Bug in generate_data()

[numbersman77]

To do:

• Write a test that can show that there is this bug
• Fix the problem (using suggested code below)
• Ensure all tests pass
• Add a NEWS entry

Tobias Mielke found a (likely) bug in our package:

I might have detected a small bug in your code, which I would like to raise to your awareness. The function currently has:

result$type_of_event <- 1L + stats::rbinom(n = n, size = 2,prob = haz/haz_all)

If you test this code, you’ll see that there will be some confusion within R on the length of your generated data (n), the probability statement (prob) and the size. An alternative to your implementation might be the one below.

typeOfEvent=runif(n,min=0,max=1)
probs=cumsum(haz/haz_all)
result$type_of_event=rep(3,n)
result$type_of_event[which(typeOfEvent<probs[1])]=1
result$type_of_event[which(typeOfEvent>probs[1]&typeOfEvent>probs[2])]=2

I think he has a point here, and Regina Stegherr agress as well. We propose to replace

result$type_of_event <- 1L + stats::rbinom(n = n, size = 2, prob = haz/haz_all)

through

result$type_of_event <- sample(1:3, size = n, prob = haz / haz_all, replace = TRUE).

Simple exploration like

haz <- c(0.2, 0.3, 0.5)
haz_all <- sum(haz)
table(sample(1:3, size = 10^6, prob = haz / haz_all, replace = TRUE))/10^6

seems to show that this does the right thing.