SNA is a powerful tool in quantitative analysis. Social networks are comprised of nodes—which are the actors, or members, of the network—and edges—which are the ties, or relations, linking the nodes in the network. Nodes may have one or more relation, and types of relations, with each other (Marin and Wellman 2011). For example, a farm might sell produce to consumers at a farmers’ market. However, the same farm might also utilize their booth at the same farmers’ market as a CSA pickup site. As such, there would be two edge connections between the farmers’ market and the farm: one denoting DTC sales via farmers’ market sales unrelated to the CSA, and another denoting DTC sales through a CSA-based relationship. This distinction is important because, as Hinrichs (2000) notes, CSAs and farmers’ markets offer differently embedded social relationships. Although farmers’ markets enable face-to-face interactions between farmers and consumers, they are not necessarily developing longer-term continuous relationships (Hinrichs 2000). On the other hand, the CSA model can foster greater trust and value-driven relationships between customers, who buy shares for the growing season, and CSA farmers, who are commonly motivated by non-economic factors and set share prices that are not exclusively profit-driven (Galt 2013). Such relationships may have different staying power over time, or allow for different evolutions across the network as farms transition from one form of marketing to another. We are able to explore both relationships over time using SNA.
Growth and death
To start, we provide a descriptive comparison of both counties and the proportion of network actors and ties, then we explore change over time and network architecture. Although Chester County has a larger local food system network, both in terms of nodes and edges, the overall local food network of Chester County is shrinking, while the local food network of Baltimore County is growing (Table 1). During the 6-year study period, Baltimore County saw the addition of 284 new nodes and 495 new edges in the network. During the same time period, Chester County saw the addition of 360 new nodes, and 684 new edges, but lost 393 nodes and 738 edges (Table 1). One possible explanation is that local food systems may reach a point beyond which added growth is very difficult, due to plateauing consumer interest (Low et al. 2015) or market saturation. However, when delineated by category (Table 2), all sectors within the Chester County local food system are growing. The one exception is the “Other” category which is primarily comprised of sales and donations to institutions and civic organizations. This category relied more heavily on 2012 survey data to uncover the many farm-to-food bank donations across Chester County. Such donations are not as readily advertised on farm websites and may therefore lead to under-counting in the 2018 dataset. This finding points to nuances in how local food system growth is tabulated both in research, such as this, and by the agricultural census, where categories are broad and may overlook central connections like that of the Chester County Food Bank.
Both networks show substantial change from 2012 to 2018, with a relatively high rate of turnover of actors within the network (Table 2). When examined by node or edge category, both counties show nearly equal rates of growth and death in network actors (nodes) and their marketing relationships (edges). Despite growth in many categories, more than half of the participants in the local food system changed over the 6-year period, with only 40% of Baltimore County’s 2012 nodes found in the 2018 data, and only 35% of Chester County’s 2012 nodes found in the 2018 data. More telling, the connections across the network changed even more than the actors themselves, with only 18% of edges staying the same across both 2012 and 2018 in both counties. The fluctuation in edges indicates that, while actors may be stable, their relationships with one another evolve.
The rates of endurance by category varied. In the Chester County dataset, the following nodes endured: 91 farms, 23 schools involved in farm-to-school and food bank connections, 18 farmers’ markets, 18 grocery stores, 15 restaurants, 11 churches involved in food bank gardening and distribution, and 3 food banks. These locations accounted for 85% of the actors that endured from 2012 to 2018. The rest of the actors were CSA drop-off locations, community gardens, and food hubs. By comparison, the Baltimore County dataset showed 37 farms, 30 restaurants, 20 grocery stores, and 14 farmers’ markets active in the network in both 2012 and 2018. These actors made up 87% of the actors that endured within the dataset. The remaining enduring actors include CSA drop-off locations, two schools, two catering companies, and two churches.
Generalizations across categories are shown in Table 2. In 2012, the Chester County “Other” node category included 80 civic organizations (e.g., schools, churches, and retirement communities), many with gardens that donated food to other civic organizations. These gardens largely catered to schools or the Chester County Food Bank. The Chester 2012 data in the “Other” category also included 88 CSA drop-off locations. While the number of restaurants, farmers’ markets, farms, and grocers increased over the 6-year period, the miscellaneous category decreased, with a decrease in both civic organizations and CSA drop-off locations (Tables 1 and 2). This change is likely because the number of gardens associated with the food bank and other civic organizations were not as readily found online in 2018. Similarly, the 2018 Baltimore County “Other” category included 15 churches and 3 food banks.
Importantly, the “Other” category is larger than any other category across both counties. This indicates the variety of actors beyond farms, farmers’ markets, restaurants and grocers, which are currently the main focus of much of local food systems research. The “Other” category also captures new marketing typologies that may tap into other socio-political movements. For example, the 2018 Baltimore County dataset included a recently legalized cannabis shop, which purchases infused honey from a local beekeeper. Although the cannabis shop typology was collapsed into the “Other” category for our analysis, this represents a new aspect to local food systems that warrants further investigation, particularly as hemp-derivatives become more common in other local food spaces, such as farmers’ markets, and as local food systems spread into new spaces with their own divergent or intersectional political objectives.
Separating network actors into categories allows us to explore further properties of local food system stability. For example, farmers’ markets were the most stable nodes within the network across both counties. This may be because farmers’ markets generally have an explicit goal of providing business opportunities for local food producers, thus making them a relatively stable outlet for local food system sales. More than half (55%) of the farmers’ markets stayed open in Chester County through the 6-year study period, and nearly half of them (47%) stayed open in Baltimore County. This finding supports USDA agricultural census information, noting that in 7 years (2009–2016), the number of farmers’ markets increased by 270% (3 to 11) in Chester County and by 40% (12 to 17) in Baltimore County (USDA Food Environment Atlas). However, our data also show high rates of turnover, with over 40% of the 2012 farmers’ markets no longer in operation by 2018. This flux over the course of a 6-year period indicates a certain degree of market instability, as well as rapid evolution in how consumers interact within an ever-changing local food system.
Across both counties, grocers also appeared to be relatively stable actors in the local food system, with a little less than half (45% and 47% in each county) of the 2012 grocers remaining in the 2018 local food network (Table 2). Because grocers are important intermediaries that are often central to local food networks (Trivette 2019; Brinkley 2017, 2018), their relative stability in the network offers promise for long-term stability and growth in local food systems. The two counties in this study differ in terms of the growth of this food system actor, with grocers making up the largest growth (53%) in the actor category for Chester County, but not Baltimore County (12%) (Table 2). Baltimore County’s local food system is comparatively more reliant on restaurants. This might explain the greater growth in the restaurant category, with the addition of 84 new restaurants between 2012 and 2018. Although 30restaurants remained in the Baltimore County local food network throughout the course of the study, a nearly equal number of restaurants (35) also dropped out of the network between 2012 and 2018 (Table 2). The restaurant category had higher turnover in both counties when compared to grocers.
Our data indicate that, unlike restaurants, farms have greater staying power. They are also increasingly joining the local food system in both study counties. Although the USDA agricultural census noted a 30% decrease in the number of farms (128 to 91 farms) that sell through direct-market channels from 2007 to 2012 in Baltimore County (USDA, Food Environment Atlas nd), our data shows an 11% increase in the number of farms in the local food system (Table 2). Similarly, the USDA agricultural census notes a modest 4% increase in farms that sell through direct-market channels in Chester County (from 735 to 782) throughout 2007–2012; our research indicates that this county saw a 25% increase in the number of farms involved in the local food system (Table 2). The differences in figures could be because our data also capture farms that sell through intermediate markets. Intermediate markets account for two-thirds of local sales (USDA NASS 2017). Further, the offset in years between the USDA agricultural census data collection and this study may also explain the difference in figures. Also of note, the Baltimore dataset appears to capture a more representative sample of direct-market farms compared to the census, while the Chester County dataset captures about 30% of direct-market farms compared to the USDA agricultural census. This may partly be because Chester County has a large portion of Amish farms that may take part in the agricultural census, but may not have an online presence as a result of religious restrictions on technology use. Due to the nature of online data collection methodology employed in this study, we were not able to verify these Amish farms and, as a result, we could not access their marketing connections.
Confirmed business closures between 2012 and 2018 provide supporting evidence for the broad categorical trends above. Importantly, closure is distinct from actors simply dropping out of the network, as closure implies a complete and indefinite severing of network ties. Uniquely, SNA allows us to assess the disproportionate impact that the loss of specific actors can have on a network. Restaurants made up 60% of the 36 confirmed closures in Baltimore County. The second highest category of closures were farms, which represented an additional 16% of total closures. Similarly, half of the Chester network’s nineteen confirmed closures were restaurants (Table 1). Additionally, four grocery stores, three farms, two farmers’ markets, and one CSA distribution location closed, thus removing them from the 2018 network.
