Search code examples
rhmisc

Adding imputed data from a model to a dataset - hmisc aregImpute

I'm trying to impute values from a dataset using hmisc. I'm following this guide.

Here is a reproducible example of my code:

#Create dataset and add 0.1 NA values randomly
data <- iris
library(missForest)
library(Hmisc)
iris.mis <- prodNA(iris, noNA = 0.1)

#Calculating imputed values with aregImpute
impute_arg <- aregImpute(~ Sepal.Length + Sepal.Width + Petal.Length + Petal.Width + Species, data = iris.mis, n.impute = 5)

completeData2 <- impute.transcan(impute_arg, imputation=1, data=iris.mis, list.out=TRUE,pr=FALSE, check=FALSE) 
head(completeData2)

#creating a fit model
library(rms)
fmi <- fit.mult.impute(Sepal.Length ~ Sepal.Width + Petal.Length + Petal.Width + Species, ols, impute_arg, data=iris.mis)

My question is: How do I apply this fit model to my data and impute the NA values in my dataset (iris.mis)?

Answers with code snippets would be greatly appreciated.

Solution

  • All you need to do is get the model's predictions:

    model_predictions <- predict(fmi)

    Now you can examine the predictions at the data's missing indices:

    missing <- which(is.na(iris.mis$Sepal.Length))
imputed <- model_predictions[missing]
imputed
#>         5        22        27        32        34        35        54        60 
#> 5.073695* 5.119113* 5.182343* 4.949794* 5.381427* 4.863149* 5.565716* 5.596861* 
#>        89       102       107       117       131       135       145       149 
#> 5.950823* 6.217764* 5.757642* 6.829916* 7.116657* 6.726274* 6.738296* 6.662452* 
#>       150 
#> 6.428420*

    And see how they compare to the actual values:

    actual <- iris$Sepal.Length[missing]

plot(x = actual, y = imputed, xlim = c(4, 8), ylim = c(4, 8), col = "red",
     xlab = "Actual", ylab = "Imputed", main = "Imputed vs Actual Sepal Length")
lines(c(4, 8), c(4, 8), lty = 2)

    enter image description here

    #>  # calculate residuals
imputed - actual 
#>            5           22           27           32           34           35 
#>  0.07369483*  0.01911295*  0.18234346* -0.45020634* -0.11857279* -0.03685114* 
#>           54           60           89          102          107          117 
#>  0.06571631*  0.39686061*  0.35082282*  0.41776385*  0.85764178*  0.32991602* 
#>          131          135          145          149          150 
#> -0.28334270*  0.62627448*  0.03829600*  0.46245174*  0.52842038* 
#>
#> # sum of squared errors
sum((imputed - actual)^2)
#> [1] 2.52802

    So, if you want a new column in your set complete with the imputations you can do

    iris.mis$Sepal.Length.Imputed <- iris.mis$Sepal.Length
iris.mis$Sepal.Length.Imputed[is.na(iris.mis$Sepal.Length.Imputed)] <- imputed