Agricultural and Resource Economics Review The Impact of Four Alternative Policies to Decrease Soda ConsumptionYizao Liu, Rigoberto A. Lopez, and Chen Zhu Key Words : Yizao Liu Department of Agricultural and Resource Economics University of Connecticut 1376 Storrs Road, Unit 4021 Storrs, CT 06269-4021 Phone 860.486.1923 Email email@example.com ARER
April 2014Agricultural and Resource Economics Review Figure 1. Trends in Consumption of Leading Nonalcoholic Beverages Beverage Digest 60 50 40 30 20 10 0 2000 2002 2004 2006 2008 2010 2012 CSDs Water Milk JuicesGallons per PersonYear
Impact of Four Policies to Decrease Soda Consumption Liu, Lopez, and Zhu Empirical Model j J j j j
April 2014Agricultural and Resource Economics Review i m uijm ipjm size j i i Adjm x j i jm ijm jm ijm ijm pjm j m xj i sizej jm ijm Adjm j m m Adjt Aj,t k Ajt t k i iiii i vi vi N I vi ijm jm pjm size j Adjm x j jm uijm pjm sizej Adjm xj vi ijm i j m sijm j r J j m
Impact of Four Policies to Decrease Soda Consumption Liu, Lopez, and Zhu sjm I vi ijm Uijm Uikm k J dG v dF v viijm G v F ijm j k i f f j J f f j f pj mcj M sj p Cf pj mcj j M Cf sj p j j f sj p l f pl mcl j l J sj p mc p s p M qjm sjM.
April 2014Agricultural and Resource Economics Review mc, p, s p Data and Estimation
Impact of Four Policies to Decrease Soda Consumption Liu, Lopez, and Zhu Table 1. CSD Brand Data Summary over Seven Designated Market Areas for 2006 through 2008 Weekly Market Gross Price Share Rating Calories Sugar Sodium Caffeine Company and Brand Table 2. Summary Statistics by Container SizeBottle Size Price Unit Price Market Share
April 2014Agricultural and Resource Economics Review Empirical Results Demand Results Table 3. Demand Estimation Results Parameters Deviations Standard Standard Variable Estimate Deviation Estimate Deviation
Impact of Four Policies to Decrease Soda Consumption Liu, Lopez, and Zhu Policy Simulations
April 2014Agricultural and Resource Economics Review Table 4. Own-Price Elasticities Container Price Company Brand Size in Ounces Elasticity Television advertising ban: A soda tax at the point of consumption: j
Impact of Four Policies to Decrease Soda Consumption Liu, Lopez, and Zhu Product reformulation (reducing sugar content): Downsizing of packages: more
April 2014Agricultural and Resource Economics Review Table 5. Estimated Percent Market Shares under Alternative Policy Scenarios S1: All S2: Sales S3: S4: No Bottle Gross Tax of One Restricted 2-LiterSize in S0: Ratings Cent per Calorie Size Company and Brand Ounces Benchmark = 0 Ounce Content Bottles
Impact of Four Policies to Decrease Soda Consumption Liu, Lopez, and Zhu regular Table 6. Simulated Percentage Change per Capita Annually in CSD Calories Consumed under Alternative Policy Scenarios S1: All S2: Sales S3: S4: No Gross Tax of One Restricted 2-LiterS0: Ratings Cent per Calorie Size Bottle Benchmark = 0 Ounce Content Bottles Size in Company and Brand Ounces
April 2014Agricultural and Resource Economics Review Conclusion
Impact of Four Policies to Decrease Soda Consumption Liu, Lopez, and Zhu References American Journal of Public Health Econometrica Beverage Digest . . American Journal of Public Health RAND Journal of Economics Agricultural and Resource Economics Review International Journal of Industrial Organization Journal of Marketing Research Journal of Economics and Management Strategy Quantitative Marketing Economics Marketing Science Econometrica Quantitative Marketing and Economics Food Availability Data System Journal of Health Economics Contemporary Economic Policy Journal of the American Dietetic Association
April 2014Agricultural and Resource Economics Review Journal of Economics and Marketing Strategy . Nutrition Journal American Journal of Preventive Medicine Journal of the American Dietetic Association Applied Economics The Food Police: A Well-fed Manifesto about the Politics of Your Plate. Review of Industrial Organization Regulation Critical Reviews in Food Science and Nutrition Econometrica American Journal of Agricultural Economics American Journal of Agricultural Economi
AND SMALL FARMS: DETERMINING A RELATI ONSHIP By WILLIAM BARKER A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2014
Â© 2014 William Barker
To my f amily
4 ACKNOWLEDGMENTS I would like to acknowledge all the help and support that I have received from the members of my committee and the faculty and staff of the Food and Resource Economics department at the University of Florida . A special thanks goes out to Dr. Wysocki who early on i funding and work experience with UF/IFAS Extension . Dr. Wy s generosity combined with his offer of an opportunity for a broader agricultural educational experience within my program fundamentally changed my experience during my time here at the university. I would also like to thank Dr. VanSickle who expanded my understanding of agricultural policy, provided opportunities for real world analytical experience, and helped me ex plore fut ure opportunities in the field. A dditional ly, I am very grateful to UF/IFAS Extension and Dr. Place , Dr. Dusky, and Dr. Adams who made the effort to continue funding my assistantship during my final year in the program. Finally, I would like to ex press my gratitude to Dr. Grogan for her help at every step in this research project. Without her guidance, patience, and advice, I would never have made it through this process. Thank you to you all, I will always fondly remember this experience.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 6 LIST OF FIGURES ................................ ................................ ................................ .......... 7 LIST OF ABBREVIATIONS ................................ ................................ ............................. 8 ABSTRACT ................................ ................................ ................................ ..................... 9 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 10 Conventional Agricultural Supply Chain ................................ ................................ .. 10 Alternative Food Networks ................................ ................................ ...................... 12 Direct Marketing ................................ ................................ ................................ ...... 14 ................................ ................................ ................................ .... 15 2 LITERATURE REVIEW ................................ ................................ .......................... 19 3 SPATIAL MODEL OVERVIEW ................................ ................................ ............... 24 4 DATA ................................ ................................ ................................ ...................... 31 ................................ ................................ ............................. 31 USDA Agricultural Census Data ................................ ................................ ............. 33 Bureau of Economic and Business Research (BEBR) Data ................................ ... 34 Educational Attainment ................................ ................................ ........................... 