Conferences

Aim of the presentation is to review a number of fundamental issues in the modelling of attitudes on the basis of (ordinal) categorical indicators, with emphasis on the role of attitude measurement in survey practice. As an example, data concerning self-reported “locus of control” from two national Polish surveys, conducted in 1984 and 1998 will be (re)analyzed in reference to the original interpretations.

Conventional procedure of “summated ratings”, as well as agregation based on the assumption of pseudo-metrical category scores will be shown inadequate with respect to a number of issues arising in attitude measurement. These include choice of scale for representing the latent variable, verification of the ordering of response categories, including interpretation of the “difficult to say” category”, numerical representation of the latent factor,
inter-temporal comparisons, assessment of over-time reliability and differential item functioning.

It will be shown that these issues are adqeuately solved by a paradigmatic switch to a functional nonparametric latent factor model (Kutylowski, 1997). Such a model, based on the idea of kernel smoothing of the category probabilities as functions of the latent factor, originally developed by Ramsay (1990, 2001), will be applied to the substantive task of quantifying “locus of control” via learning from data, separately and jointly for the two
periods.

It will be shown that quantification of the “locus of control” in this flexible modelling framework leads to analytically satisfying treatment of the technical issues mentioned above, as well as to meaningful findings concerning comparisons across socioeconomic positions, associations with other attitudes, and role of the attitude in question as an intermittent variable contributing to political choices.

Wider questions concerning the use of attitude measurement will also be considered, in particular robust measurement of attitude scores in terms of ordinal rather than metrical scores, and inclusion of the attitude scores in multistage models involving several directly
or indirectly observed variables.

The presentation will aim at being self-contained, for the purpose of quiding the reader as to similar applications in other substantive settings.

References:

Kutylowski, A. J. (1997). Nonparametric latent factor analysis of occupational inventory data. Chapter 24 in: Rost, J. and Langeheine, R., eds.; Applications of latent trait and latent class models in the social sciences. New York/Münster: Vaxmann.
Ramsay, J. (1991). Kernel smoothing approaches to nonparametric item chatacteristic curve estimation. Psychometrika, 56, 611-30.
Ramsay, J. (2001). Testgraf 98. Computer program and manual. Available from www.psych.mcgill.ca.