Automated Linking of Historical Data

• DOI: 10.1257/jel.20201599
• Publication Date: 2021-09-09T14:15:46Z
• Authors: Ran Abramitzky; Leah Boustan; Katherine Eriksson; James Feigenbaum; Santiago Pérez
• Keywords: Record Linkage; Digitization; Linkage (software); TRACE (psycholinguistics); Automated method

The recent digitization of complete count census data is an extraordinary opportunity for social scientists to create large longitudinal datasets by linking individuals from one another or other sources the census. We evaluate different automated methods record linkage, performing a series comparisons across and against hand linking. have three main findings that lead us conclude perform well. First, number generate very low (less than 5 percent) false positive rates. trace out frontier illustrating trade-off between rate (true) match rate. Relative more conservative algorithms, humans tend link observations but at cost higher rates positives. Second, when human linkers algorithms use same variables, there relatively little disagreement them. Third, plausible analyses, coefficient estimates parameters interest are similar using linked samples based on each methods. provide code Stata commands implement various (JEL C81, C83, N01, N31, N32)