QR = TRUE in BRMS interface? [QR Decomposition]


Does the BRMS package offer the functionality to provide QR = TRUE? i.e. a QR Decomposition? (https://mc-stan.org/users/documentation/case-studies/qr_regression.html)

I couldn’t find anything. I ask because I have a regression model with a ton of data points, quite a few predictors, and plenty of potential correlations looming. I will centre variables.

Many thanks,


brms does currently not support QR decomposition, but it could be a nice feature to support in the feature. Feel free to open an issue for it on https://github.com/paul-buerkner/brms

1 Like

In retrospect, it might have been better to use the orthogonal factor from a polar decomposition in linear models rather than from a QR decomposition. The original priors would make more sense that way.

Thanks for adding this feature in, Paul! And in terms of doing the equivalent to the Stan command “qr_thin_Q”, will this be possible? Or is this already actually what has been added to brms with the “qr=decomp” command?

This is exactly what happens with decomp = "QR"


OK great. Thanks!