Using SaTScan to identify and map nightlife areas in Philadelphia, PA

Abstract

Many cities have promoted nightlife or entertainment districts – concentrations of restaurants, bars, and other entertainment-related businesses – in order to revitalize declining neighborhoods. While entertainment districts can boost economic growth, they can also contribute to public health risks including violent crime, traffic accidents, and other harms. With data from the National Establishment Time Series (NETS) business database, we developed methods to use SaTScan cluster detection software to identify entertainment districts, and applied the method in a case-study of Philadelphia, Pennsylvania. Using SaTScan, we identified and mapped 101 spatial clusters of entertainment businesses in the city. Our approach is scalable and does not require prior local knowledge about entertainment areas. The results add to a small but growing literature about the use of SaTScan to map neighborhood features. Placing entertainment districts in spatial context can inform how the built environment might amplify or minimize the potential health risks of these districts.

Introduction

Over the past several decades, many cities have begun to promote nightlife or entertainment districts as part of economic development efforts to revitalize urban spaces (Montgomery 1995, Florida 2002). Entertainment districts are areas with high concentrations of bars, night clubs, and restaurants, often located near sporting arenas, movie theatres or concert venues (Hollands and Chatterton 2003, Campo and Ryan 2008, Caufield 2017). These areas can emerge spontaneously (Campo and Ryan 2008, Mattson 2015), but entertainment districts have increasingly been part of strategic economic development plans formally articulated and implemented by local governments (Chatterton and Hollands 2002, Brasuell 2015, Caufield 2017, Delgadillo 2017, Sociable City 2017). Over the past decade, many U.S. cities have created governmental offices to spur the growth of the entertainment sector (Delgadillo 2017, Feargus 2017, Sociable City 2017). Cities have also expanded bar serving hours, altered zoning regulations, and relaxed noise restriction and open-container laws to spur entertainment district development (Caufield 2017, Delgadillo 2017, Feargus 2017). The planning and development of entertainment districts is supported by multiple actors: the alcohol industry and associated trade groups; corporate entertainment and branding companies; real estate development companies; design and construction trade journals; and architecture and urban design standards setting organizations (Chatterton and Hollands 2002, Hollands and Chatterton 2003, Caufield 2017, Portman Group 2017). While entertainment areas are distinct from gentrifying neighborhoods, they can promote gentrification by enhancing the amenities and cachet of surrounding neighborhoods, thus attracting affluent residents (Gladstone and Préau 2008, Burnett 2014, Nofre and Garcia-Ruiz 2023).

While entertainment districts may contribute to economic growth, studies of bars, liquor stores, and other alcohol outlets suggest these districts could also have negative effects on safety and public health. By clustering alcohol outlets together, for instance, entertainment districts may concentrate intoxicated pedestrians near intoxicated or distracted drivers (Gruenewald, Ponicki et al. 1993, Livingston, Chikritzhs et al. 2007, Long and Ferenchak 2021). Higher densities of alcohol outlets are associated with higher risk of motor vehicle crashes (Hobday and Meuleners 2018, Nesoff, Milam et al. 2018). By promoting pedestrian travel between closely situated destinations, these districts may place more individuals at risk of injury or death from being struck by automobiles (Rothman, Buliung et al. 2014, US Department of Transportation 2015). Higher densities of alcohol outlets are also associated with violent crime and inter-personal violence (Pridemore and Grubesic 2011, Zhang, Hatcher et al. 2015, Ransome, Luan et al. 2019). It is important to note that entertainment districts can include a range of businesses besides bars or night clubs; it remains to be seen whether areas with a mix of restaurants, theaters, and sporting or concert venues create the same kinds of risks documented by studies of alcohol outlets.

Any risks associated with entertainment areas may be modified by features of the built and regulatory environment, such as pedestrian safety features (e.g., crosswalks) and speed limits; changes in the built environment and other spatially focused policies could ameliorate these risks (Pridemore and Grubesic 2012). To understand and mitigate risks, it is vital to identify the locations of entertainment districts and place them in their spatial context (Trangenstein, Sadler et al. 2021). Mapping nightlife districts allows stakeholders to visualize where these districts exist and to consider what policies might be implemented to mitigate risk or drive economic development.

