Cyberspatial Proximity Metrics: Reconceptualizing Distance in the Global Urban System

Abstract

In this paper we analyze how distances between a sample of a hundred major world cities varies when measured in cyberspace. The project develops a novel spatial statistical model based upon the number of user-generated placemarks indexed by Google Maps. We demonstrate how this metric captures the “invisible” patterns of intercity information flows and helps comprehend the contours of the complex digital network that exists between large urban centers across the world. Using a specially designed software program to interrogate Google Maps, a series of keyword searches (“tourism,” “business,” “hotel”) as well as each of the city names were conducted in each of the sample places. Comparing this digital measure with the material movement of people and other relevant descriptive variables, such as national economic development and language differences, we were able to provide a cogent model that plausibly explains why certain city pairs (especially those that are physically distant) exhibit strong informational linkages. While the strength of these digital connections undoubtedly demonstrates the continued importance of physical proximity and established transport infrastructures in the twenty-first century, one can also observe significant evidence for [new?] digital “wormholes” which indicates that processes of globalization driven by online interaction also operates by its own rules.

Notes

Of course, the precision and reliability of the answers to these questions are limited to the data that are indexed and prioritized by these online databases.

The city of Nice, France (ranked 75th in terms of airline passengers) does not appear in the final database due to the alternative meaning of "nice" to convey something that is pleasant or agreeable. Due to this significant overlap of synonyms, Nice had more total placemark references (a total of 39,197) than any other city, e.g., the next three largest were New York (25,615), London (23,296), and Paris (22,887). Given this obvious discrepancy in ranking (having 50 percent more references than the global cities of New York, London, or Paris) and a clear understanding of what was causing it, the city of Nice was replaced by Moscow (ranked 101st in terms of airline passengers). While other cities also have synonym issues (e.g., Hamburg, Germany and Hamburgers; Charlotte, NC and a woman's name) they remain in the database. Although this increases issues of "noise" in the data, we prefer this approach over risks of arbitrarily “cherry-picking” the data.

The URL for this search is <http://maps.google.com/maps?f=q&view=text&hl=en&q=london+loc:+50.8371,4.3676&mrt=kmlkmz&radius=5>. The query can be tailored by adjusting various components such as replacing the text string "london" for another key word or the coordinate "50.8371,4.3676" (the city center of Brussels) for another location. However, because Google is constantly updating its index of data and tweaking its search algorithm, replicating this particular search will produce different results than shown in this figure.

A 2002 survey of 2,024 million web pages <http://www.netz-tipp.de/languages.html> determined that by far the most web content was in English (56.4 percent); next were pages in German (7.7 percent), French (5.6 percent), and Japanese (4.9 percent).

The blog <www.floatingsheep.org> provides a number of other examples of such searches at a number of different scales and topics.

ICAO (International Civil Aviation Organization), DB1B (Airline Origin and Destination Survey).

AEA (Association of European Airlines), OAG (Official Airline Guide), SRS (Schedule Reference Service).

An important nuance to the MIDT data is that cities in close proximity, e.g., Brussels and Paris, San Diego and Los Angeles, have well developed means of ground transportation (car, bus, train) between them. This means that the MIDT statistics tend to under-represent these connections between nearby cities. Fortunately, this is ameliorated in some cases by the inclusion within the MIDT database of origin/destination IATA codes that also encompass train and bus stations. For example, in addition to the air link between Brussels and Paris, Charles de Gaulle, Air France provides the Thalys High Speed train from Brussels South train station. Thus, a portion of ground travel is included in our modelling.

Note, other metrics using different generic keywords (such as "1") or combinations of keywords behaved identically to this variable in the models.

Sensitivity testing revealed no major changes in the final specification of the model. For example, the United States operates as an "early adopter" for many new technologies and can often skew any global modelling efforts. Moreover, the United States has the most cities of any country in the database (35 in total). However, removing the United States from the analysis does not change the model with a similar r-squared, and all variables are in the same direction and exhibit the same significance. Likewise, removing city pairs closer than 200 km (with alternative means of transportation) or outliers that are likely tied to data issues, e.g., the metropolitan area of Raleigh-Durham (spanning two separate cities) has 85 percent fewer placemark references than the next lowest city, Jeddah, produce no change in the direction and significance of the independent variables. In fact, the only effect was to increase the explanatory power of the model to an adjusted r-squared of 0.574.

Multi-collinearity between independent variables is always a concern within multivariate regression but the tolerance values reported by SPSS for these models, i.e., the percent of the variance of any one independent variable that cannot be explained by other independent variables, are all above 80 percent and in most cases above 97 percent, which means that our models are not unduly troubled by collinearity.