brms.mmrm 1.1.2

• Center nuisance variables which interact baseline with time and/or the subgroup within the cell of the data they apply to, rather than across the whole dataset. Centering these columns globally broke the relationship between the fixed effects of interest and the marginal means (#139, @michael-lamm).
• brm_transform_marginal() now prints an informative message (once per session) if a nuisance column of an informative prior archetype was recentered by means other than brm_recenter_nuisance(). Recentering shifts the reference level of the model, and it is the user’s responsibility to interpret the reference level and the resulting marginal means appropriately (#139, @michael-lamm).
• Add vignettes/map.Rmd, a new vignette on meta-analytic predictive priors for Bayesian MMRMs.
• Update and rerun the SBC pipeline, using the modern crew.cluster interface, as well as CSV outputs instead of fst to reduce dependencies.

brms.mmrm 1.1.1

• Use FEV data in usage vignette.
• Show how to visualize prior vs posterior in the usage vignette.
• Add a center argument to brms_formula.default() and explain intercept parameter interpretation concerns (#128).

brms.mmrm 1.1.0

• Add brm_marginal_grid().
• Show posterior samples of sigma in brm_marginal_draws() and brm_marginal_summaries().
• Allow outcome = "response" with reference_time = NULL. Sometimes raw response is analyzed but the data has no baseline time point.
• Preserve factors in brm_data() and encourage ordered factors for the time variable (#113).
• Add brm_data_chronologize() to ensure the correctness of the time variable.
• Do not drop columns in brm_data(). This helps brm_data_chronologize() operate correctly after calls to brm_data().
• Add new elements brms.mmrm_data and brms.mmrm_formula to the brms fitted model object returned by brm_model().
• Take defaults data and formula from the above in brm_marginal_draws().
• Set the default value of effect_size to attr(formula, "brm_allow_effect_size").
• Remove defaults from some arguments to brm_data() and document examples.
• Deprecate the role argument of brm_data() in favor of reference_time (#119).
• Add a new model_missing_outcomes in brm_formula() to optionally impute missing values during model fitting as described at https://paulbuerkner.com/brms/articles/brms_missings.html (#121).
• Add a new imputed argument to accept a mice multiply imputed dataset (“mids”) in brm_model() (#121).
• Add a summary() method for brm_transform_marginal() objects.
• Do not recheck the rank of the formula in brm_transform_marginal().
• Support constrained longitudinal data analysis (cLDA) for informative prior archetypes brm_archetype_cells(), brm_archetype_effects(), brm_archetype_successive_cells(), and brm_archetype_successive_effects() (#125). We cannot support cLDA for brm_archetype_average_cells() or brm_archetype_average_effects() because then some parameters would no longer be averages of others.

brms.mmrm 1.0.1

• Handle outcome NAs in get_draws_sigma().
• Improve summary() messages for informative prior archetypes.
• Rewrite the archetypes.Rmd vignette using the FEV dataset from the mmrm package.
• Add brm_prior_template().

brms.mmrm 1.0.0

New features

• Add informative prior archetypes (#96, #101).
• Add [brm_formula_sigma()] to allow more flexibility for modeling standard deviations as distributional parameters (#102). Due to the complexities of computing marginal means of standard deviations in rare scenarios, [brm_marginal_draws()] does not return effect size if [brm_formula_sigma()] uses baseline or covariates.

Guardrails to ensure the appropriateness of marginal mean estimation

• Require a new formula argument in brm_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 by brms_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 by brms_model(), and the attributes store the "brms_mmrm_formula" object in a way that brms itself cannot modify.
• Deprecate use_subgroup in brm_marginal_draws(). The subgroup is now always part of the reference grid when declared in brm_data(). To marginalize over subgroup, declare it in covariates 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 of emmeans in brm_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 to model.Rmd since inference.Rmd also discusses methods.

Other improvements

• Extend brm_formula() and brm_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 the correlation argument in favor of individual correlation-specific arguments such as unstructured and compound_symmetry.
• Ensure model matrices are full rank (#99).

brms.mmrm 0.1.0

• Deprecate brm_simulate() in favor of brm_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 by brms anyway, and by handling by handling this in brms.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(), deprecate level_control in favor of reference_group.
• In brm_data(), deprecate level_baseline in favor of reference_time.
• In brm_formula(), deprecate arguments effect_baseline, effect_group, effect_time, interaction_baseline, and interaction_group in favor of baseline, group, time, baseline_time, and group_time, respectively.
• Propagate values in the missing column in brm_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() and brm_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).

brms.mmrm 0.0.2

• Fix grammatical issues in the description.

brms.mmrm 0.0.1

• First version.