Computer-Generated Data Sets Can Demonstrate Real Statistical Pitfalls, Misleading Conclusions

Most students take marketing research courses only after passing a course in statistics. During the statistics course, the student is exposed to such topics as probability theory, estimation, and elementary decision theory. The greatest concentration of effort seems to be on tests of significance or confirmatory data analysis. Unfortunately, the data analyzed usually are either artificial or trivial. Also, students are often led to believe that for any given set of data, statistical analysis is a one-step procedure involving the application of only one accurate technique. However, consider a marketing research course in which the instructor has already covered problem definition and formulation, experimental design, and data collection. Students should now deal with data derived from realistic problems. Among the most important statistical techniques taught in marketing research are regression analysis, analysis of variance, and discriminant, factor, and contingency table analysis. The data generated should be tailored to the level of the students in the course. As the course progresses, the instructor can incorporate into the data errors of the type that are found in real-world practice. In analyzing generated data, students become familiar with statistical computer packages and the techniques of exploratory and confirmatory data analysis. However, there are advantages and disadvantages to both real and simulated data.