VarDecomp can be used for variance decomposition, model fit checks and output visualizations of brms models.
You can install the development version of VarDecomp like so:
devtools::install_github("gabewinter/VarDecomp")Full documentation website on: https://gabewinter.github.io/VarDecomp
library(tidyverse)
md = tibble(
sex = sample(rep(c(-0.5, 0.5), each = 500)),
species = sample(rep(c("species1","species2","species3","species4","species5"), each = 200))) %>%
## Create a variables
dplyr::mutate(height = rnorm(1000, mean = 170, sd = 10),
mass = 5 + 0.5 * height + rnorm(1000, mean = 0, sd = 5)) %>%
dplyr::mutate(height = height - mean(height),
mass = log(mass))
mod = brms_model(Chainset = 2,
Response = "mass",
FixedEffect = c("sex","height"),
Family = "gaussian",
Data = md)
#> [1] "No problems so far 🙌"
#> Compiling Stan program...
#> Start sampling
model_fit(mod, Group = "sex")
#> No divergences to plot.
#> Using all posterior draws for ppc type 'loo_pit_qq' by default.
#> Using all posterior draws for ppc type 'violin_grouped' by default.
#> $`R-hat and Effective sample size`
#> # A tibble: 1 × 2
#> Rhat EffectiveSampleSize
#> <dbl> <dbl>
#> 1 1.00 1650.
#>
#> $`Traceplots plot`
#>
#> $`Posterior predictive check - Density overlay plot`
#>
#> $`Posterior predictive check - LOO-PIT-QQ plot`
#>
#> $`Posterior predictive check - Group density overlay plot`
#> $`Posterior predictive check - Group density overlay plot`$GroupPlot_sex
model_summary(mod)
#> # A tibble: 7 × 6
#> variable mean median sd lower_HPD upper_HPD
#> <chr> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 Intercept 4.49 4.49 0.002 4.49 4.50
#> 2 sex 0.001 0.001 0.004 -0.006 0.008
#> 3 height 0.005 0.005 0 0.005 0.006
#> 4 R2_sex 0.001 0 0.001 0 0.002
#> 5 R2_height 0.46 0.461 0.021 0.417 0.499
#> 6 R2_sum_fixed_effects 0.461 0.462 0.021 0.418 0.499
#> 7 R2_residual 0.539 0.538 0.021 0.501 0.582
plot_intervals(mod)
PS = var_decomp(mod)
plot_R2(PS)