How Does Innovation-Collaboration Network Structure Influence Regional Economic Resilience?

Abstract

This paper examines whether different structures of co-patenting networks affect regional economic adaptation and adaptability in divergent ways. Adaptation differentiates from adaptability by recovering in a pre-shock growth path, while adaptability entails forging a new growth path. Focusing on computer, electronic, and electrical equipment manufacturing industries in U.S. metropolitan and micropolitan areas from 1990 to 2012, the study reveals that complete networks, characterized by closely connected nodes, promote adaptation but weaken adaptability. In contrast, coalitional networks, featuring loosely connected subgroups, enhance adaptability but hurt adaptation. The study contributes to regional economic resilience literature by empirically demonstrating the trade-off relationship between adaptation and adaptability from network structures’ perspective. Practically, the study helps policy makers to effectively allocate stimuli, such as grants and tax incentives, to develop targeted co-patenting networks to improve adaptation or adaptability.

Keywords

innovation social network analysis economic resilience adaptation adaptability

Regions can be viewed as networks of connected individuals, firms, institutions, and industries (Balland et al., 2015; Tóth et al., 2022). Agents in the network interact with each other and the environment and rebuild the connections to adapt to external change (Bristow & Healy, 2014). Different network structures indicate trade-off relationships in coordinated capabilities and levels of openness (Chapple, 2002). For example, a network where nodes connect with each other increases access to diverse information and promotes learning but lacks efficiency in coordinating resources and diffusing information. In comparison, a network where a hub dominates the whole network enhances efficiency but lacks diversity (Therrien et al., 2019). Meanwhile, evolutionary economic geography literature suggests that regional economic resilience has two dimensions: adaptation and adaptability (Boschma, 2015). Adaptation describes the capability of an economic system to resist external shock and to “bounce back” to the pre-shock development path (Boschma). In contrast, adaptability refers to a region's ability to adapt to long-term structural change and develop new growth paths (Boschma). Notably, other scholars have described these two concepts differently. For example, Martin and Sunley (2015) used “engineering resilience,” instead of adaptation, to describe the economic system's ability to return to its pre-shock equilibrium. Meanwhile, they used “adaptive resilience,” instead of adaptability, to describe structural and functional change. What's tricky is that these two types of resilience hardly coexist (Pike et al., 2010). When a region is efficient in returning to the pre-shock path, it is likely reluctant to develop a novel growth path, and vice versa (Martin, 2012).

Existing quantitative empirical studies examining the relationship between economic network structure and economic resilience primarily focus on adaptation (Kurth et al., 2020; Suire & Vicente, 2014; Ye & Qian, 2021), with a few exceptions examining adaptability (Statsenko, 2017). However, even Statsenko doesn’t consider the trade-off relationship between adaptation and adaptability (Pike et al., 2010). As a result, our understanding of the relationship between adaptation and adaptability is limited. Furthermore, most economic resilience studies focus on European regions (Pontarollo & Serpieri, 2020; Tóth et al., 2022), but the United States is heterogeneous in economic structures and performance.

This paper fills in the gap by being the first to connect the trade-off relationship between bonding and bridging network structures in social capital literature with the discussion of adaptation and adaptability in economic resilience literature. Meanwhile, the study is the first to empirically quantify the magnitudes of network structures’ influence on adaptation and adaptability, using co-patenting networks in the U.S. metropolitan areas as a contextual example.

Patent is a widely used proxy for innovation, and we use co-patenting networks as proxies for innovation-collaboration networks. The structures of innovation-collaboration networks emerge from the micro interactions of inventors. The different structure indicates different patterns of exchanging resources, knowledge, and capital, and thus influences the macroregional resilience capability (Berrou & Combarnous, 2012; Vicente, 2020). For example, adaptation benefits more from efficiency to bounce back within the old growth path, while adaptability needs diversity to bring new ideas and knowledge to develop new growth paths. Different network structures reflect different combinations of diversity and efficiency, thus differently affecting adaptation and adaptability.

We focus on two innovative industries, NAICS 334: Computer and Electronic Product Manufacturing (CEPM) and NAICS 335: Electrical Equipment, Appliance, and Component Manufacturing (EACM). Our study entails the collection of patent inventors’ data from the U.S. Patent and Trademark Office (USPTO) to construct co-patenting networks within these two industries across all U.S. metropolitan and micropolitan areas on a quarterly basis from 1990 to 2012. The network structures are categorized into four types: complete, factional, coalitional, and bridging networks (Crowe, 2007). Additionally, we gather employment data per quarter at the four-digit industry-code level from the U.S. Bureau of Labor Statistics (BLS) to develop measurements for adaptation and adaptability. Employing a seemingly unrelated difference-in-differences model, our analysis reveals that the complete network contributes most significantly to adaptation, while the coalitional network provides the greatest benefits for adaptability.

This paper pertains to both researchers and practitioners. Researchers can better understand how network structure nurtures adaptation and adaptability, respectively. The study contributes to the discussion in economic resilience literature of trade-off relationships between adaptation and adaptability (Hu & Hassink, 2017) from the network structure's perspective. Economic development practitioners can better understand the advantages and drawbacks of different innovation-collaboration network structures to purposely promote adaptation or adaptability in different regions and development stages. For example, the California Office of the Small Business Advocate initiated the CA Inclusive Hub Program (CA Hub) to foster collaboration across different agents, including private firms, academia, nonprofits, and governments (California Office of the Small Business Advocate, n.d.). The program focused on technology and science-based firms and offered technical assistance and mentorship, connecting research institutes, venture capital groups, and incubators with inventors and start-ups. San Francisco uses the CA Hub program to serve local inventors and boost the economy. The Humanmade CA Accelerate California Hub in San Francisco provides inventors with creative space, professional equipment, software, and skilled staff to offer advice. Also, frequent training and workshops are available for creative individuals to enhance skills and develop networking (Humanmade, n.d.). Other local administrators can learn from San Francisco's experience, utilizing state or federal programs and grants to create shared incubator space, organize networking events and workshops, and offer technical training programs to encourage inventors living in marginalized cities to start collaboration. By strategically cultivating a complete innovation-collaboration network, policy makers can increase regional adaptation. These efforts help regions purposely reshape network structure and enhance adaptation or adaptability to combat a variety of external shocks.

Literature Review

Prior studies have confirmed the role of network structure in facilitating regional economic resilience with mixed results (Crespo et al., 2014) but failed to recognize the dual dimension of economic resilience. In this paper, we use Crowe's (2007) network framework, which classifies four network typologies based on a scale from bridging to bonding. We hypothesize that each network structure corresponds to a particular combination of adaptation and adaptability. The next section gives a detailed discussion of network structures’ conceptualization.

The Typology of Four Network Structures

Bonding and bridging are two types of network structures widely discussed in the social capital literature (Agger & Jensen, 2015; Cofré-Bravo et al., 2019; Hoyman et al., 2016). Bridging networks involve sparse connections and weak ties and are open to including people across diverse social groups, but they are less reliable and less cohesive (Andrew & Carr, 2013; Callois & Aubert, 2007; Patulny & Svendsen, 2007; Poortinga, 2012; Putnam, 2000). In contrast, bonding networks embody dense, strong ties among homogeneous groups and are exclusive, but can increase intercommunity segmentation (Granovetter, 1973; Kim, 2018; Patulny & Svendsen; Putnam).

Crowe (2007) divided bonding and bridging network structures into four types (Figure 1). The bonding network structures include complete and factional typologies. The complete network refers to a bonding structure in which each agent in the network is connected to all the other agents, which is a “global” bonding. Such a network is extremely cohesive. The factional network includes two or more bonding subgroups that are not connected to each other, which is a “local” bonding. Such structure intensifies intercommunity segregation. Meanwhile, the bridging network structures include coalitional and bridging configurations. The coalitional network includes loose connections linking densely bonded subgroups. Such a structure could facilitate resource exchange across different subgroups and maintain the power to organize collective activities. The bridging network involves loosely connected ties among agents. Such a structure has access to diverse resources but lacks efficiency due to the longer distance of exchanging resources and information.

Figure 1.
Four-network structure typology. Source: Crowe (2007).

Different networks contribute to resilience through distinctive channels: a bridging network cultivates resilience by increasing access to diverse resources, while a bonding network promotes resilience by building trust and facilitating collective actions within the community (Newman & Dale, 2005). However, Newman and Dale only theoretically discussed the advantage of different network structures and qualitatively assessed it, but the framework has not been empirically tested. Our paper moves one step further by empirically examining whether different network structures contribute to different dimensions of regional economic resilience.

The Dual Nature of Adaptation and Adaptability

Evolutionary economic geography literature classifies regional economic resilience into two dimensions: adaptation and adaptability (Boschma, 2015; Bristow & Healy, 2018; Pike et al., 2010). Adaptation describes a system's ability to accommodate shocks in the short term and within an established development path (Pike et al.). Adaptability focuses on a system's long-term capability to reconstruct the industrial structure and develop a new growth path (Boschma, 2015; Martin, 2012).

