I need your advice regarding the appropriateness of following statistical procedures:
One of our studies aimed to assess indicators of metabolic syndrome at birth in two groups of neonates. As there are no established cut offs for these parameters yet, we used median values of control group as reference values and performed single sample t-test to check whether the values for cases were significantly different from controls.
Cases and controls were from the same cohort except for diffidence in parameter of interest i.e. nutritional status of mothers.
I would like to know if there could be any issues in using this reference value.
Hello Rubina Hakeem, I would just like to query why you use median instead of mean? Is there skew values? Best Wishes, Deborah Hilton Statistics Online
http://sites.google.com/site/deborahhilton/
Dear Rubina:
I may not correctly understand your question. This is what I understand. Please correct me where I am wrong.
(1) From the control group having N_ctrl subjects, you obtained the median, say Md_ctrl.
(2) From the case group having N_case subjects, you obtained the mean and standard deviation, say Mn_case and SD_case, respectively.
(3) Using only the data from the case group, you tested
H0: True Mn_case = Observed Md_ctrl.
(4) Your t-statistic was (Observed Mn_case - Md_ctrl) * sqrt(N_case) / SD_case.
If this is what you did, then I would advise against this. In my opinion...
(A) The primary reason this is a bad idea is this: You did not appropriately account for the variability of Md_ctrl; you assumed that this was a known quantity.
(B) Another major concern: If you use the median from one sample and compare it to the mean of another sample, then an appropriate test statistic is not so easy, and likely does not have a well-defined probability distribution (like the t-statistic does).
Regards,
Reid
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis