Introduction

Small local businesses play a vital role in the dynamics of urban living, and are an essential part of both economic and cultural vibrancy of communities. The city of Boston is home to more than 40,000 small businesses, who represent 95% of all the businesses in the city, and 44% of employment and 37% of revenues for the city’s private, for-profit business sector. The vast majority (62%) of these small businesses serve local customers directly (City of Boston 2016).

Of these local businesses, food and entertainment businesses (e.g. restaurants, cafés and bars) are particularly interesting, because they not only provide basic eating and leisure services but also offer spaces for social gathering and inspire cultural innovations (Oldenburg 1999). A cluster of good restaurants and bars often implies visitor-friendliness and positive interactions between commerce and culture (Russo and van der Borg 2002; Mommaas 2004).

Most restaurant and bar owners would not refuse the chance to pour patrons a glass of alcoholic beverages, as having the choice of drinking on-site gives customers more reasons to stay and consume more, plus alcohol sales for on-premise consumption are notably lucrative. Business owners in Boston are willing to pay as much as $400,000 for a full liquor license (“Neighborhood-Based Licenses Making Slow Progress” 2015). Micro businesses started without alcohol service normally would also wish to obtain a liquor license when they plan to expand.

However, like many cities in the US, alcohol retail is highly regulated in Boston. As of August 2015, there are 4,676 food establishments and entertainment venues in the city, but only 20% (931) of them possess a liquor license1. Massachusetts General Law defines two major categories of liquor licenses: on-premise and off-premise licenses. Each category contains a few sub-categories indicating the type of alcohol beverages allowed to sell and the type of businesses the licenses are issued to. In principle, municipalities in Massachusetts are allowed to issue only one on-premise liquor license per 1,000 residents, except in Boston, where the cap numbers are restrained by much more complicated terms.

Studies repeatedly confirmed the common belief that restricting availability of alcohol is an effective measure to prevent alcohol-attributable harm (Popova et al. 2009), but the economic implications of liquor sales still gave Massachusetts legislators enough incentive to loosen alcohol regulations for small businesses. In 2014, an economic development bill embodied terms to release 75 new liquor licenses in Boston, 60 of which are restricted to designated underserved neighborhoods and the Boston Main Street districts. In December 2015, Boston City Council lifted BYOB (bring your own bottle) ban on small restaurants, despite the backlash from more established restaurants who have invested huge amounts in buying a liquor license.

In the Small Business Plan published in March 2016, the city of Boston laid down three goals in helping small businesses start and grow: to make the small business economy thrive, to enhance neighborhood vibrancy, and to foster economic and social inclusion and equity (City of Boston 2016). Various initiatives were proposed to fix the organizational, capital, real estate and human resource gaps in the small business ecosystem. For these initiatives to succeed, a thorough study on the current status of different small business segments can be vitally important.

This paper took a step in such analysis by using geographic, demographic and administrative datasets to compare the overlapping of alcohol accessibility and the clustering of small local businesses. It first constructed a measurement for on-premise alcohol accessibility, then identified the clustering of food establishments and entertainment venues throughout the city. The metrics were then examined against neighborhood characteristics, in search of answers for following questions:

The results indicated that small businesses in the urban context has a strong tendency to agglomerate and liquor licenses also tend to be needed in areas where alcohol accessibility is already good.

Methods

Data Sources

This study utilized Boston’s Business Licenses Dataset for the exact geographic locations and the typology of businesses, Tax Assessors’ Database for identifying residential parcels, and American Community Survey (ACS) data for assessing demographic characteristics of neighborhoods.

Business Licenses Dataset

The Business Licenses Dataset used in this study is a subset of business licenses issued by the City of Boston within a timeframe of roughly 7 years. It contains a raw database of 29,130 records of licensing data pertaining to 10,445 business establishments, including the contact info of the businesses and their full registration, renewal, modification and inspection history between January 1, 2007, and August 3, 2015. Licenses available for analysis are within a limited set of categories, including food establishments, on-premise liquor retail, live and non-live entertainment, swimming pools and day camps.

Acompanying these raw records are separate spreadsheet files with more processed data. Each file corresponds to a license category and contains slightly different information from the raw records (e.g. additional columns for opening hours and seating capacities; different categorization of licenses; and even missing or extra businesses).

