A colleague of mine and I are trying to reproduce an example (and we are stuck) from the following paper:
Haobo Ren , Xiao-Hua Zhou & Hua Liang (2004): A Flexible Method for Estimating the ROC Curve, Journal of Applied Statistics, 31:7, 773-784
Here the author argues that it is possible to consistently estimate a robust ROC curve using a mixed effect model. At a certain point the author states that the procedure is easy with a S-PLUS function for mixed effect model - although we are using R, whit the lme function. It seems that the author used the random effect without stating a cluster or grouping information. However, we are kind of banging our heads around for a while as it does not make much sense for us not having a "grouping" variable.
The author states that the model expression is something similar to:
Rempi = X*Beta + Z*b + e
where in our example ...
Rempi
Hi Pedro,
the fixed part is a polynomial of order 3 in your case, the random part is some sort of spline. There is no grouping in the data structure, meaning all data come from the same group/population. So to use lme, just generate
grouping
Pedro,
There's a paper on using R to do SDT/ROC analysis for multilevel designs using the lmer function: "Multilevel modeling: Beyond the basic applications" by Wright and London in the British Journal of Mathematical and Statistical Psychology. It proved very helpful to me.
Cheers,
Mike
Pedro,
I have no suggestions, however, I think everyone else have giving you some good advice. Michael thanks for the information about the "multilevel modeling." I have a course in statistics and this will help a lot.
Kudos!
RhuDee
Hi Pedro,
I suggest to use the following code for fitting the model:
Id
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis