The influence of geocoding level and definition of geographic context variables on historical demographic analyses

The quality of spatial analysis is highly dependent on the geo-referencing quality and the definition of the geographic context variables used. However, such information is seldom considered in historical demographic and epidemiological research that includes the geographic context. We investigate a suitable geographic level for geocoding of the population and definition of geographic context variables for historical demographic studies. Using longitudinal demographic data combined with unique historical geographic microdata on residential histories, we compare two geocoding levels (property unit and address unit) and two definitions of distance to wetlands (an indicator of exposure to malaria in the 19th century Sweden). We first statistically compare the differences in the distance to wetlands between the two geocoding levels. Thereafter, by analysing how distance to wetlands affected the mortality of children aged 1-15 living in four rural parishes in Sweden, 1850-1914, we study the effect of different geocoding levels and definitions of context variables on the quality of historical demographic analysis. We find that both the geocoding level and the definition of the geographic context variables strongly influence the results of the analyses. For the distance to wetland variable, decreasing its average positional accuracy by at least 100 meters affects the results. Consequently, the significant differences between the two geocoding levels indicate the importance of considering the geographic level of detail when geocoding