Mann-whitney test - Statistica General Discussion - Statistica - Dell Community

Mann-whitney test

Mann-whitney test

This question is answered

I am looking for an option to conduct a mann-whitney u test that is one-tailed. I think the default is two-tailed. Same thing for the t-test. I think older versions of statistica allowed this. I have Statistica version 13.

Verified Answer
  • Hi, Matt, Mann Whitney test is a non-parametric version of two independent samples t-test. There is supposed to be two independent groups to compare if using Mann Whitney test. If you wanna do one-sample test, there should be a target/reference value that you want to compare the one sample to. If you know the target/reference value, you might be able to still use Mann Whiteney test by creating a dummy sample group with target/reference value to compare with the one-sample. 

    Again, I don't know very much details of what you are trying to do in your project. This is only my point of view. It is best to contact our professional statistical service team to look more deeply into your project. Here is their email: DSG.Sales.Statistica.NAM.PSO@software.dell.com. You may email them to get statistical assistance. 

    Thank you very much.

All Replies
  • Hi, Mliebman, 

    For t-test, Statistica produces 2-tailed p-values. Since t-distribution (normal distribution) is symmetric, for example, symmetric about zero if comparing the means of two independent groups, thus the two-tailed p-value is twice the one-tailed p-value for alternative hypothesis (e.g. Means difference <0 or Means difference >0). So you can simply take the 2-sided p-value from t-test and divide by 2 to obtain the one-sided p-value. 

    For Mann-whitney u test, there is a column named "2*1sides exact p" in the result table. You can take that p-value and divide by 2 to get the 1-sided p-value for the test. 

    Please take note that, two-sided p-value is always used, unless you have prior knowledge before random samples are drawn from population that the mean of one group will be higher that the mean of another group. 

    Thank you.

  • Thanks Jenny. Does this apply if we want to test if one group is specifically higher than the other group, even if we didn't have prior knowledge that one would be higher? The question is this: water samples were taken after a treatment, and we wanted to know if the levels went up. We don't care if they went down.

    -- Matt Liebman

  • Hi, I don't have very much details of what project you are doing and may not give very reliable suggestions for your case. 

    This is only my personal discussion with you. Dell and Statistica will not be legally liable for these comments and suggestions. 

    Are you doing a paired/matched sample comparison of before and after a treatment?  Or comparison of two independent groups before and after? I suppose it is continuous measurement? If it is paired/matched sample comparison, I would take the difference D before and after treatment of matched subjects, and change the negative D values to 0 since the "went down" is not of interest and can be equivalent to no difference between before and after? and do a one sample t-test for D with 1-sided p-value. If it is two independent groups, I would rather do a two sided test with two sided p-value, and examine the two group means to see if after treatment mean is higher than before treatment mean. 

    Above is only my personal opinion and may not be valid for your project since I don't have a full picture and more details of what project you are doing. If you are seeking for more professional advises, we have a professional team who can assist you on statistical analysis with reasonable charges. I can contact them for you if you want to get them involved.

    Thank you very much.

  • Thanks again Jenny. The answer to your question is no, they are not matched pairs. But this gave me an idea to do a one sample t test. I wonder if there is a one sample Mann Whitney test? Perhaps it's best to talk to the statistics people.

    -- Matt Liebman.

  • Hi, Matt, Mann Whitney test is a non-parametric version of two independent samples t-test. There is supposed to be two independent groups to compare if using Mann Whitney test. If you wanna do one-sample test, there should be a target/reference value that you want to compare the one sample to. If you know the target/reference value, you might be able to still use Mann Whiteney test by creating a dummy sample group with target/reference value to compare with the one-sample. 

    Again, I don't know very much details of what you are trying to do in your project. This is only my point of view. It is best to contact our professional statistical service team to look more deeply into your project. Here is their email: DSG.Sales.Statistica.NAM.PSO@software.dell.com. You may email them to get statistical assistance. 

    Thank you very much.

  • Thanks again Jenny, I'll ask the professional stats people about this.

    -- Matt Liebman

  • Hi, Matt, you are most welcome Smile