Abstract
Parameter estimation in IRT: an evaluation of bilog in small datasets. In this report, we present a study of simulation to prove the bias of the parameter estimation in BILOG (V 3.04) with small groups (50, 100, 250 and 500) and short tests (10, 20, 30 and 40) in groups where the test was difficult (μθ = -1*), neutral (μθ = 0) or easy (μθ = +1). The bias of the item parameter estimations improve with the increase of the sample size. Also, the bias of the ability parameter estimations improve with the increase of the length test. Furthemore, the estimation of the discrimination parameter was independent of the test difficulty. It not happen the same in the difficulty or ability parameter. In these parameters, BILOG was more accurace when the test difficulty mean coincided with the grupp ability mean.
