@Reeza wrote:
When you use PROC GLM it outputs the design matrix. Did you compare this to how you dummy coded the variables?
There are dozens of equivalent ways to code dummy variables, so comparison might not help.
... View more
proc glm data= dummy;
class sex;
model loglac= mddtotal age sex;
run;
I'm not 100% sure I know what you mean by "stratify" in the context of GLM modeling, but you can add SEX into the CLASS statement; you can also include interactions of SEX with AGE and MDDTOTAL, this effectively computes different intercepts and different slopes for AGE and MDDTOTAL depending on the value of SEX.
... View more
@iressa1313 wrote:
1. I am performing a multi-linear regression analysis with 1 dichotomous and 1 continuous regressor and there is a significant interaction.
2. How can I assess coincidence from the PROC REG procedure in SAS?
3. I have read that ANCOVA is the appropriate method but what if I am only given the ANOVA and parameter estimates output?
4. Here is my current thought process: if I fit the model for each value of the dichotomous variable then the lines will differ and hence, are not coincident. Am I thinking about this correctly?
I think there are some issues that need to be clarified:
1. If you have a dichotomous regressor, you want to use PROC GLM with a CLASS statement, not PROC REG.
2. When you fit a model with PROC GLM, you can test whether the interaction term is significant. You can also test whether the coefficient for the dichotomous parameter is significant.
3. I am confused by this question. You say "what if I am only given the ANOVA and parameter estimates output." Do you have the data or not? If you only have the parameter estimates (as a table), then please post it.
4. I would fit a model that includes a CLASS variable (the dichotomous regressor) rather than fit two models, especially if you think there is interaction. To understand the difference between using a CLASS statement and fitting two different models, study the article "The difference between CLASS statements and BY statements in SAS"
... View more
Hi @iressa1313,
I would keep it simple and create a new dataset with all four variables:
data want;
call streaminit(27182818);
do _n_=1 to 20;
x=rand('norm'); /* X ~ N(0,1) */
y=rand('norm',5); /* Y ~ N(5,1) */
u=rand('bern',0.5); /* U ~ Bin(1,0.5) */
q=ifn(u,y,x); /* normal mixture */
output;
end;
run;
Or is there something special about your existing "distribution datasets" that you don't want to discard them?
... View more
yes, the group sizes between case and controls were very different. For example within one strata the case group had 25 and the control had 206. But this was also the case for the matched pair design with that a case group having 50 and the control 203.
... View more