I have the following geeglm model where X is a 3-level categorical variable (ref: X=1) and Z is a continuous variable:
fit1<- geeglm(formula=Y~ X + Z + X*Z,
id=myID,
data=dat2,
family=gaussian(link = "identity"),
corstr="independence",
waves=moments)
And output:
summary(fit1)
Coefficients:
Estimate Std.err Wald Pr(>|W|)
(Intercept) 1.7051 0.0219 6061.67 <2e-16 ***
X2 -0.0414 0.0251 2.73 0.098 .
X3 -0.0401 0.0293 1.88 0.171
Z -0.0207 0.0166 1.56 0.212
X2:Z 0.0371 0.0188 3.91 0.048 *
X3:Z 0.0304 0.0204 2.23 0.135
I want to estimate the effect of Z on Y in different levels of X, so I add the beta coefficients like this:
Change in Y associated with 1-unit increase in Z when X=1: -0.0207
Change in Y associated with 1-unit increase in Z when X=2: -0.0207 + 0.0371
Change in Y associated with 1-unit increase in Z when X=3: -0.0207 + 0.0304
How do I then get confidence intervals around these estimates? I have reviewed a similar answer here: https://stats.stackexchange.com/a/3657
It looks like I should extract the standard errors and use them in the referenced equation to get a new standard error estimate, but how/where would I find covariance in my geeglm summary to plug into that equation? Is there an easier way to obtain confidence intervals than by hand?
