The Wilcoxon signed-rank test & the sign test

Рет қаралды 3,636

Күн бұрын

In this video, we will have a look at two non-parametric tests for paired data.
1. The Wilcoxon signed-rank test
2. The one-sample Wilcoxon test (07:00)
3. The sign test (07:37)
For more videos in a logical order, go to:
www.tilestats.com

Пікірлер: 3

@idgsfree6009 Жыл бұрын

So Wilcoxon signed-rank would be most appropriate when we have fat-tailed (but symmetric) distribution right? Also is the following statement correct: "We shouldn't use Wilcoxon signed-rank test for ordinal data because ranking the difference of the ordinal data could lead to inappropriate results especially when the distance between each category is different."

@tilestats Жыл бұрын

You do not need a fat-tailed. All distributions that are symmetric will do. For example, the uniform distribution works. Yes, it is when you have different distances between the categories that it becomes problematic to use this test since you then cannot convert the categories to numbers. This is why it is recommended to use the sign-test for ordinal scale when you have paired data.

@idgsfree6009 Жыл бұрын

@@tilestats Thanks for the answer! A few days ago I couldn't think of any distribution which is symmetric but not normal, so I was wondering when we should use signed-rank test instead of t-test. But now I realized there are some non-normal symmetrical distribution like uniform or bimodal distribution.