getIC: Extract information criterion or other info from a fitted...

View source: R/MPsearchfunctions.R

getICR Documentation

Extract information criterion or other info from a fitted model

Description

Extract information criterion or other info from a fitted model

Usage

getIC(
  x,
  type = c("aic", "bic", "sic", "ll", "np"),
  N,
  usefitfunc = FALSE,
  infotype = "oakes1999"
)

Arguments

x

Fitted mxModel, e.g., from fitMP.

type

String indicating information criterion to extract. See details.

N

Sample size (used in BIC computations). Could be auto-detected, but not done yet.

usefitfunc

Logical value. Toggles how to obtain fit function (log-likelihood). At some point how these are stored may have changed.

infotype

If type is set to "sic", this determines how the information matrix is computed, if not already available from the fitted model. May require model re-fitting.

Details

Use of Bayesian priors for MP models sometimes complicates computation of some information criterion as typically done by some popular software packages, and as done by Mislevy (1986). Typically computation of AIC, BIC, and the log-likelihood is done by plugging in parameter estimates (e.g., based on the posterior mode) into the equation for the marginal log-likelihood, instead of computing the value of the log-posterior. As the latter is typically done by OpenMx, the former is done by this function.

Currently supported are AIC ("aic"), BIC ("bic"), log-likelihood ("ll"), number of parameters ("np"), and stochastic information criterion ("sic").

References

Mislevy, R.J. (1986) Bayes modal estimation in item response models. Psychometrika 51, 177–195. https://doi.org/10.1007/BF02293979


falkcarl/mpirt documentation built on July 11, 2024, 12:09 a.m.