Table 3

Techniques for identifying matched records with discrepant first, middle and/or last names

Extent of discrepancy	Source of name discrepancy	Retrieval technique	Place of application*
Slight or moderate	Misspelling, entry errors, and so on	Jaro-Winkler distance	Throughout all steps
	Phonetically similar but different spelling	Encoded all names using the Soundex function in R to allow matching of phonetically similar names36	Throughout all steps
	Shortened or expanded/hyphenated versions of same names	Allowed for substring matches between name fields	Step A: name bins 3, 4, 5 Step B: name bins 2a, 4 Step D: name bin 5
Extreme	Use of nicknames and contractions (eg, Elizabeth—Betty, Tommy Joe—TJ)	Allowed for matches to common nicknames (see section VII of online supplementary appendix)	Step A: name bins 1, 2, 6, 7, 9 Step B: blocking key; substep 1 bin 2a, 2b, 3a, 4; substeps 3(2)(a) and (b) Step C: blocking key; substep 2 Step D: blocking key; name bin 1
	Change or concatenation of last names among females	Relaxed last name matching criteria	Step C
	Change or concatenation of last names among females	Allowed for matches between current last names (in purchaser and mortality records) and former last names (in voter records)	Throughout all steps
	Switches in name order	Allowed for reverse matching of first-middle and first-last	Step A: name bins 1, 2, 3, 5, 6, 9 Step D: name bins 1, 5

*Refers to locations in the charts of linkage algorithms provided in section V of the online supplementary appendix.