Surviving Economic Adversity: Governance of Franchise Clusters

Franchise Clustering

Franchise clustering refers to the territorial arrangement of same-brand outlets in geographically proximate regions. Building on Porter's (1998, 2000) seminal concept of business clusters (i.e., groups of interconnected, proximate firms with a shared regional identity; Romanelli and Khessina 2005), recent research (e.g., Butt et al. 2018; Kalnins 2004) logically extends this concept to same-brand franchised outlets, which inherently share geographic proximity, interdependence, and brand identity (Zheng, Ji, and Su 2020).

Clustered franchised outlets benefit in two primary ways. First, clustering facilitates frequent communications, enabling the exchange of operational experiences and local knowledge (Ingram and Baum 1997; Kalnins and Mayer 2004; Kim and Jap 2022). Such communication facilitates exchange of resources and operational experiences that expands the pool of know-how, thereby enhancing their ability, adaptability, and creativity to interpret evolving market conditions and also creates a common bond due to shared risks and goals. For instance, our in-depth interviews revealed that clustered outlets share excess inventory, borrow labor, and share information on availability of “quality labor and supplies” during times of operational difficulty. Such sharing serves as a survival mechanism that is critical when market conditions are challenging. Second, clustering allows franchisors to coordinate on-site supervision and lowers the average oversight cost associated with managing each individual outlet (Lu and Wedig 2013; Shane 1998)—another vital survival mechanism because it helps ensure maintaining the “brand quality” at each franchised outlet (Heide, Wathne, and Rokkan 2007; Kidwell, Nygaard, and Silkoset 2007). This benefit, however, comes as a cost: Intensified supervision lowers the outlet's autonomy, thereby restricting its ability to respond to evolving market demand with agility (Kalaignanam et al. 2021).

Existing literature predominantly uses two methods to define clusters. The first relies on using a fixed geographic radius (e.g., 25 miles) or an administrative boundary like a state or a county but ignores DT—the number of same-brand outlets within the boundary—to define a cluster (e.g., Bell 2005; Butt et al. 2018; Lu and Wedig 2013). As such, this approach generates clusters that differ significantly in the number of outlets. For example, a franchised outlet located along a highway may have just one same-brand outlet within its 25-mile radius, whereas another situated in downtown Manhattan might have 15 within the same distance. The second method incorporates both radius and DT in defining a cluster (e.g., Alcácer and Zhao 2016; Stallkamp et al. 2018) but applies a uniform DT across all regions and firms, ignoring differences in market capacity (e.g., New York City vs. upstate New York) and the need for franchisor-specific DTs. Table 1 summarizes the representative studies. To accurately capture outlet dynamics, we hence need a more precise method for accurately identifying which outlets are part of or outside a cluster as well as identifying the boundaries of that cluster.

OPEN IN VIEWER

Table 1.

Representative Studies on Business Clustering.

Author	Research Setting	Method to Detect Clustering	A DT Considered	Firm-Specific, Demand-Adjusted DT	Governance Mechanisms Studied	Key Findings
Chung and Kalnins (2001)	Texas lodging industry	The fraction of chain-affiliated firms and the fraction of firms in the market that are larger than the focal hotel in each zip code				In rural areas, chain and larger hotels contribute to positive externalities.
Kalnins (2004)	Texas lodging industry	The distance to a specified ordinally proximate branded chain hotel (e.g., the tenth closest)	✓		Franchised outlets vs. company-owned outlets	Encroachment occurs when franchisors add new units proximately to their franchisees’ existing units.
Bell (2005)	Mutual fund companies in Canada	A dummy variable indicating whether the firm's head office is located in Toronto			Network centrality	Locating in the industry cluster as well as centrality of managerial tie network enhances firm innovation.
Kukalis (2010)	U.S. firms in semiconductor and pharmaceutical	U.S. Cluster Mapping Project	✓			Nonclustered laggards have higher financial performance than geographically clustered firms in the late stage of industry life cycle.
Lu and Wedig (2013)	Medicare- and Medicaid-certified nursing homes	Number of sibling units within 25 miles			Learning and monitoring	Clustered nursing homes achieve higher quality conditional on labor inputs and patient characteristics.
Alcácer and Zhao (2016)	A practical manual	A density-based clustering algorithm	✓			The authors propose a density-based clustering algorithm applicable to various empirical settings
Butt et al. (2018)	A large U.S.-based franchisor of automotive maintenance and repair services	Local Moran's I index to estimate clustering strength between two outlets			Shared ownership; franchisor- vs. franchisee ownership	The impact of clustering of same- brand outlets on sales is contingent on outlets’ experience and the governance context.
Lee (2018)	Firms in nine industries across six countries	Survey				Firms with greater attractiveness to resources or greater potential to exploit incoming knowledge spillovers are more likely to enjoy a positive effect of clustering on their growth.
Stallkamp et al. (2018)	Japanese foreign multinational enterprises (MNEs) in China	The density-based clustering algorithm proposed by Alcácer and Zhao (2016)	✓			Entry through a coethnic core increases the likelihood that an MNE will establish subsequent subsidiaries in other coethnic cores.
This study	Franchised outlets in food services and professional services	Firm-specific, demand-adjusted density-based algorithm adapted from Alcácer and Zhao (2016)	✓	✓	Franchisee ownership fragmentation; franchisor on-site supervision	The efficacy of governance mechanisms employed by franchisors varies across the density of outlets.

