0

I am trying to run some quantile regressions with the rq() function in the quantreg package. I have set up my data and created models etc (following the instructions on another site) and it all look good. See below; RE and WTPC are continuous variables and Armored is define below.

# Convert Treatment to 1/0 Binary factor where Armored=1
Armored<-as.numeric(Treatment=="A")

# Define variables
Y<-WTPC

X<-cbind(RE,Armored)

# Quantile Regression @ 25th
QR25<-rq(Y~X,tau=0.25)
summary(QR25)

However when I call summary() of the rq() model I get this

> summary(QR25)

Call: rq(formula = Y ~ X, tau = 0.25)

tau: [1] 0.25

Coefficients:
            coefficients lower bd  upper bd 
(Intercept)  11.09456      9.34170  18.22058
XRE          -1.81530     -3.46350  -0.78062
XArmored     -3.68480    -14.29227  -0.74389

This is nice information and all but I want the standard errors, p-values etc like in a normal linear regression. I checked the documentation and other examples and it looks like this is the way to do this. Make a model (QR25) and then call summary(QR25) and that should give you the regression output with errors and p-values.

This is my first attempt at quantile regression so maybe I am missing an earlier step or need to specify something else. Any thoughts? I am running this in RStudio Version 0.98.501.

lmo
  • 37,904
  • 9
  • 56
  • 69
wraymond
  • 295
  • 1
  • 6
  • 17
  • The documentation lists five ways to calculate the standard errors. You're using the default method. Try some of the others. (But you'll probably need to do some reading to know which ones are statistically appropriate for your data.) – joran May 07 '14 at 16:06
  • ahh, I missed that part. Thanks for the help. – wraymond May 07 '14 at 16:57
  • Asking for standard errors for quantile regression estimates seems to be an effort to ignore the fact that quantile regression is all about avoiding the assumption of normality and symmetrically constructed confidence bounds. Maybe you should go back to Normal theory methods if you are having trouble accepting this or communicating these results to an audience. Given waht appears to be a conceptual problem in the user, perhaps this should be posted at CrossValidated.com – IRTFM May 08 '14 at 00:20

0 Answers0