Gys de Jongh
GysdeJongh at compuserve.com
Thu Aug 5 15:52:35 EST 1999
<adam at tcgu.med.utoronto.ca> wrote in message
news:37a8be0a.1728968359 at netnews.utoronto.ca...
> I have a couple of stats questions. First, I am trying to use the Chi
> Square test to compare the level of gene transcription for a number of
> different genes between two different tissue samples. One problem I
> am having is I don't know what to do in the situation when one sample
> has a value of 0 and the other does not. Specifically, this results
> in an expected value of 0.
> Second, I don't even know if I can use this test for my comparison. I
> am using the following formula --- (observed-expected)^2/expected
> I assume I use a degree of freedom=1 as I am looking at individual
> genes. Also, I have seen examples where this test has been used with
> expected values <5, but I have read that this test is unreliable in
> situations when the expect is <5.
the chisquare distribution is a model which can be used to describe actual
experiments. The chi distribution is continuous and real life histograms are
discrete. One acceptable way out of this is the yates correction : if the
frequency is lower than expected add 0.5 ; when the frequency is higher than
expected substract 0.5. The model becomes a bit better. If it is not close
enough for you than you need an other model. Like precisely calculated
binominal p's or montecarlo simulations. In general the chi distribution
(with yates correction) is rather robust in the situation you describe.
More information about the Methods