Tyler James Frazier

Generally, my research focuses on advancing systematic, principled and coherent modern statistical methodology, rather than the classical ad hoc approach, to describe and infer geographical patterns. My area of research focus is the intersection of the human and environment development geographies of Sub-Saharan Africa. Following are three specific areas I am actively researching.

Generating Close-to-Reality Synthetic Populations

Motivated by the question of how to accurately infer large area population dynamics along the continuum of scale from densely populated urban areas to the more sparsely populated regional expanse, I applied and made modest extensions to a methodology for generating a close-to-reality synthetic human population. While my work in this area presents arguable advantages when compared to the traditional iterative proportionate fitting method, it still requires spatial aggregation of observations in order to specify the multinomial logistic regression models. I am working towards specifying a non-parametric, spatial model to generate a close-to-reality synthetic population beginning with the location of each survey observation and its corresponding weight while meeting the existing stated constraints. For more information please see the following paper.

Generating a Close-to-Reality Synthetic Population of Ghana

Following is a presentation I gave on the subject during the UrbanSim User conference in Berkeley, California.

Density Estimation for Boundary Definition of Human Settlements

This student centered research effort is motivated by the questions of (1) how to define the boundary of human settlements using density estimation and (2) how does the use of density estimation to describe a system of human settlements compare to political or administrative boundary definitions in the presence of Zipf's Law. This research is focused on developing and advancing basic, yet core spatial data science methods in the context of global population dynamics. My team of undergraduate researchers are working to program (1) a function that spatially describes the probability density of a random series of points located within an arbitrary window and (2) cluster those points based on density estimation. Undergraduate research team members include: Alex Fitz, Chris Eisner, Kathryn Murphy and Luke Pascual. Working paper is forthcoming in the summer 2017.

Spatial Propensity Score Matching (or other metrics)

Propensity score matching (or other metrics) is a fundamental step in the geospatial evaluation of investment effectiveness. This research is motivated by the question, what advantages are presented when using density estimation with observational studies that specify the conditional (predicted) probability of receiving treatment. Concept paper forthcoming in the summer of 2017.