Department of Statistics Graduate Seminars
Seminar in Statistics
To view a previous seminar abstracts
use the drop down menu and select the Date or Author to view Seminar Information and Abstract.
Next Seminars
Date: Wednesday, November 11th, 3:00 pm
Title: Statistical Power and Effect Size in Marketing Journals, Re-visited.
Presented by: Dr. Dwayne Ball, Associate Professor, Marketing, UNL
Location: Hardin Hall, Room 49
ABSTRACT:
In 1981, an article by Sawyer and Ball, published in the Journal of Marketing Research, investigated the extent to which authors of marketing scholarly work made prior estimates of effect size and desired power, and used these to determine the necessary sample size. It appeared that explicit prior consideration of probable power and effect size was rare among academic marketing researchers. However, it appeared from a review of published articles in the Journal of Marketing Research that power against medium and large effects was generally good. The paper went on to conclude that, since most published research designs were sufficiently powerful, the statistical tests of significance were insufficiently informative for good scholarly work; theoretically unimportant effects might be obtaining the attention of the field. It was strongly suggested that explicit calculation and publication of effect sizes was essential for the efficient scholarly growth of the discipline. This article was given a great deal of attention within the academic marketing field; it was runner-up for the most influential paper over 5 years in the Journal of Marketing Research, and has been widely used as a reading in Ph.D. programs in marketing.
Our question now, nearly 3 decades later is, “Has anything changed?” An analysis of three years of the Journal of Marketing Research and three years of the Journal of Consumer Research, two of the three most prestigious journals in Marketing, was performed. Every test of an important effect in every article using OLS regression or between-subjects ANOVA was analyzed. Where sufficient information was available, the a priori statistical power against small, medium, and large effect sizes were calculated, as was the actual power of each test, given the results. Finally, the actual effect sizes in terms of partial r-squared for predictors in regressions and the partial eta-squared for effects in ANOVAs were calculated.
The results showed generally good power for ANOVA effect tests and a very high average power for regression tests. However, as found before, there was usually no explicit calculation of effect size and frequently, no way to calculate effect sizes from the information provided in the published article. We end with yet another call, three decades later, for authors to report effect sizes.
And
Date: Wednesday, November 13th, 2:00 pm
Title: Multistage Testing Procedures with Clinical Trial Applications.
Presented by: Dr. Alex Dmitrienko, Research Advisor, Eli Lilly and Company
ABSTRACT:
This talk introduces a general approach to constructing multiple testing procedures in problems with multiple families of hypotheses(also known as gatekeeping procedures). Hypothesis testing problems of this kind arise in clinical trials with hierarchically ordered objectives (for example, multiple primary/secondary outcome variables,multiple dose-control comparisons, etc). The approach is applied to set up procedures based on popular multiple tests (p-value-based and parametric tests). The resulting procedures have a straightforward multistage structure that facilitates their implementation and communication of the results to non-statisticians. The general approach is illustrated using clinical trial examples.
