I study racial and class inequality in the physical conditions of neighborhoods in which people live. I have focused on exposure to different types of retail, especially "food deserts."
This research has also revealed some surprising findings. With colleagues from the Built Environment and Health Project at Columbia University, I found that adolescent obesity rates among New York City public school students were lower in neighborhoods with more fast food restaurants. I attributed this counter-intuitive finding to the fact that fast food restaurants follow the same retail investment patterns as other businesses, and I tested this claim by conducting a placebo test that examined the association between banks and adolescent obesity (since banks would be correlated with investment but should not have been etiologically associated with adolescent obesity). I found that banks were indeed associated with lower adolescent obesity rates. As I later told reporters who covered the article, the most disadvantaged places were those in which even fast food restaurants will not invest.
I have also shown how the infrastructure for walking and disorder varies across neighborhoods. To study these disparities, I have extended the method of systematic social observation to harness "big data" in the form of Google Street View. I led the development of a web application called the Computer Assisted Neighborhood Visual Assessment System, or CANVAS, that allows researchers to conduct systematic social observations of neighborhoods nationwide (or even world-wide). We have shown that this method can be used to generate reliable data by nationwide study of 187 different items. I have combined this work with research showing how to use the geostatistical method of kriging to improve measures of neighborhoods. We use this method to estimate the level of walkability or disorder on every block in a city based on a sample of only about 5% of blocks.
We plan to improve our prototype of CANVAS to make a more user-friendly product that will help us "Canvas America" and develop measures of the built environment in metropolitan areas across the country. I am working with computer scientists to incorporate computer vision into production that will improve the efficiency of data collection. The computer-human data collection loops that result will allow us to leverage the distinct skills of both human and computer raters. Through the computer vision and the geostatistical methods that I just described, we plan to develop a nationwide dataset of neighborhood physical environments that will allow researchers to study the factors that contribute to and result from inequality across neighborhoods.