Cross-validation is a common statistical procedure applied to problems that are otherwise computationally intractable. It is often employed to assess the effectiveness of prediction procedures. In this report, cross-validation is discussed in terms of "U"-statistics. This approach permits consideration of the statistical properties of…

Measures of agreement are compared to measures of prediction accuracy within a general context. Differences in appropriate use are emphasized, and approaches are examined for both numerical and nominal variables. General estimation methods are developed, and their large-sample properties are compared.

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].

Many assessments of writing proficiency that aid in making high-stakes decisions consist of several essay tasks evaluated by a combination of human holistic scores and computer-generated scores for essay features such as the rate of grammatical errors per word. Under typical conditions, a summary writing score is provided by a linear combination…

Subscores are reported for several operational assessments. Haberman (2008) suggested a method based on classical test theory to determine if the true subscore is predicted better by the corresponding subscore or the total score. Researchers are often interested in learning how different subgroups perform on subtests. Stricker (1993) and…

Most automated essay scoring programs use a linear regression model to predict an essay score from several essay features. This article applied a cumulative logit model instead of the linear regression model to automated essay scoring. Comparison of the performances of the linear regression model and the cumulative logit model was performed on a…

Statistical prediction problems often involve both a direct estimate of a true score and covariates of this true score. Given the criterion of mean squared error, this study determines the best linear predictor of the true score given the direct estimate and the covariates. Results yield an extension of Kelley's formula for estimation of the true…

In educational testing, subscores may be provided based on a portion of the items from a larger test. One consideration in evaluation of such subscores is their ability to predict a criterion score. Two limitations on prediction exist. The first, which is well known, is that the coefficient of determination for linear prediction of the criterion…

Recently, there has been an increasing level of interest in reporting subscores. This paper examines the issue of reporting subscores at an aggregate level, especially at the level of institutions that the examinees belong to. A series of statistical analyses is suggested to determine when subscores at the institutional level have any added value…

When a simple random sample of size n is employed to establish a classification rule for prediction of a polytomous variable by an independent variable, the best achievable rate of misclassification is higher than the corresponding best achievable rate if the conditional probability distribution is known for the predicted variable given the…

The usefulness of joint and conditional maximum-likelihood is considered for the Rasch model under realistic testing conditions in which the number of examinees is very large and the number is items is relatively large. Conditions for consistency and asymptotic normality are explored, effects of model error are investigated, measures of prediction…

Criteria for prediction of multinomial responses are examined in terms of estimation bias. Logarithmic penalty and least squares are quite similar in behavior but quite different from maximum probability. The differences ultimately reflect deficiencies in the behavior of the criterion of maximum probability.