Things are a bit different with two continuous variables. I think you now want to test the joint hypothesis that the parameters for age and age*bmi are both 0 (i.e., no effect of age at all on the response). You could do this in two steps, by fitting the model you list, and then with model lestrone = bmi; One could then do a test based on the differences in error sum of squares. As an alternative, you can use the model you list and add the contrast statement model lestrone = age bmi age*bmi; contrast 'joint' age 1, age*bmi 1; run; ... View more

Not clear what you want. One interpretation of your question: You want to know if age has ANY effect, as a main effect or interaction? Some call this a simple main effect. There are other labels. You could use a contrast statement, but the syntax depends on the number of levels of the different factors. Here is a trick to do the same thing. In a second run of the program, take out "age" from the model statement: model lestrone = smoke age*smoke / solution ...; Then the test for age*smoke combines the main and interaction regarding age. Another interpretation of your question is that you want to know the intercepts and slopes for smokers and non-smokers (for a change with age). Then, the model statement could be: model lestrone = smoke age*smoke / solution noint ...; The solution table will directly give you the intercepts and slopes if you make sure you use the "noint" option. Here, the displayed test for smoke is interpreted differently (whether the smoke means are simultaneously 0, not that they are equal to each other). ... View more

I suggest you use the simulate option, because this is based on a multivariate t vector. In its simplest form, just use adjust=simulate; ... View more

Since you are new to sas, you really should read the introductory chapters of the SAS/STAT 9.2 User's Guide (pdf available for download). ... View more

Your predictors are very large relative to the response variable. SAS is rounding the parameter estimates to 0.0000, even though they are not really 0. In a data step, add: major1=major/10000; minor1=minor/10000; Then, use major1 and minor1 in the model statement of genmod. You will get correct nonzero parameter estimates. In reporting or using the results, remember to rescale the parameters by dividing by 10000. ... View more

Also check out PROC PLAN and PROC OPTEX ... View more

Probably you have interaction means with missing cells. You might have to remove some interactions. ... View more

I realize my examples do not have a random effect. But this can be added, by using a random statement and adding a random term to the right-hand side of the equation for m. This would be a random intercept model. ... View more

Here is the code for the beta, with a logit link, and two predictors. x is continuous (0-1) response. Because I set the lower (ep) and upper (lam) to 0 and 1, respectively, I can do the same analysis in GLIMMIX. But with NLMIXED, one can choose other values for ep and lam. proc nlmixed data=two ; title2 'MLE of Beta, using NLMIXED, done in terms of logit link'; title3 'Used epsilon=0 and lambda=1. This is a regression-type problem'; parms a=-2.1 b1=.011 b2 = .1 phi = 2 to 60 by 2; ep=0.0; lam=1.0 ; bounds phi > 0; m = a + b1*x1 + b2*x2; *-logit link scale; mu=ep + lam*(1/(1+exp(-m))); *-inverse link; lbe=lgamma(((mu-ep)/lam)*phi) + lgamma(((lam+ep-mu)/lam)*phi) - lgamma(phi); lbed1 = (((mu-ep)/lam)*phi -1) * log(x-ep) ; lbed2 = (((lam+ep-mu)/lam)*phi -1) * log(lam+ep-x); lbed3= (phi-1)*log(lam) + lbe; loglik=lbed1 + lbed2 - lbed3; model x~general(loglik); predict m out=outp; run; proc glimmix data=two; title2 'GLMM fitting of beta to Griffiths data (GLIMMIX), logits'; model x = x1 x2 / dist=beta link=logit solution ; run; ... View more

I keep trying to paste a longer example of the beta and NLMIXED, but only the first third or so shows up on the preview (or posting). Not sure why I can't paste the entire example. I want to show the code for a logit link and two predictor variables, with NLMIXED and GLIMMIX. ... View more

My code got cut off. I will try again. proc nlmixed data=sbdata ; title2 'MLE of beta (NLMIXED), MU version, epsilon=0, lambda=1 '; parms mu = .3, phi = 15; ep=0; lam=1 ; *<--upper and lower bounds; bounds mu > 0; bounds phi > 0; be=gamma(((mu-ep)/lam)*phi)*gamma(((lam+ep-mu)/lam)*phi)/gamma(phi); be_dens=( ((x-ep)**(((mu-ep)/lam)*phi -1)) * ((lam+ep-x)**(((lam+ep-mu)/lam)*phi -1) )) / (be*lam**(phi-1)); loglik=log(be_dens); model x~general(loglik); estimate 'alpha' (mu-ep)*phi/lam; estimate 'beta'(lam+ep-mu)*phi/lam; estimate 'logit' log( (mu-ep)/(lam+ep - mu) ); *--Estimated logit assuming general ep & lam; estimate 'variance' lam*(mu-ep)*(lam+ep-mu)/(lam*(1+phi)); estimate 'median' betainv(0.5,(mu-ep)*phi/lam, (lam+ep-mu)*phi/lam)*lam + ep; run; *--Use of GLIMMIX to fit. To have epsilon=0 and lambda=1, must use x variable (x); proc glimmix data=sbdata; title2 'Beta, using pseudo-likelihood(GLIMMIX), epsilon=0 & lambda=1, IDentity link'; model x = / dist=beta link=id solution; run; ... View more

