Packages

library(dplyr)
library(LexOPS)

Simulating Dataset

Firstly, we will simulate the imaginary dataset of the participant pool, consisting of 10000 potential subjects, representing the kind of data we might expect to have.

n_sub <- 10000

pool <- tibble(
  subj_id = sprintf("s%04d", 1:n_sub),
  age = runif(n_sub, 18, 50),
  sex = sample(
    c("m", "f", NA), n_sub,
    replace = TRUE,
    prob = c(0.48, 0.48, 0.04)
  ),
  bmi = rnorm(n_sub, 25, 5),
  iq = rnorm(n_sub, 100, 15)
)

Choosing Subjects

Now let’s imagine we want to assign 50 subjects to an intervention group, and 50 matched subjects to a control group. We want to match by the relevant subject variables of

• Age (±1 year)
• Sex (match exactly)
• BMI (±0.5)
• IQ (±5)

In LexOPS, we could simply write this as follows:

study_subj <- pool |>
  subset(!is.na(sex)) |>
  set_options(id_col = "subj_id") |>
  split_random(2) |>
  control_for(age, -1:1) |>
  control_for(sex) |>
  control_for(bmi, -0.5:0.5) |>
  control_for(iq, -5:5) |>
  generate(50)

## Generated 2/50 (4%). 2 total iterations, 1.00 success rate.
Generated 5/50 (10%). 5 total iterations, 1.00 success rate.
Generated 8/50 (16%). 9 total iterations, 0.89 success rate.
Generated 10/50 (20%). 12 total iterations, 0.83 success rate.
Generated 12/50 (24%). 14 total iterations, 0.86 success rate.
Generated 15/50 (30%). 17 total iterations, 0.88 success rate.
Generated 18/50 (36%). 23 total iterations, 0.78 success rate.
Generated 20/50 (40%). 25 total iterations, 0.80 success rate.
Generated 22/50 (44%). 27 total iterations, 0.81 success rate.
Generated 25/50 (50%). 32 total iterations, 0.78 success rate.
Generated 28/50 (56%). 38 total iterations, 0.74 success rate.
Generated 30/50 (60%). 40 total iterations, 0.75 success rate.
Generated 32/50 (64%). 42 total iterations, 0.76 success rate.
Generated 35/50 (70%). 46 total iterations, 0.76 success rate.
Generated 38/50 (76%). 49 total iterations, 0.78 success rate.
Generated 40/50 (80%). 52 total iterations, 0.77 success rate.
Generated 42/50 (84%). 54 total iterations, 0.78 success rate.
Generated 45/50 (90%). 57 total iterations, 0.79 success rate.
Generated 48/50 (96%). 60 total iterations, 0.80 success rate.
Generated 50/50 (100%). 64 total iterations, 0.78 success rate.

This returns a dataframe, listing the subject IDs for the 50 subjects in each group. Here are the first 5 rows (10 subjects):

head(study_subj, 5)

We can see the subjects’ data in long format with the long_format() function. Here is the data for those same 10 subjects in long format. The item_nr column indicates which subjects are matched to one another.

study_subj |>
  long_format() |>
  head(10)

Checking the Results

We can use the plot_design() function to see how well our numeric variables have been controlled for. Individual points represent subjects, with matched subjects connected by lines. Variables more tightly controlled show more similar distributions, and only gentle slopes between points.

plot_design(study_subj)

We can check how many males and females we have in each group like so:

study_subj |>
  long_format() |>
  count(condition, sex)

Finally, we can use plot_sample() to see how representative our sample is of our whole participant pool.

plot_sample(study_subj)