micro@
I mean testing tens of thousands of hypotheses simultaneously, like those in microarray, fMRI, and QTL mapping data.
cran
How?
LRT?
micro@
LRT could be one of the choices.
Actually, no one knows how to do this optimally.
But many possible ways exist.
ypchen
Could you give more details?
micro@
It's an huge area. e.g. you may have 20000 dependent variables, and 3 independent experimental units from each of the two possible treatments. One of the questions is how to (optimally) find the variables that are different between the two treatments. How many of them are different? How to control error rate? How to reveal the underlying relationship between the variables?How to analyze power of the experiement? ....
cran
ANOVA?
xianhua_meng
this kind of question is usually called multiple testing. It's a hot topic now.
I'm doing research in this field.
anita_jiu
[quote]引用第4楼micro@于2006-06-24 01:02发表的“”:
It's an huge area. e.g. you may have 20000 dependent variables, and 3 independent experimental units from each of the two possible treatments. One of the questions is how to (optimally) find the variables that are different between the two treatments. How many of them are different? How to control error rate? How to reveal the underlying relationship between the variables?How to analyze power of the experiement? ....[/quote]
Not sure whether my ideas would be helpful or not.
1. 20000 DVs, 3 independent experimental units (=IVs?), then MANOVA
2. But, I would rather suggest to apply exploratory factor analysis which helps you (hopefully) to reveal the underlying relationships of the variables (God, you have 20000 and would you wanna to reduce/filter some??). Difference: discriminant analysis (DA) might be applied after factor analysis, I think. DA actually applies the same mathmetical theories as MANOVA does.
Have you got your data or not yet?
micro@
Thanks, anita_jiu. But I'm not asking for help, just listing some of the research areas.
BTW, I think MANOVA will not work under these cases. No df can be left. Sample covariance matrix will be singular. But it can be shrunk to identity so as to have full rank.
netcow
BUT I think that spss soft cannot stand the test?
xianhua_meng
To micro@:
are you doing some work or research on multiple comparisons? could you talk about it?
