Hello, I am new to proc iml and am trying to simulate multivariate data to run through statistical models and compute power.  Below is very simplified code similar to what I am do - the middle of the program where here I am simulating one variable from a normal distribution would be more complicated and would be simulating multivariate data with a specified correlation.  The part I cannot figure out is how to incorporate the do loops into iml and get out the same type of output I would get from the code below (i.e, columns for run, n, trt, plot, and y) so that I can run my statistical model by run and n and compute power.  Any help would be greatly appreciated.    Also, I am able to simulate multivariate normal data without iml.  I found code to simulate multivariate binomial both with iml and without.  For a multivariate beta distribution I found iml code using copulas.  However, I also need to simulate multivariate count data (both Poisson and negative binomial) but I have not found any code for this and was not sure if copulas could be used to these two distributions.  If anyone knows of some sample code for multivariate Poisson or negative binomial, that would be very helpful.  I am running SAS 9.4 and SAS/IML 15.2.   Thank you.  Deb   data test;    do run=1 to 1000;       do n=2 to 50 by 2;          do trt=1 to 4;             do plot=1 to n;                y=rand("Normal",0,1);                output;             end;          end;       end;    end; run; ... View more

Hello,   I am trying to code a model that is a little different than others I have done and am not sure I have the code right.  My design consists of several sites with each of these sites measured 2 years.  Each year, the sites were measured several dates.  At each survey what was recorded was the date, the presence/absence of factor a, and the response variable.  The presence/absence of factor a can vary with date for each site and year.  Dates also vary with site and year.  So I have 2 repeated measures: year and date - year would be categorical but would like date to be continuous.  I then I have a categorical (or binary) factor measured each date (this is the part I am not sure how to code).  The model I am running is given below but I am not sure this is correct.  I am using compound symmetry while trying to get the rest of the code correct, but can change that later.  So ignoring that part, do I have the model correct or am I missing some other covariance terms. PROC MIXED; CLASS YEAR SITE FACTORA; MODEL S=YEAR|FACTORA|DATE; RANDOM SITE; REPEATED / SUBJECT=SITE*YEAR TYPE=CS;   Also, would it be valid to treat year as a random factor and if yes, how would I code that model?   Thank you for your time and any help you can give me. ... View more

I was running a very simplified logistic regression using Proc Logistic, Proc Genmod, and Proc Glimmix and was surprised to find that my conclusion could change depending on the procedure I used.  My sample data set and code are as follows:   data test;    input y n x @@;    cards; 6 8 1  4 7 2  4 8 3 3 9 4  3 7 5  1 9 6 ; proc logistic data=test;   model y/n=x;   run; proc genmod data=test;   model y/n=x / dist=binomial link=logit type3;   run; proc glimmix data=test;   model y/n=x / dist=binomial link=logit s chisq;   run;   I get the same parameter estimates and standard errors from all three procedures.  However, the p-values are different for Proc Glimmix.  They match between Logistic and Genmod (however, the type3 results in Genmod do not match the test for the paramter estiamte - not sure why).  It seems like almost everything matches between Glimmix and the other two procedures except for the p-values.  I can get the same p-values if I add the Chisq option, but why are the p-values based on the F different?  My conclusions could change depending on which procedure I am using.  Therefore, if I need to use Glimmix because I have random effects, I am really concerned that the results I get will not be correct.  Any insight as to why these are different would be helpful.  (I also tried this with more comples datasets and again Glimmix differed from the other two procedure with respect to the p-values.)  Thank you.   ... View more

Hello,   I am doing a logistic regression with random effects.  I have 300 segments within the study.  At each plot the presence or absence of a particular bird species is recorded along with many continuous covariates.  Because these segments are situated within transects (not necessarily straight transects, sometimes the end point is very close to the starting point) and some transects are very close (I attached a figure showing the location of the segments relative to each other in case it would help - each circle is a segment).  I am trying to account for spatial corrleation among all segments.  My model is as follows:   PROC GLIMMIX DATA=LCSP METHOD=LAPLACE MAXOPT=50 IC=PQ; CLASS SEGMT; MODEL PRESENT_50 = WTRDEP / DIST=BINOMIAL S DDFM=RESIDUAL; RANDOM SEGMT / SUBJECT=INTERCEPT TYPE=SP(SPH)(EASTING NORTHING); RUN;   I am modeling the spatial correlation as a g-side effect and using method=Laplace because I have 34 covariates and am trying to use AIC to compare models.    My problem is that often the standard errors for the covariance parameters or the standard errors for the parameter estimates are missing.  In addition, sometimes the spatial covariance parameter is equal to zero.  The missing standard errors do not seem to be having a huge effect on the results, but I am concerned about the validity of these results.  On the other hand, when the spatial covariance parameter is equal to zero, the log likelihood is greatly affected and therefore so is the AIC value.  The log likelihood basically comes out about half of what it is for models where this covariance parameter is not zero, which in turn makes these models appear to be far superior models.   So obviously I am not doing somethiing right.  I have tried various other options that I thought would help with no luck.  The only thing that helped somewhat is to give starting parameter estimate values.  For about half the models that had a covariance parameter equal to zero, now didn't; but the other half of the models failed.  So I am not sure this is a solution or if I believe the results in this case either.   Not sure where to go from here.  Any help would be appreciated.   Thanks, Deb ... View more