A method of creating linear statistical models that incorporate assumptions implied by hypotheses is described and demonstrated by testing analysis of covariance type hypotheses in the presence of non-parallel regression lines. (SLD)

The accuracy of four empirical techniques to estimate shrinkage in multiple regression was studied through Monte Carlo simulation. None of the techniques provided unbiased estimates of the population squared multiple correlation coefficient, but the normalized jackknife and bootstrap techniques demonstrated marginally acceptable performance with…

Applied two methods for determining severe discrepancies between ability and achievement to K-12 students (n=344) to determine how a change in method might influence rates and characteristics of students meeting this criterion. Results indicate an increase when a regression method is used over a simple difference score method, but a 20% decrease…

This article describes a method of incorporating two ability estimates into a single regression analysis so that substantially fewer items exhibit differential item functioning (DIF). Comparative analysis with the Mantel-Haenszel procedure also shows more accurate matching of examinees, preventing identifying multidimensional item impact as DIF.…

Considers regression over simple difference models for assessing severe intelligence quotient (IQ)-achievement discrepancies. Identifies expected outcomes following change from simple difference to regression discrepancy model. Compares outcomes in determination of overall identification rates and identification of students by race, grade, and…

Examines methodological problems associated with analysis of event data by comparing five alternative statistical specifications: multiple regression, discriminant analysis, logistic regression, tobit regression, and event history analysis. Uses each procedure to estimate risk factors for out-of-home placement among children in families that…

Discusses models for the distribution of words in text and proposes a new model based on clustering that can be used to estimate the probability that a document contains a particular word as well as the number of distinct words in a document. Zipf's law and the Poisson approximation are also discussed. (18 references) (LRW)

Provides a comparison of centered and raw-score analyses in least squares regression. The two methods are demonstrated with constructed data in a Monte Carlo study to be equivalent, yielding identical hypothesis tests associated with the moderation effect and regression equations that are functionally equivalent. (SLD)

Argues that the conceptual and methodological foundations of both developmental and difference approaches to mental retardation are deficient. Analysis of the relationship between cognitive functioning and cognitive level within chronological age cohorts allows for more sensitive tests and suggests the need for a paradigm shift based on regression…

Explores whether flexible nonlinear models (including neural networks and genetic algorithms) can reveal otherwise unexpected patterns of relationship in typical school-productivity data. Applying three types of algorithms alongside regression modeling to school-level data in 183 elementary schools proves the hypothesis and reveals new directions…

Suggests that there is a tendency in social science and program evaluation to adhere to some methodological practices by force of custom rather than because of their reasoned applicability. These ideas include regression artifacts, random measurement error, and change or gain scores. (Author/SLD)

Agrees with L. Mohr that researchers are too quick to assume that measurement error is random, but disagrees that the idea of regression toward the mean has been a distraction and the notion that change scores analysis should be avoided in favor of regression analysis. (SLD)

Responds to C. S. Reichardt's discussion of regression artifacts, random measurement error, and change scores. Emphasizes that attention to regression artifacts in program evaluation is almost bound to be problematic and proposes some arguments in support of this position. (SLD)

Describes the use of SAS/IntrNet, a Web service rather than a Web server, to evaluate and enhance university level Web-based courses. IntrNet provides the ability to create both static and dynamic Web pages, using Web publishing tools to convert SAS output to hypertext markup language for the creation of static Web pages. (Author/LRW)

Developed a new classification method for the logistic regression (LR) procedure for differential item functioning (DIF) based on methods used in the Simultaneous Item Bias test and conducted a simulation study to determine if the effect size measure affects the Type I error and power rates for the LR DIF procedure. Results show that inclusion of…