Projects per year
The data for this project was sourced from historical property records (primarily warranty deeds) in Hennepin County using optical character recognition (OCR) and crowdsourcing. OCR was used to process the entirety of the historical property record for Hennepin County between 1900 and 1960 and identified the subset of records that contained at least one instance of a predefined list of key words and word stems that indicated the presence of a racial covenant. Because the project used OCR to identify likely covenants, there are cases where covenanted deeds were missed due to the presence of handwritten documents or when the deed scan or original document was of poor quality. Deeds containing at least one of the key words were uploaded to Zooniverse, where volunteers read the flagged deeds and transcribed the relevant attribute information. Each deed had to be transcribed by five unique volunteers and the results were checked using a variation of Cohen’s kappa statistic to assign volunteer reliability scores. The data from volunteers whose score fell below a certain threshold was then dropped from the database. For each unique deed, we then calculated the modal value for each field based on the volunteer data that remained. When a field had no agreement, the value provided by the user with the highest Cohen’s kappa score for that deed was kept. The results were joined with existing parcel boundaries where possible. In cases where the covenanted parcel had no contemporary match, the original plat map for that parcel was used to create the spatial boundaries. Covenants that only covered partial lots were also drawn in manually.
|Date made available||2020|
|Publisher||University of Minnesota Digital Conservancy|
|Temporal coverage||1910 - 1955|
|Date of data production||2016 - 2019|
|Geographical coverage||Hennepin County, Minnesota|