Assign 0 or 1 with different probabilities conditional on another column in R

Question

I am trying to assign a 0 or 1 with some stochasticity based on another column in a data frame (outcome). If outcome == 1, the new column exposure should equal 1 about 90% of the time. Conversely if outcome == 0 it should equal 1 about 20% of the time.

I am currently doing this with a for loop but wondering if there are more efficient/elegant ways to accomplish this (i.e. via vectorization).

To be clear, though the data frame is labeled example_data this is not an example - its the test data I am generating to test a series of functions related to GEE models.

set.seed(05062020)
example_data <- data.frame(id = as.factor(rep(sprintf("Record %s",seq(1:50)), each = 2)),
 outcome = as.factor(rep(sample(0:1, 50, prob = c(0.8,0.2), replace = TRUE), each = 2)))
for (i in 1:nrow(example_data)){
 example_data$exposure[i] <- ifelse(example_data$outcome[i] == 1, 
 sample(0:1, 1, prob = c(0.1, 0.9)),
 sample(0:1, 1, prob = c(0.8, 0.2)))
}

Answer 1

example_data$exposure <- ifelse(example_data$outcome == 1, 
 sample(0:1, nrow(example_data), prob = c(0.1, 0.9), replace = T),
 sample(0:1, nrow(example_data), prob = c(0.8, 0.2), replace = T))

ifelse is vectorized, so we can do this with one function call.

• jpsmith: Ack I tried this but forgot to add the replace = TRUE and so it didnt work. Thank you!

Answer 2

Sampling coin flips with data-dependent probabilities can often be done elegantly by thresholding a uniform random variable:

example_data$exposure <-
 as.numeric(runif(nrow(example_data)) <= 0.2 + 0.7*(example_data$outcome==1))

So basically the threshold is 0.2 when example_data$outcome == 0 and is 0.9 when example_data$outcome == 1. I used 0.7*(example_data$outcome==1) instead of just 0.7*example_data$outcome because example_data$outcome is defined as a factor in your data frame, and the as.numeric converts TRUE/FALSE into 1/0.