35 Soil Data ................................ ................................ ................................ ................. 36 Hardiness Zones ................................ ................................ ................................ ..... 37 Empirical Models ................................ ................................ ................................ .... 38 Weight Matri x ................................ ................................ ................................ .......... 40 Regional Analysis ................................ ................................ ................................ ... 41 5 RESULTS ................................ ................................ ................................ ............... 46 ................................ ................................ ........................... 46 Small Farms Model ................................ ................................ ................................ . 50 6 CONCLUSIONS ................................ ................................ ................................ ..... 62 LIST OF REFERENCES ................................ ................................ ............................... 66 BIOGRAPHICAL SKETCH ................................ ................................ ............................ 70
6 LIST OF TABLES Table page 3 1 Manski Model Summary of Equations (3 1), (3 2), and (3 3) .............................. 30 4 1 Summary Statistics of County level Data ................................ ........................... 43 4 2 Summary Statistics of County level Data by Region ................................ .......... 44 5 1 Results ................................ .............. 55 5 2 ................................ ........ 56 5 3 ................................ ......... 57 5 4 Statewide Small Farms Log Model Results ( Farms <50 Acres ) ......................... 58 5 5 Statewide Small Farms Log Model Results ( Farms <180 Acres ) ....................... 60
7 LIST OF FIGURES Figure page 4 1 ARS Hardiness Zone Map ................................ ................................ .................. 45
8 LIST OF ABBREVIATIONS ACS American Community Survey AFN Alternative Food Network Ag Census USDA Census of Agricultural AMS Agricultural Marketing Service ARMS Agricultural Resource Management Survey ARS Agricultural Research Service BEBR Bureau of Economic and Business Research CSA Community Supported A griculture ERS Economic Research Service FGDL Florida Geographic Data Library FM m odel FSAO Florida Statistical Abstract Online GSPRE Generalized Spatial Random effects m odel MLE Maximum Likelihood E stimation NCSS National Cooperative Soil Survey NRCS Natural Resources Conservation Service SAIPE Small Area Income and Poverty Estimates SAC Spatial Autocorrelation m odel SAR Spatial Autoregressive m odel SDM Spatial Durbin m odel SEM Spatial Error m odel SF Small Farm m odel USDA United States Department of Agriculture WSS Web Soil Survey
9 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science MARKETS AND SMALL FA RMS: DETERMINING A RELATI ONSHIP By William Barker August 2014 Chair: Kelly Grogan Major: Food and Resource Economics The past three decades have generated strong academic and public interest in Alternative Food Networks (AFNs) genera . AFNs tend to encourage shorter supply networks and emphasize social, nutritional, and environmental considerations over profit. Facing higher barriers to entry to the conventional agricultural supply chain, small far ms are increasingly turning to AFNs and diversify into alternative value added activities and sell their production . for local foods has had a dramatic impact on t two decades. The results from this research contribute to the existing literature by describing the relationship between markets and their small farm suppliers . It establishes that the location of markets does not have a statically significant impact on the number of small farms in the region . On the other hand, t he analysis suggests that recent trends influenced by the number of small farms in an area.
10 CHAPTER 1 INTRODUCTION Participants at both ends of our conventional food distribution system are seeking alternatives to mitigate real or perceived grievances with the current supply chain. Consumers are inc reasingly seeking locally sourced foods with the belief that they are more nutritious, safer, and encourage more environmentally friendly agricultural practices. Farmers, facing pressures on their production margins, are attempting to regain control of a l arger slice of the agricultural value chain. Together these forces (AFN). experienced tremendous growth in the past twenty years. While a good deal of research explores the demand less research has focused on the supply side and the economic linkage with farmers. This research explores the an effort to determine exactly how that relationship functions. The analysis here shows farms. More importantly , this research shows that this is a causal relationship ; the location of small farmers drives the location of farmers' markets. Conventional Agricultural Supply Chain The development of global conventional agricultural supply chain has diminished profit margins. Technological advancement and increased production scales exert pressure on farm resources and provoke an economic necessity to improve operating profits by increasing volume and improving
11 efficiency (Marsden, 1998). Th e increasing pressure on farm income has forced many farms to consolidate production in an effort to obtain scale economies in their operations. Increasing efficiency and volumes have led to market saturation in some areas, further reducing production marg ins (Renting, 2003). This scaling up of production in conjunction with the growth in the food processing industry has created a delinking effect in the conventional distribution system as these sections of the value chain have grown ever proportionally lar ger and more removed from consumers (Renting, 2003). Product requirements from food companies have become surprisingly complex over time. Not only do farmers have to meet minimum safety requirements and quality standards, purchasers often times require that farmers be able to supply a minimum amount of product in order to become a supplier (Sonnino, 2006). These increasingly high market entry barriers are costly for farmers , further reinforcing consolidation and specialization pressur es. Together, these forces tighten the linkage between farms, wholesalers, and processors while weakening the bond with the end consumer. In this delinked environment, food safety has increasingly relied on objective quality standards established and monit ored by experts (Renting, 2003). Increasingly intricate regulatory issues regarding environmental regulations, animal welfare standards, and sanitary measures require specialized knowledge, large capital investments, and dedicated resources. While greatly improving safety, the black box regulatory regime has greatly reduced transparency from outside the distribution network and many claim it has come at a cost to quality, nutrition, and sustainability (Sonnino, 2006).
12 Alternative Food Networks In this envir onment, farmers face two str ategic options if they want to continue to farm. They can either grow to maintain scale efficiencies and compete on volume or diversify into alternative value added activities. This has lead to increased interest in Alternative approach was an active attempt to recapture a portion of the value chain (Renting, 2003). Shortening the supply chain creates new relationships and institutional support that require ne w methods of competition that realign power structures (Sage, 2012). As such, AFNs have attracted a great deal of attention in the last twenty years emphasize social, nutritional , and environmental considerations in addition to profits. Broadly speaking, AFNs and SFSCs are an emerging network of producers, distributors, retailers, and other actors that offer a substitute to the conventional food supply chain. These arrangements pr esent an alternative to the conventional agricultural distribution system that is commonly charged with having a heavy environmental footprint, encouraging the production of less nutritious foods, and abetting the proliferation of food deserts especially i n poorer neighborhoods (Sonnino, 2006 and Jarosz, 2008 ). Diverse in nature and founded on different social constructions, AFNs proclaim a different relationship to ecology, locality, and consumption economics (Sage, 2012). As such, these networks can impac t trends in farming practices, rural development, and sustainable resources relative to more conventional supply chains (Renting, 2003). E xpanding ideological goals beyond the profit motive reestablish es the social linkages
13 between farmers and their customers , supplementing the current fragmented supermarket relationship (Jarosz, 2008). Increasingly reported outbreaks of food borne illness in the last couple of y to obscure transparency with bureaucracy has caused consumer trust in the conventional food system to wane (Conner et al., 2010). Combined with an increased concern for ecology, health, and animal welfare, consumers have begun to demand more food choice options and transparency in the food supply chain. These new public concerns prove difficult to consumer su pport. Consumers have identified three primary factors in their desire for increased access to local foods. The first is their perception that local is a higher quality food product. The second is the ability to avoid food borne illnesses; followed closely by their desire to support local farmers (Conner et al., 2010). The primary constraint to the expansion of local foods is the lack of a distribution mechanism allowing local foods ialty or desire for alternative food choices, and the barriers small farms face in acc essing the conventional distribution chain have all increased the attractiveness of AFNs. In turning to AFNs, farmers have implicitly embraced, and increasingly utilize, direct marketing practices. However, shorter supply chains and direct farmer to consum er relationships remove the middlemen, thereby requiring the farmer to take on more responsibility for
14 marketing his produce. This naturally begs the question: do these alternative marketing channels and direct marketing efforts benefit small farms? Direct Marketing Economic Research Service (ERS) collects information on sources of farm income. ERS defines direct marketing as one of seven different marketing activities including (i) r oadside stores, (ii) on state branding programs, (vi) direct sales to local grocery stores, restaurants, or other retailers, and (vii) community supported agriculture (CSA) ( ERS , 2014). O ver time, ARMS surveys have found that direct marketing to consumers accounts for an ever Direct marketing participation by farmers has dramatically increased in the last decade. The number of farms with direct marketing income has increased 17% since 2002 and direct sales have increased 50% (Uematsu, 2011). Survey data of farmers using direct marketing strategies determined that smaller farms using organic production metho ds, but without USDA certification, more often make use of these channels. Smaller farm and household size as well as high value crops all correlated with a higher percentage of direct marketing (Monson, 2008). Research has also shown that market location is key to direct marketing success (Morgan, 2001). SFSCs allow farmers to get a better price by selling directly to consumers thereby recapturing a larger share of the value chain (Darby , 2008). However, these direct to consumer efforts can impose addition al burdens on farmers. Specialty marketing techniques like value added products, such as jellies or sauces, produced on
15 a small scale impose additional labor requirements on farmers including packaging, storage, transportation, and advertising (Martinez et al., 2010). characteristics by benefitting small farms, consumers, and com munities (Brown, 2008). For farmers, it is a good market entry point because there are few upfront requirements or costs, and the markets tend to attract a desirable customer base without the expense for testing new value added products such as artisan cheeses or canned goods. Consumers benefit from the availability of local foods that, many believe, are fresher and more nutritious. The community benefits from the local reinvestment of food dollars an d from civic pride. stall market at which farmer producers sell agricultural products directly to the general public at a central or fixed location, particularly fres h fruit and vegetables FNS , 2014). managers have identified the three most important reasons their customers shop with vailability of locally grown food (Ragland, 2009). 4.5 months per year. Markets open less than six months each year typically have 25 vendors per week, earn $20,770 per month, and s erve 565 customers weekly. Markets open for more than six months each year but not year round have on average 51 vendors per week, earn an average of $57,290 per month, and serve an average of 942
16 customers weekly. Finally, year round markets reported, on average, 58 vendors weekly, earn $69,497 per month, and serve 3,578 customers weekly (Ragland, 2009). Little is know n about t he prototypical farmer beyond their sales and travel distances. The average monthly sales for a southeastern farmer market are just under $1,000 per month. However, this can vary markedly depending on also report that nearly three quarters of farmers travel less than 10 miles t o their site (Ragland, 2009) . 13.6% annual growth rate, reaching more than 8,100 markets across the country today (AMS, 2014). The concerning aspect of this growth trend is that while the number of markets is increasing rapidly, total sales are growing at a much slower rate of 2.4% (Ragland, 2009). 1 (AGMRC, 2014). The Agricultural Marketing Service (AMS) conducted a nati onal survey of than five years old. Markets open for less than five years have, on average, fewer vendors (22 versus 31 weekly), fewer customers (430 versus 959 weekly), and l ower sales volume ($15k versus $32k monthly) (Ragland, 2009). Most of these markets are still establishing themselves economically. This may help explain the disparity between 1 market sales can be attributed to the difficulty in obtaining reliable sales data from markets and the fact that the little data that is available comes f rom different sources.
