Analysis of ordinal dental data: evaluation of conflicting recommendations.

Although recommendations for the appropriate analysis of non-normal and ordinal-scaled data have appeared in the dental research literature for many years, there is no consensus. When one is conducting statistical tests for differences between groups, the central concern is whether it is safe to use parametric tests (e.g., analysis of variance), or if only non-parametric ranking tests should be considered. Relevant statistical and scientific issues associated with non-normality and measurement scale are reviewed, and three conclusions are reached regarding the analysis of dental data: (1) Parametric tests are sufficiently robust relative to typical violations of normality; (2) presumed statistical prohibitions against the application of parametric methods to ordinal data do not actually exist; and (3) 'ordinal' dental indices have sufficient quantitative meaning to be considered quasi-interval. For these reasons, parametric tests should not be avoided; they will be valid and usually more powerful and more easily applied to complex designs than non-parametric alternatives.