Economic Adversity and the Need for Agility

Macroeconomic shocks such as 2008–2009 financial crisis and the COVID-19 pandemic pose significant challenges to franchised systems, which must adapt and respond to the dynamically evolving economic conditions (Das et al. 2021; Srinivasan, Lilien, and Sridhar 2011). Changing market conditions and consumer preferences can be disruptive, and firms’ practices that work well during normal times might no longer be effective. In particular, adverse market conditions and the resulting possibility of store closure compel firms to adapt and develop innovative solutions—in essence, practice “marketing agility” where they have to “rapidly iterate between making sense of the market and executing marketing decisions to adapt to the market” (Kalaignanam et al. 2021, p. 36) to continue to succeed (Homburg, Theel, and Hohenburg 2020). By prioritizing adaptability and rapid response, agile marketing helps organizations align more closely with consumer demand and help cultivate deeper engagement and loyalty (Metcalf 2024).

We contextualize marketing agility within franchise cluster governance by examining how franchisor-deployed governance tools impact the franchisee's ability to construct local, innovative solutions that enable it to survive economic adversity. Specifically, we focus on franchisee-team-level factors, because colocated franchised outlets are not independent organizations but, instead, operate collectively as a team and address common problems (Gill and Kim 2021). As one of our franchisee interviewee put it, “While we [franchisees] usually compete over clients and talents, when the number of clients declined overall, everybody had to fight together to attract clients rather than compete for them. We were bonded to survive. If one store fails, it my spill over to mine.” In a similar vein, another franchisee explained that “during the pandemic, franchisees became more closely connected because everyone needs to survive, leading to more frequent communication. This includes supporting each other, discussing work arrangements, how to cope with difficulties, and planning strategies.” Consistent with this sentiment, we argue that certain governance characteristics of a franchisor's cluster can impact outlet survival during adversity by (1) enabling the generation of diverse ideas that get shared within the cluster (i.e., team diversity) and (2) implicitly empowering the outlets to undertake autonomous decisions (i.e., team empowerment). We turn to the discussion of these governance mechanisms next.

A Franchisors’ Governance Tool Kit

Unlike governance structures such as royalty rates, upfront franchise fees, initial training, and so on that are fixed and common across the franchised system, franchisee ownership fragmentation and franchisor on-site supervision operate primarily at the regional and outlet levels. While they may exhibit some historical path dependence and be constrained by factors such as labor agreements, union rules, or franchisor resource limitations, franchisors have the discretion to adjust these governance tools across regions and outlets. Crucially, franchisees themselves do not control these governance tools; rather, their actions manifest through the day-to-day operational responses within the boundaries set by franchisor governance. This distinction underscores that adaptation occurs primarily through the adjustments accorded by existing franchisor-led governance structures and franchisee-level operational reactions, rather than fundamental changes to the governance structure itself. As Table 1 shows, neither of these governance mechanisms have received systematic examination in the literature on either franchise clustering or resilience during economic adversity.

Franchisee ownership fragmentation refers to the extent to which ownership is distributed across multiple unique franchisees within a region (either in or out of cluster). Franchisors usually have complete discretion over the number of unique franchisees in a given area. For instance, TCBY's 2019 franchise agreement notes, “You are not granted an exclusive area or protected territory around the premises within which we or our affiliates agree not to issue franchises,” while Tropical Smoothie's 2022 franchise agreement states, “We will grant you a protected area consisting of a geographical area within one-mile radius around the site of your Tropical Smoothie Cafe Restaurant.” Ownership fragmentation expands the knowledge pool within the cluster by bringing in a diversity of ideas and solutions that complement the information sharing within a cluster and enhances the outlet's ability to adapt and create valuable solutions (Argote, Gruenfeld, and Naquin 2001; Carton and Cummings 2013; Van Knippenberg et al. 2004). A fragmented ownership structure, however, may constrain the franchisor support and resources given to each individual franchisee (Combs and Ketchen 2003; Dant, Kaufmann, and Paswan 1992; Massimino and Lawrence 2019) and increase market saturation (Kalnins 2004; Kim and Jap 2022) under normal conditions.

Franchisor on-site supervision refers to the degree of on-site monitoring and guidance franchisors provide to ensure compliance with their established business model. On-site supervision is a critical mechanism for maintaining operational standards (Brickley and Dark 1987; Combs and Ketchen 2003; Perryman and Combs 2012). While modern technologies enhance the franchisors’ ability to oversee operations remotely, on-site supervision, through the use of regional offices, remains indispensable for ensuring compliance (Bradach 1997; Dutta et al. 1995; Shane 1998; Wang et al. 2024). Enforcing compliance, however, reduces the franchisee autonomy and inhibits their ability to adapt and undertake innovative solutions. We proxy a franchisor's on-site supervision ability of an outlet by the distance of the outlet from the franchisor's regional office and investigate how such on-site supervision differentially impacts in-cluster outlets when economic conditions deteriorate.

The Role of Franchisee Ownership Fragmentation

Effective governance mechanisms can help firms mitigate the detrimental effect of cluster density on franchised outlet survival under economic adversity. Specifically, we propose that greater franchisee ownership fragmentation, where clustered outlets are owned by a larger number of unique franchisees, mitigates this joint negative impact. Each store owner entrepreneur brings in unique ideas, business know-how, and experience, enriching the cluster's knowledge diversity. This mirrors the logic in research on board of director networks, where well-connected boards aggregate diverse external knowledge to navigate uncertainty and improve innovation performance (Chang and Wu 2021). Similarly, organizational research suggests that diverse members share unique frameworks that facilitate learning and tackle complex tasks (e.g., Argote, Gruenfeld, and Naquin 2001; Carton and Cummings 2013; Van Knippenberg, De Dreu, and Homan 2004). Extending this logic, we argue that higher franchisee ownership fragmentation increases knowledge diversity within a franchise cluster, providing an information advantage that is especially valuable for denser clusters under economic adversity.

