What Is An Omnibus Test?

Question

[Pixel08]

Hi, I'm new to BioStar and new to the study of bioinformatics - so I just want to say that I'm incredibly excited to be in a community of bioinformaticians at BioStar.

I've done a google search on Omnibus tests, but there doesn't seem to have a very good explanation of it. I'm just wondering if anyone could explain the concept here? (also, I understand that the F-test is an example of it, but it doesn't really tell me what the umbrella term Omnibus test is).

Thanks in advance!

[2184687-1231-83-]

Maybe someone in http://stats.stackexchange.com will be able to answer.

Answer 1

[K_Star]

If you are looking at haplotypes: In PLINK, an omnibus association test is used to test all haplotypes combined for association with the disease/trait. If association is observed with the omnibus test, a haplotype specific test can then be used to identify which of the specific haplotypes explains the omnibus association, that is, which haplotype is ‘driving’ the association.

Check out this PLINK site: http://pngu.mgh.harvard.edu/~purcell/plink/haplo.shtml

Check out these papers: http://www.ncbi.nlm.nih.gov/pubmed/16671109 http://www.ncbi.nlm.nih.gov/pubmed/17118959

Answer 2

[David Quigley]

An F test tests a model which may have multiple explanatory variables contributing to the variance explained by the model. An F test is an omnibus test because the significance of the model is a measure of the overall significance of the explantory variables and the way they are combined, not the individual variables by themselves. Let's say I want to model how long a hamster can do the tango. My model contains "weight" and "age" as additive explanatory variables. The F test returns the significance of both of these together, even if in reality the only significant contributor is age. There is a good explanation on the wikipedia page for F-test in the "Multiple-comparison ANOVA problems" section.

[mmfansler]

Another good example of an omnibus hypothesis test is the chi-square test for categorical data. Taking independence of the categorical variables as the null hypothesis, a significant result only tells you that independence is improbable, but does not yield any information about which levels are significantly different.

I think of omnibus tests as analogous to existence proofs: they'll confirm there's something there, but won't tell you what it is.