Using SaTScan cluster detection software and national business listing data, we developed a generalizable method to identify clusters of nightlife businesses. We applied this method in Philadelphia, PA, to illustrate the approach’s strengths and limitations; the identified clusters are illustrated on the featured map. While SaTScan was developed for analysis of disease clusters, this software tool and the underlying statistical approach has also been used to characterize aspects of the built environment including gentrification as well as density of outlets selling alcohol, fast food, and other unhealthy products (Macdonald, Olsen et al. 2018, Trangenstein, Gray et al. 2020, Corrigan, Curriero et al. 2021). Our mapping of entertainment districts in Philadelphia illustrates both the utility and the limitations of this approach for identifying neighborhood features.

Background

Researchers use a variety of approaches to describe spatial access to resources such as retail outlets. In simple density measures, a count of outlets in a spatial unit such as a Census tract is divided by a measure of land area, population, or roadway distance for each spatial unit. Alternatively, distance-based measures describe the distance from a reference point to the nearest outlet or number of outlets, or describe the number of outlets within a given distance from that reference point (Trangenstein, Sadler et al. 2021). These types of measures are simple to construct and widely used, but assume that the outlet type of interest has already been operationalized. However, entertainment areas are by definition clusters of bars, restaurants, and similar outlets; a single outlet cannot be classified as contributing to an entertainment district unless nearby outlets are also considered. Count or density measures are thus feasible for entertainment venues within a given spatial unit, but identification of entertainment areas requires clustering of these outlets in proximity to one another. Likewise, distance-based measures typically consider the proximity of outlets to a reference location, such as a school or a study subject’s home, rather than to all other outlets of the same type (Neckerman, Bader et al. 2010).

Accordingly, we use cluster detection methods to identify entertainment areas. Building on research on cancer clusters (Centers for Disease Control and Prevention and Agency for Toxic Substances and Disease Registry 2022), we define an entertainment area as a greater-than-expected number of nightlife establishments present in a geographic area within a defined period of time. SaTScan provides an extensively-validated statistical approach to detect these clusters; the software is widely used in public health research (Kulldorff 2001, Heffernan, Mostashari et al. 2004, Kleinman, Abrams et al. 2005, Sonesson 2007, Cheung, Spittal et al. 2013) and performs well in comparative tests (French, Meysami et al. 2022). SaTScan’s space-scan statistics identify spatial areas with observed occurrences that are higher than those expected by chance for Bernoulli- and Poisson-distributed variables (Kulldorff 1997). Compared with density or distance measures, space-scan statistics analyses have several advantages. They adjust for inhomogeneous population density and for other confounding variables. This approach also takes multiple testing into account and identifies the location and size of the cluster that caused rejection of the null hypothesis (Kulldorff 2001, Kleinman, Abrams et al. 2005).

Our case study site, Philadelphia, did not have a formal entertainment initiative in 2014 when our business data were collected; the Philadelphia city government hired its first Nighttime Economic Director in 2022 (Department of Commerce 2023). During the time period represented by our data, the city did have a number of economic development and public space initiatives. Center City (downtown) Philadelphia was served by multiple business improvement districts and similar organizations including the Center City District, the Avenue of the Arts, Old City Special Services District, and the South Street/Headhouse District. During the 1990s and 2000s, business improvement districts were also organized in a number of neighborhoods outside the downtown, including East Passyunk Avenue (south of Center City); University City (west of Center City and anchored by the University of Pennsylvania and Drexel University); the northwest neighborhoods of Manayunk, Roxborough, Germantown, Mount Airy, and Chestnut Hill; and a cluster of neighborhoods northeast of Center City (Aramingo Avenue, Frankford, and Port Richmond) (Han, Morçöl et al. 2017).

Over this period, Philadelphia experienced growth in retail and service establishments with particularly rapid expansion in the hospitality sector, while the proliferation of sidewalk cafes promoted a lively street life in some downtown and outlying areas (Campo and Ryan 2008, Nepa 2011). Ethnographic accounts depicted a vibrant nightlife scene in parts of the city (Grazian 2008, Kavanaugh and Anderson 2017). Thus we expect to identify entertainment districts both in Center City and in neighborhoods outside the downtown. While the current analysis takes advantage of the study team’s local knowledge about Philadelphia, we expect the approach to be scalable, allowing researchers and policymakers to identify and measure entertainment districts across the U.S. without relying upon site-specific knowledge.

