1
$\begingroup$

Between the qvalue package and the p.adjust function, which is more appropriate to use when trying to calculate the q-values of a dataset? According to the manual for the q-value package, the q-value calculated is not an adjusted p-value which is what the p.adjust function would return. There seems to be an equal amount of people that use one or the other, but I wanted to see other opinions that could help clarify what would be most appropriate to use.

$\endgroup$
1
$\begingroup$

p.adjust can calculate adjusted p-values with different methods. I think you can choose Bonferroni correction and BH correction, for instance. BH correction gives you False Discovery Rate (FDR), which is one of the most commonly used I guess.

Q-value is different and cannot be calculated by p.adjust. The idea is that when you have many p-values to adjust, they form a distribution. If all p-values come from the null hypothesis, the p-value distribution is going to be uniform on [0, 1]. But usually, our p-values are a mixture of the null distribution and alternative distribution. The parameter $\pi_0$ describes the proportion of p-values (or tests) that comes from the null hypothesis. To calculated q-values, the algorithm first estimates $\pi_0$, then calculates the q-value.

Note that if you set $\pi_0=1$, which means that you believe all the p-values come from the null distribution, then q-value is equal to FDR.

I think most people do not know about the meaning and details of q-values and it might be sufficient to use BH procedure in most cases.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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