Names of regression parameters in brm_multiple

Interfaces brms
1

I just ran my first ordinal regression model with brm_multiple. The predictor variable is year, an ordinal variable whose possible values are 2007,2009,2011,2013,2015, 2017 and 2019. 2007 is the reference year, so I expected to find the parameter names to include 6 intercepts, labeled [1] through [6], then 6 change parameters named 2009,2011,2013, 2015, 2017, 2019.

Instead, the change parameters are named:
year.L
year.Q
year.C
yearE4
yearE5
yearE6

Can anyone explain this?
I’m assuming these are the parameters I expected, in the proper order, just with different names. But I have no idea where the names came from or what they refer to. There’s no L,Q,C or E in my data.

original call was:
screens_model_1a<-brm_multiple(formula=tvt_asd~1+year,data=yrbs.complex.mdimp,family=cumulative(“probit”),prior=prior_screen,combine=TRUE,chains=1,file=“screens_model1a”)

The datasets, imputed in mice, are in yrbs_complex_mdimp.
tvt_asd, the outcome variable, is a 7-level ordinal variable (multiple choice response on a survey for number of hours spent watching TV on an average school day)
The prior referred to is:
prior_screen<-prior(normal(0,5), class=“b”) + prior (normal(0,5), class=“Intercept”)

Thanks!

2

Hi!

These are contrasts based on orthogonal polynomials, as it is standard for ordinal predictor (see contr.poly).

You might:

  1. change the contrasts of the variable to forward or backward difference coding,

  2. use the emmeans package to get estimated marginal means for the change points or

  3. enclose year in mo(…) to treat is as a monotonic effect

Cheers,
Pascal

3

Amazing! I’ve just been working from the tutorial, where it looked like the names would be straightforward. I bet monotonic effects will do the trick. Thank you so much.