Marginally interpretable transformation models for clustered data. Highly experimental, use at your own risk.
mtram(object, formula, data, standardise = FALSE,
grd = SparseGrid::createSparseGrid(type = "KPU",
dimension = length(rt$cnms[[1]]), k = 10),
Hessian = FALSE, ...)A tram object.
A formula specifying the random effects.
A data frame.
Two types of models can be estimated: M1 (with standardise = FALSE)
corresponds to a marginal distribution function without direct
interpretation of the fixed effects, M2 (with standardise = TRUE)
allows a marginal interpretation of scaled fixed effects as
log-odds or log-hazard ratios (depending on object). See Hothorn
(2019).
A sparse grid used for numerical integration to get the likelihood.
A logical, if TRUE, the hessian is computed and returned.
Additional argument.
An object of class tram with coef() and logLik()
methods.
A Gaussian copula with a correlation structure obtained from a random
intercept or random intercept / random slope model (that is, clustered or
longitudinal data can by modelled only) is used to capture the
correlations whereas the marginal distributions are described by a
transformation model. The methodology is described in Hothorn (2019)
and examples are given in the mtram package vignette.
This is a proof-of-concept implementation and still highly experimental.
Only coef() and logLik() methods are available at the
moment.
Torsten Hothorn (2019). Marginally Interpretable Parametric Linear Transformation Models for Clustered Observations. Technical Report.