MDLTM, software for MultidimensionalDiscrete Latent Trait Models. mdltm is a software for Windows, Linux, and Mac OSX that allows analyseswith a wide range of latent variable models.
Such models include:
1.
unidimensional and multidimensional IRTmodels,
2.
latent class models,multiple-classification latent class models,
3.
unidimensional and multidimensional locatedlatent class models,
4.
diagnostic models with dichotomous orordinal skills variables,
5.
mixture distribution IRT and diagnosticmodels,
6.
growth mixture models,
7.
hierarchical latent class models, and
8.
hierarchical diagnostic models.
The EM algorithmis implemented in mdltm to obtainmarginal maximum likelihood estimates (MMLEs) of parameters.
This software is able to handle data withdichotomous and polytomous reponses as well as data with missingobservation.
In addition, multiplepopulations can be estimated simultaneously, thus enabling the comparison ofparameters across multiple populations.
Various constraints can be imposed upon the item parameter estimates,such as constraints across items or populations, or even fixed to values fromprevious calibrations.
This way, IRTlinking can be easily accomplished in mdltm.
The output fromthis software is comprehensive, in that it provides tabulations of observedquantities (item category frequencies and item total correlations, etc.),
as well as estimates, standard errors andexpected counts. In addition, a variety of goodness-of-fit indices aregenerated for each estimated model. Information criteria (Akaike, 1973;Schwarz, 1978), as well as related quantities such as the log-penalty (Gilula& Haberman, 1994) are available. In addition, fit diagnostics such as itemfit, based on pseudo-counts, and in the person estimates file, person fitindices based on observed quantities are generated upon user request.