval,uci); call rannor(seedval,uct); do pupil = 1 to 10; call rannor(seedval,upi); call rannor(seedval,upt); do time = 1 to 3; call rannor(seedval,e); y = 1 + usi + uci + upi + time*(1 + ust + uct + upt) + e; output; end; end; end; end;end; run; proc mixed data=hlm3; by _imputation_; class school; model y = time / s; random int time / subject=school solution;; ods output solutionr=rparms(rename=(stderrpred=stderr)); run; /*Skip this step and remove the BY statement if you do not have*/ /*the SUBJECT= option on the RANDOM statement*/ proc sort data=rparms; by school _imputation_; run; proc mianalyze parms=rparms; by school; modeleffects intercept time; run; ... View more