17 the growth in the number of markets and overall sales volume. It also raises que stions about long term viability of these new markets. administrative resources. They tend to h ave inexperienced management and experience high turnover in vendors , making revenue generation a challenge (Stephenson, 2008). Managers have identified their three top needs for improvement which include a need for additional support in market advertising , improved strategies for overcoming low customer attendance, and approaches for boosting vendor sales (Ragland, 2009). Increased competition between markets affect their ability to attract both customers and suppliers (Stephenson, 2008). Ironically, the i ncreased popularity of these markets has hurt existing markets through this increased competition. . pattern than the traditional supermarket. Since the two channels are founded on different operating principles, one could reasonably expect differences in their lo cation markets and supermarkets and found them to be similar in nature. The complementary question is: Is the location of small farms influenced by the important outlets for farmers, then their presence might influence the location of small
18 farms. This study used a spatial econometric analysis and two model formulations to markets and small farms. The results of this research showed that, during the recent pa began locating in areas with small farms and away f rom areas with larger farms. This relationship was determined to be causal whereby smalls influenced the location of did not have a significant affect on the location of small farms. It is important to understand how these factors influence citizens and their distribution value chain, it is a pa ssionate corner of the market and the community. As in the discussion about other food retailers and the impacts to the local economy and community moral e are important.
19 CHAPTER 2 LITERATURE REVIEW Historically, researchers have compared and contrasted AFNs and their marketing channels to those of the conventional agricultural supply chain. One key aspect of these contrasts, and the focus of this research, has been on the identifi cation Ragland, location is a critical factor and most successful markets are located in densely populated areas (Ragland, 2009). Unfortunately, there is not a lot of literat ure available on this subject for guidance (Berning, 2013). A natural first location comparison would be with other traditional retail food rkets to improve access inequalities, such as those created by food deserts. Using a bivariate spatial point pattern analysis, Sage (2012) those of grocery stores and super markets, the very type of retail outlets from which they tend to locate close to other food retail outlets instead of opting for a different strategy (Sage, 2012). Sage g ic d ic markets served a number of diverse economic roles for the trade of goods and services and wer e generally associated with the rural areas in economically underdeveloped regions. These markets were known to locate centrally within the community within the proximity of other services. He further points out that while ers face different economic considerations from
20 those of conventional food retailers, they still face constraints on the location of viable markets (Sage, 2012). Other research has shown that as population increases, so does the number of per person. According to Berning, total market size is a key determinate in the pursuit of co location strategies with grocery stores (Berning, 2013). Morgan measured the impact of certain economic and demographic factors on markets. This research determined that economic factors such as income, employment, and farmed acreage were all pos in a county (Morgan, 2001). Interestingly, the analysis did not identify any competitive Very small towns and cities had a negative effec population and the size of cities within its boundaries. More generally, it was determined ated near urban areas (Morgan, 2001). and zip code level analysis to determine the socio economic, structural, and competitive given area. The industrial organization approach recognizes that both demand and supply side factors impact location decisions. The study employed a two stage equilibrium outcome entry game rket and then they had to compete with other farms in the market. This approach to the analysis necessitated
21 two conditional assumptions. First, there had to be enough farms as potential market entrants in the area. Then secondly, at least some farms had t o decide that it would be profitable to compete (Berning, 2013). population size, increased education levels, households with children, and population participation in the SNAP pr ogram were all positively correlated with the number of found a positive association of grocery stores (Berning, 2013). Interestingly, increased farm size and the presence of large markets; the presence of the conventional supply chain represented an opportunity cost to farmers for adopting alternative direct marketing efforts (Berning, 2013). Jarosz (2008) proposes a slightly different approach to the location determination and development of AFNs. In ge sociopolitical process propelled by the restructuring of large scale agricultural areas into urban and rural areas nearer to larger metropolitan regions. Her hypothesis is that urbanization fuels the gro and political processes. First, urbanization attracts wealthier, middle class citizens who in turn demand more organic, seasonal, and locally sourced foods. Once relocated, this politically active demographic participates in interest groups and organizations involved with food politics in support of farmers and farmland preservation; this activity favors
22 not tend to stray far from the conventional, market oriented food distribution models. This includes their of farmers markets (Jarosz, 2008). Young, upwardly mobile urban professionals drive the demand for locally grown and organic produce, and it is this demand and their presen ce that drive This has led to reconsideration of the generally held perspective that AFNs are separate from the conventional food networks. The two market channels may not be separate supply chai ns but highly competitive retail formats. The fact that the research markets and traditional grocers lends support to the need for additional analysis along this line of reasoning. Some have argued that these networks are interconnected but and their suppliers are subject to less oversight and regulations (Sonnino, 2006). In summary, far demographic indicators like income, population, and higher education. Economic factors associated with farmers with a higher proportion of small farms. Areas with larger farms and a conventional markets in an area.
23 R e search reviewing the impact that s have on the location of small farms could not be located . Two primary contributions of this research to the body of existing literature come from this analysis. First, it identifies the positive, causal rel not previously found in the literature . Second, it also demonstrates that the revers e markets do not have an effect on the location of small farms. To the best of our knowledge, this latter effect has not been analyzed in previous literature.
