Changelog
Source:NEWS.md
brms.mmrm 0.1.0.9004 (development)
Guardrails to ensure the appropriateness of marginal mean estimation
- Require a new
formula
argument inbrm_marginal_draws()
. - Change class name
"brm_data"
to"brms_mmrm_data"
to align with other class names. - Create a special
"brms_mmrm_formula"
class to wrap around the model formula. The class ensures that formulas passed to the model were created bybrms_formula()
, and the attributes store the user’s choice of fixed effects. - Create a special
"brms_mmrm_model"
class for fitted model objects. The class ensures that fitted models were created bybrms_model()
, and the attributes store the"brms_mmrm_formula"
object in a way thatbrms
itself cannot modify. - Deprecate
use_subgroup
inbrm_marginal_draws()
. The subgroup is now always part of the reference grid when declared inbrm_data()
. To marginalize over subgroup, declare it incovariates
instead. - Prevent overplotting multiple subgroups in
brm_plot_compare()
. - Update the subgroup vignette to reflect all the changes above.
Custom estimation of marginal means
- Implement a new
brm_transform_marginal()
to transform model parameters to marginal means (#53). - Use
brm_transform_marginal()
instead ofemmeans
inbrm_marginal_draws()
to derive posterior draws of marginal means based on posterior draws of model parameters (#53). - Explain the custom marginal mean calculation in a new
inference.Rmd
vignette. - Rename
methods.Rmd
tomodel.Rmd
sinceinference.Rmd
also discusses methods.
Other improvements
- Extend
brm_formula()
andbrm_marginal_draws()
to optionally model homogeneous variances, as well as ARMA, AR, MA, and compound symmetry correlation structures. - Restrict
brm_model()
to continuous families with identity links. - In
brm_prior_simple()
, deprecate thecorrelation
argument in favor of individual correlation-specific arguments such asunstructured
andcompound_symmetry
.
brms.mmrm 0.1.0
CRAN release: 2024-02-15
- Deprecate
brm_simulate()
in favor ofbrm_simulate_simple()
(#3). The latter has a more specific name to disambiguate it from other simulation functions, and its parameterization conforms to the one in the methods vignette. - Add new functions for nuanced simulations:
brm_simulate_outline()
,brm_simulate_continuous()
,brm_simulate_categorical()
(#3). - In
brm_model()
, remove rows with missing responses. These rows are automatically removed bybrms
anyway, and by handling by handling this inbrms.mmrm
, we avoid a warning. - Add subgroup analysis functionality and validate the subgroup model with simulation-based calibration (#18).
- Zero-pad numeric indexes in simulated data so the levels sort as expected.
- In
brm_data()
, deprecatelevel_control
in favor ofreference_group
. - In
brm_data()
, deprecatelevel_baseline
in favor ofreference_time
. - In
brm_formula()
, deprecate argumentseffect_baseline
,effect_group
,effect_time
,interaction_baseline
, andinteraction_group
in favor ofbaseline
,group
,time
,baseline_time
, andgroup_time
, respectively. - Propagate values in the
missing
column inbrm_data_change()
such that a value in the change from baseline is labeled missing if either the baseline response is missing or the post-baseline response is missing. - Change the names in the output of
brm_marginal_draws()
to be more internally consistent and fit better with the addition of subgroup-specific marginals (#18). - Allow
brm_plot_compare()
andbrm_plot_draws()
to select the x axis variable and faceting variables. - Allow
brm_plot_compare()
to choose the primary comparison of interest (source of the data, discrete time, treatment group, or subgroup level).