0
$\begingroup$

I have two sets of equal sized and normally distributed data, which are output from two different experiments. I wish to do a two-sample t-test with these two data sets to see if the results from the two experiments agree.

Is it appropriate to mean (or median) centralize the data sets before doing the t-test?

The variance of the data are equal and passes the var.test. My understanding is that if I centralize the data and the mean will be the same for the two data sets and the result of a t-test will always favor the null hypothesis of equal means.

Can the same thing be said about centralizing data or not for a one-way ANOVA if I have more than two datasets?

Improve this question
$\endgroup$
1
  • 3
    $\begingroup$ So each experiment produces a single vector of numbers? or does each experiment have multiple groups? If it's just a vector of numbers then if you centralise on the mean, then by definition both vectors will have the same mean (i.e., 0). $\endgroup$
    – Jeromy Anglim
    Jun 29, 2012 at 6:37

1 Answer 1

Reset to default
2
$\begingroup$

Imagine you have two samples, one is distributed as $N(0,1)$, mean 0 and variance 1, and the second is distributed as $N(1,1)$, mean 1 and variance 1. The null hypothesis is obviously false here.

Now if we do operate a mean-centering, we will transform the second sample into something that has a distribution very close to $N(0,1)$. Intuitively, we wipe out the differences between the samples by mean- or median-centering, so this is not appropriate for testing.

Formally, you do not even need to test for equality of the means by a t-test if samples are mean-centered, because you know that they have the same mean (namely 0). Any rejection of the null hypothesis in this case is a false positive.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.