Soot n/a - "Are Sprawl and Obesity Related? Evidence from the Chicago Area"
Siim Soot, Lise Dirks, et al
"Are Sprawl and Obesity Related? Evidence from the Chicago Area"
Unpublished: Metropolitan Transportation Support Initiative (METSI) working paper 06-01
On the Web
The authors estimate the effect of urban, socio-economic, and personal characteristics on BMI using height, weight, and ZIP code data from 7 million driver's licenses and state IDs in greater Chicago. Their regression finds that population density has a slight but significant effect, dwarfed by other variables. I have many questions about their methodology, but nonetheless don't think they overturn the bulk of sprawl and obesity research.
First, the data
They use individual height, weight, and zip code data from driver's licenses and state IDs. Age data are also available, but they only use it to categorize the percentage of people under 18 (unless that comes from the Census data). They use Census block group data for all other variables, so income, college education, race, etc. are group averages, not the actual values for the individual.
Second, the quibbles
- The authors include a variable for distance from CBD--what if there are many centers of dense urbanity? The inner-ring suburbs, which are likely the most walkable, have the lowest BMI, as most BMI researchers would expect, especially given that they likely have a higher income than inner city folks. Distance from CBD may be correlated with other independent variables (income, density) and thus may weaken the regression.
- Also on distance, the implicit exluded dummy variable is people who live more than 50 miles away from the CBD. Two questions: How large is that group--is it enough to be a statistically valid separate category? Is that farming country where most residents usually engage in physical farming activity?
- The authors include journey to work mode may be erroniously included. BMI is likely affected by commute mode, but commute mode is affected by density, etc. So including it muddies the picture of how density, etc. affect BMI.
- Why do the authors include the percentage of residents who are homeowners? How does this affect BMI in a way that income doesn't capture?
- The authors mix individual-level data with aggregate data. When you don't have individual data available, this is better than nothing. But the average college-education of your neighbors may have very little effect on your own BMI, especially when neighborhood income is already included.
- Also, note that income seems to have little effect on BMI, even less than density, where college education has a very large effect. This raises (again) the specter of multi-colinearity.
- The authors report R-squared but not adjusted R-squared, a concern given the number of (questionable) variables.
Third, the verdict
Even with all these problems, density still seems to have a small but statistically significant effect. That doesn't prove anything, but it suggests that density does matter.