The odds of Y=2 given X=2 means Pr(Y=2)/Pr(Y=1 or 3) when X=2. I suspect this is not really what you want. I'm guessing what you want is an estimate of Pr(Y=2) given X=2 and similarly for the probabilities of the other Y levels at each X level. You could get that easily by using the PREDPROBS=INDIVIDUAL option in the OUTPUT statement. It provides the estimated probability of each response level for each observation. ESTIMATE statements with EXP will give you cumulative odds (such as Pr(Y=0)/Pr(Y=1 or 2) and Pr(Y=0 or 1)/Pr(Y=2)). In any case, the coefficients that you have following X in the ESTIMATE statements should be a list like 1 0 to select level 2 of X since X is categorical and has two estimated parameters. Add the E option so see the coefficients that are used to multiply the model parameters. ... View more

Thank you very much for all of your help and clarity. I moved on to work with PROC CALIS and FIML, but this was very helpful! ... View more

Thank you - it solved the convergence issue, but I am still confused as to why it is listwise deleting vs. pairwise. Can you please advise? ... View more

Hi,   You don't seem to have a specific variable identified for the standard error. The following is code that I have used successfully previously (for a GEE analysis):   proc mianalyze data = xxxx; modeleffects xxxxxxx; stderr xxxxxx; run;   I am by no means an expert on proc MI/MIANALYZE  but may be worth a try. ... View more

  I'm not sure how to troubleshoot this issue. I ran a correlation for the entire model and none of X1-X6 correlate at 1.00 with Y. The content of the variables should not linearly combine to be Y as well. "Should not" is no guarantee that they do not in your data. WARNING: An effect for variable Y is a linear combination of other effects. The          coefficient of the effect will be set to zero in the imputation. Usually means that one (or more) of the X1 to X6 can be created from some of the other X variables. For instance if X1 = 2*X2 that would be a linear combination of other affects.   The error for VAR=1 could be that you have provided a limit with Minimum or Maximum options that cannot be met for one or more of the values. Which would be that every iteration when an attempted imputation was performed that the value was outside the range [0,136]. So you might increase Minmaxiter to see if you get acceptable values, change the restriction to allow a wider range.   Perhaps you need to look at your data a bit closer ... View more

Yes, that is my Y variable and imputation variable of interest. Thank you for noting that, and I've made the appropriate adjustments in MIANALYZE. Updated code here:   proc mi data=dir.ADTrialDataCleanedNoMiss seed=123456789 out=miOutNoMiss minimum=0 maximum=120 nimpute=50 noprint; FCS; var X1 X2 X3 X4 X5 X6 X7 X8 Y; run;   proc sort data=miOutNoMiss; by _IMPUTATION_; run;   proc genmod data = miOutNoMiss; by _imputation_; class X2 X3 X4 X5 X6 X7 (ref = '0'); model Y = X1 X2 X3 X4 X5 X6 X7 X8/covb dist = normal link = identity; ods output ParameterEstimates = gmparms            ParmInfo = gmpinfo                   CovB = gmcovb; run; quit;   proc sort data = gmparms; by _Imputation_; run;   proc mianalyze parms(classvar = level)=gmparms; class X2 X3 X4 X5 X6 X7; modeleffects Intercept X1 X2 X3 X4 X5 X6 X7 X8; run;       However, now I'm getting a new, replacement error:    "ERROR: Within-imputation Estimate missing for effect X2 in _Imputation_= 2 in the input PARMS= data set."   Below is my log   109  proc mi data=data seed=123456789 out=miOutNoMiss minimum=0 109! maximum=120 nimpute=50 noprint; 110  FCS; 111  var X1 X2 X3 X4 X5 X6 X7 X8 Y; 112  run;   NOTE: The data set WORK.MIOUTNOMISS has 6150 observations and 1728 variables. NOTE: PROCEDURE MI used (Total process time):       real time           15.34 seconds       cpu time            1.85 seconds     113 114  proc sort data=miOutNoMiss; 115  by _IMPUTATION_; 116  run;   NOTE: There were 6150 observations read from the data set WORK.MIOUTNOMISS. NOTE: The data set WORK.MIOUTNOMISS has 6150 observations and 1728 variables. NOTE: PROCEDURE SORT used (Total process time):       real time           0.98 seconds       cpu time            0.37 seconds     117 118  proc genmod data = miOutNoMiss; 119  by _imputation_; 120  class X2 X3 X4 X5 X6 X7 (ref = '0'); 121  model Y = X1 X2 X3 X4 X5 X6 X7 X8/covb dist 121! = normal link = identity; 122  ods output ParameterEstimates = gmparms 123             ParmInfo = gmpinfo 124             CovB = gmcovb; 125  run;   NOTE: Algorithm converged. NOTE: The scale parameter was estimated by maximum likelihood. NOTE: The above message was for the following BY group:       Imputation Number=1         .... NOTE: Algorithm converged. NOTE: The scale parameter was estimated by maximum likelihood. NOTE: The above message was for the following BY group:       Imputation Number=50 NOTE: The data set WORK.GMCOVB has 1100 observations and 32 variables. NOTE: The data set WORK.GMPINFO has 1800 observations and 9 variables. NOTE: The data set WORK.GMPARMS has 1850 observations and 10 variables. NOTE: PROCEDURE GENMOD used (Total process time):       real time           34.36 seconds       cpu time            2.52 seconds     126  quit; 127 128  proc sort data = gmparms; 129  by _Imputation_; 130  run;   NOTE: There were 1850 observations read from the data set WORK.GMPARMS. NOTE: The data set WORK.GMPARMS has 1850 observations and 10 variables. NOTE: PROCEDURE SORT used (Total process time):       real time           0.10 seconds       cpu time            0.00 seconds     131 132  proc mianalyze parms(classvar = level)=gmparms; 133  class X2 X3 X4 X5 X6 X7; 134  modeleffects Intercept X1 X2 X3 X4 X5 X6 X7 X8; 135  run;   ERROR: Within-imputation Estimate missing for effect X2 in _Imputation_= 2 in the input        PARMS= data set. NOTE: The SAS System stopped processing this step because of errors. NOTE: PROCEDURE MIANALYZE used (Total process time):       real time           0.06 seconds       cpu time              Thanks, Luke     ... View more

Thank you very much for the response, StatDave. I appreciate the guidance.   Unfortunately, I'm not really sure how to account for the sparse data issue in a manner that preserves the intention of the analysis. We are attempting to hone in on, specifically, each of the 10 time points (since the 11th is our reference beginning point) and their proportion of "successes" in the dichotomous outcome, out of total responses. It is pertinent that we ideally preserve the separate time points as much as possible, or at the very least find an effective manner to condense them into blocks (e.g. "beginning", "middle", and "end", that end up with 3/3/4 time points in each respective block). Given that our outcome variable is binary, I'm not sure how to effectively merge time points while preserving integrity of the analysis (e.g. if an observation has 0/1/0 in the "beginning" block, it wouldn't really be appropriate to umbrella it into either values of "0" or "1").   The only solution to our problem and research goals that I can see is to do a Poisson or Negative Binomial Regression Analysis by transforming our outcome variable into count data. Since we are concerned about the proportion of successes at each particular time point, the only way I could see to model this would be to sum all of the total successes at each time point and model them as a proportion of total observations at each time point (successes/total non-missing per time point). However, my initial attempts at this analysis are showing that my specified model is saturated and is a poor fit.   I can/will post some of my coding attempts to paint a better picture of what I'm asking, but in the meantime - do you have any thoughts or advice on my interpretation and attempt to proceed? Any guidance would be very graciously appreciated.   Best, Luke     ... View more