W. Robert Stephenson

Statistics at a Distance

In 1993 the Statistics Department at Iowa State University entered into a collaborative agreement with General Motors to develop and deliver a new sequence of courses titled “Applied Statistics for Industry.” This paper describes the development and content of these courses as well as their method of delivery. In order to accommodate on campus students as well as students at a distance, the course is presented live at Iowa State University and by videotape delay at General Motors Technical Education sites in Michigan, Ohio, Arizona and Mexico, and across the country at sites of other partner industries. Some of the differences between a statistics course taught in the traditional campus setting and a statistics course taught at a distance will be highlighted. Since there are two audiences (on campus and off campus), several compromises are made in how the course is conducted. These compromises, and their possible effects on students in both environments, are discussed. A summary of how on and off campus students did in these courses over the past five years is included.

A General Class of One-Sample Nonparametric Test Statistics Based on Subsamples

Abstract A general class of nonparametric test statistics is constructed by considering the sign of the most extreme value in subsamples of size m taken from n independent observations. The test statistics can be expressed as linear rank statistics and reduce to the sign test statistic for m = 1 and a modified Wilcoxon signed-rank test statistic for m = 2. Asymptotic normality of properly standardized versions of our statistics is established under general conditions. Small sample power simulations as well as Pitman and Bahadur efficiencies indicate that members of our class do well in comparison with already existing test statistics.

Beyond Traditional Statistical Methods

Todays courses in statistical methods, for the most part, focus on the same methods that were taught 30 years ago. The actual practice of statistics has moved beyond these traditional statistical methods. Modern methods—including dynamic graphics, nonlinear estimation, resampling, and other simulation-based inference methods—are being used by many scientists and engineers. However, these methods generally are not included in courses in statistical methods, especially at the undergraduate level. This article discusses the development of a collection of instructional modules, built around actual applications from science and engineering. Each module is self-contained and includes instructional materials such as: objectives, examples, lecture materials, computer implementation of the methodology, homework, class/discussion exercises, and assignments. The modules are intended as a resource for instructors to experiment with and explore the use ofmodern statistical methodology inundergraduate statistics methods courses. Two of the modules will be presented in some detail. We also discuss the use of the modules in a new course that goes beyond our traditional methods courses.

Two sample nonparametric tests based on subsamples

The paper introduces a general class of nonparametric tests for the two-sample location problem based on subsamples. Includ- ed in this class is the Mann-Whitney (or the Wilcoxon rank sum) test. General formulas for the Pitman efficacy for different methods of subsampling are derived. A small sample power simu- lation compares the performance of members of this class

A comparison of nonparametric analysis of covariance techniques

A small sample Monte Carlo simulation provides the means of comparison for several nonparametric alternatives to the standard analysis of covariance. A completely ran-domized design with samples from K populations and one co-variate is used for this study. The simulation evaluates the robustness of the techniques as well as the power to detect differences among the populations. The study indi-cates that a modification of existing nonparametric proce-dures yields a test that performs well in many situations

An Interactive Program for the Analysis of Data from Two-Level Factorial Experiments via Probability Plotting

An interactive computer program that expedites the analysis for unreplicated two-level factorial and fractional factorial experimental designs advocated by Daniel (1976) and Box, Hunter, and Hunter (1978) is presented. The program calculates estimated effects via the Yates algorithm, identifies statistically detectable effects via normal plots and half normal plots, fits candidate models via the reverse Yates algorithm, and enables evaluation of candidate models through residual plots. The program can handle the analysis of standard 2p-q fractional factorial experiments where p q √ 7 and can be modified to allow p q > 7.

A computer program for the quick and easy analysis of unreplicated factorials

Lenth (1989) proposes a quick and easy method for the analysis of unreplicated factorial and fractional factorial experiments. This article presents a FORTRAN computer program that uses Lenths method to produce the simple graph, similar to the analysis-of-means, for determining the importance of the various effects (contrasts).

Training Statistics Teachers at Iowa State University

Graduate teaching assistants at Iowa State University develop their teaching skills through an apprenticeship-like process. First-year graduate students start out as laboratory instructors/graders. After the first year, some graduate teaching assistants teach a section of an introductory statistics course. This article describes this apprenticeship-like process and the mentoring and resources provided to graduate teaching assistants.

Does Eye Color Depend on Gender? It Might Depend on Who or How You Ask.

As a part of an opening course survey, data on eye color and gender were collected from students enrolled in an introductory statistics course at a large university over a recent four year period. Biologically, eye color and gender are independent traits. However, in the data collected from our students, there is a statistically significant dependence between the two variables. In this article, we present two ideas for using this data set in the classroom, and explore the potential reasons for the dependence between the two variables in the population of our students.

Extracting More from Dynamic Modulus Data Using Split-Plot Repeated Measures Analysis

Dynamic modulus testing and the development of master curves is an essential test used to characterize asphalt mixes. The dynamic modulus values are used in the mechanisitic-emprical pavement design guide that is being implemented by many state agencies. To compare dynamic modulus values, most researchers use master curves to analyze the values; however, the log-log scale of the master curve masks the differences between the mixes, requiring the use of an appropriate and reliable statistical data analysis approach to make accurate comparisons between mixes. In this paper, a full experimental plan is designed to investigate the effect of three different factors on the dynamic modulus, namely (1) type of mix: hot mix asphalt versus warm mix asphalt, (2) impact of reheating, and (3) moisture conditioning of specimens. An open-access statistical software package, “R,” is used for the analysis. The data showed that the type of mix and moisture conditioning were statistically significant factors. The statistical analyses presented in this paper begin with a simplified approach and then become increasingly comprehensive. The simplified approach considers a single temperature and frequency only. A more encompassing approach, with the aid of split-plot repeated measures (SPRM) analysis, is then performed in which each of the three factors is considered separately. This is followed by a full analysis on the entire data set that takes into account the interaction between all factors. The benefits of performing a full analysis in lieu of a simplified analysis are highlighted and a detailed interpretation of the results is given.