24 CHAPTER 3 SPATIAL MODEL OVERVIEW markets lead to an increase in the number of small farms in the surrounding area or if small farms lead to an increase in area. The analysis also identifies other factors that drive the spatial distribution of (FM) m odel estimates the a county as a function of possible determinants of . These determinants could include socioeconomic variables as well as the number of small farm s in the county. The Small Farm (SF) m odel estimates the number of small farms as a function o f possible determinants of small farms . These determinants could include some demographic characteristics, soil types, climate, and categories are considered for the SF model . Any geographic measure used for the analysis could potentially misrepresent the movement. A s a result, spatial models must be used to allow for correlation across the spatial unit of observation. To examine spatial relationships in the data, a formulation of the Manski model was selected (Manski , 1993) . The model expands on the concept of spati al cross -
25 sectional models to include the ability to test for spatial correlations between the variables in panel datasets. The methodology allows for the estimation of spatially correlated dependent and independent variables, the identification of spatial error autocorrelation, and the estimation of random and fixed effects. The Hausman test can then be used to determine whether the random effects specification or the fixed effects specification is more appropriate. The general specification of spatial mode l s for panel data i s explained here and summarized in Table 3 1 below. The general model is: ( 3 1) ( 3 2) ( 3 3) The weight matrix models the various spatial correlations in the panel dataset. If the dependent variable is autocorrelated , that is captured by . If the dependent variables are correlated across space, this correlation is modeled by where the spati al autoregressive coefficient is . i s the vector of unknown parameters in and is the k by n matrix of independent observations , where k is the number of independent variables in X , and n is the number of observations in each time period . If the independent variables associated with nearby observations affect the observed dependent variable , this spatial correlation is modeled by , where is a vector of fixed unknown parameters. is the unobserved county level factor that is fixed over time . If spatial correlation exists among the county level unobservable factors, this correlation is modeled by . is an unobservable effect that occurs in time t across
26 all observations . is the vector of random error terms . Any spatial correlation in the error terms is identified by where is the spatial autocorrelation coefficient. Finally, this general formulation can not be utilized with all four spatial interactions included. At least one of the spatial interactions need s formulation in order to identify all included spatial parameters . Therefore, a series of five models can be estimated to determine if the various kinds of spatial correlation exist . T he first formulation is the Spatial Autoreg ressive (SAR) m odel (Anselin , 1988) where the neighboring dependent variable s are correlated with dependent variable i . , , and are assumed to be zero thereby eliminating the autocorrelation, spatial correlation between the neighboring independent and dependent variables, spatial correlation of the unobserved county level terms , and the spatially correlated error term. The resulting model is: ( 3 4) The second formulation is the Spatial Durbin (SDM) m ode l (Anselin , 1988) where , and are assumed to be zero thereby eliminating the spatially correlated residual and the autoregressive interaction of unobserved county level interactions. In this formulation, it is the neighboring dependent and indepen dent variables that are a factor in determining the observed independent variable resulting in the following specification : ( 3 5) The Spatial Autocorrelation (SAC) m odel (LeSage , 2009) specification assumes , , and a re zero eliminating the lagged dependent variable, the spatially lagged explanatory variables, and any unobserved county level spatial correlations . For this model, the spatial correlation across the dependent variable s and residuals are tested
27 for significance. This model cannot be estimated with random effects . It is formulated as follows: ( 3 6) ( 3 7) The fourth specification assumes , , , and are zero, which eliminates all spatial correlations except for spatial correlation in the error terms . This specification is referred to as the Spatial Error (SEM) m odel (LeSage , 2009) as shown below: ( 3 8) ( 3 9) The final specification, the Generalized Spatial Random effects (GSPRE) m odel ( Belotti , 2013), assumes , , and a re zero eliminating the lagged dependent variable, the spatial correlation in the dependent variable, and the spatial effects of the explanatory variable s. The model allows for spatial correlation in the unobserved county level effects as well as the spatia lly correlated error term . This model cannot be estimated with fixed effects. It is formulated as follows: ( 3 1 0) ( 3 11) ( 3 12) All models can be balanced spatial panel data using maximum likelihood estimation (MLE) and has a number of flexible options. This subroutine also handles the five model formulations of interest to the study. More importantly, the three different panel datasets assembled were balanced and could be analyzed using the xsmle command.
28 Spatial correlation of the independent variables indicates that the o bservations of the dependent variable in a given county are affected by the independent variables of neighboring counties. For the FM model, this kind of spatial correlation could be present if the socioeconomic composition of a bordering county increase s or decrease s the barriers, there is little cost for consumers and farmers to freely travel to neighboring markets. In the SF model aphic, and environmental factors would likely be positively correlated with small farms beca u se the presence of agricultural knowledge and capital assets tend to encourage more farming in the broader area . Spatial correlation between the dependent variables may be particularly important for the FM model, as it capture s the effects of competition between markets and c ould be negatively correlated. In the S F model, the presence of this correlation would indicate the presence of infrastructure and institutionalized knowledge in the vicinity. In the case of small farms, we hypothesize that this correlat ion will be positive . Also of interest are the spatial correlations in the residuals and the unobserved county level factors. A n ex ample of the former would be a onetime event such as a hurricane which would affect multiple, neighboring counties in a given time period . An example of the latter could be soil types or environmental variables that are similar across counties in a region but relatively fixed over time . Together, these spatial characteristics will help researchers and policy makers other . This knowledge can then be used to inform rural policy consideratio ns to better support
29 sustainable small scale farm s that emphasize food and its relationship to surrounding communities (Jarosz, 2008) .
