Search

This study examined the effects of using static graphics and animated graphics to introduce statistical concepts associated with analysis of variance (ANOVA) to a group of college students. The demonstrations led the learner from the concrete to the abstract by use of dynamically adjustable variables paired with immediate feedback. Participants in this pilot study were inexperienced in statistical methods. Those in one group were exposed to static graphics, a program in which they could view before and after conditions. Those in the second group were exposed to animated graphics dynamically displaying the movement of a piston corresponding to increases in pressure or temperature. Graphs depicting the relationships were dynamically constructed on the screen. The goal of the pilot study was not to establish statistical significance, but to observe the effects of graphic presentation. Those receiving content with static graphics scored better than those receiving animated graphics, but these graphic presentations alone were not enough to add to the participants' knowledge of statistical concepts. (Contains 21 references.) (SLD)

It is argued that analysis of variance (ANOVA) and related methods should be taught using a general linear model (GLM) approach, rather than a classical ordinary sums of squares approach. The GLM approach emphasizes the linkages among conventional parametric methods, emphasizing that all classical parametric methods are least squares procedures that implicitly or explicitly use weights, focus on latent synthetic variables, and yield effect sizes analogous to "r" squared (are correlational). The case for teaching statistics using a GLM conceptual framework is based on the following four contentions: (1) a GLM instructional approach provides a unifying conceptual framework that better enables students to understand analytic methods; (2) a GLM approach provides a better match between the researcher's analytic model and the researcher's model of reality; (3) the GLM emphasis on planned contrasts helps students understand the critical role of reflective thought in good research; and (4) a GLM instructional approach helps students see that focusing on variance-accounted-for (or other) effect sizes, rather than statistical significance, is important in all analyses. Three tables present analysis examples. A 68-item list of references is included, and an appendix lists Statistical Package for the Social Sciences control cards. (Author/SLD)

Applied researchers frequently precede analyses of interest with one or more preliminary tests, used to help researchers determine which variables to examine more closely, or whether there are anomalies in the data set. These texts can be classified into three categories: omnibus tests, tests for model fit, and exploratory tests. Fifty-four applied journal articles, 31 statistical textbooks, and the statistical literature are reviewed to discuss some limitations associated with uses of some preliminary tests. The focus is on the following topics: (1) the omnibus F-test in analysis of variance; (2) tests for variance equality; (3) tests for equality of regression slopes in analysis of covariance; and (4) tests for sphericity in repeated measures designs. In general, it is concluded that many preliminary statistical tests are not useful. In many contexts, omnibus tests do not answer questions of substantive interest. Preliminary analyses in tests for model fit are unnecessary because alternative less restrictive models can be used, and because many tests for violations of data assumptions lack adequate statistical power, or are overly sensitive to another assumption violation. More focused analyses are advocated, using less restrictive analytical models, and an increased use of exploratory analyses is recommended. One table illustrates the discussion. Lists of the journal articles and textbooks reviewed are appended. (Contains 43 references.) (SLD)

It is argued that, given the importance and the increased use of multivariate techniques such as factor analysis and canonical correlation, students need to be made aware of multivariate methods and the appropriate ways in which they can be applied. As a general linear model that subsumes all other parametric measures, canonical correlation analysis provides a natural framework for instruction involving all of the various parametric procedures (e.g., analysis of variance and analysis of covariance). Furthermore, when canonical correlation analysis is used as an instructional tool, students gain an understanding of how all parametric procedures are special cases of canonical correlation analysis, that all parametric procedures involve the application of weights to derive synthetic scores, and that all parametric procedures are correlational, thus yielding a measure of effect that is important to the interpretation of one's results. A small heuristic data set (18 examples) is used to demonstrate how canonical correlation analysis can be used as an instructional device in teaching both univariate and multivariate parametric methods. Fourteen tables present data from the example and illustrate the discussion through comparison of analytic techniques. A 30-item list of references and 1 appendix are included. (Author/SLD)

