confidence intervals of the median - Statistica General Discussion - Statistica - Dell Community

# confidence intervals of the median

#### confidence intervals of the median

Hello,

After doing a krurskal wallis and median test in statistica, I  would like to calculate the confidence intervals of the median.

Please, has anyone know how we calculate the confidence intervals in statistica.

With very best wishes
GdM

• You are welcome. I modified the link ("here") on my last post. Let me know if you have problems to access it.

• Hello GdM,

Yes, the median  test gives the overall p-value for all multiple comparisons between groups, if you click on "Summary: Kruskal-Wallis ANOVA & Median test" tab.

If you click on  "Multiple comparisons of mean ranks for all groups" tab, it gives the pairwise comparison result for Rank test only. The pairwise comparison of multiple groups using median test is not available in Statistica.

If the overall median test p-value is not significant, there is no need to do post hoc test.

If it is significant and you keen to do a post hoc pairwise comparison using median test, you can alternatively create new grouping variables to perform median test on each pair of groups separately.

For example, if your original grouping variable has 3 groups, let say, 1,2,3.  To do pairwise comparison of these 3 groups, you can create dummy grouping varibable, e.g.  "Group12"  only contains values 1, 2, and "Group23" only contains values 2, 3, and "Group13" only contains values 1,3.  Then you can perform median test for "two-independent groups" to compare groups 1 to 2 using dummy grouping variable "Group12". Similarly to compare group 2 to 3 using "Group23" and to compare group 1 to 3 using "Group13". However, these p-values from individual pair-wise median test are subject to multiple test correction. Nevertheless, you can do a manual multiple-test correction of those p-values and can still tell some information of which pair of groups is significant, in the case that the overall p-value from multiple group median test is significant.

Hope it don't get you confused.

All Replies
• Hello, there is no confidence intervals generated from median test in Statistica. The median test is based on a Chi-square test with more information can be found at here

• Thanks a lot Wendy.

Best

Gdm

• You are welcome. I modified the link ("here") on my last post. Let me know if you have problems to access it.

• Hello Wendy

Please Could you confirm that the median test is for multiple comparisons.

Thanks

Gdm

• Hello, Yes. The p-value for the median test from "Summary: Kruskal-Wallis ANOVA & Median test" tab is the overall p-value for multiple comparisons. The "Multiple comparisons of mean ranks for all groups" tab only generates the pairwise comparison result for Kruskal-Wallis rank test and there is no pairwise comparison using median test.

•  Hello Jenny

Thanks a lot for your rapid response.

A question. I understand that the median  test (after Kruskal-wallis Anova)  gives the overall p-value for all multiple comparisons between groups, right ?

Please, what do you mean by  "there is no pairewise comparison using median test" ?

Thanks

GdM

• Hello, the null hypothesis for median test is : all samples come from populations with identical medians. As such, it gives an overall p-value for comparing the medians among all groups. If you click "Multiple comparisons of mean ranks for all groups" under the quick tab, this will compute post-hoc comparisons of mean ranks of all pairs of groups (see Siegel & Castellan, 1988, p. 213-215).

Click the "?" on top right of the results dialog, you can access details of explaining all test under this quick tab.

• Hello GdM,

Yes, the median  test gives the overall p-value for all multiple comparisons between groups, if you click on "Summary: Kruskal-Wallis ANOVA & Median test" tab.

If you click on  "Multiple comparisons of mean ranks for all groups" tab, it gives the pairwise comparison result for Rank test only. The pairwise comparison of multiple groups using median test is not available in Statistica.

If the overall median test p-value is not significant, there is no need to do post hoc test.

If it is significant and you keen to do a post hoc pairwise comparison using median test, you can alternatively create new grouping variables to perform median test on each pair of groups separately.

For example, if your original grouping variable has 3 groups, let say, 1,2,3.  To do pairwise comparison of these 3 groups, you can create dummy grouping varibable, e.g.  "Group12"  only contains values 1, 2, and "Group23" only contains values 2, 3, and "Group13" only contains values 1,3.  Then you can perform median test for "two-independent groups" to compare groups 1 to 2 using dummy grouping variable "Group12". Similarly to compare group 2 to 3 using "Group23" and to compare group 1 to 3 using "Group13". However, these p-values from individual pair-wise median test are subject to multiple test correction. Nevertheless, you can do a manual multiple-test correction of those p-values and can still tell some information of which pair of groups is significant, in the case that the overall p-value from multiple group median test is significant.

Hope it don't get you confused.

• Many thanks Jenny for all these precisions.

Best

GdM

• Hello GdM,

You are most welcome :)

If you have any other questions on your way of using Statistica, please feel free to post your queries in this discussion forum. We will be very happy to assist you :)