hello
I am facing problem determining which statistical test should I use to analyze my data. I have two groups of data one for dietary residue and the other for non-dietary residues they are independent and the data doesn't follow any order so it seems to be not normally distributed the mean and the median of each group is not close to each other which proved to me that the data are skewed. I used the log of the data instead of the initial data set (the mean and the median turned to be identical or very close to each other)and then used T-test (excel) to analyze the data my question: did I used the right test to compare between two different means from two different independent groups?
the P-value two-tail = 0.000084463 which is clearly lower than 0.05 which I think the T-test normally uses for the significance, does that mean that there is a significant difference between the two tested groups?
to be more precise, I found that one of the groups giving better normal distribution by using the natural log and the other group was just better when I used the square root of the data.
if the above is all right , when reporting the data should I mention that the data was transformed to be log normally distributed as it was skewed ?
kindly check the attached picture of the data
Thank you in advance
Hi Mohamed. Since the data you are testing are coming from two groups and since their values are not normally distributed, I suggest using the Mann–Whitney U test, which is a non-parametric alternative test of the T-Test. In your care, it should give you almost the same p-value and can be managed in the same way, either single tailed or two tailed. The major difference between T-Test and Mann–Whitney U test is that the later is better for testing non-normally distributed data sets.
And yes, there is a statistically significant difference between your two groups.
Thank you Mohamed Khalifa
this is really appreciated
do you mean instead of transforming the data using the natural log or the square root. Should I use the mentioned test Directly instead ?
I have never used this test before I will have to have a look where to find it and how to use it
many thanks again for your advise
regards
You can find the Mann–Whitney U test in the SPSS and many other packages. I guess Microsoft Excel also has the test.
If you took logs of one group and square roots of the other, you created a difference in the means. You should choose the same re-expression for both groups, but you are quite correct to re-express.
I assume that you've plotted the data and checked for outliers. If not, do that first.
Paul F Velleman Thank you so much this is really helpful and appreciated.
No. I mean you should re-express skewed data, but if you are comparing groups you must use the same function for both groups.
yes, I got your point right. I used the mean out of both group's square root. and that makes sense. Thank you Paul
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics