Non-parametric

Non-parametric refers to a class of statistical methods that use no or few assumptions about the distribution of characteristic(s) in the underlying population from which the data were drawn. Common non-parametric tests are Wilcoxon, Mann-Whitney and Kruskal-Wallis. Reasons for using non-parametric tests include that the sample size is very small, the data are ordinal/ranked in nature, or that there is a skewed distribution (e.g. survival, income) with extreme outliers (long ‘tail’) and a summary statistic such as median may be of more value than a mean. Non-parametric tests generally have less power (i.e. will require larger sample sizes) than the corresponding parametric tests if the data are truly normal, and interpretation of the results of such procedures can also be more difficult. Generally, non-parametric methods are of limited use for economic evaluation, where the focus is more on estimation (to support decision making) than on hypothesis testing. Nevertheless, bootstrapping is a useful non-parametric technique, and direct use of (Kaplan-Meier) survival data from source studies to estimate survival of a modelled cohort may also be considered to be non-parametric, in contrast to the use of parametric functions.