# Help! -Inference about interaction coefficients in two-factor study

1. Nov 26, 2008

### solar42

I have measurements of some response of a gene, and two factors: the gene, g=1...G and whether the patient/subject has a certain disease, t=1,2.

the full model is
$$y_{gtk}=\mu+\alpha_g+\beta_t +(\alpha\beta)_{gt}+\epsilon_{gtk}$$

I know that to see if genes have any connection at all with the disease, I just fit the reduced model without the $$(\alpha\beta)_{gt}$$ interaction and compare the two, but if I want to see if, say, gene number g=25 has anything to do with the disease... I know that the null hypothesis is $$H_0 : (\alpha\beta)_{25,t}=0, \textrm{ for all } t$$, but how do I test this hypothesis? I am confused at what to do when I can't drop the whole factor and compare.

2. Nov 29, 2008

### Enuma_Elish

You can define a separate "dummy" variable for gene 25 and interact it with the disease.

EnumaElish

Last edited: Nov 29, 2008