I suspect that there are parallels with other questions but I haven’t been able to find a combination which works in this situation.

In essence I am trying to use a for loop to do multiple imputation (I am aware that there are packages which can do the whole analysis but I need to extract interim stages).

Starting with a data set like this, where y is the outcome, x has the missing values and the imputed missing values are in a1-3.

dat <- data.frame(y = c(1,0,0,1,0),
                  x= c(1,NA,2,NA,1),
                 a1 = c(NA,1,NA,2,NA),
                 a2 = c(NA,1,NA,1,NA),
                 a3 = c(NA,2,NA,1,NA))

I thought that I could loop over a1-3, coalescing the values into the x variable and then running the anaysis but this doesn’t seem to work. I’ve tried:

for (i in 1:3) {
  dat$x <- coalesce(dat$x,dat$a[i])
  z <- glm(y ~ x, data=dat, family="binomial")
  res  <- summary(z)$coefficient
  print(res)
}

But the first line dat$x <- coalesce(dat$x,dat$a[i]) clearly isn’t doing what I think it should as all three “res” are the same and if I print dat$x after this line it hasn’t been updated with the values from A1-3.

I’d thought that using mutate from dplyr might work as in:

for (i in 1:3) {
  dat %>%
  mutate(x = coalesce(x,a[i]))
  z <- glm(y ~ x, data=dat, family="binomial")
  res  <- summary(z)$coefficient
  print(res)
}

But that gives the error “object ‘a’ not found”. I’ve attempted writing a function and then using apply instead of trying to loop but didn’t get any further with that approach. Based on similar questions, I guess this is all to do with how the loop is using the column names but I’ve not managed to implement any of the solutions successfully.

Any hints gratefully received.