Question

In general, when should you use non-parametric vs. parametric tests?

Answer 1

A non-parametric test or statistics refer to a statistical method where the data is not based on a normal distribution or any other assumptions on the data. These are often ordinal, i.e. they are generally grouped, ranked or ordered of sorts. It is easier to use as compared to parametric tests. Examples Mann-Whitney, Kruskal–Wallis, Friedman, etc. Non-parametric tests are used when:

• The parameters of the sample population is unknown.
• When the population size that is being dealt with is small.
• When the study is better represented by the use of median rather than the use of mean.
• When there are outliers, ordinal data or ranked data present in the sample that cannot be removed.
• When there are no assumptions made regarding the population.

A parametric test or statistics refer to a statistical method where the data is based on certain conditions, like a normal distribution or based on validity of assumptions on the data. These are often in nominal form, i.e. they are absolute numbers or values rather than ranks.These are trustworthy measures due to their computed samples and power of testing. Examples t-tests, z-tests, ANOVA, etc. Parametric tests are used when:

• The parameters of the sample population is known. Data is continuous, skewed and non-normal.
• When the study is better represented by the use of mean, variance or proportion (depending on population size).
• When the spread or dispersion of each group is different.
• When there are assumptions made regarding the population like normal distribution, intervals, equal or same variance, etc.
• When different samples need to be used, with each having normally distributed data.