Final Report for GS13-121
I investigated ways to connect low-resource producers and low-income consumers of fresh produce in 31 counties in NE North Carolina. My research partner and I conducted spatial, statistical, and qualitative analyses. The spatial analysis, while not part of the SSARE grant, identified three food deserts –regions of limited access to adequate food- in SE Beaufort County, NE Beaufort County, and SW Washington County. The survey and interview methodology identified major barriers to connecting producers and consumers as the decrease in the number of small farms, increasing bureaucracy, high cost of entry, and historical divisions between ethnic and socioeconomic groups.
To better characterize barriers rural producers and consumers face to produce and access healthy food in Northeastern North Carolina, my research partner and I conducted three separate analyses. Using Geographic Information Systems, we developed a spatial analysis to understand geographic patterns of food deserts and access barriers. I ran a two-model statistical analysis based on USDA Food Environment Atlas data to identify significant demographic and socioeconomic variables that affect food access. I augmented the statistical data with qualitative surveys and interviews to define barriers producers and consumers face on the intra-county scale. The qualitative and spatial analyses were focused on two low-income counties: Beaufort County and Washington County, NC. Community stakeholders, local food producers, consumers, and grocery retailers were interviewed. The statistical analysis focused both on 31 target North Carolina counties and on the entire Eastern Coastal Plain. It revealed that persistent poverty counties and counties experiencing population loss were more likely to experience little or no access to grocery stores. Race was also a factor, particularly within North Carolina where minorities are more vulnerable to food insecurity. Both Washington and Beaufort Counties exhibit a high level of economic and demographic stratification. Two-thirds of consumers from the survey had problems stretching their food budget, and identified a weekly food box at low or no-cost as the best intervention. Retail grocery stores already can and do buy local food. However, retailers buy locally according to the season and price. Major barriers to connecting low-resource producers and low-income consumers were identified as the decrease in the number of small farms, increasing bureaucracy, high cost of entry, and historical divisions between ethnic and socioeconomic groups. Using the geographic and socio-economic barriers, the spatial analysis identified three food deserts – in SE Beaufort County, NE Beaufort County, and SW Washington County and the main drivers for each.
Given that some successful models have been developed in other areas and the lack of small scale data on market barriers to entry, my research objectives were:
1) ASSESS economic and social factors that limit the transfer of sustainable food product between low income farmers and local consumers using a literature review and available datasets.
2) ASSESS local economic and social factors via interviews with local producers, consumers, government officials, and food distribution and retail businesses in the study area.
3) COMPARE within county fine scale data to regional macroscale existing data.
4) DEVELOP a set of recommendations for Resourceful Communities on investment sectors and pilot programs appropriate for the proposed project region to maximize healthy, local food choices for rural communities.
Developing a Food Desert Index using Spatial Analysis
A spatial analysis informed by the statistical analysis and qualitative data from the interviews was developed in collaboration with my research partner, Harry Zhang. Three types of non-geographic barriers—Economic, Vulnerability, and Cultural/Informational barriers—along with geographic barriers were utilized to evaluate food accessibility. In particular, following the work of Parsons (2012), 8 variables of socioeconomic characteristics were identified (Table 1 in the Appendix). Quality and resolution of the input data varied depending on the US Census Bureau. Geographic barriers were captured by two variables—travel distance to the nearest food retailer, and household availability of vehicles. The final products of the Food Desert Analysis consist of graphs and maps presenting the statistics and distribution of each barrier type. A food desert map was also created to integrate all individual barriers. The final products of the Food Desert Analysis consist of graphs and maps presenting the statistics and distribution of each barrier type. An ultimate food desert map was also created to integrate all individual barriers. The general workflow is illustrated in Figure 1 in the Appendix.
Identifying Demographic Variables that Affect Local Food Access
The initial statistical research question was “which socioeconomic variables affect the existence of food deserts in the counties in question?” To address this question, I mined the database from the Food Environment Atlas, developed by the USDA-ERS in 2012. As an initial model, I used multiple Ordinary Least Square(OLS) regression to interpret the 31 counties of interest in Northeastern North Carolina. I selected the total percent of the population with low access to a major grocery store on the county level as the response variable and various demographic, socioeconomic, and location-dependent variables as my explanatory variables (Table 2 in the Appendix).
