Using JAGS for Bayesian Cognitive Diagnosis Modeling: A Tutorial

Abstract

In this article, the JAGS software program is systematically introduced to fit common Bayesian cognitive diagnosis models (CDMs), including the deterministic inputs, noisy "and" gate (DINA) model, the deterministic inputs, noisy "or" gate (DINO) model, the linear logistic model, the reduced reparameterized unified model (rRUM), and the log-linear CDM (LCDM).The unstructured latent structural model and the higher-order latent structural model are both introduced. We also show how to extend those models to consider the polytomous attributes, the testlet effect, and the longitudinal diagnosis. Finally, an empirical example is presented as a tutorial to illustrate how to use the JAGS codes in R. Keywords: cognitive diagnosis modeling, Bayesian estimation, Markov chain Monte Carlo, DINA model, DINO model, rRUM, testlet, longitudinal diagnosis, polytomous attributes 2013), their commercialization prevents unauthorized users especially students from accessing these software without purchasing. In this article, we demonstrate how to use the freeware, JAGS, to fit several popular CDMs and present the code. It is expected that the researchers can adapt the code to fit extended CDMs,which cannot be fitted in existing software or packages for their research or application purposes. In general, JAGS makes it easy to construct a Markov chain for parameters. It does not require users to derive the posterior distribution of the model parameters by hand. Movereover,, the R2jags package (Version 0.5-7; Su & Yajima, 2015) in R could be easily used to call the JAGS. Furthermore, It should be noted that the JAGS code presented in this study can be generalized easily to other BUGS software programs by minor editing including WinBUGS and OpenBUGS 2 . The following sections first illustrate JAGS codes for five CDMs: (1) the DINA model; (2) the DINO model; (3) the LLM; (4) the rRUM, and (5) the LCDM. Besides those five models, which are based on the unstructured (or saturated) latent structuralels, the higher-order latent structural model (de la Torre & Douglas, 2004) is also demonstrated. Further, the extensions to the polytomous attributes, the testlet effect, and the longitudinal diagnosis using JAGS are presentedas well. Lastly, an empirical example analysis is conductedto illustrate how to use the R2jags package to run the JAGS code.

Joint Testlet Cognitive Diagnosis Modeling for Paired Local Item Dependence in Response Times and Response Accuracy

“…Sample code were presented in Appendix. A tutorial of using JAGS for Bayesian CDM estimation can be found in Zhan ( 2017 ).…”

Cognitive Diagnostic Models for Random Guessing Behaviors

“…Besides, for ease of use, a tutorial to introduce the RT-GDINA-RG in JAGS can be developed in future work (cf. Curtis, 2010;Zhan et al, 2019).…”

Analysing Standard Progressive Matrices (SPM-LS) with Bayesian Item Response Models

“…There are a lot of IRT specific software packages available, in particular in the programming language R (R Core Team 2019), for example, mirt (Chalmers 2012), sirt (Robitzsch 2019), or TAM (Robitzsch et al 2019; see Bürkner 2019 for a detailed comparison). In addition to these more specialized packages, general purpose probabilistic programming languages can be used to specify and fit Bayesian IRT models, for example, BUGS (Lunn et al 2009; see also Curtis 2010), JAGS (Plummer 2013; see also Depaoli et al 2016;Zhan et al 2019), or Stan (Carpenter et al 2017; see also Allison and Au 2018; Luo and Jiao 2018). In this paper, I use the brms package (Bürkner 2017(Bürkner , 2018, a higher level interface to Stan, which is not focused specifically on IRT models but more generally on (Bayesian) regression models.…”