Date:______________

The Wilcoxon Rank Sum Test

Introduction

The most commonly used methods for inference about the means of quantitative response variables
assume that the variables in question have Normal distributions in the population(s) from which we
draw our data.

In practice, few variables have true Normal distributions, but our methods have been ____________
(not sensitive to moderate lack of Normality). If the data are clearly not Normal, then using the
methods of the previous chapters will yield inaccurate results. Other approaches must be
investigated.

Options for Non-Normal Data

So what do you do when plots suggest the data are non-Normal?

• If there are outliers, it may be appropriate to ______________________________ if you have

reason to think they do not come from the same population as the other observations.
• It may be that the data follow ____________________________________________________.
If so, use methods based on that distribution.
• Modern _______________________________ and _______________________________ use
heavy computing to avoid requiring Normality or any other specific form of sampling
distribution.
• Finally, there are inference procedures that do not assume any specific form of the
distribution of the population. These are called __________________________________.

Ranking Procedures

The most useful nonparametric significance test compares two distributions. The procedure we will
learn today will replace t tests when the normality condition is not met. Analyses will be based on
the ___________ of the data rather than their actual values.

The idea of rank tests is to just look at position in an ordered list. Moving from the original
observations to their ranks retains only the ordering of the observations and makes no other use of
their numerical values.
The Wilcoxon Rank Sum Test

Hypotheses for Wilcoxon Tests

The Wilcoxon rank sum test compares any two continuous distributions, whether or not they have
the same shape, by testing hypotheses that can be stated as:

Ho: the two distributions are the same

Ha: one has values that are systematically larger

These hypotheses are “nonparametric” because they do not involve any specific parameter such as
the mean or median.

If the two distributions have the same shape, the general hypotheses reduce to comparing medians.

The Normal Approximation for W

To calculate the P-value, we need to know the sampling distribution of the rank sum W when the
null hypothesis is true.

P-values for the Wilcoxon test are often based on the fact that the rank sum statistic W becomes
approximately Normal as the two sample sizes increase.
Example

The test scores shown below were recorded by two different professors for two sections of the same
course. Test to determine if the distributions are identical or not.

Professor 74 78 68 72 77 69 71 67
A
Professor 75 85 87 81 70 73 80 76
B
Practice Problem

The following data displays the reaction times of two sets of people—one group drank alcohol, the
other group drank a placebo. Test to determine if the distributions are identical or not.

Alcohol 1.56 1.76 1.44 1.11 3.07 .98 1.27 2.56 1.32
Placebo .90 .37 1.63 .83 .95 .78 .86 .61 .38