TODO

When writing out documentation (Rd files), roxygen2 issues warnings that @s3method has been deprecated and that @export should be used instead

Note that some objects can instead be specified as TRUE as opposed to 1, e.g. adaption = TRUE instead of adaption = 1, although this isn't indicated in documentation (Rd files).

Bug when using generic update function:
Non-standard requirement that the intercept is explicitly specified might cause some confusion:
> a <- formula(normexam ~ 1 + standlrt + (1|student))
> a
normexam ~ 1 + standlrt + (1 | student)
> a <- update(a, .~. + sex)
> a
normexam ~ standlrt + (1 | student) + sex
...add note to help pages (or possible to resolve via changes to R2MLwiN?) Note on CMM R2MLwiN webpage added.

Model Specification Checks:
(IGLS) Check lower levels are non-equal standardised weights if standardised weights are specified
(IGLS) Check that an error is returned if multiple membership is specified

Additional Functionality:
Automatic calculation of VPC?
Automatic posterior predictive checks
Debugging option to extract internal model matrices (IGLS) - XSS/VMAT/ZMAT/YRES/YMAT/XMAT commands
Allow user specified design matrix (IGLS) - SETD command
Multinomial proportions - need manual calculation/reshaping - e.g. <denom> in multinomial is actually not used (or doesn't need to be used as always 1), therefore this could instead be counts per unit e.g. log(resp, counts)... D="Unordered Multinomial"
Add an option to allow the model to be set up but not run, useful for example if we just want to run a BUGS model or see the MLwiN syntax
Add more error checking
Allow display of models results in alternative measures (e.g. correlations, standard deviations, etc)
Investigate interactions with RStudio, as sometimes graphs can behave strangely
Add option to omit data from output objects, as well as modify postestimation functions to take this into account
Add option to omit chains, or allow their removal, to save on storage space needed for model objects
Store and display information regarding the Priors in the MCMC output
Model output can be too wide when using it in a document, investigate whether this can be reduced

Documentation/Demos:
Add demos for examples in "Leyland and Groenewegen (2020)"

Correctly handle predictor variables contrasts, especially for "ordered" factor variables.
Note that examples in data-fysio.R, data-bang.R and data-bang1.R (all of which use ordered factors as predictors) will need to be updated when allowcontrast = TRUE becomes the default in future releases, as well as the following demos: MCMCGuide10.R, MCMCGuide12.R, MCMCGuide23.R, MCMCGuide25.R, UserGuide09.R, UserGuide10.R.