If a local food system is more dependent on restaurants, the flux within the network could be greater, as is the case in Baltimore County. The Baltimore County dataset shows a greater loss of nodes in terms of confirmed closures, with 12% of the nodes from 2012 having closed by 2018. This resulted in a 20% loss of edge connections, as compared to a 3% loss rate for nodes and edge connections in the Chester County dataset. Restaurants have a median lifespan of 4.5 years (Luo and Stark 2014), and other network actors may have a longer business lifespan, thus translating to increased stability within the network. Many restaurants that close see the owners or chefs establish new eateries shortly thereafter. Future research could track such transitions to see if relationships are re-established with the same farmers and distributors as new spaces open up, or if restaurants that source locally have different survival rates than their non-locally sourcing counterparts. Another possible explanation is that local food systems may need to achieve critical mass in order to compete with larger-scale food supply chains. It is possible that Chester County’s large local food system has less flux compared to the still growing local food system of Baltimore County.
Another way to view the confirmed closures is that each actor is a unique contributor to the local food system. The confirmed closure of 36 actors in the Baltimore County network had a disproportionate impact on edge connections, resulting in 125 lost relationships. Conversely, while Chester County also saw the closure of a few actors (19), those closures only resulted in the loss of 30 edge connections. In Baltimore County, the closure of five actors, in particular, resulted in a substantial loss of edges. These actors included the following restaurants and farms: Simmer Rock Farm, Atwater’s Ploughboy Kitchen, Big City Farm, Woodhall Wine Cellars, and Clementine Restaurant. Simmer Rock Farm opened in 2010 and closed by 2013, resulting in the loss of 25 connections, including three farmers’ market sales locations, 15 restaurants that carried their food, one grocery store, and a CSA. The restaurant Atwater’s Ploughboy Kitchen also closed, resulting in the loss of 37connections. Big City Farm was a collection of urban farmers; its closure resulted in the loss of 14 connections, and the closure of Woodhall Wine Cellars and Clementine restaurant both resulted in the loss of seven connections. Collectively, these account for the 72% of lost connections due to closures within the network, pointing to the significant impact that a few actors can have on local food system dynamics.
Visualization of network architecture
To understand if markets are growing outward socially or if new members are incorporated at the heart of the network, we use SNA visualization to show how the web of market ties have changed over time. When visualized socially, with the most connected actors at the center of the network, Chester County’s local food system shows growth and decay concentrated along the network’s outer margins, though growth and death within the network is widespread (Figs. 1 and 2). In contrast, Baltimore County shows significant network decay amongst actors that were central to the network in 2012, with growth occurring on the network’s periphery (Figs. 3 and 4). Broadly, such patterns may be the hallmarks of a larger, more established local food system in Chester County evolving at the margins, with stable central network actors maintaining the core relationships and network architecture. Conversely, Baltimore County appears to be reinventing itself, with high turnover in actors that were once central to the network.
Basic network statistics help reinforce the findings from visualizations, while telling a more nuanced story about the evolution of the local food systems in both counties (Table 3). To quantify how connected the local food system is, we use the average degree statistic, which indicates the average number of actors to which each node is tied. Chester County had a stable average degree measure between 2012 and 2018, while the average degree of Baltimore County declined substantially from 2.023 to 1.37, meaning that actors within the local food system have fewer average connections in 2018 than they did in 2012. The clustering coefficient indicates the degree to which the neighbors of a node are connected. A coefficient of 1 would indicate that all neighbors are connected to each other, while a coefficient of 0 would indicate that none of a node’s connections have mutual ties. While the average clustering coefficient for Chester County remained stable at 0.0023 between 2012 and 2018, the clustering coefficient for Baltimore County dropped from 0.032 to 0.023. In sum, Baltimore’s network became sparser and more porous due to the many confirmed closures, mentioned above, that were central to the network architecture (Figs. 3 and 4). As central actors dropped out of Baltimore County’s local food system (Figs. 3 and 4), newer actors grew at the network’s fringe. However, this growth was not fast enough to reestablish the same level of connectivity across the network.
To understand how information might travel across the network, we use network diameter, which indicates the maximum distance between any two nodes within the network. The network diameter shrank for both networks, indicating that the overall local food system became more close-knit (Table 3) potentially enabling information to travel across market ties more quickly. Similarly, the average path length for both networks also declined. The average path length indicates the average steps needed to get from one actor in the network to another and is often used to gauge how quickly information can travel across a network. Declines in network diameter and average path length indicate the development of a more tightly integrated and consolidated local food system. Had the network split, the path across would have become disconnected or very long. Such splits can occur when social or market networks fraction, but this was not the case in either county. Finally, graph density shows the total number of edges within the network relative to the possible number of edges within a network. In other words, if every node within a network were connected to every other node in the network the density value would be 1, while if no nodes were connected to each other the density value would be 0. Both networks saw graph density decline between 2012 and 2018. As both local food systems are maturing, they are consolidating and reducing the redundancy in connections.
Centrality of actors
The perseverance of actors and ties across both years could be interpreted as strong ties among actors, while new connections and nodes may represent innovation and “weak ties.” Between 2012 and 2018 the actors most central to both networks cultivated new sales and market channel relationships, both with actors that were new to the network and with enduring actors with whom they were not previously connected. This finding indicates innovation among both enduring and new network actors. Collectively, the above statistics demonstrate that the total makeup of the network is in considerable flux.
Additionally, the data indicate that the centrality of actors is changing. Betweenness centrality indicates the extent to which a node acts as a bridge between two other nodes. As such, high betweenness centrality can suggest a node’s substantial power within a network, as it may serve as a broker between other actors. In Baltimore County, only one node (Springfield Farm) was ranked in the top ten highest betweenness centrality in both 2012 and 2018. Similarly, within the Chester County dataset, only one node (the Chester County Food Bank) was ranked in the top ten highest betweenness centrality across both years. Previous research has demonstrated the role that these specific actors have played in brokering new partnerships across the food system and influencing land-use policy (Brinkley 2017, 2018). The turnover of other actors central to the network was an unexpected finding, showing deep changes within the local food system as the constellation of people and organizations changed. These changes likely translate to shifts in the sphere of influence of these actors as well.
Scholarly literature has portrayed growing local food systems as creating enduring, embedded ties while also having a high turnover. While these claims appear paradoxical, this research helps show why such assertions may be simultaneously true. The persistence of high-centrality nodes, like the Chester County Food Bank and Springfield farm, and strength of their ties across the local food system may be especially important in an ever-changing network that is dominated by weak ties. Such weak ties foster innovation (Granovetter 1977, 1983) as new forms of market channels and associated socio-political alliances are formed across the local food system.
Network spatiality
Last, spatial trends related to network change over time help build on earlier research that considers the growth of local food systems as a response to the bow wave of urban development (Hart 1990; Zasada 2011; Brinkley 2012). The Chester County dataset shows growth of the local food network in the northern parts of the county (Fig. 5), and a simultaneous loss of food system actors in the southern portions of the county. Actor loss was clustered close to the City of Philadelphia. In Baltimore County (Fig. 6), network actors that were present across both years of the dataset were engaged in forming new edges and maintaining old connections. Similar to Chester County, actor loss is clustered in the southern portion of Baltimore County, which is closest to the City of Baltimore. Growth within the network is clustered to the north, which corresponds with Baltimore County’s more rural areas.
In both counties, the local food system experienced actor loss closer to urban areas, and new growth further from cities in peri-urban and rural areas. It is important to note that actors are not only farms, but also other nodes, such as farmers’ markets. This finding suggests that there may be spatial boundaries to the ideological objectives of the local food movement. As farms are forced further away from urban areas, the distances to get to urban markets may become too far to traverse. At the same time, suburban growth may also stretch the social distance between urbanites and rural dwellers, placing the many shared objectives of the local food movement further from people’s reach, both physically and mentally.
While the counties have many differences, the similarities across both datasets may point to larger trends regionally or nationally in local food marketing. We show that farms are joining the local food movement. This change is not captured in the USDA agricultural census for either county, though it is noted nationally. The number of farms with DTC sales increased by 5.5% from 2007 to 2012, but with no increase in DTC sales (Low et al. 2015), and then the number of farms with DTC sales declined in 2017 (O'Hara and Benson 2019). Like the USDA agricultural census, we found that the most common way of selling local food was through intermediate markets, and that online marketing appeared to be on the rise. Marketing pathways are rapidly changing. In addition, both networks are consolidating and becoming more tight knit. Such change would indicate that these local food systems are made up of weak ties, enabling rapid innovation, with ever decreasing distances from one side of the network to the other. As a result, news travels faster. The network architecture of these two cases reveals that despite these weak ties both counties have a stable central actor that maintains the core identity of the county through political engagement with land-use policy and planning. These network findings help make sense of seemingly conflicting accounts that local food systems struggle and are growing; innovate and are historic (Pretty 1990; Vitiello and Brinkley 2014); and last, that they are dominated in numbers by weak ties and in central actors with strong bonds.