When identifying food establishments and entertainment venues, I used the raw records, but when locating liquor licenses, I used mainly the separate liquor license spreadsheet. It was because for food and entertainment licenses, the raw records were in a more unified format, and the sheer amount of them made the differences between two data sources trivial; but for liquor licenses, the separate file had more up-to-date records and more accurate indications for neighborhood restricted licenses.

Auxiliary variables were created to consolidate the raw records into a list of unique food and entertainment establishments, each with flags indicating whether it was located in a Main Street district and whether it possessed a liquor license.

Other Data Sources

Boston Tax Assessors’ Database 2016 contains tax assessment attributes for all 169,199 uniquely identifiable parcels in the city for the year of 2016. This database was used to extract geographic locations of all residential and semi-residential parcels in the city, and they were used as the origin points for the calculation of alcohol accessibility.

ACS census indicators (2010-2014 estimates) as curated by Boston Area Research Initiative (BARI) was used to inspect neighborhood characteristics, including median house income, family size, race and age of residents, the proportion of owner-occupied residences, etc.

GIS data for Boston Main Streets were downloaded from the city’s GIS OpenData portal2.

Alcohol Accessibility

S L Handy and Niemeier (1997) classifies measures of accessibility into three basic types: cumulative opportunities measures count the number of opportunities reached within a given distance or travel time; gravity-based measures incorporate not only a separation factor, but also an attraction factor that discounts opportunities because of increasing time or distance from the origin; utility-based measures are based on an individual’s perceived utility for different travel choices (Susan L Handy and Clifton 2001; Bhat et al. 2000).

In accord with the data I had access to, I designed a viriant of the gravity-based measurement. It is expressed in following formulation:

\[A_{i} = \sum_{j = k} L \cdot e^{-b \cdot d_{ij}} \quad \quad [1] \]

Here \(A_i\) is the Alcohol Accessibility Index (AAI) of any given parcel, \(L\) is a weighting constant based on liquor license type, \(b\) is a constant for adjusting values of the measurement into a readable range, \(d_{ij}\) is the distance between the parcel and each one of the nearest \(k\) alcohol outlets. The cutoff number \(k\) was a subjective choice balancing between efficiency and accuracy.

For a neighborhood, its AAI is simply the median value of the AAIs of all residential parcels within its boundary. This study aggregated the final AAI to the census block group level.

The nearest alcohol outlets and their distances \(b_{ij}\) were calculated with KNN algorithm and nabor package in R. For the efficiency of computation, the geolocations were treated as Cartesian coordinates. The weighting coefficients \(L\) for different license types are as shown in Table 1.

Type Weight
Club 0.4
General On-Premises 0.6
Farmer Distillery 0.7
Common Victualler 1.0
Tavern 1.0
All Alcoholic Beverages 1.0
Malte & Wine Only 0.7

Table 1: Weight coefficients for different license types. “Malte & Wine” means this license only allows sales of malt beverages (e.g. beer) and wine, basically low-alcohol beverages.

The types of business and types of alcohol allowed to sell have different weight. The total weight is calculated with \(W_b \times 0.7 + W_a \times 0.3\), where \(W_b\) is the weight of business type and \(W_a\) is the weight of alcohol type.

Weighting is necessary because having a liquor outlet nearby does not mean residents must have easy access to it. Liquor licenses for hotels and airports were completely discarded since they do not bear the localness underlying in the purpose of this study. Clubs and general on-premise licenses (e.g. drinks provided by a Concert hall) were downplayed because they are not as accessible as regular restaurants and bars (common victuallers).

Business Clusters

Rosenfeld (1997) defines a working business cluster in the industrial sector as “an agglomeration of connected companies that are aware of their interdependence, value it, act on it, and collectively operate as a system to produce more than the sum of their individual parts”. In the urban context, such self-awareness is less patent, because the clustering of urban small businesses are often driven by careful urban planning efforts and restricted by land use and transportation conditions. In other words, they are shaped more by policymakers than by resourceless small businesses themselves.

Perhaps the most prominent public policy in the US pertaining the clustering of small businesses is the Main Street program established by National Trust for Historic Preservation in 1980. After participating in this program for 33 years, Boston now has a network of 20 Main Street districts. The program intended to revitalize historical business centers are catering budding business clusters, too.

Using a similar algorithm to Equation [1], a Business Density Index (BDI) was constructed to measure the density of businesses around a given business:

\[D_{i} = \sum_{j = k} e^{-b \cdot d_{ij}} \quad \quad [2]\]

The only difference was that the distances \(d_{ij}\) were computed from businesses to businesses, instead of residences to businesses. This metric will also be aggregated in block group level, using the median value of all businesses in a unit.

Data Analysis

AAI and BDI were both aggregated for census block groups. But BDI was examined individually, too. BDI at the business level were used to match the BDI metric with the “official” and intended business clusters as defined by Boston Main Streets.

Student’s t-test was used to examine the correlations between the distribution of small businesses and liquor licenses. They were expected to be strongly correlated.

Then I divided the businesses into two groups: those located within Main Street districts and those located outside. Linear regressions were employed to measure how likely would a business from Main Street districts possess a liquor license, and how does alcohol accessibility correlate with different neighborhood characteristics.

Results

Alcohol Accessibility Index

With a choice of \(k = 10\) and \(b = 100\) in Formula \([1]\), the AAI values for block groups are distributed between \(1.260\) and \(9.307\), with a median of \(5.237\). It was close to a t-distribution, indicating the values are reasonably representative of the differences between block groups.

**Figure 1.** The distribution of AAI at block group level

Figure 1. The distribution of AAI at block group level

The top three block groups with easiest alcohol access are all located in Downtown. Other neighborhoods that topped the list were North End (Waterfront), Bay Village, Chinatown, and Fenway. All of them are historic business districts with distinguished business vitality and compact residences (Table 2).

BG_ID_10 AAI Neighborhood
250250701018 9.307 Financial District/Downtown
250250701013 8.933 Financial District/Downtown
250250701014 8.874 Financial District/Downtown
250250304002 8.824 North End
250250701016 8.820 Bay Village
250250701017 8.819 Chinatown
250250703001 8.644 Bay Village
250250303003 8.624 Government Center/Faneuil Hall
250250101043 8.583 Fenway/Kenmore
250250303004 8.580 Financial District/Downtown

Table 2. Census block groups with Highest AAI scores

**Figure 2.** Alcohol Accessibility Index (AAI) by block group

Figure 2. Alcohol Accessibility Index (AAI) by block group

Business Density Index

The business density index, however, presented more unbalanced results. Even though results may vary if given different \(k\) values in Equation \([2]\), it is obvious that there are much more businesses getting a relatively higher BDI score, indicating businesses by their nature tend to agglomerate (Figure 3).

**Figure 3.** Distribution of Business Density Index for food and entertainment establishments

Figure 3. Distribution of Business Density Index for food and entertainment establishments

The Main Street districts overlapped with the actual business clusters reasonably well. The distribution above indicates that businesses with higher BDI scores are more likely to be located in the Main Streets. Surprisingly, though, when \(BDI > 80\), the trend of increasing BDI implying a larger likelihood in Main Streets did not continue. This is because Main Street districts are scattered regions of one or two streets, while there are much larger agglomerations of businesses in popular business areas such as the Downtown. It is also worth to mention that Main Street districts are not only about businesses, but also cultural preservation.

**Figure 4:** Business Density Index (BDI) by block groups and the boundaries of the Main Streets

Figure 4: Business Density Index (BDI) by block groups and the boundaries of the Main Streets

But when aggregated by block groups, the distribution turned closer to a t-distribution.

**Figure 5.** Distribution of BDI for block groups

Figure 5. Distribution of BDI for block groups

With \(k = 100\) (i.e. calculating with the nearest 100 businesses), the value of the business level BDI ranges from \(8.633\) to \(92.930\), at the block group level, it ranges from \(8.633\) to \(92.00\) (Table 3).

Min. 1st Qu. Median Mean 3rd Qu. Max.
Business 8.633 53.88 70.20 67.38 82.78 92.93
Block Group 8.633 41.42 52.18 53.91 67.68 92.00

Table 3: BDI summary by businesses and by block groups

Alcohol Access and the Clusters

The t-test between AAI and BDI of block groups confirmed that they strongly correlate with each other (p-value < \(2.2e^{-16}\)). This was predictable since both metrics were based on the geographic distances between businesses. What was more interesting was does and how being in the Main Streets predict the likelihood of a business owning a liquor license.

