Testing correlation in R

Question

I am working on a very large data set, where I am comparing 2 genes. Currently I am trying to test

H₀: correlation = 0.64

H_A: correlation > 0.64 at the alpha = 0.05 level.

At first I thought I could use cor.test() or cor() with an argument to set the 0.64, but I am not seeing that as an option. I'm not really sure where to go with this.

Pedantic point: Your hypotheses don't match. If correlation < 0.64 you must reject them both, and that makes no sense. — Jeremy Miles, Nov 06 '16 at 19:45
Is it possible the value $0.64$ is an estimate from other data? If so, the uncertainty in that estimate should be incorporated in your test. In other words, perhaps the null hypothesis ought to be that the two datasets have the same correlation, not that the current one actually has a correlation of $0.64$. — whuber, Nov 06 '16 at 21:10
I closed the other thread as a duplicate of this one. This thread has an answer, whereas the other doesn't & the comments are mostly about R syntax, rather than the statistical issues (as here). — gung - Reinstate Monica, Nov 07 '16 at 13:52

score 1 · Accepted Answer · answered Nov 06 '16 at 19:04

1

The function cor.test() seems to do the trick.

set.seed(1)
x <- runif(100, 0.0, 1.0)
y  <- 2*x + runif(100, 0.0, 1.0)
cor.test(x,y,alternative = "greater", conf.level = 0.95)

The output will report a confidence interval. If the desired value stated in your null hypothesis is not covered by this interval, the null hypothesis can be rejected on the respective significance level.

answered Nov 06 '16 at 19:04

33d1v

52

1

-1 Checking whether a confidence interval contains a value is not in general a good significance test. See http://stats.stackexchange.com/questions/169141/is-rejecting-the-hypothesis-using-p-value-equivalent-to-hypothesis-not-belonging – Kodiologist Nov 06 '16 at 20:33

Testing correlation in R

1 Answers1

Linked

Related