Rasch Analysis and measurement bias: modeling a response structure for the quality of teaching

The ANVUR questionnaire is aimed at evaluating the quality of teaching as perceived by the students. In order to assess this information it is necessary for the level of quality associated with different aspects of teaching (items) to be placed along a unidimensional continuum. In only this way is it possible to make a quantitative assessment on the position occupied by individuals both in absolute and relative terms. These questionnaires represent a useful qualitative description of the students' responses, however the sum of the scores relative to the various questions is quantitatively meaningless. In order to get a measurement of these responses it is necessary to move from an ordinal to a quantitative scale. A useful methodology to achieve this purpose is given by the Item Response Theory - IRT (Baker & Kim 2004), on which several statistical models have been based. The Rasch Model (Rasch 1960) is the simplest of these tools as it is characterized by a smaller number of parameters. As long as the data abides by the requirement of invariance of comparisons, this model yields a response structure through a linearization of the total scores. The intent of this study is to determine, by means of a Rasch analysis, how the various aspects of teaching (items) contribute to the shaping of the perceived quality of education (unidimensional variable). In addition it is verified if personal characteristics, such as enrolment in specific courses, would lead to a different probability of providing a certain response to the questionnaire. To this end, besides the most common diagnostic tools used when dealing with the Rasch Model, the Differential Item Functioning (DIF) is also applied. This tool verifies whether, with regard to the ith item, the probability for homogeneous groups of subjects (with respect to a given qualitative or quantitative criterion) of giving a certain response varies only according to the value of the latent trait or it also varies depending on the values and levels of the personal factors of the respondents.