A logistic regression was used to answer this question. The independent variable was a flag indicating whether a business has a liquor license, which was determined by cross-examining the liquor licenses list and the food and entertainment businesses list. The dependent variables were the neighborhood AAI, i.e. the AAI of the census block group the business belongs to, and a boolean indicating whether the business is inside the Main Streets.

term estimate std.error statistic p.value
(Intercept) -3.7244168 0.2792996 -13.3348447 < 2e-16
Neighborhood AAI 0.1580356 0.0368524 4.2883404 0.0000180
in Main Street (Boolean) 0.1175054 0.1197827 0.9809881 0.3265986

Table 4: Efficients for the regression of whether a business owns a liquor license

The model (Table 4) suggests that a restaurant or entertainment venue is more likely to possess a liquor license when it is located in a neighborhood that already has access to other alcohol outlets. And although geographically the Main Streets largely overlap with neighborhoods of easy alcohol access, being in the main streets does not necessarily mean the business is more likely to have a liquor license.

Alcohol and Business Clusters with Neighborhood Characteristics

An exploratory regression was run for the AAI metric and neighborhood characteristic indicators from ACS census data. There were in total 43 variables in the census indicators dataset, those who showed positive signals were listed in Table 5.

term estimate std.error statistic p.value
(Intercept) -30.1352386 65.8519023 -0.4576214 0.6474669
Population density 0.0000069 0.0000032 2.1392329 0.0330068 *
Age < 18 3.1119750 1.0485886 2.9677750 0.0031754 **
Age 35 to 64 2.0355409 0.9639027 2.1117701 0.0353106 *
Black -2.8676085 1.2748509 -2.2493678 0.0250190 *
Ethnic heterogeneity 0.7181549 0.4165303 1.7241362 0.0854370 .
Median house income 0.0000078 0.0000028 2.7905615 0.0055074 **
Family house percentage -1.4706555 0.4958119 -2.9661558 0.0031919 **
Less than high school 4.2712253 1.6073732 2.6572704 0.0081854 **
Professional degree 3.8204867 2.1230547 1.7995234 0.0726710 .
Go to work by walk 2.2145500 1.1026136 2.0084553 0.0452502 *
Renters percentage 1.3761247 0.4261862 3.2289280 0.0013425 **
Median Home Value 0.0000009 0.0000004 2.0945964 0.0368201 *
MedYrBuilt: < 1940 0.5594451 0.2647851 2.1128269 0.0352195 *

Table 5: Linear regression of AAI on 43 neighborhood charasteristics. Only those with significant p values are displayed.

Most of these variables are percentages of the population in a specific demographic group. This result shows that easier access to alcohol outlets is more likely to be found in neighborhoods with higher renters percentages, fewer family houses, higher median income and more young children. This basically describes residential areas near the commercial centers in the Downtown, where most young professionals tend to live.

Another regression was fitted for the BDI metric. And the results were rather different.

term estimate std.error statistic p.value
Population density 0.0000751 0.0000314 2.3915432 0.0173895 *
White -30.8697827 13.5757858 -2.2738855 0.0236745 *
Black -24.4362910 13.1733609 -1.8549777 0.0645738 .
Hispanic -21.6403225 12.1526078 -1.7807143 0.0759643 .
Family house percentage -10.1540921 5.2638831 -1.9290117 0.0546663 .
Go to work by car -20.3475203 10.3917356 -1.9580483 0.0511448 .
Renters percentage 10.6246091 4.5333995 2.3436296 0.0197455 *
Median home value 0.0000186 0.0000041 4.4887683 0.0000102 **
MedYrBuilt: 1950 to 1959 -9.6552910 4.3417983 -2.2238000 0.0269014 *

Table 5: Variables with significant impact from a linear regression of BDI on 43 neighborhood charasteristics.

The most unanticipated finding from this model was that higher percentages of the white, black and Hispanic population all contributed negatively to business density. In fact, when regressing the Asian population with BDI separately, the result was even more conclusive (\(\beta = 44.010, p = 8.2e^{-8}\)). This was coming from the high proportion of Asian population in the highly vibrant and touristic Chinatown (there are 135 food and entertainment establishments in a land of 165,000 square meters and nearly 70% of the residents are Asian). Other communities with high Asian population–Bay Village (28%), West End (20%), Fenway (18%), Allston (16%)–may have also contributed to the fit model. We can see the high density of businesses around these neighborhoods in Figure 4. It is reasonable to believe that Asian community prefers having restaurants conveniently located around where they live more than other races.