Materials and Methods

Data

The National Establishment Time Series (NETS) provides annual establishment-level data for U.S. businesses, nonprofit, and government establishments and is the most comprehensive establishment data source available, serving as an annual census of businesses and establishments in the U.S. since 1990 (Neumark, Zhang et al. 2007). NETS data are derived from Dun & Bradstreet’s (D&B) annual register of establishments, which is part of D&B’s business of developing credit ratings for banking and insurance purposes. Thus, businesses have an incentive to register with D&B, and D&B has an incentive to maximize accuracy (Neumark, Zhang et al. 2007, Currie, DellaVigna et al. 2010). Through a partnership with D&B, Walls and Associates developed the NETS database to capture the geographic and spatial dynamics of the U.S. economy. NETS attributes include entity name, street address, latitude and longitude, standard industrial classification (SIC) codes, and number of employees. NETS was originally built to support research on job growth and business relocation (Neumark, Zhang et al. 2006, Neumark, Zhang et al. 2007, Chapple and Jacobus 2009, Schuetz, Kolko et al. 2010, Kolko 2011, Schuetz, Kolko et al. 2012), but researchers have also used NETS to characterize access to health-relevant businesses and institutions (Currie, DellaVigna et al. 2010, Hirsch, Moore et al. 2014, Hirsch, Moore et al. 2014, Kaufman, Sheehan et al. 2015, Berger, Kaufman et al. 2019, Kaufman, Rundle et al. 2019).

For this analysis, we used NETS data from 2014 to identify entertainment establishments in Philadelphia, PA, which we defined here as including restaurants as well as bars and clubs. Unfortunately, data on restaurants’ alcohol licensure and BYOB (“bring your own bottle”) policies are not available in NETS or in any national data. However, restaurants are included in our nightlife measure because they often cluster in nightlife districts. In addition, they can be nightlife venues themselves, providing locations where patrons can drink alcohol. For these reasons, clusters of restaurants are expected to be indicative of the presence of nightlife districts.

NETS data provided to licensees include both business addresses and corresponding geocodes; however, geocoding methods were not consistent across time, and 19% of establishments listed in the NETS data included geocode accuracy only at the ZIP code level. To improve our ability to clearly document geocoding methods and enhance the quality of the matches, we re-geolocated addresses for NETS businesses using batch geocoding in ESRI Business Analyst software (ESRI Redwoods, CA) and a composite locator, which utilized Navteq 2014 (Q3) reference data. After this additional geocoding, 89% of the addresses were geocoded to a level of spatial accuracy (i.e., address point or street address range) deemed acceptable for further analysis. In total, 4,393 entertainment establishments were identified in Philadelphia.

Cluster detection

Spatial scan statistic analyses were applied to the 2014 NETS data using SaTScan software to identify clusters of entertainment businesses in Philadelphia. SaTScan generates a series of circles or ellipses of varying sizes and positions, known as scanning windows, and evaluates the number of observed versus expected observations in each window. Maximum likelihood ratio tests are used to identify scanning windows that cover a significantly larger than expected number of observations; these identified scanning windows represent clusters of observations. SaTScan offers a large number of parameter settings to define the scanning windows; because parameter settings can affect results (Sherman, Henry et al. 2014), we used an iterative process to optimize parameters for detecting clusters of entertainment establishments. The goal of this process was to identify a series of small statistically significant scanning windows that encompassed entertainment districts in Philadelphia, with sufficient resolution to depict the shapes of the districts. With scan radii that were too large, for instance, the entire city might be identified as an entertainment cluster; with radii that were too small, no statistically significant clusters would be identified. The count of entertainment businesses in each 2010 Census block (N=18,872 blocks) in Philadelphia was analyzed, with the count value attributed to the block centroid location. We used SaTScan’s purely spatial, discrete Poisson model with circular scanning windows centered on Census block centroids. Scanning windows were allowed to overlap, as long as the centroid of one scanning window did not fall within an overlapping window (Criteria for Reporting Hierarchical Clusters set to “No cluster centers in other clusters”).

