Hi Steve, Thanks for your response. I didn’t want to reply in haste; I thought I should take some time to read more and figure out more about SAS and GLIMMIX. In my original post I was kind of hoping that my GLIMMIX code will be torn apart because I am very new to this and don’t know all the options to be confident enough in the options I have chosen. To simplify my case, I have made my DV binomial, so instead of a 0,1, or 2 score, it is a ‘score ‘ (2 and 1 =1) or no ‘score’ (0). This will enable me to use the LSMEANS and SPLICEDIFF, which I have done successfully. However, there are a few things I am really confused about and hoping you can help clear up. I was hoping to use the Type 3 fixed effects as a global test of my hypothesis and LSMEANS comparisons to test the specific relationships between my variables. From what I have gathered in the discussion forum (albeit in a study design different from mine), it is best to use the Solutions for Fixed Effects. Will this apply in my case? With my new code (I have to include the other IVs in the class statement to be able to get LSMEANS for them), the Solutions for Fixed Effects is giving me a more detailed output at the variable level. In my training (far from SAS/GLIMMIX), for my study design you first look for a global effect, only on the condition that there is a significant global effect do you look for specific effects. Is the Type 3 Test this global effect and Solutions for Fixed Effects the specific effects? And the LSMEANS is used to further test other specific comparisons? In the Solution for Fixed Effects, I am only getting results for variable level -1, I had thought this was a problem but apparently this is standard? If I do use the Solutions for Fixed Effect I am really confused on how to interpret it . For instance, SES in my data has four levels, so I have results for levels 1 to 3, what do I say about level 4? Are the figures on the row of SES level 3 comparisons between 3 and 4, and so forth? Finally, how important is it that I should report a table of my model building process? I ask this question because with the addition of all my IVs in the class statement, I no longer have neat single variable estimate and SE for each IV in the Solutions for Fixed Effects table, but have estimates for all the levels. I wanted this table to have the values for the global effects so I can write about the specific effects, given that, at least in APA standard, you either present in a table or write about it, but not both. ` I apologize for all this questions, I have tried figuring it out myself but with the time constraint and pressure, the more I search the more confused I get. When this is over I will start learning about SAS from the fundamentals. I have included my final code below. Proc glimmix data=work.ses2 method=laplace noclprint; Class gender ses question; Model bi_res=gender ses question gender*question/ dist=binomial link=logit solution cl oddsratio(diff=first label); Random intercept / type=vc solution cl; Covtest / wald; lsmeans gender*question /slice=question slicediff=question adjust=bon; ods select slicediffs; run; The above lsmeans is to investigate the difference between questions according to gender and the lsmean below is to investigate the difference between ses according to gender. There are ten different tests, therefore I will be using a p value of 0.05/10 to determine statistical significance. Proc glimmix data=work.ses2 method=laplace noclprint; Class gender ses question; Model bi_res=gender ses question gender*ses/ dist=binomial link=logit solution cl oddsratio(diff=first label); Random intercept / type=vc solution cl; Covtest / wald; lsmeans gender*ses /slice=ses slicediff=ses adjust=bon; ods select slicediffs; run; Thanks in advance. Jane greenfield.
... View more
