PIMEX (Prediction and Identification of Mixed Effects for eXtreme clusters) works with clustered data when the focus is on identifying extreme clusters or predicting the value of cluster effects for those clusters. It uses a new class of predictors that put more weight on extreme values, and so perform better than conventional mixed models for the extremes. It can use the new predictors to identify extreme clusters, achieving better flagging rates than conventional approaches while controlling the false positive rate. And it can automatically select optimal parameters for the weighting function.
Currently, only linear mixed models are supported, and only partly.
We recommend using R >= 4.2.2 for best rendering of math in HTML
help; otherwise you may get the plain text version.
The simulation code in this package depends on a new argument to
lme4::simulate() introduced in version 1.1-37, now available on CRAN.
You can install the development version of pimex from
GitHub with:
# install.packages("devtools")
devtools::install_github("ucsf-deb/pimex")Consult Notes.md.
This package incorporates code from Simen Gaure at https://stackoverflow.com/a/50935329/4409451
We used a formula from Sasha
for msep1_num(), also derived in
msep_num.pdf.