Adaptation and adaptability are trade-offs with each another (Martin & Sunley, 2006). When a region is open and has access to diverse resources, it has high adaptability to transform its economic structure and create novel development trajectories detached from the past (Boschma, 2015; Martin, 2012). Whereas a specialized and homogeneous regional economic system is more likely to bounce back in the established development path along with its history (Hassink, 2010; Martin & Sunley, 2006; Pike et al., 2010). However, recent studies have indicated that adaptability and adaptation can move beyond a trade-off relationship and arrive at a reciprocal condition, both theoretically and empirically (Hu & Hassink, 2017, 2020). Hu and Hassink (2020) argued that high adaptability is a long-term outcome of high adaptation when agents purposely adapt to environments. One example is the Zaozhuang City in Hu and Hassink's (2017) case study. Zaozhuang is a coal city that struggled with overreliance on the coal economy. The mayor strategically collaborated with local stakeholders to cultivate a cultural tourism industry unrelated to the coal industry. They mobilized local resources, legacies, and networks to diversify the local economy and transform Zaozhuang into an attractive tourist city. Meanwhile, tourism improved the image of the region, from which the coal industry benefitted from increasing investment in the region. These new investments facilitated the local coal firms to develop coal-related businesses, such as boosting coal-related chemical products. Hence, Zaozhuang City embraces high adaptation and adaptability. The case study indicates a win-win solution between adaptation and adaptability. Our study tests whether a region can achieve high adaptation and adaptability from a network structure's perspective.

Hypotheses on the Relationships Between Adaptation, Adaptability, and Network Structures

Regional economic resilience depends on technological and economic network structures (Vicente, 2020). However, no studies have examined how different network structures affect adaptation and adaptability simultaneously. Focusing merely on adaptation risks ignoring the cost of weakening adaptability in the pursuit of adaptation, due to their inherent, often trade-off relationship. Balancing this trade-off relationship between adaptation and adaptability is important for achieving overall economic resilience. Hence, examining adaptation and adaptability concurrently is needed.

Theoretically, bonding and bridging network structures improve resilience in distinctive ways (Newman & Dale, 2005). Adaptation and adaptability have a competitive relationship and need different resources. Empirically, network structure's influence on adaptation and adaptability is found to be mixed. For example, Tóth et al. (2022) found that a bonding network structure that was less likely to collapse into segments could increase adaptation. Relatedly, Kahl and Hundt (2015) examined whether firms’ interorganizational network influenced regional adaptation. They focused on the German biotechnology industry and used the employment growth rate as a proxy for adaptation. They found that when firms’ network involved more diversified organization categories (e.g., research organization, government institutes, biomedical companies), the region exhibited stronger adaptation. Ye and Qian (2021) showed that enhancing the connections between core and peripheral cities (i.e., a bridging structure) improved regional resilience by increasing real gross domestic product (GDP), but they failed to separate adaptation and adaptability. Statsenko (2017) examined the mining equipment and technology services (METS) firms in South Australia and used the innovativeness and diversification of local suppliers as a proxy for adaptability. He found that a high-degree centrality improved adaptability. However, the study used cross-sectional supply network data and couldn’t capture the network's temporal changes and their influence on resilience. Moreover, the study failed to consider the trade-off relationship between adaptation and adaptability. In sum, our knowledge of how different network structures simultaneously affect adaptation and adaptability is still quite limited.

This paper fills in the gaps. Figure 2 summarizes the theoretical hypotheses of how different network structures influence adaptation and adaptability, respectively. We use Crowe's (2007) framework of network structure and develop four hypotheses of relationships between network structures and adaptation and adaptability.

Figure 2.
Theoretical relationships between network structures and adaptation and adaptability.

Hypothesis 1: A cohesive co-patenting network (i.e., complete network) leads to high adaptation and low adaptability. A complete network involves strong and dense within-group ties that facilitate exclusive cohesion, closure, and in-group collective action (Granovetter, 1973; Kim, 2018; Patulny & Svendsen, 2007). The complete structure favors lock-in and path-dependent specialization but weakens access to external resources and the development of novel growth paths (Boschma, 2015; Crespo et al., 2014). Thus, a complete network structure leads to high adaptation and low adaptability.

Hypothesis 2: A co-patenting network with segregated clusters (i.e., factional network) leads to low adaptation and low adaptability. A factional network comprises multiple interconnected subgroups, each of which is not linked to the others. A factional network fails to exchange resources and information to develop new growth paths and lacks the capability to integrate resources to recover along the original trajectory. Thus, it results in low adaptation and low adaptability.

Hypothesis 3: A co-patenting network with noncomplete connections of closely connected clusters (i.e., coalitional network) leads to high adaptation and high adaptability. A coalitional network refers to a noncomplete connection among densely bonded subgroups (Crowe, 2007). Such structure facilitates specialization formation within each individual bonded group and promotes intergroup learning and knowledge transfer to diversify the region's economic base. Thus, a coalitional network leads to high adaptation and high adaptability.

Hypothesis 4: A loosely connected co-patenting network (i.e., bridging network) leads to low adaptation and high adaptability. A bridging network is characterized by less density and weak ties among different groups of agents. Such characteristics contribute to high adaptability by facilitating the exchange of diverse information, knowledge, and resources from heterogeneous groups and promoting learning (Aral & Van Alstyne, 2011; Carr et al., 2017; Putnam, 2000). Meanwhile, loosely connected ties indicate less efficiency in integrating resources along the original growth path and are thus associated with low adaptation.

Data

Our spatial analysis unit is core based statistical areas (CBSAs) downloaded from the U.S. Department of Transportation (U.S. Department of Transportation, 2022). CBSAs encompass metropolitan and micropolitan statistical areas. Metropolitan statistical areas consist of urbanized areas with populations of at least 50,000, while micropolitan statistical areas include urban clusters with populations between 10,000 and 50,000 (U.S. Census Bureau, 2021). The Office of Management and Budget defined the CBSA boundaries based on the 2010 Census. The data set was updated on August 09, 2019, from the U.S. Census Bureau and is part of the U.S. Department of Transportation /Bureau of Transportation Statistics National Transportation Atlas Database (U.S. Department of Transportation, 2022).

Our data are collected from three main resources. Innovation-collaboration network structures are constructed using patent inventor data. We collect patent inventors’ data from USPTO, which includes the inventor's name, location, and patent ID that is created by the inventors.

We focus on two innovative industries: NAICS 334 (Computer and Electronic Product Manufacturing, CEPM) and NAICS 335 (Electrical Equipment, Appliance, and Component Manufacturing, EACM). CEPM includes “establishments that manufacture computers, computer peripherals, communications equipment, and similar electronic products, and establishments that manufacture components for such products” (U.S. Bureau of Labor Statistics, 2022a, website). EACM is establishment “producing lighting fixtures, electric light bulbs, and parts (except glass parts and light emitting diodes, and other lighting equipment” (U.S. Bureau of Labor Statistics, 2022b, website). CEPM and EACM are selected because they are two of the most innovative industries based on patent intensity shown in the USPTO Intellectual Property and the U.S. Economy (third edition; United States Patent and Trademark Office, 2016). We prioritize industries where innovations are active, and the number of granted patents is large, so a co-patenting network can be influential in economic resilience.

The study only includes patents that have been eventually granted. Granting patents is a long process. According to the USPTO, the average time from filling a patent application to getting a first USPTO action takes about 20 months (U.S. Patent and Trademark Office, n.d.). To avoid the unnecessary delay caused by patent processing, we use the patent application time for study. Our hypothesis is that by the time that inventors apply for the patent, the innovation is already completed, and the techniques can be diffused through collaborators’ networks.

To match patents with specific industries, we use the crosswalk between the U.S. Patent Classification (USPC) and NAICS to assign patents to specific industries (Patent Technology Monitoring Team, 2023). USPC is the official patent classification system used and maintained by USPTO before 2013 (Technology & Patent Research International, 2015). As a result, our study time is limited to 1990 until the end of year 2012. We include 7.03 million CEPM and 3.38 million EACM patent inventors’ data within the study period. We use the city-level location information to construct the inventors’ co-patenting networks in all CBSAs quarterly in the United States. In our network, nodes are cities where patent inventors are located, and ties between cities exist when inventors living in two cities are co-patented. The strength of ties is measured by the frequency of co-patenting, and it isn’t affected by the physical distances between cities.

However, the patent data have limitations. Not all innovations would be transformed into patents because maintaining a patent is costly. When an innovation doesn’t generate profits, a firm is unlikely to choose to apply for a patent. Hence, patent data are not comprehensive in measuring innovation. Nevertheless, the patent is still the most widely accepted indicator of innovation (Bottazzi & Peri, 2003; Bowen et al., 2013; Fang, 2015).

Meanwhile, we collect employment data quarterly from 1990 to 2017 at the four-digit NAICS code level in CEPM and EACM from the BLS to calculate a CBSA's adaptation and adaptability. The study spans from 1990 to 2012, with employment data from 2013 to 2017 being used to establish lagged measurements. For example, in terms of measuring 5-year adaptation, we utilize the patent data in 2012 to construct networks and use employment data in 2017 to calculate the measurement for 5-year adaptation.

Regarding external shocks, we use all the influential natural hazards to examine a region's adaptation and adaptability. The data are from the Spatial Hazard Events and Losses Database for the United States. This data set covers records of 18 kinds of natural hazards from 1960 to 2021. It includes information on the date of the hazard, the affected county, and the direct losses (e.g., property and crop losses). We focus on hazards that cause more than $100 million in losses measured by the sum of property and crop damages per incident to focus on shocks that have a significant influence. All the losses are measured using U.S. dollars in the year 2020. After matching with patent data, our analysis includes 186 (158) hazard records covering 87 (75) metropolitan areas between 1990 and 2012 in CEPM (EACM). Meanwhile, we include 257 (261) metropolitan areas and 2 (2) micropolitan areas that haven’t suffered from natural hazards during the study period in CEPM (EACM) as control regions for comparison. The spatial distributions of CBSAs are shown in Figure 3.

Figure 3.
Spatial distribution of CBSAs affected by identified natural hazards versus nonaffected CBSAs.

Method

Measurement for Network Structures

We adopt social networks analysis (SNA) to quantify the co-patenting network structures of CEPM and EACM for CBSAs in the United States from 1990 to 2012, on a quarterly basis. SNA provides a conceptual framework and a set of analytical tools to visualize relational data and analyze social network structures (Dempwolf & Lyles, 2012; Reid et al., 2008). It has been applied to study knowledge flows within and across regions (Boschma, 2015). In our case, we constructed the co-patenting networks and used centrality measurements to quantify the network structure. Such co-patenting networks reflect the collaborative relationship among cities within CBSAs.

The centrality measurements in Table 1 operationalize Crowe's (2007) theoretical framework. “Betweenness centrality” measures how often a node lies on the shortest path between other nodes in the network (Freeman, 1977; Knoke & Yang, 2008). In our context, it measures how frequently a city acts as a shortest-distance bridge to connect other cities to form a patent collaboration. When a city acts as a bridge to connect the other cities, it creates paths and promotes diverse resource exchange. At the same time, it indicates the network is more fragile due to the reliance on the bridging cities. We used the standard deviation of betweenness centrality to measure the magnitude of uneven distribution of betweenness centrality across nodes. Bridging and coalitional network structures are networks with a high standard deviation of betweenness centrality, which is characterized with less cohesion and more diversity. Our study defines CBSAs with standard deviation of betweenness centralities that fall into the top 50% of the distribution as high and the remaining 50% as low. The formula is as follows:
$C_{b e t w e e n n e s s} (i) = [\frac{\sum_{s, t \in V} p (s, t | i)}{p (s, t)}] / (N - 1) (N - 2)$
where $C_{b e t w e e n n e s s} (i)$ denotes the betweenness centrality of city i; $p (s, t | i)$ denote the number of shortest paths between cities s and t that goes through cities i. $p (s, t)$ denote the total number of shortest paths between s and t. N denotes the total number of cities in the network.

Table 1.
Network Typology and Measurement.

Network Complete network Factional network Coalitional network Bridging network

Measurement

Standard deviation of betweenness centrality low low high high

Mean of closeness centrality high low high low

“Closeness centrality” measures how close each node connects with other nodes by calculating the distance from the node to all other nodes in the network (Andrew & Carr, 2013). In our study, when inventors from one city frequently collaborate with inventors from another city, we regard the collaborative distance between the two cities as shorter. Technically, the direct distance between two cities is defined as the inverse of co-patenting frequency. More frequently collaborated cities have a shorter distance. If two cities are not directly connected but use other cities as bridges to connect, the distance between them is the sum of the accumulated distance along the shortest path from the origin city to the destination city. Shortest path refers to the most efficient route from one city to another.

If the network includes more than one disconnected subgroup, the closeness centrality of a single city in the whole network equals the proportion of cities within the CBSAs that can be reached, relative to the average distance from the reachable cities (Wasserman & Faust, 1994). The formula of closeness centrality is as follows:
$C_{c l o s e n e s s} (i) = \frac{n - 1}{\sum_{j = 1}^{n - 1} l (i, j)} * \frac{n - 1}{N - 1}$
where $C_{c l o s e n e s s} (i)$ denotes the closeness centrality of city i; $l (i, j)$ denotes the length of the shortest path from i to j; n denotes the number of cities in the reachable subgroup including city i. j denotes the city that in the same reachable subgroup as city i. N denotes the number of cities in the whole network.

A complete or coalitional network structure is a network with high average closeness centrality, which is characterized by good connections. Whereas the factional network structure is segregated on average and the bridging network structure is loosely connected, the average distance from one node to other nodes is large and the average closeness centrality is low. Like the measurement of betweenness centrality, we define CBSAs with mean of closeness centrality that fall into the top 50% of the distribution as high and the remaining 50% as low.

Measurement for Adaptation and Adaptability

Regional economic resilience studies develop different measurements for adaptation and adaptability. For example, Tóth et al. (2020) and Muštra et al. (2020) used the employment rate ratio before and after the 2008 economic crisis as the measurement for adaptation. Xiao and Drucker (2013) used the annual growth rate of income per capita and employment to reflect regional adaptation after a flood.

Regarding the measurement of adaptability, Carlsson et al. (2014) used the new business growth rate after the restructuring programs compared with the national average to reflect regional adaptability. Mardaneh et al. (2020) used changes in the income-generating industry classification before and after a shock to measure adaptability. Overall, the measurement for adaptation and adaptability involves employment, income per capita, business growth rate, and changes of industry. Nevertheless, employment data are the most widely used data source (Di Caro, 2020; Pontarollo & Serpieri, 2020). Our crosswalk between USPC and NAICS is at the three-digit NAICS-code level, so we can know which patent belongs to which three-digit NAICS code. To examine how the patent network affects adaptation and adaptability within the three-digit NAICS industries, we used the four-digit NAICS code to construct measurements and distinguish the growth path that is along with the original growth path or is detached from the original one.

“Adaptation” reflects an economic system's ability to resist external shock and recover to the preshock condition. First, we identified local pillar industries using location quotients more than 1.2 (Garnick, 1969) at one quarter before the natural hazard shock. Second, we traced the development trajectory of the prenatural-hazard pillar industry (PI) by collecting their location quotients after the natural hazard from BLS. Third, we calculated the average location quotient of the pillar industries one quarter before the natural hazard and q quarter after, and used the average change rate to reflect adaptation at time t + q. A large value indicates that the prenatural-hazard pillar industries bounce back effectively, thus indicating a high adaptation. The formula is as follows:
$a d a p t a t i o n_{i, t + q} = (M e a n P I_{i, t + q}) / (M e a n P I_{i, t - 1})$
where $a d a p t a t i o n_{i, t + q}$ denotes the adaptation in the CBSA i, q quarters after the disaster of time t. $M e a n P I_{i, t + q}$ denotes the average value of the location quotient of prenatural-hazard pillar industries in CBSA i; time t + q, where q is a set of constants equal to 1, 4,12, and 20, representing one quarter, 1 year, 3 years, and 5 years after the natural hazard. $M e a n P I_{i, t - 1}$ denotes the average location quotient of pillar industries in the CBSA i, one quarter before the natural hazard.

“Adaptability” reflects an economic system's ability to depart from the existing industrial structure and develop a new growth path. We used the percentage of new pillar industries after the natural hazards to reflect adaptability. We calculated what is the percentage of new pillar industries after the natural hazards compared with that in the prenatural hazard to define adaptability. A higher value indicates greater adaptability, signifying the emergence of more new pillar industries following natural hazards. The formula is as follows:
$a d a p t a b i l i t y_{i, t + q} = (N e w P I_{i, t + q}) / (A l l P I_{i, t - 1})$
where $a d a p t a b i l i t y_{i, t + q}$ describes the percentage of new pillar industries compared with those in the prenatural-hazards period. $N e w P I_{i, t + q}$ denotes the number of newly appeared pillar industries in the CBSA i, time t + q. $A l l P I_{i, t - 1}$ denotes the number of all pillar industries in the CBSA i, time t-1.

Measurement for Damages in Natural Hazards

Studies indicate that large natural hazards disrupt economic activities and generate long-term negative economic influence. For example, Hsiang and Jina (2014) adopted tropical cyclones from 1950 to 2008 to identify the influence of environmental hazards on a country's long-term economy. They found that national income declined 20 years after the hazards. Berlemann and Wenzel (2015) examined the relationships between natural hazards and economic growth using drought data from 1960 to 2022 and found that droughts created long-term negative effects on a country's growth. Huang et al. (2024) found that earthquakes significantly decreased GDP per capita in the long run in China.

Given these pieces of empirical evidence, we used accumulated damage to measure the hazards’ influence on regions. Specifically, to reduce skewness, the study includes the natural logarithm of how many million 2020 dollars loss a CBSA i had from 1990 to time t in the regression model.

Seemingly Unrelated Difference-in-Differences Regression

We examined adaptation and adaptability simultaneously. We used seemingly unrelated regression combined with difference-in-differences regression to approach causal inference. The system of regression is as follows:
$a d a p t a t i o n_{i, t + q} = α_{0} + α_{1} n e t w o r k_{i, t - 1} + α_{2} h a z a r d_{i, t} + γ^{D D} h a z a r d_{i, t} * n e t w o r k_{i, t - 1} + α_{3} Y e a r Q r t_{t + q} + α_{4} M S A_{i} + ε 1_{i, t + q}$
(1)
$a d a p t a b i l i t y_{i, t + q} = β_{0} + β_{1} n e t w o r k_{i, t - 1} + β_{2} h a z a r d_{i, t} + δ^{D D} h a z a r d_{i, t} * n e t w o r k_{i, t - 1} + β_{3} Y e a r Q r t_{t + q} + β_{4} M S A_{i} + ε 2_{i, t + q}$
(2)
where $a d a p t a t i o n_{i, t + q}$ and $a d a p t a b i l i t y_{i, t + q}$ reflect the two dimensions of regional economic resilience in either CEPM or EACM in CBSA i, q quarter after hazard at time t. $n e t w o r k_{i, t - 1}$ denotes the three dummy variables representing different types of networks (i.e., factional network, coalitional network, and bridging network), and we used complete network structure as a baseline. $h a z a r d_{i, t}$ reflects the natural logarithm of how many million 2020 dollars loss at CBSA i from 1990 to time t due to natural hazards. $Y e a r Q r t_{t + q}$ and $M S A_{i}$ reflect year-quarter fixed effects and metropolitan-area fixed effects. $ε 1_{i, t + q}$ and $ε 2_{i, t + q}$ are two correlated error terms. We considered the covariance structure of the error terms and ran the two equations simultaneously. We assumed adaptation and adaptability were correlated through the two error terms. $h a z a r d_{i, t} * n e t w o r k_{i, t - 1}$ are the focused variables and $γ^{D D}$ and $δ^{D D}$ are the focused coefficients, which reflect the moderating effect of network structures between natural hazards and adaptation or adaptability, respectively.

The model has limitations. The network structure may evolve toward a more resilient direction as a response to the hazards. The current results can identify which network structure is more helpful to adaptation or adaptability but cannot tease out the possibility that high resilience is due to the evolution of network structure in response to natural hazards. Second, the model focuses on the co-patenting network across cities within CBSAs but overlooks the firm's internal networks. Larger firms own their internal network, such as the connections between headquarters and the branches, which might affect the regional economic resilience. Third, we used the accumulated damage as a proxy for the long-term influence of hazards, which can be inequivalent to the real long-term influence. However, studies indicate that the precise long-term influence of hazards is nonlinear and very complicated to quantify (Botzen et al., 2019). In this case, we used the accumulated damage as a feasible proxy for the long-term influence of hazards and added sensitivity analysis, making the influence of each natural hazard depreciate in 10 years. Fourth, industry age varies across regions. Our model doesn’t control industry age due to data limitations. Fifth, labor-saving innovation can reduce the location quotient of an industry and, therefore, the value of adaptation and adaptability. This paper cannot fully separate labor-saving from nonlabor-saving innovation due to the limitation of the patent data.

Results

Descriptive Analysis

The prerequisite of using difference-in-differences regression is to identify the prenatural-hazards parallel trend. Figure 4 contrasts quarter-to-quarter average prenatural-hazards adaptation and adaptability in identified natural-hazards affected areas and unaffected areas during the study period. It focuses on one-quarter adaptation and adaptability, and the figures of 1-, 3-, and 5-year adaptation and adaptability are shown in the supplementary online appendix.

Figure 4.
The average trend of one-quarter adaptation and adaptability in the affected areas versus unaffected areas.

Before natural hazards occur, unaffected areas exhibit higher adaptation in CEPM but a lower adaptation in EACM. They display a consistent parallel trend over time. During the 1990s, both affected and unaffected areas saw a slight increase in their trend. Around 2000, both groups experienced a sharp decline. After 2002, the two groups’ adaptation stabilized at approximately half of their 1990s values. Regarding adaptability, affected and unaffected areas closely follow each other over time. In summary, the two groups share a general parallel trend in their average adaptation and adaptability during the period before natural hazards occur.

The Network Impact on Adaptation

Table 2 shows the estimation results for Equation (1). Columns 1 to 4 describe the moderating effect of network structure between natural hazards and adaptation in CEPM, and columns 5 to 8 focus on EACM. All the results use the complete network as a baseline. Compared with the complete network, the factional network shows a 1.5% decrease in the average location quotients of prenatural-hazards pillar industries one quarter after the occurrence of natural hazards in CEPM, corresponding to a 1% increase in accumulated damage. The result shows that the original pillar industries are less concentrated after natural hazards compared with the prehazard period when the network structure transforms from a complete network to a factional network. For example, the affected CBSAs have 4,850 jobs on average in CEPM pillar industries in one quarter during the study period. A 0.015-point decline in one-quarter adaptation signifies that, keeping the overall regional employment as a constant, the pillar industries shrank by 73 jobs one quarter after the natural hazard if the network structure changes from complete network to factional network.

Table 2.
The Moderating Effect of Network Structure on Adaptation in CEPM and EACM.

Dependent variable: adaptation

Computer and electronic product manufacturing (CEPM) Electrical equipment, appliance, and component manufacturing (EACM)

One quarter 1 year 3 year 5 year One quarter 1 year 3 year 5 year

(1) (2) (3) (4) (5) (6) (7) (8)

Million population 0.0892* 0.0893* 0.0784* 0.0749* 0.0446* 0.0561* 0.0684* 0.113*

(0.0126) (0.0128) (0.0129) (0.0130) (0.0125) (0.0126) (0.0128) (0.0131)

Factional network 0.0657 0.0375 −0.0174 −0.110* 0.0748 0.0954* 0.0673** 0.0274

(0.0245) (0.0250) (0.0252) (0.0253) (0.0250) (0.0253) (0.0256) (0.0261)

Coalitional network 0.0206 0.0151 0.0119 0.00492 0.0800* 0.118* 0.135* 0.100

(0.0256) (0.0261) (0.0263) (0.0264) (0.0322) (0.0326) (0.0330) (0.0336)

Bridging network 0.110* 0.0752* −0.0102 −0.107* 0.411* 0.425* 0.397* 0.321***

(0.0306) (0.0313) (0.0314) (0.0316) (0.0386) (0.0391) (0.0396) (0.0403)

Log accumulated damage 0.00702* 0.00838 0.0110* 0.00837 0.00968 0.0133* 0.00988 0.00732

(0.0029) (0.0030) (0.0030) (0.0030) (0.0036) (0.0037) (0.0037) (0.0038)

Log Damagefactional −0.0148 −0.0149* −0.0150* −0.00752 −0.0108 −0.0145*** −0.00852* −0.0125**

(0.0037) (0.0038) (0.0038) (0.0039) (0.0042) (0.0042) (0.0043) (0.0044)

Log damage* coalitional −0.0042 −0.0061 −0.00904** −0.00800* −0.0140* −0.0177* −0.0155*** −0.00936*

(0.0034) (0.0034) (0.0035) (0.0035) (0.0039) (0.0039) (0.0040) (0.0041)

Log damage* bridging −0.0064 −0.00571 0.00118 0.00384 −0.0438* −0.0443* −0.0366* −0.0269*

(0.0051) (0.0052) (0.0052) (0.0053) (0.0064) (0.0065) (0.0066) (0.0067)

Constant 0.00521 0.0187 0.0723 0.115* −0.0728 −0.0632 −0.0185 0.0339

(0.0506) (0.0516) (0.0519) (0.0522) (0.0509) (0.0515) (0.0522) (0.0532)

CBSA fixed effect Yes Yes Yes Yes Yes Yes Yes Yes

Year-Quarter fixed effect Yes Yes Yes Yes Yes Yes Yes Yes

Number of observations 21801 21801 21801 21801 19008 19008 19008 19008

R-squared 0.655 0.633 0.615 0.596 0.651 0.633 0.616 0.598

Notes: Standard errors in parentheses. , , denotes statistical significance at the 0.1%, 1%, and 5% levels, respectively.

The differentiating influence of factional and complete networks in EACM is similar, with a 0.011 decline in a factional network. The largest gap between a complete network and a factional network's moderating effect occurs 3 years after natural hazards in CEPM, while in EACM, this effect reaches its peak at 1 year. Overall, the factional network has a smaller adaptation after natural hazards across all study periods, compared to the complete network in both CEPM and EACM. The results verify our hypothesis that a factional network has a low adaptation among the four network typologies.

The coalitional network presents a smaller adaptation than the complete network. Consistent with the factional network, the largest gap between the complete network and the coalitional network occurs at 3 years after hazards in CEPM, while, in in EACM, 1 year after natural hazards witnesses the most significant difference between the coalitional and the complete network. The results reject our hypothesis that coalitional network has a high adaptation among the four networks.

A bridging network has a statistically insignificant smaller adaptation than a complete network after natural hazards in CEPM, whereas the difference is strong and significant in EACM. Consistent with factional and coalitional networks, EACM has the largest difference between networks’ moderating effect at 1 year after the natural hazards. The results verify our hypothesis that a bridging network has low adaptation among the four network structures.

Despite the heterogeneous influence of network structure across industries and time, we can see that a complete network has the strongest adaptation. Our hypotheses about the impact of complete, factional, and bridging network structures on adaptation are verified, whereas the hypothesis that a coalitional network has a high adaptation is rejected.

The Network Impact on Adaptability

Table 3 shows the estimated results for Equation (2). The complete network witnesses a low adaptability after natural hazards, compared with the other three network structures. CEPM and EACM with complete networks experience the lowest adaptability 3 years after natural hazards, whereas EACM has the lowest adaptability 5 years after natural hazards. When accumulated damage increases by 1%, the 3-year adaptability of complete network decreases by 0.007 in CEPM. It suggests that the likelihood of developing new pillar industries compared with pre-natural-hazard time is reduced by 0.7%.

Table 3.
The Moderating Effect of Network Structure on Adaptability in CEPM and EACM.

Dependent variable: adaptability

Computer and electronic product manufacturing (CEPM) Electrical equipment, appliance, and component manufacturing (EACM)

One quarter 1 year 3 year 5 year One quarter 1 year 3 year 5 year

(1) (2) (3) (4) (5) (6) (7) (8)

Million population 0.0330* 0.0447* 0.0362* −0.0155 −0.0136 −0.0215 0.018 0.0325*

(0.0091) (0.0119) (0.0143) (0.0154) (0.0087) (0.0111) (0.0133) (0.0144)

Factional network −0.0161 −0.0067 −0.0196 −0.0234 −0.0302 −0.0493* −0.0482 −0.0208

(0.0178) (0.0232) (0.0279) (0.0301) (0.0175) (0.0223) (0.0266) (0.0287)

Coalitional network −0.0388* −0.0389 −0.0292 0.004 −0.041 −0.0371 −0.0843* −0.0657

(0.0186) (0.0242) (0.0292) (0.0314) (0.0225) (0.0286) (0.0343) (0.0370)

Bridging network −0.0376 −0.048 −0.0473 −0.0287 0.00729 0.021 −0.0146 0.00613

(0.0223) (0.0290) (0.0349) (0.0376) (0.0270) (0.0343) (0.0411) (0.0443)

Log accumulated damage −0.00315 −0.005 −0.00718* −0.00147 −0.00844* −0.0142* −0.0167* −0.0168*

(0.0021) (0.0027) (0.0033) (0.0035) (0.0025) (0.0032) (0.0039) (0.0042)

Log damagefactional 0.000476 0.000399 −0.000545 −0.00341 0.00584 0.00858* 0.0113* 0.0194***

(0.0027) (0.0035) (0.0043) (0.0046) (0.0029) (0.0037) (0.0044) (0.0048)

Log damage* coalitional 0.00459 0.00959 0.0178* 0.0199* 0.00828 0.0143* 0.0208* 0.0222***

(0.0025) (0.0032) (0.0038) (0.0041) (0.0027) (0.0035) (0.0041) (0.0045)

Log damage* bridging 0.000736 0.00497 0.0156 0.0199 0.00241 0.00627 0.0109 0.0121

(0.0037) (0.0048) (0.0058) (0.0062) (0.0045) (0.0057) (0.0068) (0.0074)

Constant −0.00992 −0.0468 −0.0467 −0.0638 −0.0458 −0.0501 −0.0686 −0.0695

(0.0368) (0.0478) (0.0576) (0.0620) (0.0356) (0.0453) (0.0542) (0.0585)

CBSA fixed effect Yes Yes Yes Yes Yes Yes Yes Yes

Year-Quarter fixed effect Yes Yes Yes Yes Yes Yes Yes Yes

Number of observations 21801 21801 21801 21801 19008 19008 19008 19008

R-squared 0.655 0.633 0.615 0.596 0.651 0.633 0.616 0.598

Notes: Standard errors in parentheses. , , denotes statistical significance at the 0.1%, 1%, and 5% levels, respectively.

Compared with the complete network, factional network structure is positively associated with adaptability after natural hazards in EACM, with 5-year adaptability experiencing the largest positive impact. Specifically, 5-year adaptability after natural hazards has fully recovered in areas with factional networks. In comparison, the factional network has negligible influence on adaptability in CEPM. The results reject our hypothesis that the factional network exhibits low adaptability.

A coalitional network significantly promotes adaptability in both industries and across different periods in comparison to the complete network. The most significant influence occurs 5 years after natural hazards in both industries. Specifically, every 1% increase in million-dollar damage is associated with an additional 2% increase in 5-year adaptability in both CEPM and EACM, compared with a complete network. It indicates that network structure has a long-term influence on reshaping regional industries. It is sensible that developing a new growth path takes time, and network structures’ moderating effects between natural hazards and adaptability peak 5 years after the natural hazards. The results verify our hypothesis that the coalitional network sees high adaptability within the four network structures.

A bridging network witnesses larger adaptability than a complete network. In CEPM, the difference between bridging and complete networks is statistically significant 3 and 5 years after natural hazards, whereas, in EACM, the differentiating influence between bridging and complete networks is positive but statistically insignificant. Five years after natural hazards witnesses a 2% increase in adaptability in CEPM, compared with the complete network. The positive moderating influence of a bridging network is as strong as that of a coalitional network, which verifies our hypothesis that a bridging network presents high adaptability. Overall, our hypotheses that coalitional and bridging networks see high adaptability and factional and complete networks exhibit low adaptability are verified.

Sensitivity Analyses

To examine whether the results are robust, we conducted the following sensitivity analyses. First, we focused on CBSAs that only experienced one natural hazard and those that hadn’t experienced any natural hazards. Our study period is from 1994 to 2008. We restrain to the samples that no natural hazards existed from 1990 to 1993 and from 2009 to 2012. This analysis minimizes the influence of historical natural hazards and focuses on the resilience after a single hazard. The results of CEPM are partially consistent with our primary findings, indicating that complete networks demonstrate superior adaptation and adaptability than factional and coalitional networks (Table A1 in the supplementary online appendix). It might be because complete networks are more advantageous in safe places and places suffering from one-time rare shocks, whereas coalitional networks are superior in CBSAs under multiple shocks. Meanwhile, we find that the coefficient of interaction between hazards and bridging network drops out due to the small number of unique values among the observations.

Second, to account for the large CBSA bias (Wassmer, 2021), we adopted the natural logarithm of damage per capita to account for the population effect. The results are consistent with our main results (see Tables A2 and A3 in the supplementary online appendix). That is, a complete network presents a better adaptation than other types of networks, whereas a coalitional network performs the best in adaptability.

Third, we adopted spatial error model (SEM) regression to account for the spatial spillover effect. When a CBSA suffers from natural hazards, the neighboring CBSAs might be affected due to geographically proximate supply chains. SEM assumes that the error term is spatially correlated. We used the inverse of the distance between each pair of CBSAs to construct the spatial variance-covariance matrix to account for the spatial correlation of CBSAs. The regression involves a spatial matrix of 22,000*22,000, which requires a large amount of computing resources. To make the calculation practically feasible, we split the data into 10 subsamples using an interval of a decade, which include: 1) 1990, 2000, 2010; 2) 1991, 2001, 2011; 3) 1992, 2002, 2012; 4) 1993, 2003; 5) 1994, 2004; 6) 1995, 2005; 7) 1996, 2006; 8) 1997, 2007; 9) 1998, 2008; 10) 1999, 2009. We ran 20 SEM regressions for the 10 subsamples of CEPM and EACM industries, respectively. The results are mostly consistent with our main results (see Tables A4 and A5 in the supplementary online appendix). Specifically, the 10 results are all consistent with the main results within the 10 subsamples of EACM. Coalitional and bridging networks perform worse than a complete network in adaptation, whereas a complete network is weaker in adaptability than a coalitional network. Regarding CEPM, 6 out of 10 are consistent with the main results. That is, a complete network presents the strongest adaptation, whereas a coalitional network embraces stronger adaptability. Three out of 10 results are statistically insignificant, and the remaining one (i.e., 1995 and 2005 group) shows a different result.

Fourth, we added a recession dummy variable to reflect whether the state GDP per capita grows or declines. The outcomes are consistent with the main results (Tables A6 and A7 in the supplementary online appendix). A complete network exhibits greater adaptation than the other three types of networks in CEPM and EACM, whereas a coalitional network outperforms a complete network in adaptability.

Fifth, we changed the threshold for identifying high adaptation and high adaptability. Originally, we used the median value to separate adaptation and adaptability into the high and low categories. To check for sensitivity to this specific threshold, we changed the threshold for high adaptation and adaptability to the 75^th percentile and that for low adaptation and adaptability to the 25^th percentile. The results are shown in supplementary online appendix Table A8. A fractional network exhibits lower adaptation than a complete network, while the coefficients for coalitional and bridging networks cannot be reliably estimated due to the small number of observations. Next, we tried the threshold of above 60% as high and less than 40% as low (Table A9 in the supplementary online appendix). We find that the coefficient of interactions between natural hazards’ damage and bridging network drops out due to a small number of observations, whereas the coefficients of interactions between coalitional and factional networks are not statistically significant.

Sixth, hazards occurring immediately prior to 1990—the start of the study period—may bias the results. We excluded areas that suffered from identified natural hazards in 1989; the results are shown in Tables A10 and A11 in the supplementary online appendix. The results are consistent with our main results. A complete network exhibits greater adaptation in CEPM and EACM, and a coalitional network performs better in adaptability.

Seventh, larger firms may enhance regional economic resilience through their own internal network. Though we don’t have data on large firms’ internal networks, we collected employment and establishment numbers from the BLS to control for firm size in CBSAs. We calculated the average employment number per establishment each quarter for each CBSA in CEPM and EACM, respectively. We added average firm size as a control variable in the regression (the results are shown in Tables A12 and A13 in the supplementary online appendix). The results are consistent with our main results. A complete network has better adaptation than the other three types of networks in CEPM and EACM, whereas a coalitional network outperforms a complete network in adaptability.

Eighth, we changed the measurement of adaptation to test sensitivity. We used the average change of location quotients of all industries in CEPM and EACM before and after hazards to measure adaptation. The results are shown in Tables A14 and A15 in the supplementary online appendix. A complete network witnesses better adaptation than factional and coalitional networks, whereas coalitional and bridging networks present better performance than a complete network in adaptability.

Ninth, we filtered CBSAs that own industries with a location quotient of more than 1.2 in at least one quarter during the study period to run regressions. The CEPM results are shown in supplementary online appendix Table A16 and are consistent with our main results.

Additionally, to account for the recovery of previous natural hazards, we changed the measurement of natural hazards by making the influence of previous natural hazards depreciate 10% per year (Table A17 in the supplementary online appendix). The results are consistent with our main results. A complete network exhibits better adaptation than the other three types of networks, whereas the coalitional network shows better adaptability than the complete network.

Conclusion

The innovation-collaboration network structure is an important indicator of the innovation industry's economic resilience. Prior studies fail to consider the two dimensions of economic resilience simultaneously, and primarily focus on European contexts. Our paper addresses the gaps by adopting social network analysis to capture the role of network structure and examine the influence on two dimensions of economic resilience in U.S. CBSAs. Using natural hazards causing more than $100 million in direct property damage per incident as examples, we found that a complete network most promotes adaptation, whereas a coalitional network contributes to adaptability. The results are consistent across four time periods and two innovative industries. The results take the lens of network structure and add to the findings of a trade-off relationship between adaptation and adaptability in Martin and Sunley (2006). We find that a complete network structure embraces high adaptation but lower adaptability, whereas a coalitional network owns low adaptation but high adaptability. Furthermore, our findings complete recent efforts to examine network robustness and economic resilience in the European context (Tóth et al., 2022). We go beyond network robustness and operationalize four types of network typologies. Also, we examined economic resilience in a more detailed dimension, dividing it into adaptation and adaptability.

Policy makers could prioritize different network typologies based on the place-based industrial contexts. Policy can promote adaptability and bridging networks when the traditional industry is no longer competitive due to its resource depletion, competitive disadvantage in the global market, or technological obsolescence (Fornahl et al., 2012; Hu & Hassink, 2017). In contrast, adaptation and bonding networks can be supported when the existing industries possess capabilities that are still competitive in the broader economic context. Innovation policies can be designed to strategically foster a certain type of collaboration to promote adaptation or adaptability. For example, the U.S. Economic Development Administration has initiated the Planning and Local Technical Assistance Program to promote regional economic prosperity and resiliency (U.S. Economic Development Administration, 2023). Local governments could utilize these funds to facilitate the construction of a bridging network by establishing regional coalitions and organizing workshops and networking events, which could connect high-tech companies, research institutes, knowledge workers, entrepreneurs, and other related entities to form new collaborations. Such events can be inclusive and encourage interested organizations or individuals. Meanwhile, local governments could work together to create online forums for exchanging information and matching collaborative interests, thus facilitating new regional collaborations. Besides, creating public open spaces and amenities for entrepreneurs and creative classes to interact and network can promote the formation of new collaborations. To strengthen the bonding network, local governments could dedicate funds to encourage long-term and in-depth collaboration by organizing frequent in-person events to create a dense sense of community. Such events could be invitation-only and target key stakeholders.

Industries are heterogeneous. This study focuses on CEPM and EACM, both of which are innovative industries. The results may apply to similar high-tech industries where innovation plays a more significant role in influencing economic resilience. In terms of the service industry and traditional manufacturing industries where innovation plays a negligible role, the influence of co-patenting network structure may be small or insignificant. Nonetheless, the same methodology can be applied to analyze other network structures and potentially aid industrial policies.

Supplemental Material

sj-docx-1-edq-10.1177_08912424241287776 - Supplemental material for How Does Innovation-Collaboration Network Structure Influence Regional Economic Resilience?

Supplemental material, sj-docx-1-edq-10.1177_08912424241287776 for How Does Innovation-Collaboration Network Structure Influence Regional Economic Resilience? by Yijia Wen and Li Fang in Economic Development Quarterly

Network	Complete network	Factional network	Coalitional network	Bridging network
Standard deviation of betweenness centrality	low	low	high	high
Mean of closeness centrality	high	low	high	low

	Dependent variable: adaptation
Million population	0.0892***	0.0893***	0.0784***	0.0749***	0.0446***	0.0561***	0.0684***	0.113***
	(0.0126)	(0.0128)	(0.0129)	(0.0130)	(0.0125)	(0.0126)	(0.0128)	(0.0131)
Factional network	0.0657**	0.0375	−0.0174	−0.110***	0.0748**	0.0954***	0.0673**	0.0274
	(0.0245)	(0.0250)	(0.0252)	(0.0253)	(0.0250)	(0.0253)	(0.0256)	(0.0261)
Coalitional network	0.0206	0.0151	0.0119	0.00492	0.0800*	0.118***	0.135***	0.100**
	(0.0256)	(0.0261)	(0.0263)	(0.0264)	(0.0322)	(0.0326)	(0.0330)	(0.0336)
Bridging network	0.110***	0.0752*	−0.0102	−0.107***	0.411***	0.425***	0.397***	0.321***
	(0.0306)	(0.0313)	(0.0314)	(0.0316)	(0.0386)	(0.0391)	(0.0396)	(0.0403)
Log accumulated damage	0.00702*	0.00838**	0.0110***	0.00837**	0.00968**	0.0133***	0.00988**	0.00732
	(0.0029)	(0.0030)	(0.0030)	(0.0030)	(0.0036)	(0.0037)	(0.0037)	(0.0038)
Log Damage*factional	−0.0148***	−0.0149***	−0.0150***	−0.00752	−0.0108**	−0.0145***	−0.00852*	−0.0125**
	(0.0037)	(0.0038)	(0.0038)	(0.0039)	(0.0042)	(0.0042)	(0.0043)	(0.0044)
Log damage* coalitional	−0.0042	−0.0061	−0.00904**	−0.00800*	−0.0140***	−0.0177***	−0.0155***	−0.00936*
	(0.0034)	(0.0034)	(0.0035)	(0.0035)	(0.0039)	(0.0039)	(0.0040)	(0.0041)
Log damage* bridging	−0.0064	−0.00571	0.00118	0.00384	−0.0438***	−0.0443***	−0.0366***	−0.0269***
	(0.0051)	(0.0052)	(0.0052)	(0.0053)	(0.0064)	(0.0065)	(0.0066)	(0.0067)
Constant	0.00521	0.0187	0.0723	0.115*	−0.0728	−0.0632	−0.0185	0.0339
	(0.0506)	(0.0516)	(0.0519)	(0.0522)	(0.0509)	(0.0515)	(0.0522)	(0.0532)
CBSA fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Year-Quarter fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Number of observations	21801	21801	21801	21801	19008	19008	19008	19008
R-squared	0.655	0.633	0.615	0.596	0.651	0.633	0.616	0.598

Dependent variable: adaptability
Million population	0.0330***	0.0447***	0.0362*	−0.0155	−0.0136	−0.0215	0.018	0.0325*
	(0.0091)	(0.0119)	(0.0143)	(0.0154)	(0.0087)	(0.0111)	(0.0133)	(0.0144)
Factional network	−0.0161	−0.0067	−0.0196	−0.0234	−0.0302	−0.0493*	−0.0482	−0.0208
	(0.0178)	(0.0232)	(0.0279)	(0.0301)	(0.0175)	(0.0223)	(0.0266)	(0.0287)
Coalitional network	−0.0388*	−0.0389	−0.0292	0.004	−0.041	−0.0371	−0.0843*	−0.0657
	(0.0186)	(0.0242)	(0.0292)	(0.0314)	(0.0225)	(0.0286)	(0.0343)	(0.0370)
Bridging network	−0.0376	−0.048	−0.0473	−0.0287	0.00729	0.021	−0.0146	0.00613
	(0.0223)	(0.0290)	(0.0349)	(0.0376)	(0.0270)	(0.0343)	(0.0411)	(0.0443)
Log accumulated damage	−0.00315	−0.005	−0.00718*	−0.00147	−0.00844***	−0.0142***	−0.0167***	−0.0168***
	(0.0021)	(0.0027)	(0.0033)	(0.0035)	(0.0025)	(0.0032)	(0.0039)	(0.0042)
Log damage*factional	0.000476	0.000399	−0.000545	−0.00341	0.00584*	0.00858*	0.0113*	0.0194***
	(0.0027)	(0.0035)	(0.0043)	(0.0046)	(0.0029)	(0.0037)	(0.0044)	(0.0048)
Log damage* coalitional	0.00459	0.00959**	0.0178***	0.0199***	0.00828**	0.0143***	0.0208***	0.0222***
	(0.0025)	(0.0032)	(0.0038)	(0.0041)	(0.0027)	(0.0035)	(0.0041)	(0.0045)
Log damage* bridging	0.000736	0.00497	0.0156**	0.0199**	0.00241	0.00627	0.0109	0.0121
	(0.0037)	(0.0048)	(0.0058)	(0.0062)	(0.0045)	(0.0057)	(0.0068)	(0.0074)
Constant	−0.00992	−0.0468	−0.0467	−0.0638	−0.0458	−0.0501	−0.0686	−0.0695
	(0.0368)	(0.0478)	(0.0576)	(0.0620)	(0.0356)	(0.0453)	(0.0542)	(0.0585)
CBSA fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Year-Quarter fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Number of observations	21801	21801	21801	21801	19008	19008	19008	19008
R-squared	0.655	0.633	0.615	0.596	0.651	0.633	0.616	0.598

Footnotes

Acknowledgements

We thank the reviewers and editor for their insightful comments on the manuscript. We appreciate helpful comments from the faculty and students at Florida State University, particularly Karina Amalbert.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Yijia Wen

Li Fang

Supplemental Material

Supplemental material for this article is available online.

Author Biographies

Yijia Wen is a PhD candidate in the Department of Urban and Regional Planning at Florida State University. Her work examines the driving force of innovation and the relationship between innovation and economic resilience.

Li Fang, PhD, is an associate professor in the Department of Urban and Regional Planning at the University of Illinois at Urbana-Champaign. She specializes in economic development and land use policy, with a particular emphasis on urban innovation and entrepreneurship.

References

Agger

Jensen

J. O.

(2015). Area-based initiatives—and their work in bonding, bridging and linking social capital. European Planning Studies, 23(10), 2045–2061. https://doi.org/10.1080/09654313.2014.998172

Andrew

S. A.

Carr

J. B.

(2013). Mitigating uncertainty and risk in planning for regional preparedness: The role of bonding and bridging relationships. Urban Studies, 50(4), 709–724. https://doi.org/10.1177/0042098012455718

Aral

Van Alstyne

(2011). The diversity-bandwidth trade-off. American Journal of Sociology, 117(1), 90–171. https://doi.org/10.1086/661238

Balland

P.-A.

Rigby

Boschma

(2015). The technological resilience of US cities. Cambridge Journal of Regions, Economy and Society, 8(2), 167–184. https://doi.org/10.1093/cjres/rsv007

Berlemann

Wenzel

(2015). Long-term growth effects of natural disasters-empirical evidence for droughts. CESifo Working Paper Series No. 5598, Available at SSRN: https://ssrn.com/abstract=2701762 or https://doi.org/10.2139/ssrn.2701762.

Berrou

J.-P.

Combarnous

(2012). The personal networks of entrepreneurs in an informal African urban economy: Does the ‘strength of ties’ matter? Review of Social Economy, 70(1), 1–30. https://doi.org/10.1080/00346764.2011.577347

Boschma

(2015). Towards an evolutionary perspective on regional resilience. Regional Studies, 49(5), 733–751. https://doi.org/10.1080/00343404.2014.959481

Bottazzi

Peri

(2003). Innovation and spillovers in regions: Evidence from European patent data. European Economic Review, 47(4), 687–710. https://doi.org/10.1016/S0014-2921(02)00307-0

Botzen

W. J. W.

Deschenes

Sanders

(2019). The economic impacts of natural disasters: A review of models and empirical studies. Review of Environmental Economics and Policy, 13(2), 167–188. https://doi.org/10.1093/reep/rez004

10.

Bowen

W. M.

Park

Elvery

J. A.

(2013). Empirical estimates of the influence of renewable energy portfolio standards on the green economies of states. Economic Development Quarterly, 27(4), 338–351. https://doi.org/10.1177/0891242413491316

11.

Bristow

Healy

(2014). Building resilient regions: Complex adaptive systems and the role of policy intervention. Raumforschung Und Raumordnung, 72(2), 93–102. https://doi.org/10.1007/s13147-014-0280-0

12.

Bristow

Healy

(2018). Innovation and regional economic resilience: An exploratory analysis. The Annals of Regional Science, 60(2), 265–284. https://doi.org/10.1007/s00168-017-0841-6

13.

California Office of the Small Business Advocate (n.d.). Accelerate California inclusive innovation hubs. California Office of the Small Business Advocate. Retrieved April 7, 2024, from https://calosba.ca.gov/funding-grants-incentives/ihub2/ .

14.

Callois

J.-M.

Aubert

(2007). Towards indicators of social capital for regional development issues: The case of French rural areas. Regional Studies, 41(6), 809–821. https://doi.org/10.1080/00343400601142720

15.

Carlsson

Steen

Sand

Nilsen

S. K.

(2014). Resilient peripheral regions? The long-term effects of ten Norwegian restructuring programmes. Norsk Geografisk Tidsskrift-Norwegian Journal of Geography, 68(2), 91–101. https://doi.org/10.1080/00291951.2014.894565

16.

Carr

J. B.

Hawkins

C. V.

Westberg

D. E.

(2017). An exploration of collaboration risk in joint ventures: Perceptions of risk by local economic development officials. Economic Development Quarterly, 31(3), 210–227. https://doi.org/10.1177/0891242417710325

17.

Chapple

(2002). I name it and I claim it—in the name of jesus, this job is mine”: Job search, networks, and careers for low-income women. Economic Development Quarterly, 16(4), 294–313. https://doi.org/10.1177/089124202237195

18.

Cofré-Bravo

Klerkx

Engler

(2019). Combinations of bonding, bridging, and linking social capital for farm innovation: How farmers configure different support networks. Journal of Rural Studies, 69, 53–64. https://doi.org/10.1016/j.jrurstud.2019.04.004

19.

Crespo

Suire

Vicente

(2014). Lock-in or lock-out? How structural properties of knowledge networks affect regional resilience. Journal of Economic Geography, 14(1), 199–219. https://doi.org/10.1093/jeg/lbt006

20.

Crowe

J. A.

(2007). In search of a happy medium: How the structure of interorganizational networks influence community economic development strategies. Social Networks, 29(4), 469–488. https://doi.org/10.1016/j.socnet.2007.02.002

21.

Dempwolf

C. S.

Lyles

L. W.

(2012). The uses of social network analysis in planning: A review of the literature. Journal of Planning Literature, 27(1), 3–21. https://doi.org/10.1177/0885412211411092

22.

Di Caro

(2020). Quo vadis resilience? Measurement and policy challenges: Using the case of Italy. In Handbook on regional economic resilience (pp. 88–102). Edward Elgar Publishing.

23.

Fang

(2015). Do clusters encourage innovation? A meta-analysis. Journal of Planning Literature, 30(3), 239–260. https://doi.org/10.1177/0885412215589848

24.

Fornahl

Hassink

Klaerding

Mossig

Schröder

(2012). From the old path of shipbuilding onto the new path of offshore wind energy? The case of northern Germany. European Planning Studies, 20(5), 835–855. https://doi.org/10.1080/09654313.2012.667928

25.

Freeman

L. C.

(1977). A set of measures of centrality based on betweenness. Sociometry 40(1), 35–41. https://doi.org/10.2307/3033543

26.

Garnick

D. H.

(1969). Disaggregated basic-service models and regional input-output models in multiregional projections. Journal of Regional Science, 9(1), 87–100. https://doi.org/10.1111/j.1467-9787.1969.tb01444.x

27.

Granovetter

M. S.

(1973). The strength of weak ties. American Journal of Sociology, 78(6), 1360–1380. https://doi.org/10.1086/225469

28.

Hassink

(2010). Regional resilience: A promising concept to explain differences in regional economic adaptability? Cambridge Journal of Regions, Economy and Society, 3(1), 45–58. https://doi.org/10.1093/cjres/rsp033

29.

Hoyman

McCall

Paarlberg

Brennan

(2016). Considering the role of social capital for economic development outcomes in US counties. Economic Development Quarterly, 30(4), 342–357. https://doi.org/10.1177/0891242416659135

30.

Hsiang

S. M.

Jina

A. S.

(2014). The causal effect of environmental catastrophe on long-run economic growth: Evidence from 6,700 cyclones. National Bureau of Economic Research.

31.

Hassink

(2017). Exploring adaptation and adaptability in uneven economic resilience: A tale of two Chinese mining regions. Cambridge Journal of Regions, Economy and Society, 10(3), 527–541. https://doi.org/10.1093/cjres/rsx012

32.

Hassink

(2020). Adaptation, adaptability and regional economic resilience: A conceptual framework. In Handbook on regional economic resilience (pp. 54–68). Edward Elgar Publishing.

33.

Huang

Liu

Tang

(2024). Long-term economic impact of disasters: Evidence from multiple earthquakes in China. World Development, 174, 106446. https://doi.org/10.1016/j.worlddev.2023.106446

34.

Humanmade (n.d.). Community workshop and workforce development. HUMANMADE. Retrieved April 7, 2024, from https://www.humanmade.org/.

35.

Kahl

Hundt

(2015). Employment performance in times of crisis: A multilevel analysis of economic resilience in the German biotechnology industry. Competitiveness Review, 25(4), 371–391. https://doi.org/10.1108/CR-12-2014-0038

36.

Kim

H.-Y.

(2018). Effects of social capital on collective action for community development. Social Behavior and Personality: An International Journal, 46(6), 1011–1028. https://doi.org/10.2224/sbp.7082

37.

Knoke

Yang

(2008). Social network analysis (1st ed.). Sage Publications, Inc. https://doi.org/10.4135/9781412985864

38.

Kurth

Kozlowski

Ganin

Mersky

Leung

Dykes

Kitsak

Linkov

(2020). Lack of resilience in transportation networks: Economic implications. Transportation Research Part D: Transport and Environment, 86, 102419. https://doi.org/10.1016/j.trd.2020.102419

39.

Mardaneh

K. K.

Jain

Courvisanos

(2020). A guide to patterns of regional economic resilience. In Handbook on regional economic resilience (pp. 126–142). Edward Elgar Publishing. https://doi.org/10.4337/9781785360862

40.

Martin

(2012). Regional economic resilience, hysteresis and recessionary shocks. Journal of Economic Geography, 12(1), 1–32. https://doi.org/10.1093/jeg/lbr019

41.

Martin

Sunley

(2006). Path dependence and regional economic evolution. Journal of Economic Geography, 6(4), 395–437. https://doi.org/10.1093/jeg/lbl012

42.

Martin

Sunley

(2015). On the notion of regional economic resilience: Conceptualization and explanation. Journal of Economic Geography, 15(1), 1–42. https://doi.org/10.1093/jeg/lbu015

43.

Muštra

Šimundić

Kuliš

(2020). Does innovation matter for regional labour resilience? The case of EU regions. Regional Science Policy & Practice, 12(5), 955–971. https://doi-org.proxy.lib.fsu.edu/10.1111/rsp3.12348. https://doi.org/10.1111/rsp3.12348

44.

Newman

Dale

(2005). Network structure, diversity, and proactive resilience building: A response to Tompkins and Adger. Ecology and Society, 10(1), 1–4. https://doi.org/10.5751/ES-01396-1001r02

45.

Patent Technology Monitoring Team (2023). U.S. Patenting Trends by NAICS Industry Category Utility Patent Grants Calendar Years 1963-2012. United States Patent and Trademark Office. Retrieved October 1, 2023, from https://www.uspto.gov/web/offices/ac/ido/oeip/taf/naics/doc/naics_info.htm.

46.

Patulny

R. V.

Svendsen

G. L. H.

(2007). Exploring the social capital grid: Bonding, bridging, qualitative, quantitative. International Journal of Sociology and Social Policy, 27(1/2), 32–51. https://doi.org/10.1108/01443330710722742

47.

Pike

Dawley

Tomaney

(2010). Resilience, adaptation and adaptability. Cambridge Journal of Regions, Economy and Society, 3(1), 59–70. https://doi.org/10.1093/cjres/rsq001

48.

Pontarollo

Serpieri

(2020). Ranking regional economic resilience in the EU. In Handbook on regional economic resilience (pp. 103–125). Edward Elgar Publishing.

49.

Poortinga

(2012). Community resilience and health: The role of bonding, bridging, and linking aspects of social capital. Health & Place, 18(2), 286–295. https://doi.org/10.1016/j.healthplace.2011.09.017

50.

Putnam

R. D.

(2000). Bowling alone: The collapse and revival of American community. Simon and schuster.

51.

Reid

Smith

B. W.

Carroll

M. C.

(2008). Cluster regions: A social network perspective. Economic Development Quarterly, 22(4), 345–352. https://doi.org/10.1177/0891242408322719

52.

Statsenko

(2017). The impact of regional supply network structure and governance on the resilience of resource extractive regions: The case of South Australia .

53.

Suire

Vicente

(2014). Clusters for life or life cycles of clusters: In search of the critical factors of clusters’ resilience. Entrepreneurship & Regional Development, 26(1–2), 142–164. https://doi.org/10.1080/08985626.2013.877985

54.

Technology & Patent Research International (2015, October 12). Making the Switch from USPC to CPC: What Attorneys and Searchers Need to Know. Technology & Patent Research, Retrieved August 20, 2023, from https://www.tprinternational.com/making-the-switch-from-uspc-to-cpc/.

55.

Therrien

M.-C.

Jutras

Usher

(2019). Including quality in social network analysis to foster dialogue in urban resilience and adaptation policies. Environmental Science & Policy, 93, 1–10. https://doi.org/10.1016/j.envsci.2018.11.016

56.

Tóth

Elekes

Whittle

Lee

Kogler

D. F.

(2020). Technology network structure conditions the economic resilience of regions. Utrecht University.

57.

Tóth

Elekes

Whittle

Lee

Kogler

D. F.

(2022). Technology network structure conditions the economic resilience of regions. Economic Geography, 98(4), 355–378. https://doi.org/10.1080/00130095.2022.2035715

58.

U.S. Bureau of Labor Statistics (2022a). Industries at a Glance: Computer and Electronic Product Manufacturing: NAICS 334. Retrieved October 2, 2022, from https://www.bls.gov/iag/tgs/iag334.htm.

59.

U.S. Bureau of Labor Statistics (2022b). Industries at a Glance: Electrical Equipment, Appliance, and Component Manufacturing: NAICS 335. Retrieved October 2, 2022, from https://www.bls.gov/iag/tgs/iag335.htm.

60.

U.S. Census Bureau (2021). Housing Patterns and Core-Based Statistical Areas. Census.Gov. Retrieved June 10, 2024, from https://www.census.gov/topics/housing/housing-patterns/about/core-based-statistical-areas.html.

61.

U.S. Department of Transportation (2022, June 9). Core Based Statistical Areas. Retrieved June 9, 2024, from https://data-usdot.opendata.arcgis.com/datasets/usdot::core-based-statistical-areas/about.

62.

U.S. Economic Development Administration (2023). Planning, Technical Assistance, Research funding opportunities. U.S. Economic Development Administration. Retrieved October 29, 2023, from https://www.eda.gov/funding/funding-opportunities/category?category=266288.

63.

U.S. Patent and Trademark Office (2016). Intellectual Property And the U.S. Economy: Third Edition. United States Patent and Trademark Office. Retrieved October 1, 2023, from https://www.uspto.gov/ip-policy/economic-research/intellectual-property-and-us-economy.

64.

U.S. Patent and Trademark Office (n.d.). Patents Pendency Data February 2024 First Office Action Pendency. Retrieved March 10, 2024, from https://www.uspto.gov/dashboard/patents/pendency.html.

65.

Vicente

(2020). Unravelling the driving forces of networks on regional resilience capabilities. In Handbook on regional economic resilience (pp. 209–225). Edward Elgar Publishing. https://doi.org/10.4337/9781785360862

66.

Wasserman

Faust

(1994). Social network analysis: Methods and applications. Cambridge University Press. https://doi.org/10.1017/CBO9780511815478.

67.

Wassmer

R. W.

(2021). Do higher land costs for new single-family housing inhibit economic activity in US metropolitan areas? Economic Development Quarterly, 35(4), 325–337. https://doi.org/10.1177/08912424211006174

68.

Xiao

Drucker

(2013). Does economic diversity enhance regional disaster resilience? Journal of the American Planning Association, 79(2), 148–160. https://doi.org/10.1080/01944363.2013.882125

69.

Qian

(2021). The economic network resilience of the guanzhong plain city cluster, China: A network analysis from the evolutionary perspective. Growth and Change, 52(4), 2391–2411. https://doi.org/10.1111/grow.12530

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.34 MB

Network	Complete network	Factional network	Coalitional network	Bridging network
Measurement	Complete network	Factional network	Coalitional network	Bridging network
Standard deviation of betweenness centrality	low	low	high	high
Mean of closeness centrality	high	low	high	low

	Dependent variable: adaptation
	Computer and electronic product manufacturing (CEPM)				Electrical equipment, appliance, and component manufacturing (EACM)
	One quarter	1 year	3 year	5 year	One quarter	1 year	3 year	5 year
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Million population	0.0892***	0.0893***	0.0784***	0.0749***	0.0446***	0.0561***	0.0684***	0.113***
	(0.0126)	(0.0128)	(0.0129)	(0.0130)	(0.0125)	(0.0126)	(0.0128)	(0.0131)
Factional network	0.0657**	0.0375	−0.0174	−0.110***	0.0748**	0.0954***	0.0673**	0.0274
	(0.0245)	(0.0250)	(0.0252)	(0.0253)	(0.0250)	(0.0253)	(0.0256)	(0.0261)
Coalitional network	0.0206	0.0151	0.0119	0.00492	0.0800*	0.118***	0.135***	0.100**
	(0.0256)	(0.0261)	(0.0263)	(0.0264)	(0.0322)	(0.0326)	(0.0330)	(0.0336)
Bridging network	0.110***	0.0752*	−0.0102	−0.107***	0.411***	0.425***	0.397***	0.321***
	(0.0306)	(0.0313)	(0.0314)	(0.0316)	(0.0386)	(0.0391)	(0.0396)	(0.0403)
Log accumulated damage	0.00702*	0.00838**	0.0110***	0.00837**	0.00968**	0.0133***	0.00988**	0.00732
	(0.0029)	(0.0030)	(0.0030)	(0.0030)	(0.0036)	(0.0037)	(0.0037)	(0.0038)
Log Damage*factional	−0.0148***	−0.0149***	−0.0150***	−0.00752	−0.0108**	−0.0145***	−0.00852*	−0.0125**
	(0.0037)	(0.0038)	(0.0038)	(0.0039)	(0.0042)	(0.0042)	(0.0043)	(0.0044)
Log damage* coalitional	−0.0042	−0.0061	−0.00904**	−0.00800*	−0.0140***	−0.0177***	−0.0155***	−0.00936*
	(0.0034)	(0.0034)	(0.0035)	(0.0035)	(0.0039)	(0.0039)	(0.0040)	(0.0041)
Log damage* bridging	−0.0064	−0.00571	0.00118	0.00384	−0.0438***	−0.0443***	−0.0366***	−0.0269***
	(0.0051)	(0.0052)	(0.0052)	(0.0053)	(0.0064)	(0.0065)	(0.0066)	(0.0067)
Constant	0.00521	0.0187	0.0723	0.115*	−0.0728	−0.0632	−0.0185	0.0339
	(0.0506)	(0.0516)	(0.0519)	(0.0522)	(0.0509)	(0.0515)	(0.0522)	(0.0532)
CBSA fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Year-Quarter fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Number of observations	21801	21801	21801	21801	19008	19008	19008	19008
R-squared	0.655	0.633	0.615	0.596	0.651	0.633	0.616	0.598