Comprehension of MICE package in R: influence of parameter maxit

Question

I have problems in understanding maxit parameter of the mice package:

library(mice)
data(nhanes)
str(nhanes)
imp = mice(nhanes, m=5, printFlag=FALSE, maxit = 30, seed=2525)

As described here I want to check the plausibility of the imputation results by considering the density plot densityplot(imp) (expecting similarity) and the convergence plot(imp) .

I know plot(imp) shows as lines all m Imputations (giving different results based on Gibbs Sampling) over the iterations (defined by maxit). When the trace lines reach a value and fluctuate slightly around it, convergence has been achieved (from here). But how does the algorithm improve the predictive value in each Iteration?

Answer 1

Here is the MICE algorithm, from the book that one of the MICE authors wrote about using MICE. A detailed explanation is above my pay grade, but the gist is:

• Initially populate the missing values with random draws of non-missing values.
• Iterate (up to maxit):
  • For each variable:
    • Take draws of new and (hopefully) improved values, conditioned on this variable's observed data, current "complete"-by-imputation other variables' data, and a draw from the phi distribution (not sure about that part... I think it's the currently estimated distribution of the variable currently being imputed. Like I said, above my pay grade.).

Also, another section from the book, about convergence, which also discusses iterations and the order that the variables are visited for the imputation looping.

Reference: Flexible Imputation of Missing Data, Second Edition. 2018. Stef van Buuren.