I have data from one clinical test for 450 patients. Depending upon my objective, I divided data into 7 different categories to compare two groups, in each category (inherited from parents and not inherited). I used Mann Whitney test to compare two independent groups in all of those seven categories. At the end, I have seven P values, findings showing that either there are differences in groups or not, for each category.

7 different categories are based on gender. There is only one clinical test say C1 for which I have scores.
Gender    C1   group

Male     20    inherited

Male     40    not inherited

Female 45    inherited

In first category I compared all inherited and not inherited (C1 and group). In 2nd category I compared only inherited and not inherited males. And in 3rd I compared inherited and not inherited females. I have 7 seven different categories like that.

My question is that “Is there any way to validate these findings?” As I made seven different categories, so differences can emerge by chance as well.

  • 1
    $\begingroup$ What do you mean by "validate"? I think your biggest issue here is multiple comparisons. Your 7 different categories are 7 different responses on which the 2 groups are compared, right? Do you think of these responses as different measures of the same thing, somewhat related, or unrelated? $\endgroup$ Commented Dec 22, 2015 at 16:16
  • $\begingroup$ thanks for reply. I updated my question with more detail. Validate means I want to make sure that results are not by chance. I am from Biology, not familiar with statistical terms, so, sorry for that. I hope I made it clear. $\endgroup$
    – Asif
    Commented Dec 22, 2015 at 17:04

1 Answer 1


You might be better served by using a combined linear model to evaluate your results, rather than the multiple types of comparisons you have been performing. You would evaluate all at once the relation of C1 to gender and to group and to the interaction of gender with group. In one form of syntax for models, this would be expressed as:

C1 ~ gender + group + gender:group

Standard statistical software for this type of linear model will provide p-values for the overall model, for each of gender and group, and for the interaction. If you want to test further how well this might generalize, you can try bootstrapping. These matters are covered, for example, in Faraway's freely available "Practical Regression and Anova using R."


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.