Rankit
|
|
In statistics, the rankits of the data points in a data set consisting simply of a list of scalars are expected values of order statistics of the standard normal distribution corresponding to data points in a manner determined by the order in which the data points appear.
This is perhaps most readily understood by means of an example. If an i.i.d. sample of six items is taken from a normally distributed population with expected value 0 and variance 1 and then sorted into increasing order, the expected values of the resulting order statistics are:
- <math>-1.2816,\ \ -0.64335,\ \ -0.20189,\ \ 0.20189,\ \ 0.64335,\ \ 1.2816<math>
Suppose the numbers in a data set are
- 65, 75, 16, 22, 43, 40.
The corresponding ranks are
- 5, 6, 1, 2, 4, 3,
i.e., the number appearing first is the 5th-smallest, the number appearing second is 6th-smallest, the number appearing third is smallest, the number appearing fourth is 2nd-smallest, etc. One rearranges the expected normal order statistics accordingly, getting the rankits of this data set:
- <math>\begin{matrix}
\mbox{data}\ \mbox{point} & & \mbox{rankit} \\ \\ 65 & & 0.64335 \\ 75 & & 1.2816 \\ 16 & & -1.2816 \\ 22 & & -0.64335 \\ 43 & & 0.20189 \\ 40 & & -0.20189\end{matrix}<math>
A graph plotting the rankits on the horizontal axis and the data points on the vertical axis is a rankit plot or normal probability plot. Such a plot is necessarily nondecreasing. In large samples from a normally distributed population, such a plot will approximate a straight line. Substantial deviations from straightness are considered evidence against normality of the distribution.
The word rankit was introduced by the statistician Chester Bliss (not to be confused with the politician Chester Bliss Bowles).