Interpreting "quantile" parameter in BRMS quantile regression

Hi,

I am trying to fit a simple quantile regression model using BRMS with the asym_laplace distributional family. The command that I am using is as follows:

qr.mdl <- brm(bf(lbio.fit ~ s(HadISST.C), quantile = 0.9), 
              data = dat.fit, 
              family = asym_laplace())

…where I am using a spline as my response curve to describe the 90% percentile.

The model appears to fit ok. However, doing a comparison between the input data and the fitted() values looks like so:

Straight away this doesn’t seem to look like 10% of the points are above the line. Checking this confirms it - 65% of the observations are above the line (rather than the 10% I expect). Binning the x-axis and looking at it in chunks gives similar results.

So, my question is, what is going wrong? Have I misconfigured (or misunderstood) my model specification? Or is there something I am missing?

Best wishes,

Mark

  • Operating System: Manjaro Linux
  • brms Version: 2.80

[ deleted because my answer was way off (see Paul)]

fitted always returns the mean of the distribution, no matter what. you have to specify dpar = mu to get what you want

1 Like

Brilliant! Now it works:

For the record, when I extracted the fitted values to plot, I did it like so:

fitted(qr.mdl,dpar="mu")

Thanks Paul - here’s something for you seeing that it is Friday afternoon: https://beeroverip.org/

Mark

Thanks :-)