1
$\begingroup$

I have a dataframe with 6 columns, X1_t1, X2_t1, X3_t1, X1_t2, X2_t2, X3_t2, each of them have some missing values, which one of the following options is recommended if I want to examine the change of X1, X2, X3 from t1 to t2?

  1. impute with the variables that are involved in each test only
library(mice)
library(MKmisc)
imp <- mice(df[,c(1,4)],5,seed=2375)
dataset1 <- complete(imp,1)
dataset2 <- complete(imp,2)
dataset3 <- complete(imp,3)
dataset4 <- complete(imp,4)
dataset5 <- complete(imp,5)
dataset.imp <- list(dataset1, dataset2, dataset3, dataset4, dataset5)
mi.t.test(dataset.imp, x = "X1_t1", y = "X1_t2", paired = TRUE)

change the column index and repeat the same steps for X2 and X3

  1. impute with all the variables
imp <- mice(df,5,seed=2375)
dataset1 <- complete(imp,1)
dataset2 <- complete(imp,2)
dataset3 <- complete(imp,3)
dataset4 <- complete(imp,4)
dataset5 <- complete(imp,5)
dataset.imp <- list(dataset1, dataset2, dataset3, dataset4, dataset5)
mi.t.test(dataset.imp, x = "X1_t1", y = "X1_t2", paired = TRUE)
mi.t.test(dataset.imp, x = "X2_t1", y = "X2_t2", paired = TRUE)
mi.t.test(dataset.imp, x = "X3_t1", y = "X3_t2", paired = TRUE)

I have tried both, but they give quite different results, and I don't know which one to trust.

$\endgroup$
2
  • $\begingroup$ I found the answer from the guidelines in this book. bookdown.org/mwheymans/bookmi/… $\endgroup$
    – Ashley
    Mar 1, 2021 at 0:58
  • $\begingroup$ Variables that are part of the analysis should be included. For the rest of the variables, I should examine whether are related to missingness. If related, they should be included. Otherwise, they should not be included. $\endgroup$
    – Ashley
    Mar 1, 2021 at 1:50

1 Answer 1

0
$\begingroup$

In general it's better to use other variables in the imputations. These other variables can have information used to improve the imputation models. But if the results depend on the nature of the imputations you've used then you should definitely describe that as a limitation and consider including a sensitivity analysis.

$\endgroup$
1
  • $\begingroup$ Hi David, thank you! Do you have any recommendations for materials for sensitivity analysis? I'm new to multiple imputations. $\endgroup$
    – Ashley
    Feb 28, 2021 at 8:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.