Technological Un ivers it y. Delfi A N D P I E T E R V I J N Un iversi t y ofA msterda m. A msterda mThis article presents some critical comments on the validity generalization procedure which has been presented by Schmidt, Hunter and others. They have put forward a method for testing the hypothesis that the variance in validity coefficients across situations for job-test combinations is due to what they consider to be statistical artifacts. The Schmidt-Hunter approach is criticized on the following points: the compilation of validity data, the use of criterion measures, and the test of the hypothesis of no situational specificity. Further, the relation between the concepts ' situational specificity ' and validity generalization ' is considered. In addition, it is noted that Schmidt, Hunter and others have defined the concept ' situation ' in a different way than classical writers. It is concluded that the Schmidt-Hunter approach to validity generalization shows fundamental shortcomings. As a consequence their far-reaching conclusions for the practice of personnel selection should be considered premature. b

The Rasch model is formulated as a loglinear model. The goodness of fit and parameter estimates of the Rasch model can be obtained using the iterative proportional fitting algorithm for loglinear models. It is shown in an example that the relation between the estimates of the iterative proportional fitting algorithm and the unconditional maximum likelihood Rasch algorithm are almost perfectly linear. The Rasch model can be extended with a design for the items, which can be formulated as a loglinear model. In the Rasch model for binary scored items the probability that person j gives a response scored one to item i is written as (Rasch, 1960) where a, represents the person's ability and cm the item's difficulty. From the model in Equation 1, it follows that the natural logarithm of the response ratio is Brogden (1977) has shown the relation between the Rasch model, the law of comparative judgment, and additive conjoint measurement; and Perline, Wright, and Wainer (1979) have presented the Rasch model as a special case of additive conjoint measurement. Additive conjoint measurement applies to mental test data when a monotonic transformation of the 7~'s yields an additive representation :where f is a monotonic function. When f is the inverse logistic transformation of Equation 1, the Rasch model follows. This shows that the Rasch model is a special case of additive conjoint measurement.In the Bradley-Terry model for paired comparisons, the probability that stimulus j is preferred to stimulus i is written as

No abstract

ordered Dirichlet prior, ordered cumulative probabilities, ordered scales, order statistics, latent population, coverages, log oddsratio, stochastic ordering, complementary log-log scale, logistic scale,