Frequently Asked Questions
What is Streetchange?
Streetchange is a new way of measuring changes in the physical appearances of neighborhoods using a computer vision algorithm. A large positive Streetchange value is typically indicative of major new construction, while a somewhat smaller positive value of Streetchange is typically indicative of improvements to existing building structures. A large negative value of Streetchange is typically indicative of demolition or abandonment of housing stock, while a somewhat smaller negative value of Streetchange is typically indicative of building decay.

How is Streetchange calculated?
Streetchange is calculated by algorithmically comparing street-level images of the same location captured in different years. In the current work, we obtain Google Street View images of the same location, captured from the same viewpoint in 2007 and 2014. We use a set of computer vision algorithms to remove over-exposed, blurred, and occluded images. Next, we compute the Streetscore for each image—a metric of how safe street block looks to a human observer, derived in some of the authors’ previous work.1 Finally, we calculate Streetchange as the change in each street block’s Streetscore.

Why did you develop Streetchange?
For decades, urban planners, economists, sociologists, and architects have advanced theories on why certain neighborhoods develop over time, while others decay. However, thus far, it has been challenging to study the physical change in neighborhoods at a large scale and across cities, due to a lack of data. Streetchange enables researchers to harness Street View imagery from hundreds of cities across the world and generate data on urban change at the street block-level. This data can be used to study the causes and consequences of cities’ physical evolution.

How did you use Streetchange to understand the evolution of cities?
We calculated the Streetchange between 2007 and 2014 in five major American cities—Baltimore, Boston, Detroit, New York, and Washington DC—from Google Street View images of more than 1.6 million street blocks. Next, we aggregated the Streetchange values at the census tract level. We combined the Streetchange data for these 2,514 census tracts with socioeconomic data from the 2000 US census.

Using multivariate regressions, we tested which socioeconomic indicators from 2000 are statistically significant predictors of Streetchange. We found that population density and the share of college-educated adults (and not income, housing costs, or ethnic composition) were the most important predictors of high Streetchange. Moreover, the physical appearance and location of the neighborhood is important for physical growth as well. Neighborhoods that had a better appearance to begin with (“2007 Streetscore”) experienced significant physical growth, as did those surrounded by neighborhoods with better appearances, and those close to downtown. In sum, we found that physical improvement occurred in geographically and physically attractive neighborhoods with dense, highly-educated populations.

How do your findings with Streetchange relate to traditional theories of urban change?
We found support for three classical theories of urban change. First, neighborhoods that are densely populated by college-educated adults are more likely to experience physical improvements—an observation that is consistent with human capital agglomeration theories.2,3,4,5  Second, neighborhoods with better initial appearances experience larger improvements—an observation that is compatible with “tipping” theories of urban change.6,7  Third, neighborhood improvement correlates positively with physical proximity to the central business district and to other physically attractive neighborhoods—an observation consistent with the “invasion” theories of urban sociology.8

How accurate is Streetchange in calculating the physical change in neighborhoods?
We validated the accuracy of Streetchange using three sources: a survey conducted on Amazon Mechanical Turk, a survey of graduate students in MIT's School of Architecture and Planning, and data from Boston's Planning and Development Authority. We found strong agreement between Streetchange and both human assessments and new urban development. Please see our paper for details.

Why do some image pairs with high positive Streetchange do not show real physical change?
While the Streetchange algorithm is fairly accurate, it sometimes mistakenly judges image pairs with no real physical change to have large changes just because of poor image quality or the limited training set size. Additionaly, some locations in our maps may show insignificant physical change due to the discrepancies introduced by Google's periodic updates for the panorama associated with a given location and time. Since we obtain images on the maps directly from Google's servers, the section of the street that is displayed in our map image may not be the section of the street on which Streetchange was evaluated by us in 2015. In such cases, we encourage you to visit the location directly on Google Street View using the address displayed below the image, and observe the physical change in the street block.

What’s next?
We have released the Streetchange data in the form of interactive maps for 2,514 census tracts from the five cities in our study. We hope that our work will be of interest to policymakers, urban planners, social scientists, and the general public. Policymakers can use these tools to understand the impact of new policies on the evolution of private and public infrastructure. Urban planners can use Streetchange and its underlying technology to study the impact of urban design on neighborhood change. And social scientists will be able to improve their understanding of the relationship between the socioeconomic composition of neighborhoods and the built environment.

1. Naik N, Philipoom J, Raskar R, Hidalgo CA (2014) Streetscore – Predicting the per-ceived safety of one million streetscapes. IEEE CVPR Workshops, pp 793–799
2. Glaeser EL, Scheinkman JA, Shleifer A (1995) Economic growth in a cross-section of cities. J Monet Econ 36:117–143.
3. Bettencourt LM (2013) The origins of scaling in cities. Science 340:1438–1441.
4. Ciccone A, Hall RE (1996) Productivity and the density of economic activity. Am Econ Rev 86:54–70.
5. Glaeser EL, Gottlieb JD (2009) The wealth of cities: Agglomeration economies and spatial equilibrium in the United States. J Econ Lit 47:983–1028.
6. Schelling TC (1969) Models of segregation. Am Econ Rev 59:488–493.
7. Grodzins M (1957) Metropolitan segregation. Sci Am 197:33–41.
8. Burgess EW (1925) The Growth of the City (Univ of Chicago Press, Chicago).