Logo image
Consequences for social measurement of collapsing adjacent categories within items with three or more ordered categories
Thesis   Open access

Consequences for social measurement of collapsing adjacent categories within items with three or more ordered categories

John Harris
Other, Murdoch University
1991
pdf
JHarris19912.65 MBDownloadView
Open Access

Abstract

Educational tests and measurements Evaluation research (Social action programs) Educational evaluation
The use of ordered categories for social measurement is discussed and associated requirements for measurement are considered. The scaling of responses onto a linear continuum is examined in terms of the Thurstone cumulative probability model and the Rasch cumulative threshold model, and requirements which might be imposed on the scaling process are considered. The approaches used by the two models in scaling responses are described and contrasted, and their application to data derived from the use of a performance scale is demonstrated. The issue of collapsing adjacent categories and the consequences for scaling in terms of the two models are considered, together with an examination of competing theoretical principles. A possible fundamental principle, that a respondent's choice of category should be identical whatever the actual boundaries between categories, is put to an empirical test, with all conditions constructed to provide maximum opportunity for its manifestation. All analyses of the response data indicate that the principle is untenable. As post hoc pooling of data is intended to predict the effects of a priori collapsing of adjacent categories, analyses using such post hoc pooling are brought into question. The study provided support for the view that if a set of ordered categories works as a scale for measurement, this does not imply that a set based on collapsed categories would also work. The study also provided support for the use of the Rasch cumulative threshold model in the scaling of item responses.

Details

Metrics

14 File views/ downloads
41 Record Views
Logo image