BookmarkSubscribeRSS Feed
milda
Calcite | Level 5

I am trying to test for the equality of coefficients in the quantile regression setting, but keep getting the following warning:

 

WARNING: The test for equal coefficients across quantiles is ignored because the joint covariance is not computed.

 

The sample size is 1745, with 2 independent indicator variables, 8 continuous variables, and 2 class variables. The specification is as follows: 

 

proc quantreg ci=rank algorithm=interior(tolerance=1.e-4)
data=sample;
class year ind_code;
model dep = dummy1 dummy2 cont1--cont8 year ind_code /

quantile= 0.1 to 0.9 by 0.1;
test dummy1 /QINTERACT;
test dummy2 /QINTERACT;
run;

 

What could be the reasons for this? I tried alternative ci= specifications and lower number of quantiles but nothing seems to help.

4 REPLIES 4
Rick_SAS
SAS Super FREQ

Your code works perfectly on simulated data. There must be something peculiar about your particular data, such as an abundance of missing values, dependencies in the independent variables, or some other degeneracy.

 

data sample;
do i = 1 to 100;
   x1 = rand("Normal");
   x2 = rand("Normal");
   c = rand("Bernoulli", 0.3);
   dummy1 = rand("Bernoulli", 0.5);
   dummy2 = rand("Bernoulli", 0.5);
   dep = 3 + x1 + x1*x2/4 + dummy1 - dummy2 + dummy1*x1 + c + c*x2;
   output;
end;
run;

proc quantreg ci=rank algorithm=interior(tolerance=1.e-4) data=sample;
   class c;
   model dep = dummy1 dummy2 x1-x2 c /
   quantile= 0.1 to 0.9 by 0.1;
   test dummy1 /QINTERACT;
   test dummy2 /QINTERACT;
run;
sparsityBlues
Calcite | Level 5

@Rick_SAS 

 

i am having this same issue in a data set of ~4000 patients with almost no missing data. any other thoughts on checking the data for what might be causing this and/or how to fix it?

 

a related question is what are the implications, if any, of this error on the veracity of the model outputs?

 

thanks.

Rick_SAS
SAS Super FREQ

Since I cannot reproduce this issue, I do not know the answer to the question about what causes it or how to fix it. 

 

Based on the warning message, it seems likely that this message is related to the QINTERACT option because the message uses the phrase "The test for equal coefficients across quantiles." That would mean that procedure cannot obtain the estimates for that test. I would guess that the other tests (Wald and LR) and the parameter estimates for the model are still valid. If the rest of the model were invalid, I would expect an ERROR, but this message is a WARNING. Typically, a WARNING tells you that SAS cannot produce something that you requested.

 

Without any evidence, I wonder whether this error is related to comparing extreme quantiles. This is pure conjecture, but quantile regression methods sometimes fail for extreme quantiles (eg, 0.05 and 0.95) when they succeed for less extreme quantiles (eg, 0.25 and 0.5). You might try rerunning your program with fewer or less extreme quantiles and see if the warning goes away.

sparsityBlues
Calcite | Level 5
thanks for your reply!

for clarity, yes, i only get this warning with the QINTERACT request (ie, not present when asking for wald or LR).

i tried your suggestion of less extreme quantiles, which did work for 0.4 to 0.7. i had originally log-transformed my DV and in this case i only got qinteract for some IVs. however, when i used the DV without the transformation, it worked for all the IVs in the model. i am trying to work through all of this...

thanks for your assistance!

sas-innovate-2024.png

Join us for SAS Innovate April 16-19 at the Aria in Las Vegas. Bring the team and save big with our group pricing for a limited time only.

Pre-conference courses and tutorials are filling up fast and are always a sellout. Register today to reserve your seat.

 

Register now!

What is ANOVA?

ANOVA, or Analysis Of Variance, is used to compare the averages or means of two or more populations to better understand how they differ. Watch this tutorial for more.

Find more tutorials on the SAS Users YouTube channel.

Discussion stats
  • 4 replies
  • 1324 views
  • 0 likes
  • 3 in conversation