sample.naive.posterior: Samples from quadratic posterior densities of models

In rmcelreath/rethinking: Statistical Rethinking book package

Samples from quadratic posterior densities of models

Description

Samples from the posterior density of a fit model or models, assuming multivariate normal density.

Usage

sample.qa.posterior(model, n = 10000, clean.names = TRUE, model.weights =
                 "AICc", nobs = 0, add.names = FALSE, fill.na = 0,
                 verbose = FALSE)

Arguments

model	A fit model object
models	A list of fit models of the same class
n	Number of samples to draw from joint posterior
model.weights	If passing a list of models, method for computing posterior probability of each model family. Can be "AIC","AICc","BIC" or a vector of numeric weights.
nobs	Number of observations used to fit model or all models in list. Sometimes needed for model.weights values, like AICc.
add.names	Adds a column of model names, when passing a list of models
fill.na	Fills missing values with 0, by default, for model families that do not contain a given parameter. Useful for linear models. Hazardous for non-linear ones.
verbose	If TRUE, prints various debugging information

Details

This is a legacy function and is no longer supported nor unit tested.

This function provides a way to draw parameter values from a multivariate normal posterior density, estimated from the maximum a posterieri (MAP) estimates and variance-covariance (vcov) of a fit model or models.

When passing a single fit model object, the function returns a data frame in which each row is a sample and each column is a parameter.

When passing a list of fit model objects, the function returns a data frame containing samples from the joint posterior across model families. The fraction of rows drawn from a specific model family is determined by the model.weights parameter. BIC, AIC, or AICc are used to compute approximate predictive probabilities of each model family, and the total samples n is proportioned according to these estimates. The user can also supply a numeric vector of model weights, computed by any method. This vector should sum to 1.

Author(s)

Richard McElreath

See Also

mvrnorm

rmcelreath/rethinking documentation built on Sept. 18, 2023, 2:01 p.m.