serp: Smooth Effects on Response Penalty for CLM
A regularization method for the cumulative link
models. The smooth-effect-on-response penalty (SERP) provides
flexible modelling of the ordinal model by enabling the smooth
transition from the general cumulative link model to a coarser form of
the same model. In other words, as the tuning parameter goes from zero
to infinity, the subject-specific effects associated with each
variable in the model tend to a unique global effect. The parameter
estimates of the general cumulative model are mostly unidentifiable or
at least only identifiable within a range of the entire parameter
space. Thus, by maximizing a penalized rather than the usual
non-penalized log-likelihood, this and other numerical problems common
with the general model are to a large extent eliminated. Fitting is
via a modified Newton's method. Several standard model performance and
descriptive methods are also available. For more details on the penalty
implemented here, see, Ugba (2021) <doi:10.21105/joss.03705> and
Ugba et al. (2021) <doi:10.3390/stats4030037>.
Documentation:
Downloads:
Reverse dependencies:
Linking:
Please use the canonical form
https://CRAN.R-project.org/package=serp
to link to this page.
[aut, cre, cph]