Changelog • brmsmargins

brmsmargins 0.3.0

Now supports response variables with a binomial distribution. Predictions that marginalise over random effects will use the number of trials. For contrasts, such as using brmsmargins() you can use the at = argument to force trials to 1 to get marginal probabilities rather than counts of successful trials.
Experimental support for integrating out random effects from a student t distribution. Currently only supports this for location parameters. When dpar is not empty in brms only Gaussian distributions are supported. Some caution is needed here as brms uses the same degree of freedom parameter for all random effects of the same grouping variable, so for complex random effect structures, there may be errors in how they are integrated out.

CRAN release: 2025-09-26

CRAN release: 2022-05-18

Fixed a bug when using prediction() with option effects = "integrateoutRE" when smooth terms were present. As prediction() underpins other functions, such as brmsmargins() this issue also impacts those other functions.
New function: marginalcoef() which calculates population averaged (marginal) coefficients for the fixed effects coefficients from mixed effects models using a method described by Donald Hedeker, who joins the author team. Currently, only the main location parameter is supported. That is, marginal coefficients for the scale part of a model, in location and scale models, is not currently supported.
New argument, wat, added to brmsmargins() to support including calculating average marginal effects for multilevel centered categorical predictors.
Updates to vignettes demonstrating: (1) the use of marginal coefficients;
1. marginal effects for centered categorical predictors; and
2. ‘simple’ marginal effects when models include interaction terms.
Revised documentation for bmrsmargins() and prediction() to be clearer around which arguments users must directly specify and which are optional or have sensible defaults.
Added more unit testing and vignettes.

CRAN release: 2021-12-16

Fixed a bug preventing predictions integrating out random effects for mixed effects models with a random intercept only (reported in Issue#1). Thanks to @ajnafa for reporting.
Added support for Gamma and Beta regression models.
More extensive testing added.