Importance of the Geocoding Level for Historical Demographic Analyses: A Case Study of Rural Parishes in Sweden, 1850–1914

Abstract

Geocoding longitudinal and individual-level historical demographic databases enables novel analyses of how micro-level geographic factors affected demographic outcomes over long periods. However, such detailed geocoding involves high costs. Additionally, the high spatial resolution cannot be properly utilized if inappropriate methods are used to quantify the geographic factors. We assess how different geocoding levels and methods used to define geographic variables affects the outcome of detailed spatial and historical demographic analyses. Using a longitudinal and individual-level demographic database geocoded at the property unit level, we analyse the effects of population density and proximity to wetlands on all-cause mortality for individuals who lived in five Swedish parishes, 1850-1914. We compare the results from analyses on three detailed geocoding levels using two common quantification methods for each geographic variable. Together with the method selected for quantifying the geographic factors, even small differences in positional accuracy (20-50 m) between the property units and slightly coarser geographic levels heavily affected the results of the demographic analyses. The results also show the importance of accounting for geographic changes over time. Finally, proximity to wetlands and population density affected the mortality of women and children, respectively. However, all possible determinants of mortality were not evaluated in the analyses. In conclusion, for rural historical areas, geocoding to property units is likely necessary for fine-scale analyses at distances within a few hundred metres. We must also carefully consider

Infant mortality in mid‐19th century Amsterdam: Religion, social class, and space

“…Extending the dataset by following up on the cohort would also allow scholars to study long‐term health and mortality effects of early life circumstances. However, this not only requires the addition of demographic life course events but also, if one wants to take the spatiotemporal context into account as well, implies dealing with the complex matter of geocoding the longitudinal demographic data (Hedefalk, Pantazatou, et al, ). Adding new birth cohorts would allow the study of urban infant mortality changes over time, requiring digitising population register data for later years from for instance Amsterdam Population Register scans, available for the periods 1854–1863 and 1874–1893.…”

“…Findings of Hedefalk, Pantazatou, Quaranta, and Harrie (), Hedefalk, Quaranta, and Bengtsson (), and Olson (), for instance, show that the choice of geographical level is important for demographic analyses using historical individual‐level data. Hedefalk, Pantazatou, et al () and Hedefalk, Quaranta, et al () combine micro‐level geographical factors, such as soil conditions, with individual‐level historical demographic data in a case study of rural parishes in Sweden. Studies analysing micro‐level geographic factors for urban environments have also become more common; see, for example, Olson () and Thornton and Olson () for Montreal, and Ekamper () for the Dutch town of Leeuwarden.…”

Regional differences in historical diphtheria and scarlet fever notification rates in The Netherlands, 1905–1925: a spatial-temporal analysis

“…lower infant mortality means relatively more susceptible living children aged older than 1 year, all else being equal) or points to an unmeasured additional factor [ 28 ]. In addition, the geographical regions used in this analysis, while already of a high resolution, may nevertheless hide underlying local associations with our covariates [ 29 , 30 ]. For example, while overall population density is highest in urban regions, people living in rural parts probably also live close together in small communities or as families sharing a farmhouse.…”

Unequal lands: Soil type, nutrition, and child mortality in southern Sweden, 1850-1914

“…Moreover, some individuals who owned several property units were only linked to the property unit in which they lived, which means that the soil type and property unit area measures do not represent the size of the total land for these individuals. Thus, these uncertainties may introduce biases and errors in the models used in this study, e.g., by overestimating or underestimating the effects on mortality from soil, especially if the errors are nonrandom (see e.g., Zandbergen 2007; Hedefalk et al 2016). Therefore, in future studies it is important to study the propagation of the uncertainty.…”