City-REDI Seminar Series 2021 - With Dr Levi Wolf, University of Bristol

Space and place in predictive models

Models are representations of reality. Generally, we all want models that are detailed "enough," but also models that are easy to interpret and estimate. Sometimes, we get lucky: multilevel models provide incredibly powerful specifications that are easy to understand and easier to fit thanks to some of their intrinsic mathematical properties. In geographical applications (when a "level" is a "place") however, these intrinsic mathematical properties conflict with some well-known effects of spatial dependence. Who wins out, "space" or "place?"

It depends! This talk will provide a high-level, conceptual overview of the tensions between how "spatial" and "platial" (i.e. multilevel) models work. I'll discuss some of the results from Wolf et al. (2021) about how these tensions get resolved, and what it means for practitioners working with classical multilevel models that involve geographic levels.

Find out more