Predictions of a model parameter for easystats and marginaleffects

With these methods, insight and marginaleffects, and through them modelbased::estimate_means(), modelbased::estimate_contrasts(), modelbased::estimate_relation() and the see plots of their results, predict a model parameter on its native scale, as native_parameters() computes it: with the inverse link applied draw by draw and the mixture weights softmaxed jointly, so that predict = "thetat" returns the probability of a target response.

Usage

# S3 method for class 'bmmfit'
get_predicted(
  x,
  data = NULL,
  predict = NULL,
  iterations = NULL,
  ci = NULL,
  ci_method = NULL,
  include_random = "default",
  ...
)

# S3 method for class 'bmmfit'
get_predict(model, newdata, type = "response", ...)

Arguments

x, model: A bmmfit object (created by bmm()).
data, newdata: The data to predict for. get_predicted() also takes newdata in ..., and predicts for the data of the fit if both are NULL.
predict: The name of a model parameter, or one of the prediction types of insight::get_predicted(). A type in ... replaces it.
iterations: The number of posterior draws to predict from, all of them if NULL.
ci: The probability mass of the credible intervals.
ci_method: "quantile" (the default), "eti" or "hdi".
include_random: FALSE or NA for population-level predictions, a formula for the random effects to include, or "default" or TRUE for all of them if the data contain the grouping factors.
...: For a model parameter: draw_ids, allow_new_levels and sample_new_levels, passed to native_parameters(); for get_predicted() also newdata, type and centrality_function, and for get_predict() dpar, ndraws and re_formula (or re.form). For any other prediction, passed to the methods for brms fits.
type: The prediction type of marginaleffects, used only when dpar is not a model parameter.

Value

get_predicted(): a get_predicted object, as described in insight::get_predicted(). get_predict(): a data.table of the estimates with the draws in the posterior_draws attribute, as marginaleffects expects.

Details

predict (or dpar for get_predict(), which is how modelbased passes it on) names the model parameter. Without it, get_predicted() predicts the first parameter the model estimates and says which. The prediction types of insight ("expectation", "link", "prediction", ...) and of marginaleffects are left to the methods for brms fits; the expected response is not defined for most bmm models.

Random effects enter the predictions only if every grouping factor is a column of the data to predict for and holds only levels of the fit, as insight does for brms fits, or new levels with allow_new_levels = TRUE. Otherwise the predictions are for the population level, with a warning if you asked for include_random = TRUE. A grouping factor that is NA, as in the grids of modelbased, always means the population level. For a mixed model, modelbased::estimate_means() averages the predictions over the participants in its grid, so its means differ from those of native_parameters(re_formula = NA).

The point prediction of get_predicted() is the centrality_function of the draws, base::mean by default, as for brms fits.

Examples

if (FALSE) { # \dontrun{
insight::get_predicted(fit, predict = "c")
modelbased::estimate_means(fit, by = "set_size", predict = "c")
modelbased::estimate_contrasts(fit, contrast = "set_size", predict = "c")
} # }