I have data on biochemical parameters ( like CK, AST, LDH, BUN, glucose, total protein, albumin etc) for 150 patients. The patients are divvied into two groups based on treatment response i.e survivors and non-survivors. The data for most of the parameters in both the groups are not normally distributed. Can I use non parametric tests instead of transforming the data.
With two groups, a Mann-Whitney U test is appropriate, although with such a large sample size, you could probably get away with ignoring the skewed distributions.
I agree with Adrian, but would add that if skewness is very high, 150 samples may not be enough - a better alternative is to use bootstrap.
Thank you for your valuable comments.
@ Adrian...By checking the distribution of data, mean does not seem to be representative of data for many parameters, so would prefer Mann-Whitney U test...
@Ahmed...Although I have means and SD for these parameters but the data is skewed so in my opinion parametric analysis will not be an appropriate choice.
If you email me the data, I can run bootstrap analysis for you if you are familiar with this method.
A.K. Singh...if you are willing to help, your mail id please....
If your data are skewed Mann Whitney U test is the right choice (but you will see, Student test is so robust that sometimes the results with parametric and non-parametric statistic are very similar)
@Marcela Velez.. Thank you for your suggestion...you are correct...even in present case there not much difference for many parameters. So i had to go with Mann Whitney U test...
What are you trying to predict? I would think a logistic regression would work if you want to test the different variables and their effect on survival. If you have data on when the non-survivors passed away, you can use survival analysis.
- Badges
- Science topic
- Similar topics
- Medicine
- Disease
- Infectious Diseases