Up to now I have used the Kolmogorov-Smirnov test for comparing various groups of ordinal data, in surveys with answers/opinions given between 1 and 5 [completely disagree ... up to ... completely agree] and grades given between 1 and 10. Visual inspection showed that the data was not normally distributed. According to the literature I consulted, the Kolmogorov-Smirnov test requires both groups to have the same size in case the size of each group is =40, size equality is not mandatory. I calculated everything using a self-made test-bank in Excel 🙂. DATAtab suggests usage of the Mann Whitney U-test in these cases, at least, that is the only option DATAtab provides if I upload my data to DATAtab. What would be the advantage of MW-U test versus the KS test?
Thank you for the tutorial video. This explains much better than the study notes in my course. Just to highlight at the 6:00 mark, that there is a typo error for the word "Rang", which is "Rank".
12 Simply derives from the formula and is independent of the sample size. There is already 12 in the formula without the values being inserted. Therefore 12 is a fixed number.
Using your great tool DATAtab, I applied the one-tailed Mann-Whitney U Test to one of my data sets. It is unfortunate that the p-value apparently is not shown in the statistics overview in the hypothesis test calculator. It seems that I separately have to use the z-distribution tutorial for calculating the p-value. I do see a value for Asymptotic Significance (1-tailed) and Exact significance (1-tailed) however. How do these two values relate to the p-value?
Hi, many thanks for you r comment! Both are the p-Values, once calculatetd with Asymptotic Significance and once the exact! Therfore: significance = p-Value
Nice video! But could I ask what to do if several data points are tied and have the same rank? Because of that my test is yielding just slightly too high p values, what should I do by hand to resolve this? Thanks!
Hi Matthew. Imagine the reaction times for group A={3, 5, 5} and for group B={5, 5, 8}. If we sort all the data in ascending order (3, 5, 5, 5, 5, 8), we note that the corresponding ranks appear to be (1, 2, 3, 4, 5, 6). However, we are assigning to 5 four different ranks, namely, {2,3,4,5}. Hence, we take the average, i.e., Rank(5) = (sum of ranks) / (# of repetitions) = (2+3+4+5)/4 = 3.5. Consequently, the vector of ranks is (1, 3.5, 3.5, 3.5, 3.5, 6), the ranks for A are Rank(A) = {1, 3.5, 3.5}, and Rank(B) = {3.5, 3.5, 6}. The corresponding rank sums are T_A = 8 and T_B = 13. The U statistic is 2, and the corresponding p-value is 0.3017, which implies that we cannot reject the null hypothesis that the two groups are statistically different.
Not clear how you start with non parametric testing and end up assuming normal distribution to pick the p-value from a table. Perhaps it relies on the central limit theorem but in the examples given, the sets are so small that it cannot be postulated. I understand that the idea is to illustrate the concepts, and this is very well done as usual, but I wish there was also a more detailed version of these videos.
9:06 in order to calculate the standard error of u, do you always need to divide by 12 in the formula? Or is it because 12 is n1+n2+1 Thank you in advance
In Datatab, I very frequently use the Mann-Whitney U-test. Now I wonder how to correctly interpret the measured effect size r. For the t-test, a table is shown in Datatab, for help on interpreting the value of d (the t-test effect size). Is there a comparable table for r (the effect size of the Mann-Whitney U-test)?
@datatab I performed Mann Whitney U test and achieved statistical significance however my median score is the same. How do I intepret and explain this?
can i use this statistical treatment if I want to interpret the data of the simulated short circuit currents that are connected by electrical buses? I want to compare two electrical systems with its dependent variable being the short circuit current
Hello, the distributions of the samples should be about the same, how quickly this becomes a problem or how robust the method is against it I can not tell you unfortunately! I would also have to google. Regards, Hannah
Dear madam, I compared two independent groups (health science and non-health science students) by using the Mann-Whitney test, but the results showed significant with same median, is there any way to interpret this result or recommend a test that could replace it?
The z stat from the z test is dervied from the CLT and it allows for probabilstic inferences. Where does the U statistic come from and why is it defined this way?
Hello, I have a question: if I only have two biological samples, namely animal A and B, and they are in condition A and B. For each animal, I have 100 neurons records. In this case, what is the meaning if I conduct Mann-Whitney test on A and B? It seems I cannot make any conclusion for conditions A and B since there is only one sample.
Great video! Is this test valid for proportions (0 -> 0.72)? All my data points are average percent cover of seagrass with about 53% of the data being zeros across 8 group levels (4 years x 2 seasons per year). I'd like to see if the median percentage is the same between dry and wet season, for each year (n=47 values (1 value per site) per season).
Hello, thanks for your question. There are 6 female participants therefore 6 and there are 5 male participants therefore 5. I hope that helped! Regards, Hannah
Dear Madam, I am comparing two groups in terms of different research aspects using the Mann-Whitney test. But I want to add a categorical covariate (discipline) How do I take it into account with this test as a covariate? thank you
Nice video! in case of evaluation of percentage data (0-100%) as the outcome measured between two groups, could Mann Whitney U be used? e.g. dry matter content (%) in feed samples of two different kinds of diets are compared to see if the diets differ in dry matter content. Assuming assumptions are otherwise met for Mann Whitney U.
Dear, u r mixing groups with variables. We don’t classify groups as nominal as all groups are so, we classify variables under these groups as nominal or quantitative. If quantitative then we describe their distribution as normal or not.
@@datatab plz any teacher of statistics that i can contact? I having trouble in some questions and i am not able to find anyone from web plz if possible can someone help me? Its about probability and statistics i can send some questions that i am not able to do
We use the non-parametric test because the sample is not normally distributed but then use a normal distribution to calculate the p-value? It strikes me although maybe these are independent things. The explanation was great!
If you like, please find our e-Book here: datatab.net/statistics-book 😎
Sure thanks
h
I love you! This is what I exactly need for my tomorrows presentation! Thank you so much for simplified explanation :)
Gread! if you like please share the video : ) Regards Hannah
I wanted to say I love you to this channel too! I scrolled down to type this and You beat me to it :D
Up to now I have used the Kolmogorov-Smirnov test for comparing various groups of ordinal data, in surveys with answers/opinions given between 1 and 5 [completely disagree ... up to ... completely agree] and grades given between 1 and 10. Visual inspection showed that the data was not normally distributed. According to the literature I consulted, the Kolmogorov-Smirnov test requires both groups to have the same size in case the size of each group is =40, size equality is not mandatory. I calculated everything using a self-made test-bank in Excel 🙂. DATAtab suggests usage of the Mann Whitney U-test in these cases, at least, that is the only option DATAtab provides if I upload my data to DATAtab. What would be the advantage of MW-U test versus the KS test?
Thank you for the tutorial video. This explains much better than the study notes in my course. Just to highlight at the 6:00 mark, that there is a typo error for the word "Rang", which is "Rank".
Oh many many thanks!!!! Regards, Hannah
Fantastic explanation. Helped me understand exactly what I was doing in my computational data science class. Also, I love your accent :)
Thanks : )
I love your teaching. Easy to understand. Keep it up! and thank you!
Glad you liked it and many thanks for the nice feedback! Regards Hannah
@@datatab Preferably, one should give reasons for those formulas. Otherwise it is not educative, just encyclopedic.
Wonderfully explained with visuals. Excellent!
Many thanks! Regards, Hannah
God bless your soul I was so so confused by this and your explanation and visuals were so helpful! thank you!
Glad it was helpful!
The more of these videos I watch, the more I believe I should take a six sigma yellow belt course.
Perfect : ) Glad you like them! Regards Hannah
Thank you very much for your time and teaching, Madam.
Welcome!
You literally saved my life right now
I have a lot of ties in my dataset. Does DATATAB take a correction for ties into account when calculating the p-value?
Sorry for the late answer! Yes it handels ties!
@@datatab that's great!
for standard error of U where did the number 12 come from? there were 11 numbers in the total set so I wasn't sure.
12 Simply derives from the formula and is independent of the sample size. There is already 12 in the formula without the values being inserted. Therefore 12 is a fixed number.
Using your great tool DATAtab, I applied the one-tailed Mann-Whitney U Test to one of my data sets. It is unfortunate that the p-value apparently is not shown in the statistics overview in the hypothesis test calculator. It seems that I separately have to use the z-distribution tutorial for calculating the p-value. I do see a value for Asymptotic Significance (1-tailed) and Exact significance (1-tailed) however. How do these two values relate to the p-value?
Hi, many thanks for you r comment! Both are the p-Values, once calculatetd with Asymptotic Significance and once the exact! Therfore: significance = p-Value
@@datatab thank you! 👍
Brilliantly explained. Thanks ❤
Thank you for the intuitive explanation!
Thanks for your feedback! Regards, Mathias
Muchas gracias from Seattle Wa.
Nice video! But could I ask what to do if several data points are tied and have the same rank? Because of that my test is yielding just slightly too high p values, what should I do by hand to resolve this? Thanks!
Hi Matthew. Imagine the reaction times for group A={3, 5, 5} and for group B={5, 5, 8}. If we sort all the data in ascending order (3, 5, 5, 5, 5, 8), we note that the corresponding ranks appear to be (1, 2, 3, 4, 5, 6). However, we are assigning to 5 four different ranks, namely, {2,3,4,5}. Hence, we take the average, i.e., Rank(5) = (sum of ranks) / (# of repetitions) = (2+3+4+5)/4 = 3.5. Consequently, the vector of ranks is (1, 3.5, 3.5, 3.5, 3.5, 6), the ranks for A are Rank(A) = {1, 3.5, 3.5}, and Rank(B) = {3.5, 3.5, 6}. The corresponding rank sums are T_A = 8 and T_B = 13. The U statistic is 2, and the corresponding p-value is 0.3017, which implies that we cannot reject the null hypothesis that the two groups are statistically different.
Not clear how you start with non parametric testing and end up assuming normal distribution to pick the p-value from a table. Perhaps it relies on the central limit theorem but in the examples given, the sets are so small that it cannot be postulated. I understand that the idea is to illustrate the concepts, and this is very well done as usual, but I wish there was also a more detailed version of these videos.
9:06 in order to calculate the standard error of u, do you always need to divide by 12 in the formula? Or is it because 12 is n1+n2+1
Thank you in advance
In Datatab, I very frequently use the Mann-Whitney U-test. Now I wonder how to correctly interpret the measured effect size r. For the t-test, a table is shown in Datatab, for help on interpreting the value of d (the t-test effect size).
Is there a comparable table for r (the effect size of the Mann-Whitney U-test)?
This video is very clear and very easy to understand! thx
@datatab I performed Mann Whitney U test and achieved statistical significance however my median score is the same. How do I intepret and explain this?
Hi, is Mann-Whitney U Test the same as Mann-Whitney-Wilcoxon Test?
Sorry for the late Answer, Yes it is!
Excellent explanation!!! Thank you!!!!
Glad you liked it and many thanks for the nice feedback! Regards Hannah
Hi, wonderful video but i have a question, at 10:35, how do you translate the z value to a p value? What happens? What is the formula? Thanks!
you use the z distribution! datatab.net/tutorial/z-distribution
can i use this statistical treatment if I want to interpret the data of the simulated short circuit currents that are connected by electrical buses? I want to compare two electrical systems with its dependent variable being the short circuit current
Is this method robust when variances are unequal? Thanks for the video!
Hello, the distributions of the samples should be about the same, how quickly this becomes a problem or how robust the method is against it I can not tell you unfortunately! I would also have to google. Regards, Hannah
Dear madam, I compared two independent groups (health science and non-health science students) by using the Mann-Whitney test, but the results showed significant with same median, is there any way to interpret this result or recommend a test that could replace it?
The z stat from the z test is dervied from the CLT and it allows for probabilstic inferences. Where does the U statistic come from and why is it defined this way?
Many thanks for this lecture. Perfect!
Many thanks for the feedback and for watching! Regards Hannah & Mathias
Hello, I have a question: if I only have two biological samples, namely animal A and B, and they are in condition A and B. For each animal, I have 100 neurons records. In this case, what is the meaning if I conduct Mann-Whitney test on A and B? It seems I cannot make any conclusion for conditions A and B since there is only one sample.
Is Mann-Whitney U Test a correlational test?
No, not that I know of! Hannah
Can this test be used if there are some equal rank?
Yes please have a loo at our home page at the tutorial for the Mann Whitney U test! Regards Hannah
Great video! Is this test valid for proportions (0 -> 0.72)? All my data points are average percent cover of seagrass with about 53% of the data being zeros across 8 group levels (4 years x 2 seasons per year). I'd like to see if the median percentage is the same between dry and wet season, for each year (n=47 values (1 value per site) per season).
Oh sorry, that I can not answer you in a hurry unfortunately!
@@datatab That's ok, thanks anyway
Why does she continue with the z value and all the other videos on youtube I've seen about this test just finish when they get the U value.
What happens you have two values the same, how do you rank?
You use tied ranks datatab.net/tutorial/mann-whitney-u-test
@@datatab Super, thx.
What if two measures have the same value? How do you then rank them?
Please find the answer here if you scroll down a bit: datatab.net/tutorial/mann-whitney-u-test
Help! At 8 minutes, where does the 6.5 (n1 and n2) come from?
Hello, thanks for your question. There are 6 female participants therefore 6 and there are 5 male participants therefore 5. I hope that helped! Regards, Hannah
@@datatab thanks! I didn’t realize the dot was a ‘point’. So if there were 4 male participants and 6 female participants it would be 6.4?
@@lisafuckinlonergan Yes exactly
@@lisafuckinlonergan Should be 6x4, or 24. Not 6.4! Big difference.
Nice lecture of nonparametric test and best explanation with examples. Thank you for this video
Glad you liked it!!!
Dear Madam, I am comparing two groups in terms of different research aspects using the Mann-Whitney test. But I want to add a categorical covariate (discipline) How do I take it into account with this test as a covariate? thank you
Nice video! in case of evaluation of percentage data (0-100%) as the outcome measured between two groups, could Mann Whitney U be used? e.g. dry matter content (%) in feed samples of two different kinds of diets are compared to see if the diets differ in dry matter content. Assuming assumptions are otherwise met for Mann Whitney U.
But this test has a table.
Why did not you use table?
Where is 12 coming from for the SE of U?
CRYSTAL CLEAR!
why use z distribution if data is not normally distributed
It is assumed that the distribution of ranks are normally distributed!
Very helpful.. 💙
Glad it was helpful! Regards Hannah : )
THANK YOU!
You're welcome! : ) Regards Hannah
madam, I want to know in Standard error of U , where we can find the value 12. pls explain ma'am
What happens incase of a tie?
Pleas have a look here: datatab.net/tutorial/mann-whitney-u-test It is explained if you scroll down a littel bit
Thank you.
thank you so much!
Many Thanks! Regards, Hannah and Mathias
Dear, u r mixing groups with variables. We don’t classify groups as nominal as all groups are so, we classify variables under these groups as nominal or quantitative. If quantitative then we describe their distribution as normal or not.
Hello Miss at 8:06 where did u get 6.5 plz
I can't finde the value 6.5 in the Video??? Sorry Regards, Hannah
@@datatab hi look at 8:11 just above the answer 14 there is 6.5
@@Ilemaurice687 Ah, that is "times", 6 times 5! n1 times n2, number of cases groupe 1 times number of cases groupe 2. Regards Hannah
@@datatab thx you miss
@@datatab plz any teacher of statistics that i can contact? I having trouble in some questions and i am not able to find anyone from web plz if possible can someone help me? Its about probability and statistics i can send some questions that i am not able to do
Thanks alot, I love you
Glad it helped
What if you have more than 25 cases?
You can also use the Mann Whitney U Test but don't need to read the cretical value from a table! Regards
@@datatab thank you
You are Welcome, Cheers Hannah
Thank you
Mam, How could you calculate p value plzz elaborate
The value of Z is not negative take absolute value means + value
PERFECT
Great.. learned alote
We use the non-parametric test because the sample is not normally distributed but then use a normal distribution to calculate the p-value? It strikes me although maybe these are independent things. The explanation was great!
11:22 But where is the significance level???
the best
Thanks!!! Regards, Hannah
You know how to teach madam
Nice German accent.
Thanks . ) Regards Hannah