Parameter estimation in latent trait models (Q791268)

Consider following mixed model for binary responses Y to k test items: \(\theta_ i\) is a real valued ability parameter of the ith examinee; \(\beta_ j\) a vector valued item parameter for the jth item; \(\beta\) fixed, \(\theta\) random has a prior distribution indexed by \(\Phi\). The couple (\(\beta\),\(\Phi)\) is estimated using an EM algorithm (i.e. an iterative procedure for finding ML estimates), here maximizing E[Log f(y,\(\theta| \beta,\Phi)]\) where f(y,\(\theta| \beta,\Phi)\) is the joint pdf of \((y,\theta)\) given \(\beta\),\(\Phi\). The procedure is applied to the Rasch model and compared to other procedures on real and computer generated data.