Q-Matrix Learning via Latent Variable Selection and Identifiability
Much of the research and application in cognitive diagnostic assessments to date has been centered on a confirmatory approach where a Q-matrix is pre-identified using content experts’ opinion or test developers’ knowledge on test items. As opposed to the traditional methods, which require prior knowledge about latent dimensions and underlying structure of test items, the approaches described in this chapter attempt to identify a Q-matrix solely relying on the observed test response data and thus avoid probable decision error. There are several important aspects to consider when estimating a Q-matrix from the observed data. First, a fundamental question of identifiability arises, that is, whether and to what extent Q can be estimated from data. The second aspect to consider in learning Q concerns the computational intensity that arises from estimation. The third aspect pertains to the presence of missing data, more precisely, the latent attributes underlying the observed data. The completeness of a Q-matrix, the other important aspect to consider in identifying Q, is beyond the scope of the present chapter.
- Chen, Y., Culpepper, S. A., Chen, Y., & Douglas, J. (2017). Bayesian estimation of the DINA Q matrix. Psychometrika. https://doi.org/10.1007/s11336-017-9579-4
- Fang, G., Liu, J., & Ying, Z. (2017a). Latent variable selection via overlap group LASSO with applications to cognitive assessment. Preprint.Google Scholar
- Templin, J. L., & Henson, R. A. (2006). A Bayesian method for incorporating uncertainty into Q-matrix estimation in skills assessment. In Symposium Conducted at the Meeting of the American Educational Research Association, San Diego, CA.Google Scholar
- Tibshirani, R. (1996). Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society, Series B, 58, 267–288.Google Scholar
- Tseng, P. (1988). Coordinate ascent for maximizing nondifferentiable concave functions (Technical Report LIDS-P, 1840). Massachusetts Institute of Technology, Laboratory for Information and Decision Systems.Google Scholar
- von Davier, M. (2005, September). A general diagnostic model applied to language testing data (Research report No. RR-05-16). Princeton, NJ: Educational Testing Service.Google Scholar