Specifically, under deteriorating economic environments, past practices may not address the evolving market demands, and franchisees are forced to continuously interpret the evolving market conditions and seek innovative, out-of-the-box solutions (Kalaignanam et al. 2021). While a denser cluster enables better communication and knowledge-sharing among the outlets, what also matters is the diversity of know-how within that cluster that enables a faster interpretation of the evolving market conditions and agile generation of feasible solutions. Compared with a cluster with low level of ownership fragmentation, a cluster with high level of ownership fragmentation has more franchisees with unique know-how, experience, and operational acumen that increases the diversity of know-how within the cluster, allowing them to pool ideas and strategies to navigate adversity (Carton and Cummings 2013). This increase in knowledge diversity, in turn, complements the knowledge-sharing ability of a denser cluster and enables the franchisees to make valuable and agile adaptations. Consequently, this reduces the outlet failure rate. Thus,

H₁: Under more adverse economic conditions, a higher level of franchisee ownership fragmentation mitigates the effect of cluster density on the failure hazard of franchised outlets.

The Role of Franchisor On-Site Supervision

Conversely, we propose that during economic adversity, a high level of franchisor on-site supervision amplifies the impact of cluster density on franchised outlet failure. While supervision normally ensures brand compliance and operational effectiveness (e.g., Agrawal and Lal 1995; Antia, Mani, and Wathne 2017; Zheng, Ji, and Su 2020), particularly for denser clusters, during economic downturns strict oversight hinders franchisees from making critical local adaptations. For example, during the COVID-19 pandemic, a franchisee avoided closure by temporarily sourcing ingredients from an unauthorized supplier to overcome severe supply disruptions. This survival tactic was only possible due to minimal franchisor supervision at the time. As economic adversity increases, franchisees require operational flexibility to adjust to market conditions, and on-site supervision restricts this adaptation in two main ways: First, enforcing rigid compliance erodes decision-making autonomy and slows response times; second, mandating standardized procedures prevents franchisees from innovating tailored solutions to local challenges. Ultimately, these local challenges—and solutions—are likely to differ across clusters with varying densities.

In sum, centralized franchisor-designed policies bolstered by on-site supervision, which are beneficial in normal times, exacerbate the misalignment of interests among parties during adverse economic conditions by stifling flexibility, innovation, and collaboration. While larger clusters improve monitoring efficiency during normal times, during times of economic adversity, more intense supervision is detrimental to the survival of the franchisee outlet. As one franchisee interviewee stated, “[During times of economic adversity,] we do not count on the franchisor for much operational guidance because they are not familiar with the local market. But we hope that they can provide an exchange platform for us to talk and exchange ideas.” Therefore, we propose the following hypothesis:

H₂: Under more adverse economic conditions, a higher level of franchisor on-site supervision amplifies the effect of cluster density on the failure hazard of franchised outlets.

Empirical Context and Data Sources

We focus on the business format franchising sector in the United States, where rich and reliable sources of archival information exist on franchisor–franchisee governance. We focused on two major industry sectors that require direct customer contact and have been significantly influenced by economic adversity: Accommodation and Food Services (two-digit North American Industry Classification System [NAICS] 72) and Professional, Scientific, and Technical Services (NAICS 54). We constructed our database using multiple sources, including the franchise disclosure documents (FDDs), the U.S. Census Bureau, and the Bureau of Economic Analysis. Specifically, we used FDDs to identify and record the geographic locations of franchised outlets from their exhibits and convert the text information (addresses) to geographic coordinates (longitude and latitude) using the Bing Maps application programming interface, together with R packages “RCurl” and “RJSONIO.” We delineated the contours of franchise clusters using our proposed algorithm after mapping the outlets. The franchised systems in our dataset included familiar brands like American Leak Detection, Buffalo Wings & Rings, Decor & You, Famous Dave’s, Fastsigns, Pancheros Mexican Grill, Property Damage Appraisers, Schlotzsky's Deli, Steak ’n Shake, TCBY, and Tropical Smoothie Cafe.

As our research focuses on franchised outlet survival, we checked whether an individual outlet was operating in a particular year and recorded its survival longitudinally. We then match-merged these location and survival data with other outlet-, system-, and market-level variables, including governance mechanisms and control variables obtained from FDDs and company websites. Economic and demographic data were acquired from the U.S. Census Bureau and the Bureau of Economic Analysis. To ensure the accuracy of our analysis, we included only those outlets for which we could determine their years of inception within our examination window. Compiling and merging those data sources yielded 35,911 observations (8,677 franchised outlets across 18 franchisors) located in all 50 states over a 14-year time window (2008–2021).

Unit of Analysis and Dependent Variable

Our unit of analysis was an individual outlet i (i = 1 … n) observed across t years since its inception (t = 1 … T). All the outlets in our sample started their franchised operation within our observation window of 2008–2021. Given the distinctive characteristic of survival analysis that the dependent variable be measured as the time duration until the failure event occurs (i.e., the inverse of franchised outlet survival; Cleves et al. 2010; Del Rio Olivares et al. 2018), our dependent variable was the outlet-specific Failure, which we coded as 1 if the outlet i was terminated in year t, and 0 otherwise (Kalnins and Mayer 2004; Winter et al. 2012).

Delineation of Clusters

To test the joint impact of cluster density and governance on outlet survival under different economic conditions, we first delineated the contour of the same-brand franchise clusters. There are several factors to consider when selecting an appropriate identification method. First, the method must incorporate a DT (i.e., a minimum number of franchised outlets) in defining a cluster. Second, it should allow us to delineate the contours of franchise clusters which are often irregular in shape. Third, outlet density varies considerably across franchises; hence, a firm-specific DT is necessary to define and identify franchise clusters. Fourth, market capacity also varies by region and fluctuates with the economic environment; hence, the DT for defining and identifying clusters should vary with market capacity. For example, if a firm-specific DT is set at 5, then an area with a higher market capacity warrants a higher threshold (e.g., DT = 10), requiring an adjustment to the DT.