The iterative process of investigating parameters began with the Maximum Spatial Cluster Size option set to the recommended 50% of the population at risk, which identified several very large scanning windows that provided poor spatial resolution for defining entertainment areas. This setting was iteratively reduced with and without setting bounds for the maximum radius of the scanning windows. Consistent with our work on neighborhood walkability as well as prior analyses of alcohol and other retail establishments, radii in the range of 0.5 to 1.0 Km were tested (Macdonald, Olsen et al. 2018). Ultimately, the minimum number of establishments required to define a cluster was set to two and the Maximum Spatial Cluster Size was set to 0.5% of the underlying population “at risk” (i.e. Census blocks in the city). When the maximum radius of the scanning windows was set to 0.5 Km, as in Macdonald et al. (Macdonald, Olsen et al. 2018), the results were essentially identical to setting the Maximum Spatial Cluster Size to 0.5%: the same number of statistically significant scanning windows was identified and the spatial distribution of the identified windows was similar.

To create maps of the clusters of entertainment establishments, we joined the spatial output from SaTScan to the Topologically Integrated Geographic Encoding and Referencing (TIGER) Line shapefiles for 2010 Census blocks in Philadelphia using ArcGIS. All blocks that fully or partially overlapped the identified scanning windows were identified and depicted on the map.

A note about terminology: we will use “cluster” as a synonym for the significant scanning windows identified by SaTScan, while entertainment “areas” will refer to larger aggregations of adjacent or overlapping clusters. Entertainment or nightlife “districts” will refer to areas identified by city planning or economic development agencies.

Results

SaTScan identified 101 statistically significant scanning windows, representing clusters of entertainment businesses in Philadelphia. These clusters encompassed 1,510 establishments, of which 86.0% were restaurants, 11.5% were bars, and 2.6% were other types of businesses. Of the city’s 18,872 Census blocks, 1,041 were overlapped by one or more significant scanning windows. As shown in Figure 1, in Center City and South Philadelphia (Box 1) and University City (Box 2), multiple clusters overlapped, forming large contiguous entertainment areas. SaTScan also detected entertainment clusters in outlying neighborhoods including Manayunk/Roxborough, Germantown, and Chestnut Hill in northwest Philadelphia and Bustleton and Somerton in northeast Philadelphia. In total, the 101 statistically significant clusters depicted 32 distinct entertainment areas. Table 1 describes the distribution of these significant clusters and their associated establishments across different parts of the city. Like the map in Figure 1, these data indicate the concentration of entertainment-related businesses in Center City and South Philadelphia.

Figure 1.

Entertainment areas in the city of Philadelphia.

Table 1.

Clusters of entertainment establishments for 2014 in Philadelphia, PA, as detected by SaTScan

	Total	Center City/ South Philadelphia	University City	Northwest Philadelphia	Northeast Philadelphia
Number of clusters	101	61	10	15	15
Number of establishments	1,510	1,129	92	156	133
Establishments by SIC code (%)
Restaurants	86.0	86.8	88.0	83.3	80.4
Bars	11.5	10.6	8.7	14.7	16.5
Other	2.6	2.6	3.3	1.9	3.0

Note: Clusters of entertainment establishments were identified using SaTScan analyses of NETS establishment data.

Figure 2 (a close-up of Box 1 in Figure 1) provides a more fine-grained depiction of the clusters in Center City and South Philadelphia, with Census blocks color-coded to indicate the number of entertainment establishments. While nearly the entire downtown was included in at least one cluster, Figure 2 shows that many blocks within this area did not contain any entertainment businesses at all, while other blocks contained multiple establishments.

Figure 2.

Detailed map of entertainment area in Center City and South Philadelphia neighborhoods.

Figures 3 and 4 display identified clusters and nightlife business overlaid on Census block geographies. Figure 3 (a close-up of Box 2 in Figure 1) depicts the University City neighborhood around the University of Pennsylvania and Drexel University. University City has multiple large, overlapping scanning windows that cover most of the neighborhood. As in Center City, there is substantial clustering within the largest University City entertainment district: some blocks contain no entertainment outlets at all, while others contain multiple outlets.

Figure 3.

Detailed map of entertainment areas in the University City neighborhood.

Figure 4.

Detailed map of entertainment clusters in northeast Philadelphia.

Figure 4 (a close-up of Box 3 in Figure 1) depicts three distinct clusters in Northeast Philadelphia, a section of the city with a lower density of both population and commercial development. Two clusters comprise only one Census block each, while the third encompasses more than a dozen blocks. These clusters are a reminder that the geographic extent of an underlying cluster of establishments may not correspond exactly to the size and location of SaTScan’s scanning windows, but rather emerges empirically in ways that reflect street architecture and urban form. It is worth noting that in neighborhoods characterized by lower density and higher segregation of land uses, the location and size of the SaTScan circles do not indicate establishment locations with much precision. The spatial mismatch we observe in Figure 4 indicates SaTScan should be used with caution in these types of neighborhoods.

