Problem 6: Failure to check relevan

Problem 6: Failure to check relevant assumptions of a statistical model led to incorrect inferences.Example: A student was analyzing data on angular rotation of the head across four different conditions in both healthy controls and individuals with mild traumatic brain injury. Using a repeated-measures ANOVA, the student found that the main effect of condition was statistically significant (P =.032). While the manuscript was in preparation, a co-author asked to see some of the raw data, at which point the student and their advisor realized that the variance in angular rotation in two of the conditions was about three times larger than in the other conditions, and angular rotation also had a positive skew. The data violated two assumptions of the ANOVA: homogeneity of variance (more specifically sphericity in this case) and normality of the residuals. Violations of sphericity can lead to inflated Type I error rates. When a non-parametric test was conducted on the data, the resulting P value was much higher (P=.24).Solution: Researchers should fully understand the variables in their dataset before running any formal statistical models and tests. Descriptive statistics and distributional plots, such as histograms, should be generated for all relevant variables. Researchers should also test the assumptions of any formal statistical tests that they plan to run to determine whether alternative statistical approaches are required.