This semester I took an econometrics class because I got an MA with the bare minimum of quantitative classes. For the class, I wrote a paper asking the question, “Are consumers willing to pay a premium to live in dense urban areas?” It’s easy to see that urban density is correlated with higher housing prices, but this could come from many factors such as people having to live in dense cities to find jobs or to earn higher salaries or from supply restrictions that impact dense cities more than suburbs.
As a proxy for cities’ urban qualities, I used Walk Score. Walk Score is based on residential distance to amenities, block length, and road connectivity and ranks cities on a scales of 100. It is designed to test the feasibility of living in a city without a car, but it excludes some factors that are often considered relevant to facilitating pedestrianism, including street width, sidewalk width, and population density. Still, I think Walk Score provides a pretty good measure of a city’s urbanist quality. The correlation between Walk Score and median house price is pretty striking:
To test demand for urban living, I wanted to control for the economic factors that drive demand to live in a given city. I tested the impact of Walk Score on median house prices controlling for household income, unemployment, and cost of living. The sample includes 259 cities for which I had Walk Score data and house price data from Kiplinger. The results suggest that for a one-point increase in Walk Score, we can expect a .5% increase in a cities’ median house price, and this result is statistically significant.
In another way of measuring the same question (an IV regression using the year the city was founded as the instrument), I found that a one-point increase in Walk Score can be expected to increase home prices by 3%. This result is also statistically significant, but I have less faith in this model.
For the most part, the other studies that I’ve seen of Walk Score’s relationship to house prices look at one city or a few cities and control for variables like a neighborhood’s crime rate and housing quality. While there are obvious advantages to these more detailed, local studies, I think the national view gets around the sample selection problems that make other results ungeneralizable.
I’d be happy to hear your criticisms of this model — what important variable are omitted, etc. I think there is a lot of room to study people’s preferences for urban form. As Stephen has said previously, looking at where people live without controlling for other factors gives us a better sense of allowable land use than free market revealed preferences, but looking at home prices while controlling for important variables can remove some of this bias.
Thanks to Eli Dourado for helping me think through this model, but of course its problems are my fault.
[Note: I had originally said that the house price data came from the Census. I realized that Kiplinger does not get this data from the Census as their Statistical Abstract only covers select MSAs. The data was collected by Clear Capital, but I haven't seen it publicly available from them.]