In: Statistics and Probability
What is nonparametric statistics? Provide an examples of a research hypothesis where researchers can use nonarametric statistical techniques.
We perform z-test, t-test, one-way ANOVA etc. in which it is
required for the data to follow normal distribution. These are
called parametric tests.
But in reality, the data you have usually doesn't follow normal
distribution. Hence, non parametric statistics are statistical
methods which are used in which data is not required to follow
normal distribution.
It does not rely on numbers but on a ranking or order (ordinal
data). Hence, when the data has a ranking but no numerical
interpretation, non-parametric methods are often used. These
methods are simple and easier to use than parametric methods
because they make fewer assumptions.
Some of the non-parametric statistical methods are Kendall's W,
Cohen's Kappa, Mann Whitney U (alternative of t-test), sign test
etc.
Consider a dataset of two drugs where we need to find whether
which drug is more effective to increase the sleep of a
person.
The dependent variable we have is the difference in the number of
hours of sleep between Drug 2 and Drug 1.
Ho: The difference in number of hours of sleep is equally likely
to be + or -
Note that in the above hypothesis, we haven't used the numerical
interpretation but we have used the sign of the difference.
The above hypothesis can be tested using non-parametric technique, Wilcoxon signed rank test. If the drugs are equally effective in increasing the sleep of a person, there should be the same number of positives and negatives.