The Linkage of Historical Records by Man and Computer: Techniques and Problems

Abstract

This article reports the evolution of a method to computerize record linkage in the "Hamilton Project" under the direction of Michael B. Katz at the Ontario Institute for Studies in Education. By record linkage is meant the bringing together of information concerning a particular individual (or even a historical event) from independent sources. In this case there was an attempt to link the names on a census roll of 1851 with those on an assessment roll of 1852 in Hamilton, records which differed in time by only three months. The problem is more difficult than might appear to the uninitiated because an individual referred to in one document as J. H. P. G. Yelverton (turner) is the same as J. H. P. G. Silvester (sawyer) in another. The first attempt at linkage reported here was done by hand, took five months, and netted 1,863 matches out of a possible 2,550. In the second linkage experiment the same two sources were employed but the data were placed on Hollerith 80-column punch cards and preliminary sorting was done by a mechanical card sorter. This second project, conducted by the same research assistant, took three months, and resulted in 1,955 linked pairs. Part of the time saved might have been due to the greater expertise of the assistant and cannot all be credited to the use of the card sorting techniques.Using the 1,955 linked pairs as a model file of linked pairs, a computer simulation linkage was designed by trial and error. In the first attempt a quasi-alphabetic sequence known as the Soundex code, which has been used in the field of medical record linkage, was utilized. The Soundex system sets aside the least reliable portions of Anglo-Saxon surnames, thus Smith/Smyth/Smythe all are codified the same. The effect is to reduce all surnames to a code consisting of one alphabetic character (the initial letter) and three numeric characters. To compress them in this way vowels are not coded and neither are W nor H. The remaining letters are grouped into six categories with, for example, B, P, F, V, all grouped together. Unfortunately the data proved somewhat resistant to such coding although all but 11 percent of those already linked were linked by computer. The system did not account for the fact that McKay often became McCoy, or Hicken, Hicker, or Suten, Sutor. In a second effort a pre-treatment system was utilized, based in part upon the SINGS code, designed for Scottish surnames, and a visual analogue of the Soundex code devised by Ian Winchester and John Tiller which they called a Viewex code. With this

Conceptualizing and Constructing the Canadian Century Research Infrastructure

“…Indeed, it was record-linkage projects that, in the 1960s, first pointed to the need to contextualize census data because researchers found that, to an unanticipated extent, different sources did not cumulatively or consistently support the same interpretations of individual or collective lives. The most important scholar working in Canada on the epistemological and philosophical aspects of this topic was Ian Winchester (1970), who collaborated with Michael Katz at the Ontario Institute for Studies in Education.…”

Past, present and future of historical information science

“…The extensive literature about this subject, dating back to the early 1970s (Winchester, 1970;Wrigley, 1973), deals with questions as how spelling variants in sources are to be standardised, to what degree linkage can be automated, and what level of relative confidence is acceptable with automated procedures. A final consensus has not yet been reached, in spite of thorough debates, experiments and testing.…”

“…For a great deal this may be explained by the variety of sources and historical pro blems involved, covering different times and cultures, thus creating an endless range of peculiarities: poll books, census and tax registers, baptism and death records, used for a diversity of research purposes as studying political behaviour, land holding, family reconstruction, life course analysis, regional demo graphy, and prosopography (i.e. using individual data to describe the history of a group 51 ) (Adman, 1997;Adman, Baskerville and Beedham, 1992;Davies, 1992;Harvey and Green, 1994;Harvey, Green and Corfield, 1996;King, 1992King, , 1994Ruusalepp, 2000;Tilley and French, 1997;Vetter, Gonzalez and Gu tmann, 1992;Winchester, 1970).…”

Concept and Rule Based Naming System

“…Guth, 1976 andLevenshtein, 1965 algorithms are spelling/string analysis based, whereas Soundex (Winchester, 1970) and Phonex (Lait & Randell, 1998) are phonetic/sound based algorithms. However, most researchers have tried to implement a method which can deal with the culture of names and naming system which are meant to overcome the ethnic problems, e.g.…”