BUSI 820 Week 3 Discussion Forum 3 Main Thread Example

Discussion Forum 3: Main Thread and Forum 2 Reply

BUSI 820 Quantitative Research Methods, Discussion Forum 3 Main Thread (with the Forum 2 reply)

Thread by [Student Name]

The study, the classmate and the prompt are composites standing in for the course's own; the books and the published research are real; the week, points and readings follow Liberty's public course guide for BUSI 820, which keeps each forum's detailed instructions inside the course.

My turnover study has four research questions, and Morgan et al. (2020) offer a consistent way to choose a statistic for each. For every question, I identified the number of independent variables, whether it compares groups or relates variables to one another, and the level of measurement of the variables. Table 1 summarizes the choices.

Two choices deserve comment. For RQ1, the three groups are unequal in size, 115, 44 and 27, so Levene's test for equal variances will matter; if it is significant, I will use the Welch version of the test. For RQ4, work arrangement is nominal and must enter the regression as two dummy variables, with on-site staff as the reference group. Before running it, I will check that the predictors are not so highly correlated with one another that their separate contributions cannot be estimated, since branch size and work arrangement may overlap; larger branches are more likely to offer hybrid schedules. If the tolerance values Morgan and colleagues describe fall too low, I will drop or combine predictors rather than report unstable coefficients.

The second half of chapter 6 changes how I will report results. A statistically significant result tells us that a difference or relationship this large would be unlikely if there were no effect in the population, but it is silent on whether the effect is big enough to act on. With 186 respondents, even a small correlation can reach significance. A bank deciding whether to retrain its supervisors needs to know how much support matters, not merely that it matters. I will therefore report an effect size with every test: r for the correlation, d for the t test, eta squared for the ANOVA and R squared for the regression. Cohen (1988) proposed benchmarks, such as .10, .30 and .50 for small, medium and large correlations, but he also warned that they should give way to judgment about what is meaningful in a particular field. For our bank, a medium effect of supervisor support would justify a training investment; a small one might not.

Part III of Keller and Alsdorf (2012) argues that the gospel gives work a new compass, including honesty. In statistics, that compass rules out reporting only the significant results or dressing a trivial effect in the language of significance.

Reply to Ahmed (Forum 2)

Ahmed, your descriptive statistics for the warehouse safety survey were clear, and your table made the results easy to read. One choice may need changing. You reported a mean of 2.4 for supervisors' ranking of five safety priorities, where 1 means most important. Because ranks are ordinal and the distances between them are unknown, Morgan et al. (2020) would recommend reporting the median and the frequency of each rank rather than the mean. The same issue will matter in your Forum 3 thread: if you plan to relate the ranking to injury counts, a Spearman rank-order correlation fits your data better than a Pearson correlation. Would the survey let you report how many supervisors ranked each priority first?

References

Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Lawrence Erlbaum Associates.

Keller, T., & Alsdorf, K. L. (2012). Every good endeavor: Connecting your work to God's work. Dutton.

Morgan, G. A., Barrett, K. C., Leech, N. L., & Gloeckner, G. W. (2020). IBM SPSS for introductory statistics: Use and interpretation (6th ed.). Routledge.