Difference between printed and specified predictor matrix

[kasiakasia93]

Hello, I am new to github discussions and mice. I have looked through the main van Buuren paper and the function documentation as well as the discussion threads and still need some help.

My goal is to impute 2 continuous variables (bounded between 0-100) in a longitudinal data set that is in long format. The data set is unbalanced without fixed measurement occasions.

It is my understanding that I will use the 2L.norm method to impute these variables.

I have edited a prediction matrix to include my covariates of interest (specified at '1') and the identification variable (specified as '-2'). I have not specified any random effects.

When I run my mice object it does output that 2l.norm has been used but the predictor matrix is different than the one if i do imp$predictorMatrix.

Is this a bug or something wrong with my process? Which predictor matrix is being used?

Thank you for any help!

• Katherine

#simpleMI
imp1 <- mice(dataset1, maxit=1,seed = 123)

#edit predictor matrix for multilevel
pred<-imp1$pred
pred["Score1","id"] <- c(-2)
pred["Score2","id"] <- c(-2)
pred["Score1",c("c1","c2","c3","c4","c5","c6","Score1","c8","c9","c10","c11","c12","Score2")]<- c(1,1,1,1,1,1,0,1,1,1,1,1,0)
pred["Score2",c("c1","c2","c3","c4","c5","c6","Score1","c8","c9","c10","c11","c12","Score2")]<- c(1,1,1,1,1,1,0,1,1,1,1,1,0)
pred["c1",c("id","c1","c2","c3","c4","c5","c6","Score1","c8","c9","c10","c11","c12","Score2")]<- c(0,0,0,0,0,0,0,0,0,0,0,0,0,0)
pred["c2",c("id","c1","c2","c3","c4","c5","c6","Score1","c8","c9","c10","c11","c12","Score2")]<- c(0,0,0,0,0,0,0,0,0,0,0,0,0,0)
pred["c3",c("id","c1","c2","c3","c4","c5","c6","Score1","c8","c9","c10","c11","c12","Score2")]<- c(0,0,0,0,0,0,0,0,0,0,0,0,0,0)
pred["c4",c("id","c1","c2","c3","c4","c5","c6","Score1","c8","c9","c10","c11","c12","Score2")]<- c(0,0,0,0,0,0,0,0,0,0,0,0,0,0)
pred["c5",c("id","c1","c2","c3","c4","c5","c6","Score1","c8","c9","c10","c11","c12","Score2")]<- c(0,0,0,0,0,0,0,0,0,0,0,0,0,0)
pred["c6",c("id","c1","c2","c3","c4","c5","c6","Score1","c8","c9","c10","c11","c12","Score2")]<- c(0,0,0,0,0,0,0,0,0,0,0,0,0,0)
pred["c8",c("id","c1","c2","c3","c4","c5","c6","Score1","c8","c9","c10","c11","c12","Score2")]<- c(0,0,0,0,0,0,0,0,0,0,0,0,0,0)
pred["c9",c("id","c1","c2","c3","c4","c5","c6","Score1","c8","c9","c10","c11","c12","Score2")]<- c(0,0,0,0,0,0,0,0,0,0,0,0,0,0)
pred["c10",c("id","c1","c2","c3","c4","c5","c6","Score1","c8","c9","c10","c11","c12","Score2")]<- c(0,0,0,0,0,0,0,0,0,0,0,0,0,0)
pred["c11",c("id","c1","c2","c3","c4","c5","c6","Score1","c8","c9","c10","c11","c12","Score2")]<- c(0,0,0,0,0,0,0,0,0,0,0,0,0,0)
pred["c12",c("id","c1","c2","c3","c4","c5","c6","Score1","c8","c9","c10","c11","c12","Score2")]<- c(0,0,0,0,0,0,0,0,0,0,0,0,0,0)
pred["id",c("id","c1","c2","c3","c4","c5","c6","Score1","c8","c9","c10","c11","c12","Score2")]<- c(0,0,0,0,0,0,0,0,0,0,0,0,0,0)

## confirm that the matrix is what I want 
pred

#multi-level imputation

imp2 <- mice(dataset1,m=5, seed = 123, meth = c("","","","","","","","2l.norm","","","","","","2l.norm"), pred = pred, maxit=1)

imp2
# here the predictor matrix is incorrect 

#but here it is correct
imp2$predictorMatrix

mice.data<-complete(imp2)
#describe(mice.data) it also does NOT respect the bounds of the variable so I will have to use the squeeze function somehow

[stefvanbuuren]

Welcome to MICE discussions.

There is a difference between the real object imp2$predictorMatrix and the version that is printed on the terminal by imp2. Here's what mice does to produce the printed version:

library(mice, warn.conflicts = FALSE)
mice:::print.mids
#> function (x, ...) 
#> {
#>     cat("Class: mids\n")
#>     cat("Number of multiple imputations: ", x$m, "\n")
#>     cat("Imputation methods:\n")
#>     print(x$method, ...)
#>     cat("PredictorMatrix:\n")
#>     print(head(x$predictorMatrix), ...)
#>     if (!is.null(x$loggedEvents)) {
#>         cat("Number of logged events: ", nrow(x$loggedEvents), 
#>             "\n")
#>         print(head(x$loggedEvents), ...)
#>     }
#>     invisible(x)
#> }
#> <bytecode: 0x10f1e5af8>
#> <environment: namespace:mice>

^{Created on 2023-08-18 with reprex v2.0.2}

The line print(head(x$predictorMatrix), ...) prints only the first six lines in order to keep output manageable. Internally, mice always uses imp2$predictorMatrix.