30 Table 3 1. Manski Model Summary of Equations (3 1), (3 2), and (3 3) Parameter/Term Interpretation Example Dependent variable spatial correlation Effects of neighboring farmers' markets on the number of markets in a county may dissipate for counties farther apart Independent variables spatial correlation Effects of neighboring infrastructure on the number of small farms in a county may dissipate for counties farther apart Spatial correlation of unobserved county level effects O netime event such as a hurricane which would affect multiple, neighboring counties Spatial correlation of error terms E nvironmental variables that are similar across counties in a region but relatively fixed over time. Lagged dependent variable correlation Effects of the number of small farms in the previous period on the observed number of farms Vector of unknown parameters and matrix of independent observations Independent agricultural, demographic, and environmental variables Unobservable effect that occurs in time Economic trends such as inflation or recessions
31 CHAPTER 4 DATA Several spatial levels of analysis were considered. Due to limitations in data availab ility at a smaller spatial scale , county level units of observations were used. In addition to data availability, this unit of measure may be most appropriate for two reasons. First, each county in Florida has an extension office, and the activities of the extension service in the county may influence the number of small farms. The county level fixed effect will account fo markets are organized by county level organizations. Although appropriate for this analysis, county level data limit the number of observations in each time period to sixty seven. Three types of data were p ooled for analysis in the two models and each panel element was designated by county and year. The information collected included county level agricultural and demographic data that were then combined with environmental d climate. Due to changes in methodology used by the USDA in collecting and tabulating the Census of Agriculture (Ag Census) the study was limited to the period of 1996 through 2013. Fortunately this time period coincides nicely with the period of most int erest. Table 4 1 summarizes the data sixty seven counties for all years in the study. Farmers Market Data In 1994, the USDA began publishing a self markets from across the United States entitled the National Directory of Farmers Markets . The directories were compiled every two years until 2008 when they became an annual event. Directories for the years 1996, 1998, 2000, 2002, 2006, and 2008
32 While the lists may not be inclusive of every market that existed in every county for every year covered by the analysis, any bias should be consistent throughout the state and across time. The basis for this assessment rests on the assumption that a well one that would be effective as a direct to consumer channel for small farms , would be motivated to register in the national directory for its own self promotion. Markets that did not self register would likely have been of a short lived nature or suffered from poor management. For both the FM and S F models, data from the National Directories of Farmers Markets ) in each county. Four modifications were made to the published data. First, wholesale assess the impact of direct to consumer marketing channels for small farms. These markets were state spons ored wholes ale markets that, while allow ing some direct to consumer retail sales, primarily served as wholes ale markets. 1 Second, duplicate registrations or registrations for companion markets or incomplete registrations that could not be verified were rem oved. 2 These removals amounted to fifteen markets from the 2011 directory and nine markets in the 2013 directory. The third modification was to the 1996 directory, which consisted of only fourteen state operated and supported wholesale markets. Therefore, the 1996 USDA directory needed to be supplemented existing 1 This is consistent with the literature where only markets that conducted at least 51 percent of their retail sales directly with consumers where included in the analysis (Ragland, 2009) 2 Companion markets are defined as markets located in the same location and run by the same management team but operating during a different time period of the year.
33 markets was conducted during February and March of 2014 by phone and email. When it could be established that a market ha d beg u n operations prior to 1997, it was included in the data for the year 1996. The final modification was to interpolate the number of markets for the years 2001 and 2003. Since the USDA had originally conceived of the national directory as a biennial ex ercise, the National Directory of Farmers Markets did not exist for the years 2001 and 2003. Moreover, AMS could not find the 2004 National Directory of Farmers Markets and a published copy could not be located. On a county by county basis, the number of halfway between the number 200 0 and 200 2 directories. For 2003, a weighted average of i n the 2002 and 2006 directories , with 2002 receiving 75% of the weight and 2006 receiving 25% of the weight. In general, there is an upward trajectory of farmers markets acros s time, so these extrapolations should approximate the actual number well. USDA Ag ricultural Census Data Every five years the USDA conducts the Census of Agriculture (Ag Census) to determine key agricultural statistics by state and county across the United States. The Ag Census was the limiting resource in this study because it was the only source for determining the number and size of farms in e seven counties ( NASS , 2014). Although the Ag Census dates back more than 150 years, the USDA changed its accounting methodology for farms in 2002. The department adjusted the 1997 data, but not prior periods. Thus, farm count compari sons prior to 1997 were incompatible and were excluded from the study. Data are utilized from the 1997, 2002, 2007, and 2012 censuses ( NASS , 2014).
34 The USDA defines small farms as those with sales of less than $250,000. A consistent alternative definition for small farms based on acreage could not be identified ( NASS , 2014). Therefore for completeness, two different categories of small farms were created from the Ag Census. The first small farm designation was less than 50 acres ( Farms <50 Acres ) and the s econd was less than 180 acres ( Farms <180 Acres ). This second measure also includes the farms that are less than 50 acres. Farms that were Farms >180 Acres ). An intermediate category of medium sized farms was created for farms between 50 and 180 acres ( Farms 50 to 180 Acres ). Finally, the variables for all the agricultural acreage in the county ( Farmed Acres ) and the total number of farms in a county ( Total Number of Farms ) were also collected from the Ag Census. For two counties, data were not available for all years. The number s of acres farmed in Franklin and Nassau counties were not explicitly reported for 2002. Using the number of farms and total state farmed acreage from the 2002 census and the trend in the average size of farms in these two counties from the 1997 and 2007 Ag Censuses, the number of farmed acres in these counties was interpolated for 2002. Bureau of Economic and Business Research (BEBR) Data Several categories of demographic informatio n were obtained from the Florida Statistical Abstract , which is published by the Bureau of Economic and Business Research (BEBR) at the University of Florida. The bureau publishes annual intercensal estimates of state and county demographic data by using t decennial census counts and augmenting those with intercensal estimates and revisions. The county level data for population ( Population ) used in this analysis was cated on its
35 website (BEBR, 2014) . Bureau's Small Area Income and Poverty Estimates (SAIPE) program generating the most reliable annual estimates of county level median income. The county level data for median income ( Medium Income ) used in th e study w ere downloaded from the FSAO located on its website (BEBR, 2014) . BEBR calculates population density as the number of people per square mile . 3 T he county level data for population density ( Population Density ) used in this a nalysis w ere downloaded from the FSAO located on its website (BEBR, 2014) . Educational Attainment An important demographic statistic not published by BEBR was a county level indicator of educational attainment. Throughout much of the literature, higher levels of education have been linked with more along with higher spending levels and greater usage ( Gallons , 1997). The only Florida county level estimates found were the Annual Educational Attainment Est imates for U.S. counties 1990 2005 (Bode , 201 1 ). However, these estimates, generated by an algorithm calibrated for the entire country, were inconsistent with the American Community Survey (ACS) estimates for Florida counties for the years 2009, 2010, 2011, and 2012 (Census Bureau, 2014). seven counties were imputed using a basic fixed effects panel regression model : ( 4 1 ) 3 Land area is not adjusted for undevelopable or uninhabitable land such as parks, reserves, or water bodies.
