4
$\begingroup$

What would be an appropriate statistical test to test whether one sample is more significantly spread out around zero than other?

I believe that what I am looking for is a kind of f-test but instead of comparing variances (variations around the mean of each sample) I would want to compare "variances" around a fixed predefined value (in my case zero) between the two samples.

Here's a pic in an attempt to depict qualitatively how the test I seek should work:

enter image description here

$\endgroup$
6
  • $\begingroup$ Why are you interested in zero? $\endgroup$ Dec 13, 2016 at 15:12
  • $\begingroup$ I am interest in allelic expression. And I would like to be able to test if two samples (homozygous and heterozygous) have expression values sufficiently different from zero (in log-scale, that is). So I would like to be able to detect if the spread from zero of heterozygous samples is greater than that of the homozygous samples. $\endgroup$ Dec 13, 2016 at 16:03
  • $\begingroup$ Can you post an example dataset to clarify your situation? $\endgroup$ Dec 13, 2016 at 17:41
  • $\begingroup$ @gung: Just added a pic which I hope clarifies what I am looking for. $\endgroup$ Dec 13, 2016 at 18:55
  • $\begingroup$ What would you guys say about using Mann–Whitney U test (en.wikipedia.org/wiki/Mann%E2%80%93Whitney_U_test) on samples A' and B', where A' and B' are absolute-value-transformed samples of A and B, respectively? $\endgroup$ Dec 13, 2016 at 19:41

1 Answer 1

1
$\begingroup$

Perform the Mann–Whitney U test on the absolute values.

Here's some code in R that does the job:

stat.test <- function(sample_A, sample_B, difference = 0) {
  sample_A_abs <- abs(sample_A)
  sample_B_abs <- abs(sample_B)

  wilcox.test(x = sample_B_abs, y = sample_A_abs,
              alternative = "greater", mu = difference)

}
$\endgroup$
0

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.