With reference to the questionnaire adopted within the Italian project
"Ulisse" to assess health condition of elderly people, we investigate two
important issues: discriminant power and actual number of dimensions measured
by the items composing the questionnaire. The adopted statistical approach is
based on the joint use of the latent class model and a multidimensional item
response theory model based on the 2PL parametrization. The latter allows us to
account for the different discriminant power of these items. The analysis is
based on the data collected on a sample of 1699 elderly people hosted in 37
nursing homes in Italy. This analysis shows that the selected items indeed
measure a different number of dimensions of the health status and that they
considerably di?er in terms of discriminant power (effectiveness in measuring
the actual health status). Implications for the assessment of the performance
of nursing homes from a policy-maker prospective are discussed.