36 is a scalar representing th e individual county level fixed effects , and is the vector of unknown parameters in where is a k x 1 vector of independent variables in time t . is the error term representing the differences between the observed and predicted value of the dependent variable . Equation (13) was estimated using county demographic data and both linear and quadratic forms of explanatory variables were considered. Two educational categories were imputed : p ercent of individuals 25 years old or older wi th a high school degree or higher and p ercent of individuals 25 years old or older with a bachelor's degree or higher . Decennial census data from the 1990 and 2000 censuses were combined with ACS five year estimates for 2009, 2010, 2011, and 2012 for the regression analysis (Census Bureau, 2014) . The predicted values from the linear and quadratic models were compared to the ACS five year estimates and the difference between the predicted and actual values were squared and summed. A model for each education al attainment category was then chosen based on minimizing the sum of the squared errors. For both the high school ( >High School Degree >Bachelor Degree ) educational attainment categories, a linear model using year and medium inco me and individual county intercepts generated the most accurate estimates of educational attainment. 4 Soil Data Soil types are a prominent factor when determining crop types and subsequently the location of major farming regions. To aid with local planning, the USDA Natural 4 BEBR did not publish medium income data for 1996. In its place, 1995 and 1997 medium income data were averaged and substituted for 1996 medium income data. Estimates included calculated County intercepts when significant; otherwise intercept was zero.
37 Resources Conservation Service (NRCS) conducts the National Cooperative Soil Survey (NCSS) de NRCS has also developed a farmland classification system from the soil data to help land area ( Total Acr es ) data can be accessed through the Web Soil Survey (WSS) link on the NRCS home webpage (NCSS, 2014). Three of the NRCS soil designations were used to describe different types of useful farmland in a county. The information was used in the small farms m odel ( S F ) as a possible determina nt of the location of small farms in the county. The best classification for agricultural land is prime ( Prime Acres ) and is defined as has the best combination of physical and chemical characteristics for producing food, feed, forage, fiber, and oilseed crops and t hat is available for these uses (NRCS, 2014). The other two classifications were farmland of local importance ( Local Importance Acres ) and unique importance ( Unique Importance Acres ). Both of th ese designations indicate that, while the soil characteristics may not be nationally recognized as especially good for all crops, it has been identified as regionally or locally important for farming (NRCS, 2014). Hardiness Zones In addition to soil data, regional climate affects the probable success and location of farms as well as the type of crops grown. For the purposes of this research, an index USDA Agricultural Research Service (ARS) publishes the Plant Hardiness Zone Map (Figure 4 1) which is the standard used by farmers to assess crop suitability for their location. The ARS constructs the hardiness zone map in 5 degree Fahrenheit
38 increments based on the average annual e xtreme minimum winter temperatures during the 1976 to 2005 period. A copy of the ARS Plant Hardiness Zone Map is included in Figure 4 1 (ARS, 2014). These zones were used to create a hardiness index value ( Hardiness Zone Index hardiness zones range from 8a to 11a corresponding to minimum extreme temperatures of 10 15 degrees Fahrenheit and 40 45 degrees Fahrenheit respectively. To develop the hardiness index value, each hardiness zone was assigned an increasing val ue based on rising minimum temperatures beginning with 8a being assigned the value 1 up to 11a receiving the value 7. An average value for each county was then calculated based on all zones ness index values ranged from 1.5 for Holmes and Okaloosa, the coldest counties in the state, up to 6.0 for Monroe County. Empirical Models Market (FM) m odel and the Small Farms (SF) m odel. The FM model includes the number of at the county level as the dependent variable. The explanatory variables include two categories: agricultural variables and demographic variables. The agricultural variables include Farmed Acres a nd the number of farms by size: Farms <50 Acres , Farms 50 to 180 Acres , and Farms >180 Acres . It is hypothesized that the agricultural characteristics of the observed county, and its neighbors, will be positively correlated with the number of ets in a county.
