In this guide, I will explain how to perform a Mann-Whitney U test in GraphPad Prism. I will also show you how to interpret and report the results.

# Assumptions of a Mann-Whitney U test

Before performing the test, it is important to check that your data satisfies the assumptions of a Mann-Whitney U test. The assumptions are:

- The variables of interest contain continuous or ordinal data.
- There should be two groups of data present.
- Each data point should be independent of each other.
- The distribution and shape of the data in the two groups should be similar.

# Setting up the GraphPad sheet

To perform a Mann-Whitney U test in GraphPad Prism you will need to enter two groups of data into separate columns. Upon opening GraphPad Prism, select the ‘**Column**’ type for the ‘**New Table & Graph**’ option. Then select ‘**Enter replicate values, stacked into columns**’ as the ‘**Enter/import data**’ choice.

Alternatively, you can go to ‘**File > New > New Data Table & Graph …**’.

# Example dataset

I will use a theoretical example of comparing the concentrations of a certain protein between a control and treated group of individuals. In total, I have 20 data points for each group. Each group’s data is separated into different columns.

The null hypothesis for this example will be:

**“There is no difference in protein levels between the control and treated group”.**

And the alternative hypothesis will be:

**“There is a difference in protein levels between the control and treated group”.**

## Performing the test

- To perform a Mann-Whitney U test in GraphPad Prism, first, go to ‘
**Insert > New Analysis …**’.

2. This will open a new window. Here you need to tell GraphPad which test to perform. Select the ‘**t-tests (and nonparametric tests)**’ analysis and make sure the two datasets are ticked on the right window. Click the ‘**OK**’ button.

3. The next window will ask you to specify which test to perform. In this case, select the ‘**unpaired**’ option, under the ‘**Experimental design**‘ header. Next, select ‘**No. Use nonparametric test**’ under the ‘**Assume Gaussian distribution**’ header below. Finally, ensure the ‘**Mann-Whitney test. Compare ranks**‘ is selected under the ‘**Choose test**‘ header.

4. If you want to change any other settings, such as the confidence level, go to the ‘**Options**’ tab. For our purpose, we are going to leave everything in the default settings but will select ‘**Descriptive statistics for each data set**’ so GraphPad will produce the mean and standard deviation for each group. Click ‘**OK**’ to perform the test.

# Output

The great thing about GraphPad Prism for statistical testing is that the output is very user-friendly and self-explanatory. All you need to do is to refer to the ‘**Tabular results**’ sheet of the ‘**Mann-Whitney test of Data 1**‘ option on the left-hand side.

Focus on the output under the Mann Whitney test header. Each output is as follows:

**P value**– The P value (of course!).**Exact or approximate**P value – Whether the P value was approximated or is exact.**P value summary**– A summary of the p-value as represented by asterisks. These are useful to signify the level of significance on graphs, for example.**Significantly different? (P<0.05)**– A straight yes or no answer to was there any significance.**One- or two-tailed P value?**– Whether a one- or two-tailed test was performed.**Sum of ranks in column A,B**– The sum of ranks for the two groups in the test.**Mann-Whitney U**– The U statistic.

Since we also selected the descriptive statistics to be reported, you can also find further information about each dataset in the test in the ‘**Descriptive statistics**‘ results sheet. This will be useful when writing the results since we need the median, 25th percentile and 75th percentile values.

# Interpretation

By looking at the ‘**Significantly different? (P<0.05)**’ output, a ‘**No**’ is given which essentially means that our results are not significantly different from each other. The exact P value is given next to ‘**P value**’. In this case, a P value of 0.0798 is obviously larger than our level of significance threshold of 0.05.

Therefore, we accept the null hypothesis and reject the alternative hypothesis.

# Reporting

If we were to report the results of this test in a paper, for example, we could write the following summary.

This includes the median, 25th percentile and 75th percentile (this information can be found in the ‘**Descriptive statistics**’ sheet) for each group, as well as the U value and the P value.

GraphPad Prism version used: 6