R help - Jul 2022 - mice: selecting small subset of variables to impute from dataset with many variables (> 2500)

Hello,

I am looking for some advice on how to select subsets of variables for
imputing when using the mice package.
>From Van Buuren's original mice paper, I see that selecting variables to
be
'skipped' in an imputation can be written as:

ini <- mice(nhanes2, maxit = 0, print = FALSE)
pred <- ini$pred
pred[, "bmi"] <- 0
meth <- ini$meth
meth["bmi"] <- ""

With the last two lines specifying the the "bmi" variable gets skipped
over
and not imputed.

And I have come across other examples, but all that I have seen lay out a
method of skipping variables where EVERY variable is named (as "bmi"
is
named above). I am wondering if there is a reasonably easy way to select
out approximately 30 variables for imputation from a larger dataset with
around 2500 variables, without having to name all 2450+ other variables.

Thank you,

Ian

	[[alternative HTML version deleted]]

If I understand your query correctly, you can use negative indexing to
omit variables. See ?'[' for details.
> dat <- data.frame (a = 1:3, b = letters[1:3], c = 4:6, d = letters[5:7])
> dat
  a b c d
1 1 a 4 e
2 2 b 5 f
3 3 c 6 g
> dat[,-c(2,4)]
  a c
1 1 4
2 2 5
3 3 6

Of course you have to know the numerical index of the columns you wish
to omit, but somethingh of the sort seems unavoidable in any case.

Cheers,
Bert

On Thu, Jul 14, 2022 at 11:00 AM Ian McPhail <ivmcphail at gmail.com>
wrote:
>
> Hello,
>
> I am looking for some advice on how to select subsets of variables for
> imputing when using the mice package.
>
> From Van Buuren's original mice paper, I see that selecting variables
to be
> 'skipped' in an imputation can be written as:
>
> ini <- mice(nhanes2, maxit = 0, print = FALSE)
> pred <- ini$pred
> pred[, "bmi"] <- 0
> meth <- ini$meth
> meth["bmi"] <- ""
>
> With the last two lines specifying the the "bmi" variable gets
skipped over
> and not imputed.
>
> And I have come across other examples, but all that I have seen lay out a
> method of skipping variables where EVERY variable is named (as
"bmi" is
> named above). I am wondering if there is a reasonably easy way to select
> out approximately 30 variables for imputation from a larger dataset with
> around 2500 variables, without having to name all 2450+ other variables.
>
> Thank you,
>
> Ian
>
>         [[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.

Hello,

You can use mice() argument predictorMatrix to tell mice() which 
variables/blocks are used when imputing which column. If the column 
vector is set to zeros, no column or block will used in its imputation.


library(mice)

predmat <- matrix(1L, ncol(nhanes2), ncol(nhanes2),
                   dimnames = list(names(nhanes2), names(nhanes2)))
diag(predmat) <- 0L
predmat[, "bmi"] <- 0L
predmat
#>     age bmi hyp chl
#> age   0   0   1   1
#> bmi   1   0   1   1
#> hyp   1   0   0   1
#> chl   1   0   1   0




Then use argument where to skip the variables you do not want imputed.
Note that this is not the same as not being imputed according to 
variables shown above as rownames of predmat.

The default of where is the matrix is.na(nhanes2) so make a copy of this 
matrix then set column "bmi" to FALSE. Then call mice().



predmat <- matrix(1L, ncol(nhanes2), ncol(nhanes2),
                   dimnames = list(names(nhanes2), names(nhanes2)))
diag(predmat) <- 0L
predmat[, "bmi"] <- 0L
predmat
#>     age bmi hyp chl
#> age   0   0   1   1
#> bmi   1   0   1   1
#> hyp   1   0   0   1
#> chl   1   0   1   0

not_bmi <- is.na(nhanes2)
not_bmi[, "bmi"] <- FALSE

ini_all <- mice(nhanes2, print = FALSE)
ini_bmi <- mice(nhanes2,
                 predictorMatrix = predmat,
                 where = not_bmi,
                 print = FALSE)


cmpl_all <- complete(ini_all)
head(cmpl_all)
#>     age  bmi hyp chl
#> 1 20-39 28.7  no 187
#> 2 40-59 22.7  no 187
#> 3 20-39 30.1  no 187
#> 4 60-99 27.5 yes 284
#> 5 20-39 20.4  no 113
#> 6 60-99 20.4  no 184
cmpl_bmi <- complete(ini_bmi)
head(cmpl_bmi)
#>     age  bmi hyp chl
#> 1 20-39   NA  no 187
#> 2 40-59 22.7  no 187
#> 3 20-39   NA  no 187
#> 4 60-99   NA yes 206
#> 5 20-39 20.4  no 113
#> 6 60-99   NA yes 184


Hope this helps,

Rui Barradas

?s 18:59 de 14/07/2022, Ian McPhail escreveu:
> Hello,
> 
> I am looking for some advice on how to select subsets of variables for
> imputing when using the mice package.
> 
>  From Van Buuren's original mice paper, I see that selecting variables
to be
> 'skipped' in an imputation can be written as:
> 
> ini <- mice(nhanes2, maxit = 0, print = FALSE)
> pred <- ini$pred
> pred[, "bmi"] <- 0
> meth <- ini$meth
> meth["bmi"] <- ""
> 
> With the last two lines specifying the the "bmi" variable gets
skipped over
> and not imputed.
> 
> And I have come across other examples, but all that I have seen lay out a
> method of skipping variables where EVERY variable is named (as
"bmi" is
> named above). I am wondering if there is a reasonably easy way to select
> out approximately 30 variables for imputation from a larger dataset with
> around 2500 variables, without having to name all 2450+ other variables.
> 
> Thank you,
> 
> Ian
> 
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.