On the finiteness and uniqueness of the weighted likelihood estimator of ability in polytomous IRT models

In item response theory (IRT), the weighted likelihood (WL) estimator has become a central method to estimate ability levels of respondents. Primarily introduced with dichotomous IRT models (Warm, 1989), it was later extended to polytomous IRT models (Samejima, 1998). However, very few information is available about the behavior of the WL estimator, and especially about the uniqueness of the ability estimates as well as their finiteness. The purpose of this talk is to establish that with polytomous item response models, the WL estimator of ability always returns finite values. This result is valid for the class of difference models and divide-by-total models, independently of the number of items and the response patterns. However, such estimates may not necessarily be unique, as the WL solving equation may provide several optimal values. Some examples are considered to illustrate both findings