Mann-Whitney U-Test

statisticslectures.com - where you can find free lectures, videos, and exercises, as well as get your questions answered on our forums!

Category:

Education

Tags:

• mann-whitney u
• null hypothesis
• alternative hypothesis
• decision rule
• test statistic
• results
• conclusion

[jacksite2007]

You say it like DAY-ta americans! Say it right xD

[statslectures]

@jacksite2007

haha, I think I switch between both ways depending on which video you're watching. I can't make up my mind!

[merdirafiei]

go to the part where you average two data points' ranks (4+5)/2=4.5. Where else in the video do you use this information to calculate the U statistic? You do calculate the ranks, but you never use the information to calculate the U. Maybe I'm missing something because your example is too simplistic in that data sets A and B are strictly disjoint in that all a(i) is bigger (or smaller- I forget which) than b(i)

[statslectures]

@merdirafiei

You are right about that second part, I could have used a better example for this. Points are given to each score from the B sample for all the A sample scores that are ranked above it, and vice-versa. The sum of points for both the B sample scores and the A sample scores are then found. The smaller of those two values becomes our "U". Let me know if it's still not clear.

[merdirafiei]

My original question stands: where in the video do you use the ranks to calculate the z statistic? Nowhere, right?

[statslectures]

@merdirafiei

The "U" statistic we calculate is used in calculating the z score. We can't find the U score without first finding the ranks.

[crocodilerachel]

are you able to use this test if your two samples (A and B) each have a different n number? for example, i have 18 data points in my A group and 11 in my B group

[mysoukouss]

Thanks for this review. Very helpful.'

God bless you Bother.

[angela1894]

What are the assumptions on the data that must hold for Mann-Whitney to be usable ?

[statslectures]

@merdirafiei

The "U" statistic we calculate is used in calculating the z score. We calculate ranks because we're dealing with ordinal data - that is, data that has order, but no set difference between values. Look up "variable measurement scales" for an explanation that's a little bit better than that.