Putting science into data science

Spatial Reasoning and Pitfalls

These open teaching materials are under development at the University of Chicago’s Center for Spatial Data Science, in collaboration with partners. They are designed to teach data science with scientific reasoning, with a specific focus on spatial data and spatial methods. This is generally missing in existing courses.

The Spatial Reasoning and Pitfalls course is about how to reason -and not to reason- with spatial methods and data. It adopts the framework developed for the statistical pitfalls course and is currently under development. It will be taught by UChicago’s GIS Librarian Rob Shepard in spring 2025 at the University of Chicago.

To start, the courses will teach you to recognize and avoid spatial data gotchas like null island, misaligned projections or geocoding errors (see The Immigrant Paradox in Chicago: Real or Artifact?). You will also learn about spatial methods pitfalls such as the modifiable areal unit problem (MAUP), selecting on the dependent variable (NYC’s geocoded stop and frisk data), Simpson’s Paradox / Spatial Regimes, and spatial patternicity (spatial signal vs noise). Also included are pitfalls re. the interpretation of spatial results, such as ecological fallacy and spatial cluster cores vs neighbors. Last but not least, the course will cover spatial examples of classic cognitive and sampling biases, such as confirmation bias and selection bias.

Partners: This course is being co-developed with UChicago’s GIS Librarian Rob Shepard.