Discussion

This paper designed two novel metrics to measure alcohol accessibility and business density, and employed correlation and regression tests to verify the implications of these metrics. It has found that businesses have a strong tendency to agglomerate, although I was not certain of whether it is because of purposeful design of city planners or incentives driven by small businesses themselves. Alcohol licenses for food and entertainment establishments not only go with clusters of businesses but also follow the footprints of existing on-premise alcohol outlets.

An exploration in neighborhood demographics suggested that certain demographics may prefer living in areas with easier on-premise alcohol access, but the accessibility to alcohol venues is hardly the determining factor of why they live there. However, some race groups, namely, Asians, do prefer having more food and drink services very near where they live.

These findings are helpful in understanding the centripetal force that drives the clustering of small local businesses and facilitate policy designs that make use of it. For example, we see an overlap of the Main Street districts and the actual business clusters, but a lot of businesses agglomerating around the clusters were not covered by the Main Street program. Resources should be invested to help those businesses succeed as well, or some other programs should be put forward to foster new business clusters.

The paper demonstrated a method of converting geographic information into analysable metrics and exemplified the potential of using administrative data in spatial analyses. The novel accessibility metric or a more sophisticated vesion of it can be used to monitor neighborhood equity and ensure equal access to important resources. This is inline with the second goal in the Small Business Plan: to forster economic and social inclusion and equity.

Importance of data quality

Working on this paper also underscored the importance of data quality in policy researching. There are huge potentials in the business licenses dataset, but I was not able to proceed many of my ideas because of its poor data quality.

For example, although the dataset contained issuing and expiration dates of the licenses, the information was highly incomplete–13% of the records did not include dates or had the expiration dates earlier than issuing dates. The deficiency of reliable temporal data was an excruciating impediment to any effort of longitudinal analysis.

The lack of a unique identifier was also a big issue. It has been very difficult to track businesses with multiple licenses. This study used a virtual identifier constructed from the unique address key and the name of the business owner. But the names of business owners were remarkably inconsistent across license categories. Even after fixing various spelling and capitalization inconsistencies that were easily identifiable, I was only able to map 366 out 1,009 liquor licenses with licenses in other categories. In fact, not being able to incorporate all available data might have undermined the trustworthiness of my previous prediction on liquor license ownership.

References

Bhat, Chandra, Susan Handy, K Kockelman, H Mahmassani, Q Chen, and Lisa Weston. 2000. “Urban Accessibility Index: Literature Review.” Center of Transportation Research, University of Texas at Austin, Springfield.

City of Boston. 2016. “Small Business Plan.” www.cityofboston.gov/news/Default.aspx?id=20545.

Handy, S L, and D A Niemeier. 1997. “Measuring Accessibility: An Exploration of Issues and Alternatives.” Environment and Planning A 29 (7). SAGE Publications: 1175–94. doi:10.1068/a291175.

Handy, Susan L, and Kelly J Clifton. 2001. “Evaluating Neighborhood Accessibility: Possibilities and Practicalities.” Journal of Transportation and Statistics 4 (2/3): 67–78.

Mommaas, Hans. 2004. “Cultural Clusters and the Post-Industrial City: Towards the Remapping of Urban Cultural Policy.” Urban Studies 41 (3). SAGE Publications: 507–32. doi:10.1080/0042098042000178663.

Oldenburg, Ray. 1999. The Great Good Place: Cafes, Coffee Shops, Bookstores, Bars, Hair Salons, and Other Hangouts at the Heart of a Community. Da Capo Press.

Popova, S., N. Giesbrecht, D. Bekmuradov, and J. Patra. 2009. “Hours and Days of Sale and Density of Alcohol Outlets: Impacts on Alcohol Consumption and Damage: A Systematic Review.” Alcohol and Alcoholism 44 (5). Oxford University Press (OUP): 500–516. doi:10.1093/alcalc/agp054.

Rosenfeld, Stuart A. 1997. “Bringing Business Clusters into the Mainstream of Economic Development.” European Planning Studies 5 (1). Informa UK Limited: 3–23. doi:10.1080/09654319708720381.

Russo, Antonio P, and Jan van der Borg. 2002. “Planning Considerations for Cultural Tourism: A Case Study of Four European Cities.” Tourism Management 23 (6). Elsevier BV: 631–37. doi:10.1016/s0261-5177(02)00027-4.

  1. As counted with the Business Licenses Dataset updated by that time. The numbers were counted by unique license IDs in the associated license categories. Liquor licenses owned by hotels and airport facilities were excluded.↩

  2. http://bostonopendata.boston.opendata.arcgis.com/↩