Pakillo / DHARMa.helpers

Helper functions to check Bayesian brms models with DHARMa
https://pakillo.github.io/DHARMa.helpers/
Other
3 stars 0 forks source link
bayesian r-package statistics

DHARMa.helpers

R-CMD-check HitCount HitCount

https://pakillo.github.io/DHARMa.helpers

DHARMa.helpers is an R package that facilitates checking fitted statistical models via the DHARMa package. By now, only Bayesian models fitted with brms are implemented. See this blogpost for a detailed explanation of the approach.

Installation

# install.packages("remotes")
remotes::install_github("Pakillo/DHARMa.helpers")

Example

library(brms)
library(DHARMa.helpers)

Poisson regression

Fit model:

# Example model taken brms::brm()
# Poisson regression for the number of seizures in epileptic patients
fit1 <- brm(count ~ zAge + zBase * Trt + (1|patient),
            data = epilepsy, family = poisson(), refresh = 0)
#> Compiling Stan program...
#> Start sampling

Check with DHARMa:

simres <- dh_check_brms(fit1, integer = TRUE)

Note that we use integer = TRUE in this case as we are modelling a discrete response (counts).

Now check residuals against a predictor (zAge):

plot(simres, form = epilepsy$zAge)

Test overdispersion:

DHARMa::testDispersion(simres)
#> 
#>  DHARMa nonparametric dispersion test via sd of residuals fitted vs.
#>  simulated
#> 
#> data:  simulationOutput
#> dispersion = 1.1747, p-value = 0.194
#> alternative hypothesis: two.sided

See https://pakillo.github.io/DHARMa.helpers/reference/dh_check_brms.html for more examples.