Conferences

Latent characteristics of respondents, such as degree of trust, religiosity, or job satisfaction, are often measured in survey research by interpreting item responses on an interval scale, based on Likert’s (1932) defense of the five-point rating scale as measurement at interval level. This allows the calculation of means, standard deviations and correlations, which in turn are used in Reliability analysis, Factor analysis and Structural Equation Models. For parallel test measurement, items that indicate the same latent trait ideally have the same distribution.

Item Response Theory, starting with ideas of Thurstone (1925) and Guttman (1944), uses the differences in distribution of the items to allow for a measurement of the (dichotomous) items, along with the measurement of respondents. Later developments allowed for the measurement of the response categories of polytomous items. In a Guttman scale the measurement of respondents and items is based on interpreting the response of a respondent as a deterministic dominance relation between respondent and item.

Two types of development will be described:
First, probabilistic dominance models have been developed for latent variables at the ratio level (e.g., the one parameter logistic (e.g., the Rasch model, 1960), the ordinal level (e.g., the Mokken scale, 1971), and the nominal level (Latent Class Analysis; e.g., Lazarsfeld & Henry, 1968).

Second, models have been developed in which the response of a respondent is interpreted as a proximity relation between respondent and item. The most important ones are the linear unfolding model (e.g., Coombs, 1950, Andrich and Luo, 1993, and Van Schuur, 1993 , and the circular unfolding model (e.g., Leary, 1956, Brown, 1992, and Mokken, Van Schuur and Leeferink, 2001).

Examples of these developments will be given in the other papers in this session.