Mann Whitney U Test

Text-only Preview

  1. Comparing medians: the Man Whitney U-test The Mann Whitney U-test is a fairly complicated statistical test to understand, though it is quite easy to apply to a set of data. So, while the calculation is relatively easy, knowing when to apply it, and what the calculation actually means, is a little more difficult. It is also important not to be put off by the formula.
    • What is the Mann Whitney U-test?
    • The Mann Whitney U-test is a nonparametric test which is used to analyse the difference between the medians of two data sets. By using the critical values tables, it is possible to assess the degree to which any observed difference is a result of chance or fluke.
    • If the answer to each of the following questions is ‘yes’ then you may use the Mann Whitney U-test.
    • Are you investigating the difference between two samples of data?
    • Is the data nonparametric?
    • Is the data ordinal ?
    • Are there more than five pieces of data in each sample?
    • Are there 20 or fewer pieces of data in each sample (recommended)?
    KEY TERMS Nonparametric test : statistical test that assumes that the data is not normally distributed. Ordinal data : data that can be ranked, i.e. put into order from highest to lowest
  2. Though this is a nonparametric statistical test, both samples should have a similar distribution. You can plot the data for each set on a simple graph to check this. Like many of the other statistical tests, you have to start with a null hypothesis (H o). However, unlike some of the other tests, the null hypothesis (H o) is always the same: There is no significant difference between the two samples.
  3. Applying the Mann Whitney U-test Comparing two traffic flows in a town centre A student was interested in finding out if a new retail development had an impact upon traffic (and therefore congestion) in the local area near to the development. There were two parts to the primary data collection. The first part was conducted before the construction of the planned development (sample x). Methodology for primary data collection She recorded the time of day and date. She counted traffic (in both directions) on 10 streets around the development selected randomly). She counted for 10 minutes. She used a stopwatch for timing and a simple tally chart for recording the data. She completed the tally at different times of the day. TIP Is it a one-tailed or a two-tailed test? This relates to the difference between the data sets. If you assume one specified data set will be larger than the other, you are investigating a one-tailed distribution. If you assume differences can operate in both directions, i.e. up or down, you are investigating a two-tailed distribution. This is important when you interpret you findings using the critical values. In this example, the student is assuming traffic can go up or down in study 2 for all sites, despite the fact that more customers are likely to be attracted to the development. In this case, this makes it a two-tailed test
  4. For the second study (sample y), she waited until 2 months after the development had been completed. She went to another 10 sites (selected randomly) and repeated the test. She then devised the following null hypothesis (H ?): ‘ There is no significant difference in traffic flows before and after the development.’ Now let’s take a look at the formula: U ? = N?.N ? + N ?(N? + 1) 2 - ? r? U ? is the Mann Whitney calculation for sample x n is the number in the sample ? r? is the sum of ranks for sample x (‘sum of’ just means added together)
  5. The best way to proceed is to incorporate the findings into a table that also allows you to calculate the result. When you get two or more equal values, use the mean rank. Here are the student’s findings: Complete the table by ranking all the data from highest to lowest. Total traffic flow in 10 minutes ( ? ) Rank r ? Site Number Total traffic flow in 10 minutes ( ? ) Rank r ? 126 11 1 194 148 7 2 128 85 15.5 3 69 61 19 4 135 179 4 5 171 93 12.5 6 149 45 20 7 89 189 3 8 248 1 85 15.5 9 79 93 12.5 10 137
  6. ? r? = 120 ?r ? = Ranking puts values in order from highest to lowest . Next, she substituted the data into the formula : U ? = 10x10 + U? = 100 + U? = 100 + 55 – 120 U? = 35
    • Summary
    • Mann Whitney U-test can be used to compare any two data sets that are not normally distributed . As long as the data is capable of being ranked, then the test can be applied.
    • Other possible uses:
    • Investigating differences in questionnaire responses relating to a new development.
    • Investigating differences in species diversity near to footpaths.
    • Investigating differences in vegetation cover between two different slopes
    U ? = N?.N ? + N ?(N? + 1) 2 - ? r? 10(10 + 1) 2 - 120 110 2 - 120 TIP There is a useful way of checking the accuracy of your calculations. U ? + U ? should be the same number as N?.N ? (which in this case is 10x10 = 100). If it is not, you have made a mistake somewhere
  7. TASKS
    • You now have the figure for U ?. Use the same method of calculation to work out the figure for U ? . Here is the formula
    • You now need to select the smaller of the two figures and use the critical values table to decide on the statistical significance of your result. For this test, if your result equal to or smaller than the critical value as the 0.05 level of significance, then you can reject the null hypothesis (H?)
    • For 10 figures in each sample, the critical value at the 0.05 level of significance is 23.
    • Do you accept or reject the null hypothesis (H?)? Give reasons for your answer.
    • What do these findings suggest about the difference in traffic flow before and after the retail development in this study?
    U ? = N?.N ? + N ?(N ? + 1) 2 - ? r ?