I want to perform a quantile regression on two continuous variables; Y (DV) and X (IV). I want to find out if there is an significant association between Y and X.

When doing this in R like:

fit2 <- rq(Y ~ X,tau=c(.05, .25, .5, .75, .95))

If say, the 75% quantile of X is significant with a p-value < 0.05 but rest is not, can I say that X is significant in total? If none of the quantiles are significant, is X not significant in total?


As long as you correct for multiplicity (e.g. Bonferroni-Holm), this is one of many possible ways to test for association. Of course it can capture only linear aspects of the association. A non-significant result can thus be due to low power, lack of linear aspects in the true association or due to conservativeness of the correction for multiplicity.

| cite | improve this answer | |
  • $\begingroup$ Thank you for your answer. In this case, would it be enough to multiply a p-value with the number of quantiles tested, in this case 5 (for correction of multiplicity)? $\endgroup$ – eXpander Oct 23 '13 at 16:22
  • $\begingroup$ Yes. However, if the effects of $x$ on different quantiles are quite similar, the corresponding test decisions will be strongly related and thus your overall test very conservative. $\endgroup$ – Michael M Oct 23 '13 at 16:55

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.