A New Method for Estimating Model Parameters for Multinomial Data

Recognition memory models and binary-response ROCs: A comparison by minimum description length

“…The NML index thereby addresses the concerns raised by Roberts and Pashler (2000) and others (e.g., Chechile, 1998;Myung, 2000) in a head-on fashion: A model' s ability to fit the observed data should be put in relation to its ability to provide good fits in general. A model recovery study confirmed that model selection based on NML outperforms model selection based on AIC, as well as BIC, for the models and kind of data sets considered here.…”

New multinomial models for the Chechile–Meyer task

“…For Bayesian analyses with more parameters than degrees of freedom, Chechile (1977a) pointed out that the variance of posterior distribution does not approach 0 as the sample size increases to infinity, i.e., the estimates are not consistent. Furthermore, Chechile (1998) has developed a new Bayesian method for estimating the latent model parameters for multinomial data-the method is called population-parameter mapping (PPM). The PPM method does not require the evaluation of numerical integrals.…”

“…In Section 6, the maximum likelihood estimates (MLE) for the parameters of the 6P model are provided. In Section 7, an alternative method for fitting the model parameters for multinomial data, which is called population parameter mapping (PPM) (see Chechile, 1998), is applied to the 6P model. Although the parameters of the 6P model can be estimated by either MLE or PPM methods, the relative accuracy of these estimates are not equivalent.…”

“…Riefer and Rouder (1992) and Rouder and Batchelder (1998) also developed MPT models for storage/retrieval measurement that was tied to a testing procedure consisting of a combination of free and cued recall. Chechile (1998) and Chechile and Soraci (1999) generated MPT models for storage/retrieval measurement linked to a testing procedure that has an initial free recall test that is followed by a series of forced-choice recognition tests.…”

Minimum description length model selection of multinomial processing tree models

“…For example, it does not address the issues of model misspecification and individual differences. To allow for model misspecification, exact equality constraints should be replaced by fuzzy equality constraints, as was done in an elegant sampling-based method known as populationparameter mapping proposed by Chechile (1998). Regarding individual differences, one way of incorporating this important factor is through hierarchical modeling in which parameter values corresponding to different individuals are assumed to be sampled from a common distribution (see, e.g., Klauer, 2010;Smith & Batchelder, 2010).…”