Skip to content

Issue with get_variance() handling of intercept-only sigma model in brms #967

@matschmitz

Description

@matschmitz

It appears that get_variance() in insight (v0.99.0.13) doesn’t handle brms models with an intercept-only sigma term correctly. When comparing models with and without sigma, the expected residual variance (var.residual) and distribution variance (var.distribution) should match, as both models are essentially equivalent. However, the results differ.

library(brms)
library(insight)

# Model without sigma
mdl1 <- brm(mpg ~ hp + (1 | cyl), data = mtcars, seed = 123)
get_variance(mdl1)
# Output:
# $var.fixed
# [1] 4.911032
# $var.random
# [1] 22.40697
# $var.residual
# [1] 10.93045
# $var.distribution
# [1] 10.93045

# Model with intercept-only sigma
mdl2 <- brm(bf(mpg ~ hp + (1 | cyl), sigma ~ 1), data = mtcars, seed = 123)
get_variance(mdl2)
# Output:
# $var.fixed
# [1] 4.910691
# $var.random
# [1] 23.4812
# $var.residual
# [1] 1
# $var.distribution
# [1] 1

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions