Following is a brief overview of Item Response Theory (IRT) analysis in Mplus,
a list of IRT examples in the Mplus Version 4 User's Guide, and links to technical
descriptions of IRT modeling in Mplus.
Starting with Version 1
released in 1998, analysis using the "normal ogive" model of probit regression for
binary and ordered polytomous items was possible with limitedinformation weighted
least squares estimation using the WLSMV estimator. The analysis was implemented in
the general framework of Muthen (1984; see website References), including covariates,
multiplegroup analysis, and other model parts. Factor score estimation uses the
maximum a posteriori (MAP) approach. For a description of these approaches, see the
Technical appendices.
The release of Mplus Version 3 in March 2004 added analysis with a 2parameter
logistic model and maximumlikelihood estimation. As a special case, this gives the
1parameter Rasch model, where factor loadings (item discriminations) are held equal
across items. The items can be binary, ordered polytomous, censored, and counts.
Factor score estimation uses the expected a posteriori (EAP) approach. ML estimation
is also available with the normal ogive probit model for binary and ordered polytomous
items. The analysis is implemented in the general Mplus framework. As such, this
also includes finite mixture IRT analysis, multilevel IRT analysis, and multilevel
mixture IRT analysis.
Mplus Version 4.1 released in May 2006 added IRT plots to its graphics features.
These are item characteristic curves and information curves for individual items,
sets of items, and all items. They are available for mixture models, multilevel models,
and multilevel mixture models. In addition to presenting parameter estimates and standard
errors in the regular factor analysis metric, they are also given in conventional IRT 2PL
metric with a single latent variable and binary items.
The 2PL IRT examples in the Mplus Version 4 User's Guide are as follows. Ex 5.5 presents a standard 2PL model. Ex 7.26 presents a
mixture approach to a nonparametric alternative of the normality assumption for the latent variable. Ex 7.27 presents a mixture IRT model.
Ex 7.29 presents a 2group 2PL IRT analysis for monozygotic and dizygotic twins. Ex 9.6 presents a 2level analysis with a random
intercept factor and covariates. These examples are also available in Monte Carlo simulation form on the Mplus website and CD.
Mplus Version 7.4 introduced 3PL, 4PL, and generalized Partial Credit Modeling (PCM); see
Asparouhov and Muthén (2015). Due to the general modeling framework, this means, for example,
that multilevel and mixture PCM with observed and latent predictors are available as well.
