brms

The brms package provides an interface to fit Bayesian generalized (non-)linear multivariate multilevel models using Stan, which is a C++ package for performing full Bayesian inference (see http://mc-stan.org/). The formula syntax is very similar to that of the package lme4 to provide a familiar and simple interface for performing regression analyses. A wide range of distributions and link functions are supported, allowing users to fit -- among others -- linear, robust linear, count data, survival, response times, ordinal, zero-inflated, hurdle, and even self-defined mixture models all in a multilevel context. Further modeling options include non-linear and smooth terms, auto-correlation structures, censored data, meta-analytic standard errors, and quite a few more. In addition, all parameters of the response distribution can be predicted in order to perform distributional regression. Multivariate models (i.e. models with multiple response variables) can be fitted, as well. Prior specifications are flexible and explicitly encourage users to apply prior distributions that actually reflect their beliefs. Model fit can easily be assessed and compared with posterior predictive checks and leave-one-out cross-validation.

How to use brms

library(brms)

As a simple example, we use poisson regression to model the seizure counts in epileptic patients to investigate whether the treatment (represented by variable Trt) can reduce the seizure counts. Two group-level intercepts are incorporated to account for the variance between patients as well as for the residual variance.

fit <- brm(count ~ log_Age_c + log_Base4_c * Trt + (1|patient) + (1|obs), 
           data = epilepsy, family = poisson())
#> Compiling the C++ model
#> Start sampling

The results (i.e. posterior samples) can be investigated using

summary(fit, waic = TRUE) 
#>  Family: poisson 
#>   Links: mu = log 
#> Formula: count ~ log_Age_c + log_Base4_c * Trt + (1 | patient) + (1 | obs) 
#>    Data: epilepsy (Number of observations: 236) 
#> Samples: 4 chains, each with iter = 2000; warmup = 1000; thin = 1; 
#>          total post-warmup samples = 4000
#>     ICs: LOO = NA; WAIC = 1146.29; R2 = NA
#>  
#> Group-Level Effects: 
#> ~obs (Number of levels: 236) 
#>               Estimate Est.Error l-95% CI u-95% CI Eff.Sample Rhat
#> sd(Intercept)     0.37      0.04     0.29     0.46       1744 1.00
#> 
#> ~patient (Number of levels: 59) 
#>               Estimate Est.Error l-95% CI u-95% CI Eff.Sample Rhat
#> sd(Intercept)     0.51      0.07     0.38     0.67       1828 1.00
#> 
#> Population-Level Effects: 
#>                  Estimate Est.Error l-95% CI u-95% CI Eff.Sample Rhat
#> Intercept            1.74      0.11     1.53     1.96       2680 1.00
#> log_Age_c            0.48      0.37    -0.26     1.21       2445 1.00
#> log_Base4_c          0.88      0.14     0.59     1.15       2355 1.00
#> Trt1                -0.34      0.16    -0.64    -0.04       2740 1.00
#> log_Base4_c:Trt1     0.35      0.21    -0.08     0.77       2602 1.00
#> 
#> Samples were drawn using sampling(NUTS). For each parameter, Eff.Sample 
#> is a crude measure of effective sample size, and Rhat is the potential 
#> scale reduction factor on split chains (at convergence, Rhat = 1).

On the top of the output, some general information on the model is given, such as family, formula, number of iterations and chains, as well as the WAIC, which is an information criterion for Bayesian models. Next, group-level effects are displayed seperately for each grouping factor in terms of standard deviations and (in case of more than one group-level effect per grouping factor; not displayed here) correlations between group-level effects. On the bottom of the output, population-level effects are displayed. If incorporated, autocorrelation effects and family specific parameters (e.g., the residual standard deviation 'sigma' in normal models) are also given.

In general, every parameter is summarized using the mean ('Estimate') and the standard deviation ('Est.Error') of the posterior distribution as well as two-sided 95% credible intervals ('l-95% CI' and 'u-95% CI') based on quantiles. The last two values ('Eff.Sample' and 'Rhat') provide information on how well the algorithm could estimate the posterior distribution of this parameter. If 'Rhat' is considerably greater than 1, the algorithm has not yet converged and it is necessary to run more iterations and / or set stronger priors.

To visually investigate the chains as well as the posterior distributions, you can use

plot(fit) 

An even more detailed investigation can be achieved by applying the shinystan package:

launch_shinystan(fit) 

There are several methods to compute and visualize model predictions. Suppose that we want to predict responses (i.e. seizure counts) of a person in the treatment group (Trt = 1) and in the control group (Trt = 0) with average age and average number of previous seizures. Than we can use

newdata <- data.frame(Trt = c(0, 1), log_Age_c = 0, log_Base4_c = 0)
predict(fit, newdata = newdata, allow_new_levels = TRUE, probs = c(0.05, 0.95))
#>      Estimate Est.Error 5%ile 95%ile
#> [1,]  6.87775  5.412128     1     17
#> [2,]  4.93950  4.020806     0     13

We need to set allow_new_levels = TRUE because we want to predict responses of a person that was not present in the data used to fit the model. While the predict method returns predictions of the responses, the fitted method returns predictions of the regression line.

fitted(fit, newdata = newdata, allow_new_levels = TRUE, probs = c(0.05, 0.95))
#>      Estimate Est.Error    5%ile   95%ile
#> [1,] 6.823163  4.518070 2.032872 15.42288
#> [2,] 4.883973  3.255885 1.433088 10.86055

Both methods return the same etimate (up to random error), while the latter has smaller variance, because the uncertainty in the regression line is smaller than the uncertainty in each response. If we want to predict values of the original data, we can just leave the newdata argument empty.

A related feature is the computation and visualization of marginal effects, which can help in better understanding the influence of the predictors on the response.

plot(marginal_effects(fit, probs = c(0.05, 0.95)))

For a complete list of methods to apply on brms models see

methods(class = "brmsfit") 
#>  [1] add_ic                  add_loo                 add_waic               
#>  [4] as.array                as.data.frame           as.matrix              
#>  [7] as.mcmc                 bayes_factor            bayes_R2               
#> [10] bridge_sampler          coef                    control_params         
#> [13] expose_functions        family                  fitted                 
#> [16] fixef                   formula                 hypothesis             
#> [19] kfold                   launch_shinystan        log_lik                
#> [22] log_posterior           logLik                  loo                    
#> [25] LOO                     loo_linpred             loo_predict            
#> [28] loo_predictive_interval marginal_effects        marginal_smooths       
#> [31] model.frame             neff_ratio              ngrps                  
#> [34] nobs                    nsamples                nuts_params            
#> [37] pairs                   parnames                plot                   
#> [40] post_prob               posterior_interval      posterior_linpred      
#> [43] posterior_predict       posterior_samples       posterior_summary      
#> [46] pp_check                pp_mixture              predict                
#> [49] predictive_error        print                   prior_samples          
#> [52] prior_summary           ranef                   residuals              
#> [55] rhat                    stancode                standata               
#> [58] stanplot                summary                 update                 
#> [61] VarCorr                 vcov                    waic                   
#> [64] WAIC                   
#> see '?methods' for accessing help and source code

Details on formula syntax, families and link functions, as well as prior distributions can be found on the help page of the brm function:

help("brm") 

More instructions on how to use brms are given in the package's main vignettes.

vignette("brms_overview")
vignette("brms_multilevel")

FAQ

How do I install brms?

To install the latest release version from CRAN use

install.packages("brms")

The current developmental version can be downloaded from github via

if (!require("devtools")) {
  install.packages("devtools")
}
devtools::install_github("paul-buerkner/brms", dependencies = TRUE)

Because brms is based on Stan, a C++ compiler is required. The program Rtools (available on https://cran.r-project.org/bin/windows/Rtools/) comes with a C++ compiler for Windows. On Mac, you should install Xcode. For further instructions on how to get the compilers running, see the prerequisites section on https://github.com/stan-dev/rstan/wiki/RStan-Getting-Started.

What is the best way to ask a question or propose a new feature?

Questions can be asked in the google group brms-users. To propose a new feature or report a bug, please open an issue on github. Of course, you can always write me an email (paul.buerkner@gmail.com).

How can I extract the generated Stan code?

If you have already fitted a model, just apply the stancode method on the fitted model object. If you just want to generate the Stan code without any model fitting, use the make_stancode function.

Can I avoid compiling models?

When you fit your model for the first time with brms, there is currently no way to avoid compilation. However, if you have already fitted your model and want to run it again, for instance with more samples, you can do this without recompilation by using the update method. For more details see

help("update.brmsfit")

How can I specify non-linear or distributional models?

Specification of non-linear or distributional models requires multiple formulae. In brms, the function brmsformula (or short bf) is used to combine all formulae into one object, which can then be passed to the formula argument of brm. More help is given in

help("brmsformula")

For a detailed discussion of some examples see

vignette("brms_nonlinear")

vignette("brms_distreg")

What is the difference between brms and rstanarm?

rstanarm is an R package similar to brms that also allows to fit regression models using Stan for the backend estimation. Contrary to brms, rstanarm comes with precompiled code to save the compilation time (and the need for a C++ compiler) when fitting a model. However, as brms generates its Stan code on the fly, it offers much more flexibility in model specification than rstanarm. Also, multilevel models are currently fitted a bit more efficiently in brms. For a detailed comparison of brms with other common R packages implementing multilevel models, see

vignette("brms_overview")