CHOICE OF TESTS
Nonparametric Statistics
When calculating the random distribution of some variable it is
easy to visualize the values as clustered around some center value in the well-known
bell-shaped (termed "normal") distribution. However, as noted earlier, not all data
have such a nice distribution. Data that are not in interval format, such as nominal
data (e.g., red, black; male, female) or ordinal data (first, third), cannot be described
by mathematical distributions with parameters such as the mean or SD. Such a situation
calls for nonparametric statistics. Nonparametric statistical methods are typically
used for the analysis of nominal or ordinal data. Data that are distributed in a
very atypical skewed fashion (a "non-normal" distribution) will not be properly handled
by analyses that have the mathematical assumptions of normality; such data will need
nonparametric analysis.
The first step in considering the choice of a statistical test
is to decide whether statistical methods that assume a normal distribution are appropriate
or whether nonparametric methods are needed ( Table
23-3
). A test for normality is included in typical statistical computer
packages.
