Methods and validity of a panel study using record linkage: matching death records to a geographic census sample in two Massachusetts towns, 1850-1912.

ongitudinal panel studies and computerized record linkage are both now common in historical research. L Panel study research designs are found in nearly every quantitative historical work and underlie even such simple comparisons across time as those of decennial census data within given geographic areas such as regions, states, counties, or towns. In addition, the construction of multiple-source databases through record-linkage techniques is central to historical studies ranging from European family reconstitutions to cross-census studies of American occupational mobility. Despite their widespread use, even simple panel studies confront methodological concerns that are profound yet seldom addressed. The lesscommon combined use of linked records within panel studies (Bideau and Brunet 1993) also presents several unique problems and concerns. Our objective is to discuss methods used by the Connecticut Valley Historical Demography Project (CVHDP) and to evaluate the validity and reliability of the longitudinal database the project constructed. We will consider methods and general issues of primary data quality, research design, geographic sampling, and computerized record linkage. The objectives of the CVHDP are to examine various aspects of urban and industrial change significant to the continuing high levels of mortality in select rapidly growing Massachusetts towns-Northampton and Holyoke-of the late nineteenth century.' Initial findings of this project have, for example, suggested the importance of rapid population growth and aging population composition in maintaining high levels of New England mortality (Hautaniemi, Swedlund, and Anderton 1999). The database required for our individuallevel mortality analysis was constructed through a computerized record linkage of census records and corresponding death records from years immediately following the census. Our longitudinal research design thus links census and death records to analyze mortality variation across periods, or panels, defined by the timing of decennial censuses.