Mann Whitney U test or Scheffe? - (May/21/2008 )
I am beginner for statistical analysis of gene-expression profiling. and I would like to ask the statistical anaylisis.
After normalization of raw data, we used the nonparametric analysis, Kruskall Wallis test because the data were not normally distributed. And then when there is a difference (P<0.05),exsamination the differences between the individual groups (3 groups)? Mann Whitney U test or Scheffe?
Thank you in advance for your kind cooperation.
Neco
After normalization of raw data, we used the nonparametric analysis, Kruskall Wallis test because the data were not normally distributed. And then when there is a difference (P<0.05),exsamination the differences between the individual groups (3 groups)? Mann Whitney U test or Scheffe?
Thank you in advance for your kind cooperation.
Neco
See if you find something useful in these results:
http://search.vadlo.com/b/q?keys=Mann+Whit...mp;sn=158621799
..
Thank you for comments, Cell counter.
..well, in fact, I did both ways and Scheff test tended to show more strictly result compared to M-U test. I have checked several articles and both tests are used in these published articles.
I'm still comfused.
Neco
After normalization of raw data, we used the nonparametric analysis, Kruskall Wallis test because the data were not normally distributed. And then when there is a difference (P<0.05),exsamination the differences between the individual groups (3 groups)? Mann Whitney U test or Scheffe?
Thank you in advance for your kind cooperation.
Neco
Scheffé is only for data for which can use an AMOVA before (i.e. data that fit to its basic assumptions on distribution etc)
Mann Whitney-test you can use, but it is as the t-test only for two samples to compare!
For multiple non-parametric comparisons use Nemenyi or Steel-Dwass tests.
Dear hobglobin,
Thank you for your reply.
I'd like to confirm.
In my case,
normalization of raw data--Kruskall Wallis test of each gene expression (each gene expression is compared to 3 individual groups)-- the gene expression data showing significant differences (P<0.05), Mann Whitney U is tested by each individual groups.
Is it ok?
Thank you in advance for your kind coopeartion.
Neco
Thank you for your reply.
I'd like to confirm.
In my case,
normalization of raw data--Kruskall Wallis test of each gene expression (each gene expression is compared to 3 individual groups)-- the gene expression data showing significant differences (P<0.05), Mann Whitney U is tested by each individual groups.
Is it ok?
Thank you in advance for your kind coopeartion.
Neco
I'm not so sure.
First are the gene expressions independent or dependent experiments?
Second each gene expression is compared to 3 individual groups, this would IMO increase enormously the error probability because all the 5% (or 1%) error probabilities add up (thats the reason not to do many t-tests to compare for many samples but an ANOVA (this page) gives a better and exacter explanation).
The easiest way what you can do is to adjust the p-values with the Bonferroni-correction to avoid the mentioned problem or use more complicated methods as bootstrapping, permutation tests etc.