To ensure a more accurate regression, I extended the statistical analysis outside of the original 31-county study area in North Carolina, and included 55 additional regional coastal plains counties in Virginia, South Carolina, and Georgia as defined by the Environmental Protection Agency’s definition of an ecoregion. The Eastern coastal plain runs along the East Coast from Maine to Georgia and is defined by lowlands dominated by woodland, urban land, or marshland with less than 20% used as pasture land or crop land (EPA, 2000). Two subsets of Ecoregion VIII, the Level III Middle Atlantic Coastal Plain (number 63) and the Southern Coastal Plain (number 75), were used to define counties included in this study from Virginia to Georgia with similar climates and land use histories to the North Carolina counties of interest (EPA, 2013). Every county fit in or near the ranges defined by the original 31 county study area in terms of the percentage of the population experiencing low access to healthy food (2.8-24.6%), median household income ($30,586-$59,522), and poverty rate (9-27%). Notably the data from the North Carolina counties of interest is extremely variable, reflecting historical poverty in the area and tourism development that has decreased poverty and increased income in counties such as Camden and Dare.
I applied the original model (Model 1) to the larger sample size of 85 coastal plain counties and independent cities across Virginia, North Carolina, South Carolina, and Georgia to develop Model 2. The entire list of counties and MSAs (metropolitan statistical areas) included in the regional analysis are in Table 3 in the Appendix.
Determining Barriers using Interview and Survey Data
To augment the statistical and spatial analyses and provide some fine scale resolution on what consumers, retailers, and producers identify as barriers, I conducted surveys and interviews of local retailers, producers, and community stakeholders in both Washington and Beaufort Counties collected over three weekends and weekdays in February and March 2014.
Unfortunately, I was unable to find a suitable community partner in Washington County to disseminate the consumer surveys. Thus I only have data for Beaufort County, courtesy of Jared Cates, Community Mobilizer, Carolina Farm Stewardship Association, who completed a similar study for Beaufort County. However, Beaufort County contains the town of Little Washington (population about 10,000) which serves as a regional center for shopping, doctor’s appointments, and other community needs. Some of the respondents (approximately 8 percent) lived in counties other than Beaufort according to their zip codes. Based on some demographic similarities, we can assume some of the barriers and community needs identified are applicable to Washington County as well.
Jared Cates collected data from July to December 2013 with several community partners in Beaufort County – Eagle Wings Food Pantry, Beaufort County Department of Social Services, Beaufort County Health Department, Vidant Family Medicine Aurora, and others.
I conducted additional analyses of the consumer survey results to determine correlations between different variables using STATA, statistical analysis software. The summary of the survey results is available in Table 4 in the Appendix.
My research partner identified possible grocery stores, roadside stands, and specialty markets that would meet the definition of “a grocery store or supermarket” using Google maps. ReferenceUSA data using the NAICS (North American Industry Classification System) code 541105 for both Washington and Beaufort Counties was also used to identify retail grocery stores and to determine the accuracy of both Google maps and ReferenceUSA.
I drove to the identified sites over two weekends in February 2014 using the addresses provided by my research partner and cross-referencing them with ReferenceUSA. If a site was primarily a gas station, restaurant, or convenience store/corner market, I marked it as such and took it off the list of surveyed grocery stores. If a site could not be found or had changed owners or store type, I marked it as such. If a site had the wrong address or geocoordinates, I marked it as such and corrected the information. Nineteen survey sites were identified – three sights were not surveyed due to time constraints, one site had an absent manager, two sites were based on original data from Jared Cates of Carolina Farm Stewardship Association, and twelve sights were successfully surveyed. Of the thirteen sites approached, only one store refused to be surveyed.
I determined the barriers producers face via structured interviews with local producers in Washington and Beaufort Counties, via primary data from Jared Cates’ interactions with local producers, and via interviews with community stakeholders including representatives of the North Carolina Public Health Foundation, the Washington, Beaufort, and Martin County Extension Offices, Center for Disease Control (CDC) Community Transformation Grant representatives, North Carolina Department of Health and Human Services, North Carolina Department of Agriculture, and others. I recorded the interviews using an Olympus voice recorder, partially transcribed using InqScribe software, and analyzed using NVivo software to identify common themes and barriers.
Nearly all people identified as possible interview subjects were interviewed, save the Center for Disease Control Community Transformation Grant Healthy Eating Lead from Region 9. Farmers were identified via a contact of Resourceful Communities, Jared Cates. Of 14 local producers, three were contacted and two responded. Notably, eleven of these producers are only part-time or seasonal producers for one crop such as “u-pick” strawberries. While there are a dozen local fruit and vegetable producers in Beaufort County not including producers of processed products such as honey, there are only two identified producers in Washington County. Primary data from Jared Cates, including survey data and notes from community meetings, was used to supplement the interviews.