I don't have a macro, but here is some code for the beta. The first program works with the mu/phi parameterization of the beta. It assumes the proportion data is in a variable x. The parameter estimates of mu and phi (if ep=0 and lam=1) are equivalent to the intercept and scale parameters obtained from GLIMMIX. MY CODE GOT MESSED UP WHEN PASTING. DON'T USE THIS CODE. SEE MY SUBSEQUENT MESSAGES... proc nlmixed data=sbdata ; title2 'MLE of beta (NLMIXED), MU version, epsilon=0, lambda=1 '; parms mu = .3, phi = 15; ep=0; lam=1 ; *<--ep is the lower bound parameter, and lam is upper bound param.; bounds mu > 0; bounds phi > 0; be=gamma(((mu-ep)/lam)*phi)*gamma(((lam+ep-mu)/lam)*phi)/gamma(phi); be_dens=( ((x-ep)**(((mu-ep)/lam)*phi -1)) * ((lam+ep-x)**(((lam+ep-mu)/lam)*phi -1) )) / (be*lam**(phi-1)); loglik=log(be_dens); model x~general(loglik); estimate 'alpha' (mu-ep)*phi/lam; *<--to get the conventional alpha/beta parameters; estimate 'beta'(lam+ep-mu)*phi/lam; estimate 'logit' log( (mu-ep)/(lam+ep - mu) ); *--Estimated logit assuming general ep & lam; estimate 'variance' lam*(mu-ep)*(lam+ep-mu)/(lam*(1+phi)); estimate 'median' betainv(0.5,(mu-ep)*phi/lam, (lam+ep-mu)*phi/lam)*lam + ep; run; *--Use of GLIMMIX to fit. To have epsilon=0 and lambda=1, must use x variable (x); proc glimmix data=sbdata; title2 'Beta, using pseudo-likelihood(GLIMMIX), epsilon=0 & lambda=1, IDentity link'; model x = / dist=beta link=id solution; run; Now for a logit link and two predictor variables (x1 and x2). proc nlmixed data=two ; title2 'MLE of Beta, using NLMIXED, done in terms of logit link'; title3 'Used epsilon=0 and lambda=1. This is a regression-type problem'; parms a=-2.1 b1=.011 b2 = .1 phi = 2 to 60 by 2; ep=0.0; lam=1.0 ; *<--For these data, use epsilon=0 and lambda=1; bounds phi > 0; m = a + b1*x1 + b2*x2; *<--This is logit; mu=ep + lam*(1/(1+exp(-m))); *<--Inverse link to get expected (mu); lbe=lgamma(((mu-ep)/lam)*phi) + lgamma(((lam+ep-mu)/lam)*phi) - lgamma(phi); lbed1 = (((mu-ep)/lam)*phi -1) * log(x-ep) ; lbed2 = (((lam+ep-mu)/lam)*phi -1) * log(lam+ep-x); lbed3= (phi-1)*log(lam) + lbe; *be_dens= ( ((x-ep)**(((mu-ep)/lam)*phi -1)) * ((lam+ep-x)**(((lam+ep-mu)/lam)*phi -1) )) / (be*lam**(phi-1)); loglik=lbed1 + lbed2 - lbed3; model x~general(loglik); predict m out=outp; *<--Put logits in a file; run; proc glimmix data=two; title2 'GLMM fitting of beta to Griffiths data (GLIMMIX), logits'; model x = x1 x2 / dist=beta link=logit solution ; *<--Note the logit link here; run; MY CODE GOT MESSED UP WHEN PASTED. SEE MY SUBSEQUENT MESSAGES FOR PROPER CODE. Message was edited by: lvm Message was edited by: lvm Message was edited by: lvm ... View more

The references I mentioned deal with the subject. ... View more

I am sorry, but your questions are much too general to be addressed here. You are basically asking for an entire course in empirical modeling. I suggest you find a consulting statistician who can work with you on this problem. ... View more

You obviously have not learned how to use SAS yet. You actually specified the data file as AA, so I am guessing that this is just followed from an example. NLP is one of the most sophisticated procedures in SAS, requiring a knowledge of SAS and programming, as well as numerical optimization methods. You need to learn how to use sas data steps and procedures first. ... View more