Based on these considerations, we adapted a density-based algorithm for cluster detection proposed by Alcácer and Zhao (2016), which uses a uniform DT but allows for clusters with irregular shapes. To incorporate firm-specific heterogeneity in outlet density and local market capacity in defining and constructing clusters, we modified the algorithm by adding a firm-specific and demand-adjusted DT to our clustering approach. Specifically, we calculated the firm-specific average number of outlets per area as the base for that firm's DT and then used GDP at the county level as a weight to adjust the DT. This enabled us to capture the influence of both the population size and individual income (i.e., GDP per capita). Web Appendix B provides the details of this method.

Figure 2 provides an intuitive illustration of the clustering process. First, the algorithm selects a random outlet A and calculates the number of outlets (density) within a given neighborhood radius (NR), e.g., 25 miles (Figure 2, Panel A). If the density exceeded the DT, all outlets within the NR received a cluster ID. Next, the algorithm moved to another arbitrary outlet B within the identified cluster for A (i.e., within 25 miles of A) and assessed whether the number of outlets within the given NR of B was equal to or larger than the DT (Figure 2, Panel B). If yes, the two clusters merge, expanding the original cluster's boundaries and sharing its ID. This iterative expansion continues until the cluster stops growing. This usually forms an irregular contour (Figure 2, Panel C). Finally, the algorithm evaluates the remaining unvisited outlets to form new clusters (Figure 2, Panel D) and repeats the process until no unassigned outlets meet the DT criteria.

OPEN IN VIEWER

Figure 2.

The Process of Cluster Detection.

Note that this technique also enables us to track the historical pattern of expansion of the franchise system. Hence, the clusters in our analysis are not static. Web Appendix C shows the evolution of the average number of franchised outlets per cluster and the average distance between two outlets within store clusters over the examination window. Specifically, the average cluster size across all franchise firms in our sample, as defined by our technique, peaked at around 120 in 2013–2014. During the COVID-19 pandemic, the average size of those clusters dropped to around 90 (Figure C1 in Web Appendix C). Across the United States, some states had multiple clusters for the same franchised brand, whereas others had none. For example, using our technique, we identified more than one cluster of Tropical Smoothie Cafe in Oklahoma in 2021 but none in Colorado that year (Figure C2 in Web Appendix C).

Independent Variables and Moderators

Control variables

We controlled for a variety of outlet-, system-, and market-level variables that might simultaneously affect a franchised outlet's survival. First, at the outlet level, we controlled for outlet age (AGE_it) to capture the potential effect of a franchisee's experience on its likelihood of failure (Kalnins and Mayer 2004). We also controlled for outlet embeddedness of the focal outlet because it affects knowledge exchange among outlets. Specifically, building on the concept of tie strength from network literature, we measured the embeddedness using the inverse average distance between the focal outlet and other outlets within the same cluster (or within the same counties for out-of-cluster outlets) as the shorter distance indicates greater tie strength with other outlets (Rowley, Behrens, and Krackhardt 2000; Zhang and Guler 2020).

Second, at the franchise system level, we controlled for franchisor age (FRAGE_jt), as franchisor experience may affect outlet performance (Ingram and Baum 1997). Franchise system size (FRSIZE_jt), measured as the total number of outlets in a given year t, was included in the model as larger franchise systems have scale advantages and greater resources, which may impact the closures of outlets within the system (Ingram and Baum 1997; Kalnins and Mayer 2004; Shane 1998). We controlled for royalty rate (ROYAL_jt), as it directly impacts franchisee returns, in turn affecting the likelihood of outlet closure (Michael and Combs 2008). We also controlled for total assets (TOTASS_jt) and system net income (SYSINC_jt), which indicate the operational efficiency of the franchise system, thus influencing the outlet failure. The inclusion of SYSINC_jt ensures that the failure-risk-reducing role of governance mechanisms does not come at the expense of reduced franchise system performance. Franchisor-provided services (SERVICE_jt), measured as the number of services provided by franchisor j in a given year t, and contract completeness (CONTRACT_jt), measured as the number of clauses specified in the contract of franchisor j in a given year t, were also controlled for in our model. These variables reflect an outlet's access to resources that can influence their survival rate (Antia, Mani, and Wathne 2017; Butt et al. 2018; Kalnins 2004; Kashyap and Murtha 2017).

Third, at the market level, we accounted for interbrand competition intensity (COMP_kt), as the number of outlets owned by other franchisors in the same industry and the same county, to control for potential competitive effects on outlet failure (Kalnins and Mayer 2004). We also controlled population (POP_kt), area (AREA_kt), and per capita income (INCOME_kt) in year t of county k in which the focal outlet is located (Butt et al. 2018). Finally, we added the industry dummy (INDUS_it) to control for industry-specific effects (Shane 1998). The variance inflation factors range from 1.06 to 2.32, which were all lower than the cutoff benchmark (10.0), indicating that multicollinearity was not a concern (Hair et al. 1995). Table 2 provides the measures and data sources for all variables, and Table 3 provides the descriptive statistics and correlation matrix for all variables included in the study.

OPEN IN VIEWER

Table 2.

Definition, Measurement, Data Sources, and Literature Support for All Variables.

