Tuning multiple imputation by predictive mean matching and local residual draws

Table 1 Summary of existing software implementations of PMM and LRD

Software	Method	Command/instructions	Match	Option to	Default	Option to specify k*	Source of information
			types	specify	value
			available	match type†	of k*
R	PMM	mice.impute.pmm(mice package)	1	–	5	–	v2.18 documentation [12]
R	PMM	aregimpute (hmisc package)	1, 2	pmmtype = #	n _h	kclosest = #	v3.13-0 documentation [22]
R	PMM	bbpmm (Baboon package)	?	–	?	–	v0.1-6 documentation [13]
R	PMM	mi.pmm (mi package)	?	–	?	–	v0.09-18.03 documentation [14]
SAS	PMM	regpmm (statement within proc mi)	2	–	?	K = #	SAS website [15]
SAS	PMM	midas[31]	?	–	n _h	N/A donor selected from all h with probability proportional to a function of \|δ _hj\|	Reference [31]
Solas	PMM	Analyze → Multiple Imputation → Predictive Mean Matching method…	0	–	10	Select ‘Use # closest cases’ option in ‘Donor pool’ tab.	Solas website [16]
SPSS	PMM	Analyze → Multiple Imputation → Impute Missing Data Values. Under the ‘Method’ tab select ‘Custom’, and under the menu for ‘Model type for scale variables’ select ‘Predictive Mean Matching (PMM)’.	?	–	1	–	SPSS website [17]
Stata	PMM	mi impute pmm	2	–	1	knn(#)	Help file for mi impute pmm[18]
Stata	PMM	ice, match	1, 2	matchtype(#)	10	matchpool(#)	Help file for ice
Stata	LRD	ice, match uvisopts(lrd)	1, 2	matchtype(#)	10	matchpool(#)	Help file for ice

^*Type 0 matches linear predictors for observed and missing values; type 1 uses a draw of parameters for missing values before matching; type 2 uses a draw of parameters for both observed and missing.
^† k is the size of the donor pool.

ISSN: 1471-2288