Developing a Food Desert Index using Spatial Analysis
Non-Geographic Barrier Scores
The Economic and Cultural/Informational barrier scores have realized ranges of 2-100 and 1-99, respectively. The vulnerability score range was only 4-65, indicating that the vulnerability barrier was not a significant stressor. The final normalized score of all three types of barriers had a realized range of 11-87. Its moderate lower end suggested that the lowest of the three barriers scores did not coincide spatially. In other words, every place was stressed by at least one type of barrier. On the other hand, the realized higher score suggests that severely stressed areas have all barrier types. The mean values of barrier scores and the final score fell on the lower half of their ranges, indicating larger areas of relatively low scores than of high scores. This observation is consistent with the positively skewed final score distribution observed in Figure 2 in the Appendix.
Spatially, the greatest non-geographic barriers were found in urban areas near Plymouth, Washington County in the north, and Washington, Beaufort County in the southeast (Figure3). Significant barriers were also observed in northeastern Beaufort County along the Beaufort-Hyde border, as well as in lower Beaufort County south of the Pamlico River, especially the southeastern region centered roughly at town of Aurora (Figure 3 in the Appendix). Also worth noting is northern Washington County which has stressful Economic and Cultural/Informational conditions, but moderate Vulnerability stresses. In general, Vulnerability barriers do not show the same spatial distribution patterns exhibited by Economic and Cultural/Informational barriers.
Geographic Barrier Scores
Geographically, three areas of distance-based food deserts were located in SE Beaufort County, NE Beaufort County, and SW Washington County (Figure 4 in the Appendix). Clusters of food retailer locations tended to occur around urban centers and populous areas. However, some parts of Beaufort County still suffer geographic barriers despite the presence of a number of grocery stores, e.g., northeastern Beaufort County and southeastern Beaufort County near the town of Aurora. Limited access to vehicles is the primary factor in these high-stressed areas although insufficient road connections are also important. The limited food access area immediately north of Pamlico River is well connected by roads and has adequate vehicle access. However, the region lacks food retailer stores. Additionally, the vast areas across the Beaufort-Washington boundary categorized as distance-stressed are results of few retailer stores, sparse road connections, and, to some degree, low vehicle availability.
A downloadable Food Desert Locator Tool was developed to reproduce the geographic barrier analysis. Optional population density and urban/rural constraints are also available. The tool requires ArcGIS for Desktop v. 10.1 or higher and the Network Analyst extension.
Identifying Demographic Variables that Affect Local Food Access
Statistically, the North Carolina analysis of 31 counties and the regional analysis of 85 Eastern coastal plain counties were very similar in terms of rural/urban divides, the negative impact of population loss on food availability, and the negative impact of persistent poverty. However, the North Carolina analysis indicates more stratification by ethnicity, meaning individuals who are part of minority groups are more likely to have low access to grocery stores and fresh produce. Age is also an important factor in food access – children under 18 and residents over age 65 are particularly vulnerable. This may be because seniors in Eastern North Carolina live in isolated, rural areas without access to transportation, while children are dependent upon their parents or guardians.
Table 5 in the Appendix indicates selected pairwise correlations for Model 1 (North Carolina Analysis) and Model 2 (Regional Analysis).
In Model 1, median household income actedas a proxy for education in my final model. As seen below, there remains some clustering in the residuals plot, but the overall distribution of fitted values is relatively normal but heteroskedastic. I used non-white variables in my final model. I found that the model is statistically significant at the 5% level with an F-statistic p-value of 0.043 and an R-squared value of 0.5272. Table 6 in the Appendix includes relevant correlation coefficients and p-values. The final model equation is:
y =-87.1*Log a + 3.9b – 3.5c –2.4d – 0.5e – 0.002f – 2.1g + 10.2 h – 11.5i+ 1081.4
where y is: % of the population with low access to store
a: median household income
b: % nonwhite population
c: % of people 65 and older in 2010
d: % of people 18 and younger in 2010
e: % nonwhite*log of income
f: grocery store change * % nonwhite
For Model 2, the poverty rate actedas a proxy for education in my final model. As seen below, there remains some clustering in the residuals plot, but the overall distribution of fitted values is relatively normal and not heteroskedastic (Figure 8 and Table 7). I used non-white variables and robust standard errors to account for multi-collinearity in the final model. I found that the model is statistically significant at the 1% level with an F-statistic of 4.44. The R-squared indicates that the variables explain 0.2746 or about 27% of the variation within the response variable, percentage of people in a county experiencing low access to grocery stores. Table 7 in the Appendix includes relevant correlation coefficients and p-values. The final model equation is:
y = -5.368a – 5.540b – 45.340*Log c –4.205d + 0.252e +22.308f + 0.00265f + 1.287g– 3.938h + 11.076i – 7.674j + 677.242
where y is: % population, low access to store
a: % white population
b: % nonwhite population
c: % of people 65 and older in 2010
d: % of people 18 and younger in 2010
e: poverty rate in 2010
f: % white * % nonwhite
g: Log of % 65older in 2010 * % 18younger in 2010
Determining Barriers using Interview and Survey Data
Survey data from Jared Cates, Carolina Farm Stewardship Association, was used to identify barriers consumers face. He collected 687 surveys, 94% of them from zip codes within Beaufort County. Approximately two-thirds of respondents noted they had problems stretching their food budget, and approximately one-half of respondents believed they did not receive proper nutrition. While 85% of all respondents primarily shopped at retail grocery stores, approximately 10% used discount stores like Dollar General or Family Dollar as their primary source of food. 80% of consumers traveled via car or truck; however, 15% of consumers, nearly all of them African-American, rode with friends or family. More than half of all respondents were on some kind of government assistance, but most still experienced food insecurity. 40% of respondents indicated they would be interested in a subsidized weekly food box program. The survey data more or less accurately represents the low-income community of the region. Figures 5-8 in the Appendix are graphical representations of the results of the 13-question survey.
