multiple imputation diagnostics for non-iterative methods

[Daniel Rosler]

Dear Statalist users,

I am conducting multiple imputations by predictive mean matching for 3 continuous variables with a monotone missingness pattern in a multilevel dataset. Since this procedure is non-iterative, I can not use trace plots to assess the imputed values. What are conventional methods to diagnose (visually and descriptively) non-iterative multiple imputations?

Here is the full imputation command (and the analysis model) I am using:

egen incrank = xtile(inc), n(100) by(country)
gen incr_pdi = incrank * pdi
gen incr_idv = incrank * idv

mi set mlong
mi register imputed incrank incr_pdi incr_idv
mi register passive inc
set seed 46352

mi impute monotone (pmm, knn(5)) incr* = i.sex i.age i.educ i.empstat i.country, add(70) rseed(46352)

mi estimate: mixed sa incrank i.educ i.age i.sex i.empstat pdi idv gdp gini_disp incr_pdi incr_idv || country: incrank, mle

Thank you in advance for your help!

Best regards,
Daniel Rösler

[Rich Goldstein]

I'm not entirely sure what some of your language above means but I think that -midiagplots- will at least get you started; use -search midiagplots- to find and install

[Daniel Rosler]

Thank you for answer Rich. Miadiagplots indeed looks very useful.

To clarify, I was wondering whether for non-iterative imputations there was an alternative method to check for convergence through trace plots and auto-correlation plots or if this is not necessary in this case. Also, is it common to use non-iterative imputation methods for variables with monotone missing patterns or are there better alternatives for that?