MAXIMUM LIKELIHOOD ESTIMATION IN THE GENERALIZED CASE OF THURSTONE'S MODEL OF COMPARATIVE JUDGMENT

Abstract

In light of the empirical evidence against simple scalability a generalized case of Thurstone's model of comparative judgment was proposed, in which covariance terms in discriminal processes are not necessarily assumed zero or constant. In order ·to avoid the indeterminacy (over-parametrization) pnillem of the model, the covariance matrix was assumed to have a prescribed rank. A parameter estimation. procedure based on the maximum likelihood principle was developed and implemented in the fonn ot It FORTRAN program. An· example. was given to illustkte the empirical relevance of the proposal.