bblocks is designed to prepare data for blocked, or stratified, randomization. Given a minimum block size and a set of covariates, bblocks efficiently partitions the covariate space into subgroups with reduced heterogeneity.

You can install the released version of bblocks from CRAN with:

install.packages("bblocks")

And the development version from GitHub with:

# install.packages("devtools")
devtools::install_github("notsaragong/bblocks")

bblocks(data, min_block_size, blockby)

data is a dataframe with variables describing background characteristics.

min_block_size is a number indicating the desired minimum number of observations in each block.

blockby is a list of specifications for partitioning the covariate space, in order of decreasing importance.

bblocks iteratively sorts data in the order of covariates described by blockby, then works backwards to combine subgroups until there are at least min_block_size observations in each block. Within each covariate, blocking is performed according to the specifications given by the elements of blockby. Observations that do not match any listed specification are also grouped together. Note that if there are outliers, there may not be a solution to the sorting algorithm. In this case, try reordering the specifications in blockby or using broader specifications.

Calling bblocks on a dataframe of N observations will return an N-length column vector assigning each observation to a block. Randomization may then be carried out using any preferred randomization package.

The following code performs blocking by three variables in the built-in dataset mtcars. Within the variable carb, carb == 1 and carb == 2 are grouped together, as are carb == 3, carb == 4, and carb == 5.

library(bblocks)

# load built-in dataset and create useful variable
data("mtcars")
mtcars$mpg.over.18 <- ifelse(mtcars$mpg > 18, "Yes", "No")

# perform blocking
mtcars$block <- bblocks(mtcars, 4, blockby = list(mpg.over.18 = "Yes",
                                                  mpg.over.18 = "No",
                                                  gear = 4,
                                                  carb = c(1, 2),
                                                  carb = c(3, 4, 5)))
#> [1] "Successfully created 5 blocks of minimum size 4."

In this case, bblocks returns a vector with a maximum of 3x3x2 = 18 unique blocks.