Issue with 'Number of Logged Events' in Multiple Imputation Using MICE Package

[Isabella678]

Hello, I am encountering the "Number of logged events" issue while performing multiple imputation using the mice package. I have tried using the quickpred() function, but the problem persists.
loggedEvents_test.csv

Let me provide a detailed description of my variables. The sample size is over 30,000, and there are 101 variables in total. The first three columns have been excluded. Among the remaining variables, 57 are binary (with imputation method specified as "logreg"), 7 are ordinal (with imputation method specified as "polr"), 3 are unordered categorical (with imputation method specified as "polyreg"), and 31 are numeric (with imputation method specified as "pmm"). The data has been factorized first, followed by specifying the imputation methods according to the data types. The following code is used for imputation:

``
col.methods[c("sfid", "father", "mother")] <- ""

predictors <- quickpred(data)

imputed_data <- mice(data,
method = col.methods,
predictorMatrix = predictors,
m = 1,
maxit = 2,
print = FALSE)
``

My understanding is that the 'out' column represents the predictor variables that are excluded during imputation. So, should these logged events be handled or not?

I would greatly appreciate any help you can offer. Looking forward to your response. Thank you!

[stefvanbuuren]

This is not really a software issue, so I transfer to discussion.

The loggedEvents are warnings that some variables are removed due to multicollinearity. They are generally not problematic, but it's good practice to try to minimize them. In your case you could

1. set the predictorMatrix columns for sfid, father and mother to zero because micesees these as constants
2. set the predictorMatrix columns for eating_disorder1 and pers_dis1, as well as some others, to zero because these appear to have redundant information.

Hope this helps.

[Isabella678]

Thank you for your reply.

I tried excluding the variables eating_disorder and pers_dis, and indeed the number of logged events decreased. However, as the number of iterations increases, the logged events still tend to rise.

What I understand is that this is not an error, and the process can still continue with further analysis, such as comparing the data quality before and after imputation, calculating correlations to exclude highly correlated variables, and so on.

Is that correct?

Thank you for your patient response.