Preliminary Programme

Hitherto it has been very difficult to explore internal migration of women in nineteenth century Denmark. Migrating historical individuals tend to disappear in studies based on family reconstition and migrating women are not prominently featured in the aggregated published statistics. This paper, however, examines migratory trajectories of women, I have build using ALA (Assisted Linkage Application); a tool developed by the Link-Lives project for human linkage of person registrations in two different sources, e.g two fully transcribed nation wide censuses.
My dataset consists of all women 5 to 50 years of age residing in three provincial towns (Sorø, Næstved and Holbæk) in Zealand, Denmark, in the 1850-census (app. 2300 women). I have reconstructed their previous and later migration trajectories by linking these women to the censuses of 1845, 1860 and 1880. Through a life-course approach, this paper tests whether female migration conformed to existing hypotheses, among them: Were female migrants restricted to areas in close geographical distance to their place of birth thus not far away from home? Were women’s migratory experiences dependent on family migration? (Show less)

Carefully analysing data for a large number of locations, Bourgeois-Pichat
found that deaths during the last eleven months of infancy were uniformly
distributed on a log-cube transform of age, independent of the level of
mortality. He found that, When plotted on the log-cube time scale, the
cumulative number of deaths after the first month tend to follow a straight line, similar to the development of body weight (Bourgeois-Pichat 1951a, b). The relationship is surprising, since one would expect the cumulative deaths risks, and not cumulative deaths, to follow the increase in body weight. Therefore, we instead suggest that the cumulative hazards function on the same time transformation is linear during the last eleven months of infancy, implying an exponential distribution on the time transformed data. We compare the Bourgeois-Pichat model with our own for several locations in Sweden going back to the eighteenth century, as well as for the entire country. Doing so, we find that the two models produce similar results for populations with low or moderately high infant mortality but diverge when infant mortality is high. In addition to being more intuitive and directly related to progress in body weight, the advantages using our model are threefold: First, the proposed model fits better, especially in populations with high infant mortality, and second, the estimation of exogeneous and endogenous infant mortality is easily performed with standard survival analysis programs. Third, the proposed model allows easy estimation in the case of right censored and/or left truncated data. (Show less)

Recent computing advances have permitted the creation of large-scale population databases based on historical records, of interest not only to historical demographers and historians but also to economists and those in the social and health sciences. This was formerly a time- and human resource-heavy endeavor, linking records in necessarily restrictive geographies and chronologies. But the construction of historical population databases is now increasingly reliant on automated methods of linkage, resulting in large datasets with greater potential for research. Several linking approaches have emerged from a number of linking projects, but as yet little is known about how their linking quality varies. While automation has undoubtedly opened up new avenues of research by speeding up the process of linking populations, there are challenges to their use. The potential for bias is very real in both the original sources and during the linking process, compounded by the use of certain identifying characteristics. Occupations and indicators of migration, for example, are strong identifiers which tend to favour men and family groups. Any methodologies which perpetuate population bias will render any subsequent linked data flawed and unrepresentative, a danger of particular importance when using automated machine learning methods.

In this paper we test a number of historical source linking methods for bias, using women as a test case. How well does each method capture their presence, and how far is gender bias reduced, reproduced or magnified using each approach? We use learning from Link-Lives, a research project which is creating life courses for more than 60 million person appearances in Danish national censuses, parish records and burial records for the period 1845-1911. Firstly we review the key methods for creating linked datasets, both manual and automated. Secondly, we analyse the impact of several linking methods on the gender bias, including the Link-Lives rule-based algorithm, tailored to the specific needs of our sources; an implementation of the method proposed by Abramitzky et al; and an in-development Link-Lives machine learning approach using support vector machine. Lastly, we compare these results to the Link-Lives benchmark dataset of domain-expert generated data, consisting of 40,000 manual links. For all approaches, we measure the differences in link, precision and recall rates, which we calculate over different time periods and geographies. (Show less)

Urbanization can have a significant impact on the demographic profile of rural regions. Large scale rural to urban migration will over time lead to an overall population decline in rural areas but has also been shown to affect other population structures such as the composition of age, gender and socioeconomic status. In Sweden the process of industrialization and subsequent urbanization began in the late 19th century, one region particularly affected by increased outward migration was the largely rural northern inland region. The overall aim of this paper is to analyse how migration flows influenced the population structure of the northern Swedish inland, 1900-1950.

The study is based upon historical population data comprised of digitalised parish registers from the county of Västerbotten. The county was divided up into three regions based on occupational structure of the inhabitants: rural (consisting of all inland parishes and one coastal parish), semi-industrial rural (consisting of coastal parishes in close proximity of urban centres) and urban (consisting of the two larger coastal town parishes in the region, Umeå and Skellefteå). These regions are compared with regards to the aforementioned factors of age structure, gender composition and socioeconomic status. This longitudinal comparison will show the characteristics of the migration patterns as a result of early urbanization and how these affected long term population structures in regions with close geographical proximity but different economic structures. (Show less)