Parametric pitfalls: The overlooked importance of testing for normality in dental research

Respected Editor, I wish to highlight a persistent but under-recognized methodological concern in biomedical research, the widespread use of parametric statistical tests without verifying whether the underlying assumptions, particularly normality of data, are met. Tests such as the independent t-test and one-way ANOVA are commonly employed in medical and life science literature. However, their reliability hinges on certain prerequisites, most notably the assumption that data are normally distributed. In practice, this assumption is often overlooked, especially in studies with small sample sizes where violations of normality can meaningfully distort results. At the same time, it should be recognized that the aforementioned tests are often robust to mild deviations from normality in moderate to large samples with equal group sizes.[1,2] Nonetheless, robustness is not absolute, and smaller or skewed datasets remain vulnerable to biased estimates and inflated error rates, making it critical that assumption checks are still performed and reported. Ghasemi and Zahediasl[3] noted that researchers frequently apply t-tests without checking for normality, a practice that undermines statistical validity. Altman and Bland[4] cautioned that parametric methods may yield misleading conclusions if their assumptions are not satisfied. Notwithstanding this, numerous publications either neglect assumption testing entirely or report it inadequately, as emphasized in a meta-research review by Strasak et al.,[5] which found that nearly 40% of biomedical papers using t-tests failed to report any distribution checks. There are accessible and well-validated methods to assess normality: the Shapiro–Wilk test, Kolmogorov–Smirnov test, Q–Q plots, and histograms, to name a few.[6,7] A concise summary of commonly used tests, along with their recommended applications in Table 1. As recently noted by Habibzadeh,[10] the Shapiro–Wilk test is advised for smaller sample sizes (