Best linear prediction (BLP) and penalized best linear prediction (PBLP) are techniques for combining sources of information to produce task scores, section scores, and composite test scores. The report examines issues to consider in operational implementation of BLP and PBLP in testing programs administered by ETS [Educational Testing Service].

The concepts of elevation, scatter, and shape are investigated as parameters in assessing the similarity among profiles of individuals. A computational strategy for differentiating the independent contribution of each parameter to more global indices of resemblance is presented. (Author/JKS)

Perspective on article by P. Isaac published in the Psychological Bulletin, 1970, 74, 213-18. (CB)

Most currently used measures of inter-rater agreement for the nominal case incorporate a correction for "chance agreement." The definition of chance agreement is not the same for all coefficients, however. Three chance-corrected coefficients are Cohen's Kappa; Scott's Pi; and the S index of Bennett, Goldstein, and Alpert, which has…

The Werts-Linn procedure for dealing with categorical errors of measurement in "Comments on Boyle's 'Path Analysis and Ordinal Data'" in The American Journal of Sociology, volume 76, number 6, May 1971, is shown to be inappropriate to the problem of ordered categories. (For related document, see TM 002 301.) (DB)

Study compares the Rasch and the Guttman models of measurement and thus adds to the description of the characteristics of Rasch's logistic model. Such knowledge is of importance in making decisions as to which model and which statistics should be used in evaluations of tests. (Author/CB)

The importance of regression diagnostics in detecting influential data points is discussed, and five statistics are recommended for the applied researcher. The suggested diagnostics were used on a dataset of 24 subjects, and effects were analyzed. Colinearity-based diagnostics and diagnostics for a variety of procedures are discussed. (SLD)

The Feldt-Gilmer congeneric reliability coefficients make it possible to estimate the reliability of a test composed of parts of unequal, unknown length. The approximate standard errors of the Feldt-Gilmer coefficients are derived via a method using the multivariate Taylor's expansion. Monte Carlo simulation is employed to corroborate the…

Procedures are presented for equating simultaneously several tests which have been calibrated by the Rasch Model. Three multiple test equating designs are described. A Full Matrix Design equates each test to all others. A Chain Design links tests sequentially. A Vector Design equates one test to each of the other tests. For each design, the Rasch…

A stochastic disturbance term appears to be essential for structural models in the social sciences. The analysis of such models is considered from the perspective of Joreskog's (1970) general model for the analysis of covariance structure. (For related document, see TM 002 301.) (Author)

Optimal appropriateness indices, recently introduced by Levine and Drasgow (1984), provide the highest rates of detection of aberrant response patterns that can be obtained from item responses. These optimal appropriateness indices are used to study three important problems in appropriateness measurement. First, the maximum detection rates of two…

A method of estimating the operating characteristics of a new test item added to an established test without assuming a prior model has been proposed by the author. In this paper, the author extends this logic by proposing a different hypothesized distribution of the resulting estimate. The former method used a normal approximation; the present…

The framework and procedures of generalizability theory are introduced and illustrated in this instructional module that uses a hypothetical scenario involving writing proficiency. Generalizability analyses are useful for understanding the relative importance of various sources of error and for designing efficient measurement procedures. (SLD)

LISREL-type structural equation modeling is a powerful statistical technique that seems appropriate for social science variables which are complex and difficult to measure. The literature on the specification, estimation, and testing of such models is voluminous. The greatest proportion of this literature, however, focuses on the technical aspects…