b_Intercept and Intercept: how are they related?

Interfaces brms
1

Please also provide the following information in addition to your question:

  • Operating System: macOS
  • brms Version: 2.11.1

Following one of the recent brms updates (2.11.0, I believe), many functions (e.g., fixef(), posterior_samples(), posterior_summary()) now return information for both b_Intercept and Intercept. Sometimes these appear the same. Other times they appear different. What is their relationship? I looked around in the documentation and didn’t readily see an answer.

2

Intercept should should not be shown. I will take a look.

1 Like
3

I think I messed it up only in a recent github version of brms. Can you try it again with the latest github version. If it fails, can you provide a reproducible example?

4

Just updated to the github version (2.11.5). Still not quite right.

library(tidyverse)
library(brms)

set.seed(1)
d <-
  tibble(x = rnorm(200),
         y = rnorm(200, 0.6 * x))

fit <-
  brm(data = d,
      y ~ 1 + x,
      cores = 4,
      seed = 1)

The fixed() function appears to work.

fixef(fit)

            Estimate  Est.Error        Q2.5     Q97.5
Intercept 0.04094791 0.07085068 -0.09793115 0.1790767
x         0.57683302 0.07737151  0.42293581 0.7279272

But we still get two intercepts for posterior_summary().

posterior_summary(fit)

                 Estimate  Est.Error          Q2.5        Q97.5
b_Intercept    0.04094791 0.07085068   -0.09793115    0.1790767
b_x            0.57683302 0.07737151    0.42293581    0.7279272
sigma          1.02001843 0.05127489    0.92435009    1.1259122
Intercept      0.06144835 0.07081172   -0.07706498    0.2000552
lp__        -292.76052962 1.19581758 -295.80919614 -291.3828067

Similar issue with posterior_samples().

posterior_samples(fit) %>% 
  glimpse()

Observations: 4,000
Variables: 5
$ b_Intercept <dbl> 0.092151574, 0.015175760, 0.052049185, -0.051800379, 0.08…
$ b_x         <dbl> 0.5802608, 0.5863573, 0.4986585, 0.5511931, 0.6093724, 0.…
$ sigma       <dbl> 0.9753384, 1.0698012, 1.0501172, 0.9832180, 1.0370508, 1.…
$ Intercept   <dbl> 0.112773838, 0.036014691, 0.069771329, -0.032211171, 0.10…
$ lp__        <dbl> -291.8098, -291.9726, -292.0529, -292.4099, -291.6936, -2…
5

Thanks, should be fixed now. FYI, “Intercept” refers to the centered intercept, that is, when all standard predictors (fixed effects) have mean zero. The centering happens internally inside Stan to speed up computation and has no user visible effects otherwise (except that Intercept was accidentally exported).

6

Success! Thanks, @paul.buerkner.