12 retail grocery stores out of a possible 19 grocery stores were surveyed via in-person directed surveys. 11 out of 12 stores noted they sourced fresh produce locally, often but not always from local farmers. One chain (Food Lion) noted they could not source locally directly because they were a corporation. However, their statewide distributor in Lumberton buys North Carolina produce. Almost all stores except very small grocery stores accepted SNAP/EBT (the official name for a federally-funded food assistance program). Figure 9-11 in the Appendix are graphical representations of the results of the 8-question survey.
Interviews were conducted with eight community stakeholders involved in public health and local food and two producers. They identified barriers producers faced due to the high cost and high level of risk associated with growing food. Interviewees noted local producers are older with more financial and other resources. However, lack of farm labor was identified as the biggest issue for food production. While Washington County lost its farmer’s market in 2009, Beaufort County has more active investment and two active farmer’s markets as of 2014. Figures 12-15 are flowcharts of the local food supply chain and indicate barriers producers and consumers face.
Based on the barriers identified via qualitative, statistical, and geospatial methods, I have several strategic recommendations for Resourceful Communities as both a funder and a capacity builder. These recommendations are applicable to a wide range of non-profit organizations, foundations and trusts, and local, regional, and state government entities.
- Support organizations that address consumer transportation issues. The consumer survey indicates that people without a car are more likely to shop at places that are not grocery stores. They are more likely to suffer poor nutrition and food insecurity. Innovative delivery programs such as Produce Ped’lers in Goldsboro deliver fresh produce from the farmer’s market straight to people’s doors via bicycle delivery. A similar program to address food insecurity in the town of Washington, North Carolina could be easily applied. Geographically, food deserts identified in Southeast Beaufort County are primarily limited by transportation access and could benefit from similar programs that use cars instead of bicycles. Programs that encourage ride sharing from rural areas and/or expanded use of county van services should be encouraged.
- Support subsidized or low-cost CSA programs. Approximately 40% of respondents to the consumer survey indicated they would be interested in a low-cost or free box of produce from a local farm. At least two farms in Washington County have similar CSA (community-supported agriculture) programs already. Focus could be given to food desert regions of Southwest Washington County and Southeast Beaufort County where large populations experience economic stressors.
- Support organizations that address SNAP/EBT purchases at farmer’s markets and roadside stands. Given that over half the respondents on the consumer survey indicated they were on a food assistance program, doubling their dollars at farmer’s markets or allowing them to apply food assistance money toward a CSA encourages healthier eating and better connections with local farmers. This approach will be especially meaningful in the city of Washington, the city of Plymouth, and food desert areas in Southeast Beaufort County given high local poverty rates and rates of people dependent on SNAP benefits.
- Encourage producers to be “retail ready” and provide technical support and training. Gary Bullen of the North Carolina State University Cooperative Extension in Raleigh developed the report Retail Ready for Local Farm Products in 2013. Trainings from NCSU extension offices are ongoing throughout the state.
- Connect producers, farmer’s groups, and business alliances with information about Pitt County’s “Healthy Corner Store Initiative.” (Pitt County 2012).
- Encourage and support small scale programs and events that connect producers and consumers, even if there is not a long-term benefit. For example, Resourceful Communities could support an organization that arranges for a local food dinner at a food pantry or an organization that arranges for schoolchildren to visit local farms. While these small gestures do not have immediate effects, they open the door to further connection and collaboration between producers and consumers.