Experimenters sometimes wish to estimate for a particular dependent variable the proportion of total group variance that is associated with mean differences among fixed treatments or subject classifications. Hays (1981) represents this proportion by the parameter omega squared. A point estimate may be easily computed as a function of the number of treatments, the total sample size, and the mean squares between and within treatments. This paper presents tables which facilitate the construction of 90 or 95 percent confidence intervals for omega squared. (Author)

This study investigated the effects of computer and print text density on learning, and the nature and effects of learner preference for different density levels in both print and computer presentation modes. Subjects were 48 undergraduate teacher education majors, who were assigned at random to six treatment groups in which a statistics lesson was presented in either of two modes (computer or print) crossed with one of three text density-level conditions (high, low, or learner control). Dependent variables were different types of learning achievement, lesson completion time, attitudes, and learning efficiency. Overall, the experimental findings were not supportive of computer based instruction relative to print as a delivery medium for the current statistics lesson. This lesson's dependency on mostly narrative descriptions of rules and operations and allowance of self-pacing remained constant regardless of mode, but print offered the possible advantage of representing the text in a more readable and accessible form. Furthermore, the newness of the computer mode may have caused students to perceive it as more difficult or challenging. In the attitude survey, subjects were generally favorable to computer based instruction (CBI). In learner control selections, high density material was favored for CBI 75% of the time, as opposed to 25% for print. The overall impression is of a less confident and more conservative computer based instruction group, which generally worked as a disadvantage for achievement and learning efficiency. The text is supplemented by one figure and one table. (39 references) (EW)

This research project was designed to explore ways of making computer text presentations more readable as a way of compensating for the constraints that computers impose on the display of text as compared with print-on-paper displays, e.g., limited screen resolution and display area. Forty-eight undergraduate teacher education majors enrolled in an instructional technology course received either a print or computer-based instruction (CBI) statistics lesson containing low-density (concise) narrative text, high-density (conventional) text, or the density type they preferred (learner-control). Results showed that the low-density display reduced completion time relative to the high-density version, while yielding equivalent achievement. Subjects receiving the lesson in the print mode had shorter completion times and higher achievement than their CBI counterparts. Attitude results showed that CBI was favorably regarded, but perceived as longer and slower-moving than print. A discussion of these results and their implications for designing instructional material in accord with CBI attributes and learner characteristics concludes the report. A 52-item list of references is provided, as well as three tables and samples of low-density and high-density frames from the CBI lesson. (Author/MES)

Although computer-assisted instruction (CAI) simulates a conversation between a tutor and student, it does not permit the quality of interaction so desirable in the tutorial process. This study attempted to see if team learning techniques might be one answer to the lack of interaction in programmed instruction. What would happen if pairs of students took a CAI course cooperatively? Thirty liberal arts college students were selected for this experiment on a predictor variable, College Entrance Examination Board (CEEB) scores (verbal only), to take a computer-guided statistics course. Students were divided into low- and high-scoring groups. In each group, 10 students formed pairs and 5 worked separately as controls. After the final exam, students were administered a questionnaire designed to measure their perceptions of their performance and those of their partners, if they were paired. The results from this study seem to indicate that students paired on CEEB verbal scores as a predictor variable will do as well as their controls on a final exam in a CAI course. In addition, they can complete the course in the same amount of time as their controls. The economic advantage is quickly realized since the cost of the educational terminal device has been cut by a factor of two in the process. (MLF)

Defines and illustrates basic concepts of dichotomous logistic regression (DLR) and presents strategies for teaching these concepts. Strategies include using analogies between ordinary least squares regression and logistic regression; illustrating concepts with contingency tables; and linking logistic regression concepts to interpretation of statistics in computer outputs. (MJP)

Describes a study conducted at the University of Southern Queensland (Australia) that investigated problem-solving behavior in statistics by documenting differences between successful and unsuccessful students. Two methodological approaches are described: one required students to identify key structural features of common problems, and the second used verbal protocol procedures. (Author/LRW)

Comparisons of the Wilcoxon test, Friedman test, and repeated-measures analysis of variance (ANOVA) on ranks in a computer simulation show that the Friedman test performs like the sign test whereas the ANOVA performs like the Wilcoxon test. Classification of these tests in introductory statistics textbooks should be revised. (SLD)