RESPONSE TO ANOTHER STUDENT DISCUSSION-YOUR EXPERT ADVICE

Robert Laukaitis

Top of Form

U08D1 – Expert Advice

In the hypothetical situation in which a researcher proposes using multiple ANOVA to evaluate effects of three training methods on two outcome variables, it would be reasonable to assume the researcher had experience using the ANOVA approach. Therefore, the situation suggests that the researcher has screened the data to ensure data used has met the assumptions of the ANOVA: the normality of the data in the outcome variables, homogeneity of variance of the data collected for use the outcome variables, that the sample size is sufficient to produce sufficient statistical power from which assumptions and inferrences might be drawn and that there exists an independence of observations for each case in the study (Warner, 2013).

However, doing so could result in a lack of detection of intercorrelations between variables and an increase in the potential for a Type I error when reaching inferences (George & Mallery, 2013; Warner, 2013). Given the suggestion that the desired outcome variables were found to be highly correlated (r = .4), Warner (2013) suggested that when several correlated dependent variables are known to exist, the omnibus test of the MANOVA is appropriate for use. This could help the researcher determine if there is any influence of the different types of training on each other while detecting their influence on the outcome variables. In addition, exploration of the three types of training (i.e., a traditional model, a computer-based model, and a video-based model) using the MANOVA approach can help reduce compounding errors that might be created when using multiple statistical tests not optimized for multiple use on themselves.

Next, the researcher might also benefit from exploring how any independent variable might influence the vector of the dependent variables in terms of response patterns (George & Mallery, IBM statistics 21 step by step: A simple guide and reference, 2013; Field, 2009). While the ANOVA approach provides an omnibus test to determine the equality of means for the groups being tested, the approach lacks the robustness to detect differentiation of the groups themselves. However, if the MANOVA is not robust to detection of variables responsible for differences in means, contrast must be used to detect dependent variable sensitivity to independent variable groups (George & Mallery, 2013; Warner, 2013). Lastly, the MANOVA is not subject to the assumption of sphericity as is the ANOVA. This helps the researcher is not subject to the assumption that the correlation matrix is not an identity matrix and therefore sensitive to the behavior of the variables considered in the multivariate construct.

Thus, in order to help the researcher detect the affect of three training approaches on math anxiety and public speaking, while reducing the probability of Type I errors over individual ANOVA being conducted and compounded, the MANOVA could yield more robust detection of significance. If used with post hoc tests such as Wilks' Lambda, Pillai's Trace, Hotelling-Lawley Trace and Roy's Greatest Root, potential areas for which the researcher would need to validate statistical power and research validity could be reduced.

References

Field, A. (2009). Discovering statistics using SPSS (3rd ed.). Thousand Oaks, CA: Sage.

George, D., & Mallery, P. (2013). IBM statistics 21 step by step: A simple guide and reference (13th ed.). Pearson. Retrieved from http://online.vitalsource.com/books/9781269627795

Warner, R. M. (2013). Applied statistics: From bivariate through multivariate techniques (2nd ed.). Thousand Oaks, CA: Sage. Retrieved from http://online.vitalsource.com/books/9781452268705

Bottom of Form