Jonas Helgertz (Sweden): Using methods of machine learning to create databases for (historical) demographic research

During the past decades, scholars have produced a vast amount of research using linked historical individual-level data, shaping, and changing our understanding of the past. This has been enabled by developments in record digitization, computational speed, and machine learning, allowing researchers to replace time-intensive human matching with large-scale automation. This talk reviews recent work within this field, focusing on a range of aspects concerning what is known about the performance of different linking methods. While these methods allow researchers to construct big data at minimal cost, emerging evidence shows that the choice of method may have important implications for the researchers’ conclusions. The talk concludes with suggestions for best practices when generating, evaluating, and using linked samples, within historical demography and beyond.