New to the site and to stats here! This may be a silly question, but I haven't been able to find a satisfactory answer on the procedure for a power analysis (or general guidelines about sample size) with a quantile regression. I plan on looking at 3 predictors along a single outcome. What would a recommended sample size be, or how can I conduct a power analysis? I appreciate any help, thanks!


For general linear quantile models of the form $$F^{-1}_{Y|X}(\tau) = X'\beta_n(\tau)$$ where $\beta_n$ is allowed to depend on the sample size, Chernozhukov and Fernandez-Val (2005) construct a power analysis for conditional quantile regression models as in your case. They consider the general null hypothesis $$R(\tau)\beta_0 (\tau) - r(\tau) = 0$$ for $\tau \in (0,1)$. Their procedure allows you to test for

  • a significant effect for a given predictor
  • a constant effect for a given predictor across quantiles
  • stochastic dominance (e.g. unanimously beneficial impact of a treatment)

Victor Chernozhukov makes his R code available on his website under the section "Policy Analysis" for the paper "Subsampling on Quantile Regression Processes".

Chernozhukov, V. and Fernandez-Val, I. (2005) "Subsampling on Quantile Regression Processes", The Indian Journal of Statistics, Special Issue on Quantile Regression and Related Methods, Vol. 67 part 2, pp. 253-276


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.