We Made Up Some Data: A New Approach to Synthesizing the Lost Census of 1890

We demonstrate a method to synthesize a small but statistically valid sample of individuals for the 1890 Census, which was destroyed. This generated sample covers a subset of the questions that were asked in the 1880 Census producing records that match a subset of the aggregate statistics that did survive. Strictly speaking, this sample is "made-up" data, and as such would be useless for genealogical research, but it has the same aggregate statistical properties that the actual data is known to have had. While our work was restricted to a subset of the available variables, our implementation of a constraint satisfaction solver within the census/survey domain should provide the basis for the construction of a better dataset for historical research than the use of straight-line interpolation when the scaled to consider more variables and geographic regions.