Variables	Measures	Data Sources	Supporting Literature
Failure (FAILURE)	A dummy variable indicating whether the focal outlet i exits in year t	FDDs	Kalnins and Mayer (2004)
Cluster density (CLUSTER)	Number of outlets in the cluster for in-cluster outlets and 1 for out-of-cluster outlets	FDDs and computed based on latitude and longitude	Butt et al. (2018). Lu and Wedig (2013)
Economic adversity (EA)	Decline in GDP in the county k in year t, defined as −(GDPk,t − GDPk,t−1)	Bureau of Economic Analysis	Steenkamp and Fang (2011)
Franchisee ownership fragmentation (OWNFE)	The number of franchisees in the cluster where the focal outlet is located for in-cluster outlets and the number of franchisees not part of the cluster within the same state s for out-of-cluster outlets	FDDs and computed based on latitude and longitude	Carton and Cummings (2013), Kaufmann and Dant (1996)
	The number of franchisees in the cluster where the focal outlet is located for in-cluster outlets and the number of franchisees not part of the cluster within the same county k for out-of-cluster outlets (for robustness check)	FDDs and computed based on latitude and longitude
	The average number of outlets per franchisees (for robustness check)	FDDs	Massimino and Lawrence (2019); Kalnins and Lafontaine (2004)
Franchisor on-site supervision (FRSUP)	Logarithm of distance between the focal outlet and the nearest regional office	FDDs and computed based on latitude and longitude	Brickley and Dark (1987), Perryman and Combs (2012), Nishida and Yang (2020)
Outlet age (AGE)	The number of years since the establishment of the outlet i	FDDs	Kalnins and Mayer (2004), Shane (1998)
Outlet embeddedness (EMB)	The inverse average distance between the focal outlet i and other outlets within the same cluster, or other outlets within the same county k for out-of-cluster outlets	FDDs and computed based on latitude and longitude	Rowley, Behrens, and Krackhardt (2000), Zhang and Guler (2020)
Franchisor age (FRAGE)	The number of years since the establishment of the franchisor j	FDDs	Ingram and Baum (1997)
Franchise system size (FRSIZE)	The number of outlets in the franchise system j	FDDs	Ingram and Baum (1997), Kalnins and Mayer (2004), Shane (1998)
Royalty rate (ROYAL)	The percentage of a franchisee's gross sales that the focal outlet i must be paid to its franchisor j in year t	FDDs	Michael and Combs (2008), Shane (1998)
Total assets (TOTASS)	Total assets of the franchisor j in year t	FDDs	Zheng, Ji, and Su (2020)
System net income (SYSINC)	Net income of the franchisor j in year t	FDDs	Zheng, Ji, and Su (2020)
Service (SERVICE)	The number of ongoing service categories provided by franchisor j in year t	FDDs	Antia, Mani, and Wathne (2017)
Contract completeness (CONTRACT)	The number of clauses specified in the contract of franchisor j in year t	FDDs	Kashyap and Murtha (2017)
Interbrand competition (COMP)	Number of franchised outlets of competitor brands located in county k in year t	FDDs	Butt et al. (2018), Kalnins and Mayer (2004)
Population (POP)	Population of county k in year t	Bureau of Economic Analysis	Kalnins and Mayer (2004)
Area size (AREA)	Area of county k (in square miles)	U.S. Census Bureau	Butt et al. (2018)
Income per capita (INCOME)	Income per capita of county k in year t	Bureau of Economic Analysis	Kalnins and Mayer (2004)
Industry (INDUS)	A dummy variable indicating whether the franchise system j belongs to accommodation and food services or professional, scientific, and technical services	U.S. Census Bureau	Shane (1998)

OPEN IN VIEWER

Table 3.

Descriptive Statistics and Correlation Matrix.

	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18
1. Outlet failure	1
2. Cluster density	.003	1
3. Economic adversity	.024*	.036*	1
4. Franchisee ownership fragmentation	−.018*	.003a	−.174*	1
5. Franchisor on-site supervision	.055*	.090*	−.022*	.022*	1
6. Outlet age	−.078*	−.007	−.041*	−.015*	−.041*	1
7. Outlet embeddedness	−.018*	−.099*	−.047*	.086*	−.010*	.016*	1
8. Franchisor age	.020*	.144*	−.015*	−.061*	.022*	.088*	−.126*	1
9. Franchise system size	−.057*	−.006	−.068*	.289*	.093*	.038*	.115*	−.349*	1
10. Royalty rate	.107*	.055*	.061*	−.113*	−.025*	−.087*	−.111*	.234*	−.101*	1
11. Total assets	.024*	−.056*	−.007	−.108*	−.053*	.007	−.076*	.027*	−.197*	−.081*	1
12. System net income	−.034*	−.031*	−.032*	.018*	−.034*	.066*	.005	.090*	.078*	−.117*	−.143*	1
13. Service	−.013*	.055*	−.075*	.156*	.189*	.007	.063*	.073*	.082*	−.231*	−.008	.063*	1
14. Contract completeness	−.044*	.051*	−.062*	.001	−.041*	.108*	.067*	.086*	.154*	−.037*	−.109*	.103*	.465*	1
15. Interbrand competition	.004	−.065*	−.346*	.107*	.025*	.016*	.112*	.002	−.046*	−.048*	.004	.011**	.055*	.048*	1
16. Population	.013*	−.033*	−.546*	.264*	.013*	−.001	.051*	−.028*	.030*	−.060*	−.015*	.016*	.099*	.024*	.616*	1
17. Area size	.002	−.104*	−.165*	.071*	.036*	.022*	−.036*	−.021*	−.012*	−.007	.003	−.009*	−.004	−.036*	.236*	.343*	1
18. Income per capita	−.043*	.142*	−.252*	.151*	.032*	.120*	.085*	.039*	.172*	−.075*	−.029*	.071*	.135*	.162*	.199*	.142*	−.143*	1
Mean	.115	14.91	−2.583	23.082	.995	2.279	.034	33.180	520.787	.058	336.853	15.384	9.651	4.249	66.799	919.812	1251.898	52.542
SD	.319	47.85	7.119	37.466	.893	3.138	.048	15.077	318.363	.022	955.878	45.878	1.575	1.913	87.01	1391.561	2139.170	15.583

*p < .05.

a

Postorthogonalization correlation coefficient.