Discussion

As more cities promote entertainment districts in order to revitalize downtown or declining neighborhoods, tools to map these districts are needed to measure the impact of revitalization initiatives and to understand and mitigate any public health and safety risks of the concentration of entertainment businesses. In our case study of Philadelphia, PA, SaTScan identified 101 statistically significant, often overlapping, scanning windows representing underlying clusters of entertainment businesses. Merging overlapping or adjacent clusters resulted in a total of 32 separate entertainment areas. These identified areas are consistent with local knowledge about nightlife in the city.

The clusters of businesses identified by SaTScan, and the resulting maps, can be a valuable tool for city planning and economic development agencies and for researchers and policymakers studying social, economic or health impacts of nightlife. Applying SaTScan to business listing data allows for measurement of nightlife districts across multiple cities using standardized methods, supporting comparative analysis of how the built environment context might shape and ameliorate risk. While we have focused on nightlife businesses, the NETS data allow spatial cluster analyses of a wide range of business and not-for-profit activities, such as social services or medical offices (Hirsch, Moore et al. 2021). More broadly, SaTScan could be used to identify clusters of a wide variety of urban amenities, risks, and behaviors. For instance, spatial or space-time scan statistics are already used to identify hotspots of crime, drug overdoses, disorder, traffic accidents, and weather-related falls (Dey, Hicks et al. 2010, Shiode 2011, Linton, Jennings et al. 2014, Olsen, Mitchell et al. 2017, Samuels, Goedel et al. 2024), as well as spatial patterns of walking and other physical activity (Mazumdar, Bagheri et al. 2020) and use of urban space by tourists and local residents (Li, Zhou et al. 2018). Description of these spatial patterns can guide local policymakers seeking to efficiently reduce risk by concentrating efforts and resources in neighborhoods with particular need.

In our SaTScan analysis, we selected settings to identify clusters that covered walkable distances and were small enough that overlapping clusters would depict the shapes of larger entertainment areas. It is important to recognize that the sizes of individual scanning windows identified by SaTScan do not necessarily correspond to the optimal geographic extent of the underlying cluster of businesses due to constraints on the shape of each cluster as circular or elliptical. Detailed maps illustrate this spatial mismatch by situating entertainment businesses within both SaTScan’s scanning windows and the surrounding Census blocks. The results add to a small but growing literature about the use of SaTScan to measure and map neighborhood environmental features (Macdonald, Olsen et al. 2018, Trangenstein, Gray et al. 2020, Corrigan, Curriero et al. 2021). As spatial scan statistics become more widely used for this purpose, it will be important to understand how the urban form context shapes the properties and limitations of these measures.

Limitations of this study include that the NETS data are from Philadelphia in 2014; other settings or more recent data might be of more interest for some readers. In addition, our measure of entertainment districts includes a range of types of businesses including some that may not serve alcohol; the NETS data do not include information on whether a restaurant serves (or allows patrons to bring in) alcoholic beverages. Whether areas with a mix of restaurants, theaters, and sporting or concert venues create the same kinds of risks documented by studies of alcohol outlets is an empirical question for future research. More generally, while we believe SaTScan has significant promise for characterization of neighborhoods, we note that its measurement of clusters may not be appropriate for all studies. Despite meeting our statistical criteria as a cluster, some areas identified had very few nightlife establishments. As a result, the set of clusters identified may vary over time and be sensitive to the closure of a single establishment. Researchers should consider whether any specific statistically significant cluster, which may include only a few establishments in a largely residential area, is likely to have an impact on the dynamics and spatially patterned outcomes of interest. In addition, the NETS data may not be appropriate or feasible for all (U.S.-based) studies. A license to use the NETS data is expensive and time-limited; experience and expertise are required to work with the data; and the data may require time-consuming processing and additional geocoding. As an alternative to the use of national NETS data, restaurant inspection and alcohol sale licensure data could be used with SaTScan to similarly identify nightlife areas within a single city.

Data availability statement

The data that support the findings of this study are available from Walls & Associates. Restrictions apply to the availability of these data, which were used under license for this study. Data are available from Walls and Associates at dwalls2@earthlink.net.