Eg: i have a values of WBC count and mortality (yes or no). I have about 120 test subjects. When checking mean WBC count, i find that the SD is twice or more than the Mean value. Is this a heterogeneous data.
To compare should i use independent T test or Mann Whitney ?
In the case of T test the normality assumption should be funfiled in both compared samples. Moreover, if the determination coefficient is less than 30% then the mean is not good estimate for the sample position. In such cases it is preferable to use test Mann Whitney
Make up your mind what hypothesis you want to test. You will ind that there is rarely more than one test available, if there is a test available at all (in which case one could apply bootstrap techniques to estimate the sampling statistic).
What are the groups among which you wish to compare the mean WBC? In case of high SD, transformations like log, may reduce the coefficient of variation to a great extent. Try log transformation of WBC counts and then compare between different groups.
Thanks guys ..
I trying log transformation and then running independent T test .
You should not use t-test on count data (unless you have large sample sizes and the mean is clearly larger than the sd).
It may be interesting to know if the mean and the variance are about equal. If so, a Poisson model would be appropriate. Otherwise you should bette use a negative-binomial model.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics