easystats support for bmmfit objects

These S3 methods let the insight package, and through it the easystats packages parameters, bayestestR and modelbased, read a bmm fit as the model you specified with bmm(): its parameters, their formulas, the response variables and the data, rather than the brms model that bmm builds from them. With them, for example, parameters::model_parameters() and bayestestR::describe_posterior() summarise each model parameter in its own component, and for models such as sdm(), modelbased::estimate_means(fit, by = "set_size", predict = "c") returns the estimates of a model parameter on its native scale.

Usage

# S3 method for class 'bmmfit'
find_formula(x, verbose = TRUE, ...)

# S3 method for class 'bmmfit'
find_predictors(
  x,
  effects = "fixed",
  component = "all",
  flatten = FALSE,
  verbose = TRUE,
  ...
)

# S3 method for class 'bmmfit'
find_response(x, combine = TRUE, ...)

# S3 method for class 'bmmfit'
get_data(
  x,
  effects = "all",
  component = "all",
  source = "frame",
  verbose = FALSE,
  ...
)

# S3 method for class 'bmmfit'
find_parameters(
  x,
  effects = "all",
  component = "all",
  flatten = FALSE,
  parameters = NULL,
  ...
)

# S3 method for class 'bmmfit'
get_priors(x, verbose = TRUE, ...)

# S3 method for class 'bmmfit'
model_info(x, response = NULL, ...)

# S3 method for class 'bmmfit'
link_function(x, ...)

# S3 method for class 'bmmfit'
link_inverse(x, ...)

Arguments

x: A bmmfit object (created by bmm()).
...: Ignored, for consistency with the generics.
effects: "fixed" for the population-level part of the model, "random" for the group-level part or "all" for both: coefficients for find_parameters(), predictors for find_predictors(), columns for get_data(). find_parameters() also takes "full", "random_variance" and "grouplevel", as the methods of insight for brms fits do.
component: "all", "conditional", "location", "distributional" or "auxiliary", which all select every model parameter, or for find_parameters() the name of a model parameter; ignored by get_data().
flatten: Should a character vector be returned instead of a list?
combine, source, verbose: Ignored, for consistency with the generics.
parameters: A regular expression the names of the returned parameters must match.
response: Ignored, for consistency with the generic.

Value

See the insight generics, e.g. insight::find_formula() and insight::find_parameters().

Details

Each method reshapes what the fit already stores:

find_formula() returns one formula per model parameter, and per helper parameter of a non-linear formula, with the population-level and group-level parts as separate elements (<parameter> and <parameter>_random), plus a conditional formula of the response variables on all population-level predictors.
find_predictors() returns the predictors of these formulas: the population-level predictors for effects = "fixed", the grouping factors for "random" and both for "all". Every model parameter counts as both a location and a distributional parameter, so each component returns the same variables.
find_response() returns the names of the response columns, as get_data() returns them.
get_data() returns the data the model was fitted to as you passed them: without the columns bmm adds for its own use or brms computes from the formula (such as ID:session or log(x)), and with the response categories of a model fitted to counts (e.g. m3()) as columns of their own. Data with one row per trial come back with the response columns you named, such as the rating column of sdt_rating() or the judgment and confidence columns of sdt_cdp(). The model frame stored in the fit is always the source; source is ignored.
find_parameters() returns the names of the posterior draws in one element per model parameter (<parameter> for the population-level coefficients, <parameter>_random for the group-level ones), without the parameters that the brms family needs but the model does not have, such as the fixed mu of the response-time models. component selects a model parameter by name; "all", "conditional", "location", "distributional" and "auxiliary" select all of them. insight::get_parameters() and insight::clean_parameters() build on it.
get_priors() returns one row per population-level coefficient, group-level standard deviation and correlation, with the prior that applies to it. A parameter fixed to a constant has the distribution "constant" and the constant as its location.
model_info() describes the model as not linear and as having no dispersion parameter; the family is the one of the brms model.
link_function() and link_inverse() return the link function of the first parameter that has one and its inverse. bmm models have one link per parameter, so for anything else use model$links.

Examples

if (FALSE) { # \dontrun{
insight::find_formula(fit)
parameters::model_parameters(fit)
bayestestR::describe_posterior(fit)
modelbased::estimate_means(fit, by = "set_size", predict = "c")
} # }

Usage

Arguments

Value

Details

See also

Examples