Model Specifications

Our dependent variable was the failure of a franchised outlet; hence, we employed survival analysis as our estimation approach. To address the endogeneity of cluster density, as well as our two governance variables, franchisee ownership fragmentation and franchisor on-site supervision, we employed an instrumental variable (IV) approach in our main analysis as well as an IV-free approach as a robustness check. In the main analysis, we used a control function approach and specified a two-stage model (Danaher et al. 2015; Del Rio Olivares et al. 2018; Mani, Astvansh, and Antia 2024; Risselada, Verhoef, and Bijmolt 2014; Terza, Basu, and Rathouz 2008) to capture the endogenous part of the error term so that the independent variables no longer correlated with the error term in the survival equation (Sande and Ghosh 2018). Following prior research (Petrin and Train 2010; Sridhar and Srinivasan 2012), we first performed an auxiliary estimation with the endogenous variable as the dependent variable and IVs related to the endogenous variable but not to the focal dependent variable as independent variables. We then added the “residual” obtained from the auxiliary equation as an additional control variable to the second-stage main regression equation. We provide the model specifications for these two stages next.

First-stage regression

The first endogenous variable was cluster density. Strategically, franchisors will likely open more outlets in areas with larger populations. Therefore, we instrumented cluster density by the number of public hospitals in state s (HOS) because it correlates with the size of the local population but was beyond the control of any franchisors and therefore not likely to be correlated with the error term of the equation. Hence, this instrument met both the relevance criterion and exclusion restriction. The second endogenous variable was franchisee ownership fragmentation. We generated the instrument for this endogenous variable by measuring the corresponding peer franchisee ownership fragmentation (MeanOWNFE) across all franchise systems in the focal franchisor's industry d, excluding the focal franchisor j, averaged across its neighboring counties k^m, where m is the number of counties adjoining the focal state, and lagged the measure by one year as our instruments (Antia, Mani, and Wathne 2017; Germann, Ebbes, and Grewal 2015). The rationale for this instrument was that the focal outlet experienced similar market conditions as outlets in the adjoining counties and in the same industry (Germann, Ebbes, and Grewal 2015); as such, franchisors can be expected to demonstrate mimetic and isomorphic behavior (DiMaggio and Powell 1983; Kumar, Sunder, and Leone 2014), making similar decisions in their choice of franchisee ownership arrangement. The instrument also satisfied the exclusion restriction because those decisions by competing franchisors in the adjoining states were beyond the control of the focal franchisor, and those competing outlets did not directly compete with the focal outlet. It was therefore unlikely to be associated with the error term of the equation.

For franchisor on-site supervision, we used the number of direct flights (FLT) between the county where the focal outlet operated and the county where the headquarters were located as an instrument. The continuous nature of the instrument aligned with the continuous nature of the endogenous variable. The instrument was relevant because the number of direct flights between the focal outlet and the headquarters is negatively correlated with the likelihood of the franchisor providing on-site supervision to outlets, prompting franchisors to set up a regional office near the focal outlet (at a shorter distance to the focal outlet). It also satisfied the exclusion restriction because the availability of airline routes is determined by airline companies rather than the franchisor or franchisee, making it unlikely to correlate with the error term in the equation.

We estimated the first-stage models using ordinary least squares, where we regressed the three endogenous variables on their instruments and control variables (Controls), respectively.

Equations 1–3 show the model specifications:

DENSITY it = β 0 + β 1 HOS s , t − 1 + ∑ 2 12 β Control s i , t − 1 + ω 1 it ,

(1)

OWNFR it = β 0 + β 1 MeanOWNFE ( j ) ds m , t − 1 + ∑ 2 12 β Control s i , t − 1 + ω 2 it ,

(2)

FRSUP it = β 0 + β 1 FLT k , t − 1 + ∑ 2 12 β Control s i , t − 1 + ω 3 it .

(3)

Second-stage regression

We tested our hypotheses by employing the hazard function regression method. Since our data were grouped into discrete time intervals (i.e., yearly observations) and ownership arrangements, and because economic conditions and some of our control variables were time-varying, a discrete-time model that allowed for time-varying variables was appropriate in our research setting (Nikolaeva 2007). Following Nikolaeva (2007), Shmargad and Watts (2016), and Van den Bulte et al. (2018), we used the discrete-time hazards model in a complementary log-log parametric form to estimate the hazard of failure of outlet i in year t, given by g ( h it ) = X i ′ ( t ) β , where g(h_it) = ln(−ln(h_it)) was the complementary log-log link function, h_it was the discrete-time hazard of failure of outlet i in year t, and β captured the effects of the variable X on the hazard rate. This specification was a discrete-time alternative to the Cox proportional hazard model and allowed us to capture an underlying continuous-time process when the recorded data were at discrete-time intervals (Allison 1982; Prins and Verhoef 2007; Van den Bulte et al. 2018) and to incorporate the time-varying covariates (Chandrasekaran and Tellis 2008; Shmargad and Watts 2016). We included residuals from Equations 1–3 as covariates. The model specification, as shown in Equation 4, is as follows:

g ( h it ) = β 0 + β 1 CLUSTER i , t − 1 + β 2 EA s , t − 1 + β 3 OWNFR i , t − 1 + β 4 FRSUP i , t − 1 + β 5 CLUSTER i , t − 1 × EA S , t − 1 + β 6 OWNFR i , t − 1 × EA S , t − 1 + β 7 FRSUP i , t − 1 × EA s , t − 1 + β 8 OWNFR i , t − 1 × CLUSTER i , t − 1 + β 9 FRSUP i , t − 1 × CLUSTER i , t − 1 + β 10 OWNFR i , t − 1 × CLUSTER i , t − 1 × EA s , t − 1 + β 11 FRSUP i , t − 1 × CLUSTE R i , t − 1 × EA s , t − 1 + β 12 AGE i , t − 1 + β 13 EMB i , t − 1 + β 14 FRAGE j , t − 1 + β 15 FRSIZE j , t − 1 + β 16 ROYAL j , t − 1 + β 17 TOTASS j , t − 1 + β 18 SYSINC j , t − 1 + β 19 SERVICE j , t − 1 + β 20 CONTRACT j , t − 1 + β 21 COMP k , t − 1 + β 22 POP k , t − 1 + β 23 AREA k , t − 1 + β 24 INCOME k , t − 1 + INDUS j + ∑ p = 1 3 β 24 + p ω pi , t − 1 ,

(4)

where ω_pi,t−1 are the residuals obtained from Equations 1–3.

