NEWS.md
In msm: Multi-State Markov and Hidden Markov Models in Continuous Time

User-visible changes only. For internal changes, see Github commits.

Version 1.8.1 (2024-10-04)

Subject-level weights supported for likelihood calculation with new msm() argument subject.weights.
Tidying methods added for msm objects and most of msm's output functions (prevalence.msm, qmatrix.msm and related functions, and totlos.msm and related functions). These methods convert the output to a tidy data frame (tibble), in the manner of the broom package.

To use these methods, just call tidy(x), where x is the result of calling, e.g. msm, prevalence.msm, or qmatrix.msm.

Hence the msm package now imports the generics and tibble packages.
Subjects with one observation are no longer dropped in HMMs, since they provide information about the distribution of the outcome given the hidden state.
ppass.msm now supports pci models and other time-inhomogeneous models. Thanks to Jon Fintzi for working on this.
New function hmodel2list to extract HMM constructor function calls from fitted HMMs. Thanks to Will Hulme for working on this.
Objects returned by totlos.msm, efpt.msm and envisits.msm now have class "msm.estbystate".
Objects returned by prevalence.msm now have class "msm.prevalence".
Fix of bugs for models containing a covariate named "baseline" or "Baseline".

Fix of a bug in the Viterbi algorithm for the calculation of the fitted state at the initial time for each subject.
Bug fix for pmatrix.piecewise.msm given just intensity matrices instead of a fitted model.
Auto-generated initial values set to a small positive number rather than zero when there are no observed data for a particular permitted transition. This fixes consistency checks (e.g. for qconstraint) in this situation.
Fix when covariates come into the data as one-column matrices instead of vectors.
Modernised to use roxygen, and pkgdown website created at https://chjackson.github.io/msm
Some internal functions (e.g. deriv.msm) that clash with base S3 generics renamed

rpexp is now more efficient. Thanks to Mark Clements. Note that the values simulated by rpexp, sim.msm and simmulti.msm for a specific seed will now be different in models where the intensities are piecewise constant.
Bug fix: in HMMs with partially known initial states specified through obstrue and censor, initprobs is now accounted for.
Bug fix for bootstrapping with character subject IDs.
crudeinits.msm handled when state is a factor.
Fix for hranges on multiple parameters.
obstrue handled in forward-backward algorithm (viterbi.msm).