I have a relatively complex logistic model I built using generalized estimating equations (GEE). The model is used to predict population mortality. I have gone about estimating the mortality in the population by (1) fitting a model to my data, (2) making individual-level predictions of mortality from the data, and finally by (3) taking the mean of these predictions. I am particularly interested in examining the shape of the model as I vary one predictor, holding all other variables constant. To this end, I need to calculate a confidence band around the population prediction function from my model. I can calculate population prediction intervals just fine (using the advice given by @Ben Bolker here Parametric Bootstrap without model refitting?), but now I'm trying to figure out how I might be able to bootstrap confidence bands/simultaneous confidence intervals instead of just bootstrap point-wise confidence intervals. Can anyone point me in the right direction?
rms
package'scontrast
andPredict
functions can provide simultaneous confidence intervals on cluster sandwich covariance based estimates, by assuming multivariate normality of $\hat{\beta}$.plot.Predict
makes plots such as what you mentioned easy to do. $\endgroup$rms
will work for GEE models? I have the second edition of your book right here, but I don't see any GEE model examples. $\endgroup$rms
implements GEE with working independence covariance assumptions. Fit the multiple record per subject model using any of therms
fitting functions then run the fit throughrobcov
to get a robust cluster sandwich covariance estimator to take intra-subject correlation into account. Thencontrast
andPredict
can compute simultaneous confidence regions based on assuming multivariate normality of the GEE $\hat{\beta}$. $\endgroup$