Robustness Checks

We conducted several analyses to verify that our results were robust to alternative measures, techniques, and model specifications. First, consistent with extant literature, we used a predetermined value of NR to be 25 miles. Alternatively, we constructed clusters with different values of NR by replacing the predetermined 25-mile radius with other odd numbers between 20 and 30 (i.e., 21, 23, 27, and 29). The firm-specific and demand-adjusted DT, as well as the rest of the model specifications, remained the same as in the main analysis. Models 1–4 of Table D1, Web Appendix D, show that the results remain largely consistent, suggesting that our results were not susceptible to our choice of NR.

Second, we used alternative measures for franchisee ownership fragmentation. In our original measure, if a focal outlet is not part of any cluster, we measured it by the number of franchisees not part of any cluster within the same state s for out-of-cluster outlets. We then replaced it by the number of franchisees not part of any cluster within the county k. The three-way interaction term among cluster density, economic adversity, and ownership fragmentation remained consistent with our main findings (β = −.033, p = .002; see Model 1, Table D2, Web Appendix D). Additionally, we used the average number of outlets per franchisees as a reverse proxy for franchisee ownership fragmentation, as the larger number of outlets owned by each franchisee indicates a lower level of ownership fragmentation. The results remain consistent with the main results (β = .235, p = .053; see Model 2, Table D2, Web Appendix D).

Third, we used an alternative instrument—driving time from the headquarters to the focal outlet—in the first-stage regression for franchisor on-site supervision because supervisory travel typically occurs by road rather than by air. The driving time was computed using the Open Source Routing Machine, an open-source routing engine that calculates shortest paths based on OpenStreetMap road network data (Luxen and Vetter 2011). The results remain consistent with our findings (see Model 1, Table D3, Web Appendix D).

Fourth, we changed the model specification. Specifically, we used multilevel mixed-effects complementary log-log regression to allow the intercept to be random for each firm. Given the nested structure of our data (i.e., outlets are nested in firms), this specification captured unobserved heterogeneity and within-firm correlation by allowing outlets in the same firm to share common random effects (Karniouchina, Uslay, and Erenburg 2011; Lindstrom and Bates 1990). The results remain consistent with our findings (see Model 2, Table D3, Web Appendix D).

Fifth, to mitigate potential biases introduced by the COVID-19 pandemic, we excluded data from the year 2020 and conducted the analysis based on the subsample. The findings remain consistent with our hypotheses (see Model 3, Table D3, Web Appendix D).

Lastly, we assessed the robustness of our results by adopting a Gaussian copula approach to account for the endogeneity concerns (Burchett, Murtha, and Kohli 2023; Haschka 2022; Park and Gupta 2012). The Shapiro–Wilk tests showed that cluster density (W = .610, p = .000), franchisee ownership fragmentation (W = .812, p = .000), and franchisor on-site supervision (W = .781, p = .000) were nonnormally distributed. Additional skewness tests likewise rejected normality for cluster density (skewness = 4.143, p = .000), franchisee ownership fragmentation (skewness = 1.750, p = .000), and franchisor on-site supervision (skewness = −2.852, p = .000), satisfying the requirement of the Gaussian copula approach (Becker, Proksch, and Ringle 2022). The results remain robust (see Model 4, Table D3, Web Appendix D).

Additional Analyses

To better understand the advantage of our firm-specific, demand-adjusted, density-based clustering algorithm, we compared our method with clustering methods proposed in previous studies—the number of outlets within 25 miles and the number of outlets in predetermined geographic units (e.g., a county). We then estimated the full model using these alternative measures, maintaining the same model specification as in our main analysis. Table E1 in Web Appendix E shows the results. We find that only the three-way interaction hypothesis of franchisee ownership fragmentation is supported if we use the number of outlets within 25 miles as our measure of a cluster, whereas neither of the two hypotheses receive support if we use number of outlets in an administrative region as our measure of a cluster, suggesting that how clusters get identified and measured is crucial for understanding the effects of governance on outlet survival in a cluster. Furthermore, we compared our method with the other two approaches on key model performance metrics like positive predicted value (PPV), negative predicted value (NPV), recall based on the optimal threshold inferred from the receiver operating characteristic (ROC) curve, the area under ROC curve (ROC AUC), AIC, and BIC (Lee, Yang, and Anderson 2026; Homburg, Koschate, and Hoyer 2005). As Table 6 shows, our firm-specific, local-demand-adjusted, density-based algorithm provided a better fit than the other approaches on most criteria, suggesting that our clustering procedure represents a better fit and an improvement over existing approaches. Particularly, our method yields consistently lower AIC/BIC values, which indicates that our approach provides a more accurate and efficient explanation of the data. Moreover, the results suggest that the effects of our focal governance mechanisms are underestimated using existing clustering approaches compared with the results obtained by our proposed clustering method. In sum, our clustering technique not only provides better fit but also yields different implications regarding the role of our particularistic governance tools in helping clustered outlets survive under economic adversity.

OPEN IN VIEWER

Table 6.

Performance Metrics of Different Methods for Defining a Cluster.

