56:07
Non-Parametric Hypothesis Tests – 2-Population Median Test (Wilcoxon/ Mann-Whitney)
iTeach
Overview
This video explains non-parametric hypothesis tests for comparing the medians of two populations, focusing on the Wilcoxon signed-rank test for paired samples and the Mann-Whitney U test for independent samples. It details the steps for conducting these tests, including calculating test statistics and interpreting results, particularly for small sample sizes. The video emphasizes the importance of choosing between paired and unpaired tests based on experimental design and the nature of the data, and illustrates the concepts with practical examples. It also touches upon the distribution shapes of these non-parametric test statistics, showing how they approximate a normal distribution with larger sample sizes.
How was this?
Save this permanently with flashcards, quizzes, and AI chat
Chapters
- Non-parametric tests for two populations are used to compare population medians, similar to parametric tests for means.
- Two main types of tests exist: paired (when samples are related, e.g., before/after measurements on the same subject) and unpaired (when samples are independent).
- Paired tests reduce 'noise' by controlling for individual variability, making them more sensitive to the effect being studied.
- Unpaired tests are used when samples are from distinct groups, such as comparing day shift vs. night shift productivity.
Understanding the distinction between paired and unpaired tests is crucial for selecting the correct statistical method, ensuring the validity of your conclusions by accounting for the relationship (or lack thereof) between your data samples.
Comparing employee productivity before and after a training program (paired) versus comparing productivity between day shift and night shift employees (unpaired).
- The Wilcoxon signed-rank test is used for paired samples to test if the median difference between paired observations is zero.
- Steps involve calculating the differences between paired observations, ranking the absolute values of these differences, and summing the ranks of positive and negative differences (W+ and W-).
- The test statistic is typically the smaller of W+ and W-, or a related value.
- For small sample sizes, critical values are looked up in a table; for larger samples, software often uses approximations like the normal distribution.
This test allows you to determine if a treatment or intervention has a significant effect when you have related measurements, such as before-and-after data, without assuming the data follows a normal distribution.
Testing if a training program significantly changes employee performance by comparing performance scores before and after the training for the same group of employees.
- The Mann-Whitney U test (also known as the Wilcoxon rank-sum test) is used for independent samples to compare the medians of two populations.
- It involves combining all observations from both samples, ranking them together, and then calculating the sum of ranks for each sample (V1 and V2).
- The test statistics (U1 and U2) are derived from these rank sums and sample sizes, with the smaller of U1 and U2 typically being the test statistic.
- Similar to Wilcoxon, critical values are used for small samples, and approximations for larger samples.
This test is essential for comparing two independent groups when normality assumptions cannot be met, allowing you to infer differences in central tendency between distinct populations.
Comparing the productivity levels of two different groups of employees: one working the day shift and another working the night shift.
- For both tests, the null hypothesis (H0) is typically that the population medians are equal (or the median difference is zero for paired tests).
- Rejection of H0 depends on comparing the calculated test statistic to a critical value from a distribution table or a p-value to a significance level (alpha).
- Non-parametric tests are robust with small sample sizes but may have less power to detect differences compared to parametric tests when assumptions are met.
- The distribution of the Mann-Whitney U test statistic, especially with larger sample sizes, tends to approximate a normal distribution.
- Simulation can be used to visualize the shapes of these test statistic distributions.
Understanding how to interpret test statistics and p-values, and recognizing the behavior of non-parametric distributions, enables you to draw statistically sound conclusions from your data, even with limited sample sizes or non-normal data.
In the Mann-Whitney U test example, the calculated U statistic (6) was compared to the critical value (1), leading to a failure to reject the null hypothesis, suggesting no significant difference in productivity between day and night shifts.
Key takeaways
- Choose between paired (Wilcoxon signed-rank) and unpaired (Mann-Whitney U) non-parametric tests based on whether your samples are related or independent.
- Non-parametric tests are valuable alternatives when the assumption of data normality is violated.
- The core idea behind these tests is ranking data to assess differences in central tendency, making them less sensitive to outliers than mean-based tests.
- For small sample sizes, non-parametric tests are reliable but may have lower statistical power; larger sample sizes increase power and can lead to distributions that approximate normality.
- The Wilcoxon signed-rank test is a non-parametric equivalent for paired t-tests, and the Mann-Whitney U test is a non-parametric equivalent for independent samples t-tests.
- When interpreting results, compare your calculated test statistic to a critical value or your p-value to your chosen significance level (alpha).
Key terms
Non-parametric testsTwo-population median testPaired samplesUnpaired samplesWilcoxon signed-rank testMann-Whitney U testNull hypothesis (H0)Alternative hypothesis (H1)Test statisticCritical valueP-valueSignificance level (alpha)
Test your understanding
- What is the primary difference in data structure that dictates the choice between the Wilcoxon signed-rank test and the Mann-Whitney U test?
- How does ranking data help non-parametric tests like the Mann-Whitney U test assess differences between populations?
- Why might a researcher choose a non-parametric test over a parametric test for comparing two populations?
- What is the role of the critical value in determining whether to reject the null hypothesis in both the Wilcoxon signed-rank and Mann-Whitney U tests?
- How does sample size influence the power and the distributional properties of non-parametric tests for two populations?