 The Food Desert Locator Tool can be downloaded from http://dukespace.lib.duke.edu/dspace/bitstream/handle/10161/8570/FoodDesertLocatorTool.zip?sequence=1. Copyright © Yiduo “Harry” Zhang.
 The report can be found online at http://ag-econ.ncsu.edu/sites/ag-econ.ncsu.edu/files/faculty/bullen/RETAIL_READY_COMPLETE_2-27-2013_6pm.pdf.
 For more information, please speak with Diana Vetter Craft, Pitt County Government, or reference www.depc.org/files/Pitt_County_Healthy_Corner_Store_Presenation.pptxand http://www.cdc.gov/pcd/issues/2013/12_0318.htm.
Educational & Outreach Activities
A final report Identifying barriers to sustainable food production by low resource producers and purchase by low income consumers in Washington and Beaufort Counties, NC has been published and is available on Duke Library’s website at http://dukespace.lib.duke.edu/dspace/handle/10161/8570#files. The report includes an additional “Food Desert Locator Tool” developed by my research partner.
Additionally, a shortened 4 page executive summary of the project and its results was developed for distribution to community partners in May 2014. On 23rd July 2014, Kimberly Hill presented to Beaufort County stakeholders in the HEAL (Healthy Eating, Active Living) Collaborative and provided several more copies of the executive summary to interested members of the Beaufort and Washington County communities.
A full project description, including methods, results, and contact information for the graduate student team is available via a project website online at http://sites.duke.edu/foodbarriersinnc2013/.
The final results of the report have been distributed to more than 30 project collaborators via e-mail and via a community partner that agreed to print the Executive Summary and distribute it to community members in Washington and Beaufort Counties. Resourceful Communities, a program of the Conservation Fund that funds small-scale non-profits and provides capacity-building resources, is considering how to best fund local food initiatives – particularly discounted food boxes from local farms – in the 31-county study area. Carolina Farm Stewardship Association is aware of the study and is deciding how best to use its results to fund local producers and local “healthy eating, active living” initiatives. On 23rd July 2014, I (Kimberly Hill) presented to Beaufort County stakeholders on the project, answering any questions they may have.
While there was neither time nor money to evaluate the benefits and costs of different recommendations made to Resourceful Communities and other community stakeholders, some general economic assumptions can be made based on the literature review of similar studies and the specific information gathered for this study.
The barriers residents in Beaufort and Washington Counties face in accessing or marketing local produce are similar to barriers faced in other rural areas of the country. According to the statistical analysis, the entire Southeastern Coastal Plain is affected by decreases in the number of grocery stores per 1,000 people. According to the Center for Rural Affairs, nearly one in five rural grocery stores has gone out of business since 2006 in the Midwest, and fewer people are employed in the grocery sector (Bailey 2010). The situation is undoubtedly very similar for North Carolina, though no academic data has been collected. Rural consumers are more likely to travel further to reach farmer’s markets, according to a study conducted by Eastern North Carolina University in Greenville (Jillcott-Pitts et al. 2013). However, only two farmer’s markets are present in the two-county study area – both are struggling, both are located in Beaufort County, and neither of them has the resources to accept SNAP/EBT. Out of 687 respondents for the consumer survey, only nine noted they shopped primarily at a farmer’s market. However, retailers already do buy and market local produce. Some buy directly from farmers, while others buy local produce through wholesalers. While the retailer survey was subject to bias, it does indicate openness on the part of most grocery stores to work with local producers.
The cusp of the problem is that food does not make a profit for the small-scale producer. Since the Tobacco buyout of 2004 (Department of City and Regional Planning – UNC 2008), Eastern North Carolina lost several thousand farms. Few new farmers have started in the area and very few of these are willing or able to grow food. As noted by the Washington County Extension Office, the Washington County Farmer’s Market fell apart after 2009 when the few remaining producers passed away.
The results of this project are not meant to be adopted by farmers. Rather, the results are meant to be considered by local and state funding agencies as well as non-profit foundations such as the Conservation Fund’s Resourceful Communities Program. State-level, county-level, and municipality-level stakeholders, as well as members of civil society, have all received copies of the abbreviated report.
Areas needing additional study
The consumer survey identified a significant portion of the population (about 10%) that primarily shops for food at discount stores such as Dollar General and Family Dollar. Due to time and financial limitations, discount stores were not included in the retailer survey. We recommend either a phone or in-person survey of select discount stores in the future.
Additionally, due to the economic history of NE North Carolina, we recommend interviewing a select number of small farmers that do not produce food crops for commercial sale. This would allow the researchers to identify why producers cannot or will not grow produce and to understand why only established producers with more resources can afford to grow food.