	Our Method	Using a 25-Mile Radius Defining a Cluster	Using an Administrative Region (i.e., County) Defining a Cluster
PPV	.171	.171	.178
NPV	.918	.916	.913
Recall	.630	.616	.564
ROC AUC	.647	.641	.640
AIC	24,635.788	24,700.286	24,706.029
BIC	24,881.963	24,946.461	24,952.205

In addition, we split the data by the median distance to the nearest regional office and reran our model for the two subsamples to test if excessive franchisor monitoring impedes local adaptation despite high franchisee ownership fragmentation. As shown in Table E2, Web Appendix E, for the high-monitoring group, the three-way interaction is insignificant ( β  = −.170, p = .502), but the three-way interaction remains significant for the low-monitoring group ( β  = −.050, p = .041). This supports the assertion that stricter monitoring hinders franchised outlets’ local adaptation, even though the ownership is highly fragmented.

Lastly, we assessed whether the deployment of these two governance tools to reduce outlet failure comes at the expense of other crucial measures of franchisee performance (e.g., outlet sales). To that end, we conducted an analysis using outlet sales as the dependent variable. We utilized SafeGraph Spend data, which comprises anonymized debit and credit card transactions from approximately 10.6 million consumers, covering more than 1.1 million business locations in the United States since January 2019 (Chiong, Kim, and Kim 2025). Merging this with the FDD data based on business names and locations yielded a subsample of 4,107 observations from 1,785 outlets over the period of 2019–2021. We applied a Heckman selection approach to address potential concerns about selection bias. Table E3 in Web Appendix E shows the results. We see that the determinants of outlet survival differ substantially from those of outlet sales. More importantly, the three-way interaction effect of economic adversity, cluster density, and our two governance tools has no impact on outlet sales, suggesting that franchise managers do not have to worry that the deployment of these governance tools creates a trade-off between survival and sales.

Theoretical Contributions

Our work makes three key contributions to research. First, we identify two novel governance mechanisms that vary at the cluster or outlet level—namely, franchisee ownership fragmentation and franchisor on-site supervision—which enable the outlets to operate more effectively and lower their failure hazards in the face of significantly adverse, exogenous shocks. We show that the effectiveness of the interplay between a firm's place strategy (here, clustering) and governance is contingent on a key third factor: the external, macro-economic environment. Specifically, we provide evidence that franchisee ownership fragmentation, a hitherto underexplored and novel governance tool, can be crucial to lowering the failure hazards of outlets especially under economic adversity. A more fragmented ownership structure reflects a greater diversity in team knowledge and ability, and this diversity enables the creation of ideas and solutions that complement knowledge-sharing within larger clusters to adapt better and avoid store failure. Likewise, we show that under economic adversity, lower on-site supervision provides better autonomy to the franchisees, enabling them to adapt to local conditions, and thereby reduces the likelihood of outlet failure. Both results suggest a core dynamic—the need to balance the franchisor's authority to maintain consistency in brand reputation with the franchisee’s need to adapt and be agile to address local conditions.

Second, our research represents a novel effort to explore how external, macro-economy-wide shocks influence the role of franchise clustering on the survival of franchised outlets. Although prior research has suggested that changes in the economic environment can profoundly impact the effectiveness of marketing activities and result in a “permanent realignment in the marketplace” (Srinivasan, Lilien, and Sridhar 2011, p. 49), there remains a significant gap in understanding how adverse economic conditions that disrupt market dynamics affect a firm's “place” strategy, especially when outlets are clustered within geographically proximate regions. Existing research has documented both the benefits and drawbacks of clustering under normal circumstances; yet it remains unclear whether clustering aids or hinders franchised outlet survival during periods of economic adversity. The effect is complex because economic downturns, whether global or regional, disrupt market dynamics by simultaneously intensifying competition among franchisees and fostering cooperation among them to collectively address external challenges. Employing survival analysis and a unique, extensive dataset, we find that under adverse economic conditions, the effect of clustering on individual outlet survival is contingent on franchisor governance. These findings provide new insights into the nuances regarding the effect of clustering strategies and highlight the need for franchisors to properly govern franchise clusters under adverse economic conditions, which joins the broader discourse on how firms navigate economic challenges (e.g., Das et al. 2021; Srinivasan, Lilien, and Sridhar 2011).

Finally, our research provides an accessible method for detecting franchise clusters. Building on Alcácer and Zhao (2016), we construct a firm-specific, local-demand-adjusted DT in identifying cluster formation. Incorporating such a threshold is important because DT by definition should correlate with a region's market capacity. Two factors determine a region's market capacity: its population size and its individuals’ disposable income, which collectively yield the local GDP. Therefore, we use local GDP as a weight to reflect the role of market capacity in identifying cluster formation and to capture varying economic conditions across time periods. To our knowledge, we offer the first firm-specific, local-demand-adjusted clustering technique for business cluster detection, complementing the existing literature and providing an accessible tool for identifying outlets located within clusters that should be managed collectively.

Managerial Implications

Limitations and Future Research

As with any research effort, our study has limitations that indicate avenues for future work. First, although our hypothesis development builds on interaction dynamics among franchisees, such as knowledge-sharing and collaboration, the underlying mechanisms are not directly tested. Future researchers can examine dynamic relationships among franchisees under different economic conditions by observing their online and offline interactions and formally testing these mechanisms. Second, franchised outlet failure hazard is a downside risk aspect of performance; investors, such as franchisees, may seek to balance the risks and rewards as trade-offs. Our study does not consider this additional trade-off. Future researchers can consider rewards such as profits and sales to reflect their role in firms’ location decisions. Third, our study considers only two governance instruments: franchisee ownership fragmentation and franchisor on-site supervision. Although those two are common mechanisms that can vary at the regional and outlet levels, they do not represent an exhaustive list. Future researchers can explore additional governance mechanisms, particularly those related to the benefits and costs of clustering (Tracey, Heide, and Bell 2014). Finally, we conducted our study in two specific industries; future researchers can explore whether our findings are generalizable beyond their current context.