Reordering brms marginal_effects plots by estimate values

Hi all,

For brms marginal_effects plots for categorical or ordinal data, is there an easy way to reorder the values by their brms_fit estimate values? I know this is possible in ggplot via aes(reorder), but am not too sure how this can be done in marginal_effects.

Thank you!

  • Operating System: Windows 10 x64
  • brms Version: 2.10.0

This seems like kind of a niche question because the size of parameter estimates is scale-dependent and not always directly comparable in terms of magnitude. For example, a coefficient estimate of 3 may be large when applied to USDollars but small when applied to a binary variable like married. And what if your intercept or scale parameter has a larger magnitude than your covariates?

So in any case I’m not sure the option you’re describing should be included in brms but you could hack the result you’re looking for by specifying the order of the parameter plots using the effects parameter like so:

N <- 100
a <- rnorm(N, 0, 1)
b <- rnorm(N, 0, 1)
c <- rnorm(N, 0, 1)

# effect sizes are: b > c > a
y <- rnorm(N, 
           1 * a + 
           3 * b + 
           2 * c, .1)

fit <- brm(y ~ a + b + c, 
           data = data.frame(a,b,c,y))

# extract magnitude of fixed effects, minus the intercept
no_intercept <- fixef(fit)[-1,"Estimate"]
# get names in order of the size of the parameter estimates 
effect_names_in_order <- names(sort(no_intercept, decreasing=TRUE))

# plot marginal effects, specifying the effects in that order
marginal_effects(fit, effects = effect_names_in_order)
2 Likes

In the doc of marginal_effects you see examples of how to extract the underlying plot objects.

2 Likes