Prediction using BCF

In this vignette, we show how to use the bcf package to fit a model and use the fitted object to predict estimates for new data.

library(bcf)
library(latex2exp)
library(ggplot2)

We use the same data generating process as in the “Simple Example” vignette:

set.seed(1)

## Training data
p <- 3 # two control variables and one effect moderator
n <- 1000
n_burn <- 2000
n_sim <- 1500

x <- matrix(rnorm(n*(p-1)), nrow=n)
x <- cbind(x, x[,2] + rnorm(n))
weights <- abs(rnorm(n))

# create targeted selection, whereby a practice's likelihood of joining the intervention (pi) 
# is related to their expected outcome (mu)
mu <- -1*(x[,1]>(x[,2])) + 1*(x[,1]<(x[,2])) - 0.1

# generate treatment variable
pi <- pnorm(mu)
z <- rbinom(n,1,pi)

# tau is the true treatment effect. It varies across practices as a function of
# X3 and X2
tau <- 1/(1 + exp(-x[,3])) + x[,2]/10

# generate the response using q, tau and z
y_noiseless <- mu + tau*z

# set the noise level relative to the expected mean function of Y
sigma <- diff(range(mu + tau*pi))/8

# draw the response variable with additive error
y <- y_noiseless + sigma*rnorm(n)/sqrt(weights)

# Fitting the model
bcf_out <- bcf(y               = y,
               z               = z,
               x_control       = x,
               x_moderate      = x,
               pihat           = pi,
               nburn           = n_burn,
               nsim            = n_sim,
               w               = weights,
               n_chains        = 2,
               update_interval = 1)

Predicting using BCF

We make predictions at 10 new observations, including some extreme values:

set.seed(1)
n_test = 10
x_test <- matrix(rnorm(n_test*(p-1), 0, 2), nrow=n_test) # sd of 2 makes x_test more dispersed than x
x_test <- cbind(x_test, x_test[,2] + rnorm(n_test))
mu_pred  <- -1*(x_test[,1]>(x_test[,2])) + 1*(x_test[,1]<(x_test[,2])) - 0.1
pi_pred <- pnorm(mu_pred)
z_pred  <- rbinom(n_test,1, pi_pred)

We now predict \(y\) and estimate treatment effects \(\tau\) for these new observations based on our fitted model.

pred_out = predict(object=bcf_out,
                   x_predict_control=x_test,
                   x_predict_moderate=x_test,
                   pi_pred=pi_pred,
                   z_pred=z_pred,
                   n_cores = 1,
                   save_tree_directory = '.')
#> Initializing BCF Prediction
#> Starting Prediction 
#> Starting to Predict Chain  1 
#> Loading...
#> ntrees 50
#> p 3
#> ndraws 1500
#> done
#> Loading...
#> ntrees 200
#> p 4
#> ndraws 1500
#> done
#> Starting to Predict Chain  2 
#> Loading...
#> ntrees 50
#> p 3
#> ndraws 1500
#> done
#> Loading...
#> ntrees 200
#> p 4
#> ndraws 1500
#> done

Comparison

Let’s compare the results for our training and testing data. We will show the estimated treatment effects for training and test observations as a function of \(x_3\), which is an effect modifier.

tau_ests_preds <- data.frame(x     = c(x[,3], x_test[,3]),
                             Mean  = c(colMeans(bcf_out$tau), 
                                       colMeans(pred_out$tau)),
                             Low95 = c(apply(bcf_out$tau, 2, quantile, 0.025), 
                                       apply(pred_out$tau, 2, quantile, 0.025)),
                             Up95  = c(apply(bcf_out$tau, 2, quantile, 0.975), 
                                       apply(pred_out$tau, 2, quantile, 0.975)),
                             group = factor(c(rep("training", n), rep("testing", n_test))),
                             agroup = c(rep(0.2, n), rep(1, n_test)))
ggplot(tau_ests_preds, aes(x, Mean, color = group)) +
  geom_pointrange(aes(ymin = Low95, ymax = Up95), alpha = tau_ests_preds$agroup) +
  xlab(TeX("$x_3$")) +
  ylab(TeX("$\\hat{\\tau}$")) 

Note that the credible intervals get wider as the \(x_3\) values get closer to the end of the range, as we would hope.