First predictor is dropped when using "formulas"

[jawitte]

Hello,

I noticed that when I use the "formulas" argument of "mice", then the first predictor gets lost along the way. I think it happens within mice:::sampler.univ. My formula is "X ~ 0 + Y", which is updated to "X ~ Y - 1" in mice:::sampler.univ. Then the design matrix x is obtained, which has one column containing the values of Y. So far, so good; but then the first and only column is removed:
x <- x[, -1L, drop = FALSE]
Thus Y is not used to predict X.

The issue does not occur when I use the "predictorMatrix" argument to specify the same imputation model.

Here is an example illustrating the problem:
Example_for_GitHub.txt

I would very much appreciate if you could have a look at the issue!

[prockenschaub]

This does indeed seem to be a bug. It looks like the intercept (or what mice thinks to be the intercept) is removed twice:

1. in make.formulas by adding 0 +
2. in sampler.univ by removing the first column of the design matrix

My best guess is that this was changed at some point in sampler.univ and the corresponding code in make.formulas somehow slipped through the cracks.

Suggestion:
Change formulas[[h]] <- paste(paste(y, collapse = "+"), "~", paste(c("0", x), collapse = "+")) to formulas[[h]] <- paste(paste(y, collapse = "+"), "~", paste(x, collapse = "+")) in sampler.univ

Happy to make a PR if that seems like a reasonable solution.

[jawitte]

Yes, I think that would fix the issue! If you don't mind, would you make a pull request? Thanks :)

[stefvanbuuren]

Now merged. Thanks!