Constructing a population database using automated methods – Stockholm in 1955

Welcome to the joint seminar CEDAR and Unit of Economic History, Umeå University.

We have two guest speakers, Jonatan Andersson and Leonardo Costa Ribeiro who is an expert in AI and Machine learning in social science research.

Constructing a population database using automated methods – Stockholm in 1955

Researchers in economic history and historical demography have long relied on large micro-level databases to study, for instance, social mobility, fertility, migration, and firm performance. Sweden’s strong tradition of quantitative historical research, combined with its access to high-quality original historical records, has enabled departments to construct various excellent databases, including SEDD and the Historical Manufacturing Census of Sweden in Lund, POPLINK in Umeå, and GOPP in Gothenburg.

Yet, assembling such databases is highly labor-intensive, requiring large teams with expertise in handwriting and transcription, and generous external funding. New machine-learning tools are, however, cutting the costs associated with transcribing historical records. With only minor hand compiled training datasets, small teams can transcribe records and assemble large databases in months rather than years. In this presentation, we present the strategies used to construct a new population database, covering everybody living in Stockholm in 1955. We describe the machine-learning method, the original records, and the main challenges that we encounter and how we address them.