39 H aving a good supply of farm ers market products nearby would likely encourage the establishment for an outlet for those products. The demographic data include Population , Medium Income , Population Density , and both educational attainment categories, >High School Degree and > Degree these as drivers of demand that lead to the establishment of markets in the region (Conner et al., 2010 and Brown, 2008). In order to determine causality instead of correlation, all independent variables are lagged relative to the dependent variable . correspond to the years 1998, 2003, 2008, and 2013, while the agricultural and demographic var iables correspond to the years 1997, 2002, 2007, and 2012 , respectively . This generated 268 observations. The SF model dependent variable . Two classifications are used for small farms: less than 50 acres ( Farms <50 Acres ) and less than 180 acres ( Farms <180 Acres ), both of which are defined above. For this model, the years include 2002, 2007, and 2012, leading to 201 observations. Just like the FM model, the independent variables are lagged t o avoid endogeneity. Since the agricultural data from the Ag Census are only available every five years, the explanatory variables in the categories and Farmed Acres are lagged by five years. This resulted in the 1997 Ag Census informatio n being used as lagged independent data for the 2002 observations. The demographic data are lagged one year and environmental explanatory variables are constant across time .
40 The independent variables include agricultural, demographic, and environmental var iables . The agricultural variables were the total number of farms by size and the total Farmed Acres . Higher values of these variables indicate increased presence of infrastructure and institutional knowledge, which encourage farming. The number of Farmers was also included as an explanatory variable to test the hypothesis that the pr esence of a favorable direct to markets, would encourage more small farms. The demographic data included in the SF model were Medi an Income , Population Density , and both educational attainment categories , >High School Degree and > Degree . The goal was to determine if these factors have an influence on the choice to farm . independent variables in the SF model. Total Acres restricts the amount of land ultimately available for farming. A larger county would have m ore capacity for small farms and so should be positively correlated with the number of small farms . The environmental characteristic Hardiness Zone Index is a proxy for relative minimum temperature. A higher index value implies a longer growing season, whi ch should also positively correlate with the number of small farms. Finally, the soil designations of Prime Acres , Local Importance Acres , and Unique Importance Acres all indicate salient agricultural land. Higher availability of important farming land sh o uld positively correlate with more small farms. Weight Matrix In all of the spatial models estimated, a weighting matrix is needed for the spatial correlation components . For county i , counties closest to county i are likely most
41 spatially correlated with county i . The weighting matrix, W , contains assumptions about how this correlation dissipates as counties get farther and farther from county i . Common weighting matrixes include using a 0 1 measure to only account for spatial correlation between adjacent neighbors . Other matrixes make use of the inverse of the distance between t w o observations, or transformations of the inverse. Different weighting matrixes could have been used to model each of the spatial dependencies ( ) . H owever, the inverse of the distance s between the centers of each county was used for all weighting matrixes. The distances were determined using a county level dataset from the Florida Geographic Data Library (FGDL) . 2002 Census Counties provides an ArcGIS compatible spatial seven counties (FGDL, 2014). These spatial references are two dimensional, polygon shape files. Using the centroid command in ArcGIS to establish the geometric center of each county, the centroid and those of its sixty six neighboring counties were calculated. The inverse of these distances were used to create a 67 x 67 symmetric matrix to be used as the Region al Analysis The county level information for the entire state overlooks interesting regional South Florida regions. The South Florida region consists of the thirty one count ies that are approximately south of Marion County, Florida and the city of Ocala. The North seven counties. Table 4 2 s thirty six counties.
42 A quick review reveals that South Florida has more than twice the number of of farms and nearly two thirds few er total farmed acres. While medium incomes and graduation rates are similar between the two regions, the South has four times the population density. It is interesting that all of the best soil for farming is located in the North while a larger share of t he farmed acreage is in the South. The panel dataset was compiled from a large number of sources and provides a good view of the wide ranging a gricultural, demographic, and environmental factors present within the peninsula . Comparing North and South Florida highlights the contrast between the two regions of the state . It also indicates that the southern region looks more like the state as a whole. The results of the FM model buttress this observation by establishin g that the South Florida region drives the overall statewide results of the model s .
43 Table 4 1 . Summary Statistics of County level Data Mean SE Min Max Dependent Variables Farmers' Markets 1.40 2.76 0.00 31.00 Farms <50 Acres 461.00 537.12 11.00 2,968.00 Farms <180 Acres 594.91 625.14 12.00 3,598.00 Agricultural Total Number of Farms 690.64 675.89 15.00 3,870.00 Farms 50 to 180 Acres 133.91 123.80 0.00 663.00 Farms >180 Acres 95.74 77.76 0.00 425.00 Farmed Acres (1,000s) 148.71 150.33 0.10 652.67 Demographics Population (1,000s) 257.81 418.38 6.65 2,551.29 Medium Income (1,000s) 38.44 9.95 21.98 94.13 Population Density 312.48 506.71 8.20 3,388.90 > Degree (%) 17.86 8.74 1.40 43.19 >High School Degree (%) 78.97 7.73 54.48 93.32 Environmental Hardiness Zone Index 3.27 1.14 1.50 6.00 Prime Acres 16,615.70 41,443.94 0.00 250,570.00 Local Importance Acres 21,670.19 47,910.76 0.00 192,500.00 Unique Importance Acres 90,577.93 156,063.24 0.00 744,220.00 Total Acres 512,519.40 222,131.68 66,000.00 1,286,600.00
44 Table 4 2 . Summary Statistics of County level Data by Region North Florida South Florida Difference in Means Mean SE Mean SE t test Dependent Variables Farmers' Markets 0.88 1.35 2.01 3.71 3.42 *** Farms <50 Acres 305.50 421.90 641.58 598.43 5.37 *** Farms <180 Acres 443.06 519.97 771.24 689.68 4.43*** Agricultural Total Number of Farms 528.26 569.54 879.22 740.14 4.38*** Farms 50 to 180 Acres 137.56 126.58 129.66 120.86 0.52 Farms >180 Acres 85.19 72.57 107.98 81.98 2.41** Farmed Acres (1,000s) 81.77 67.45 226.45 180.07 8.94*** Demographics Population (1,000s) 95.53 148.63 446.27 536.34 7.52*** Medium Income (1,000s) 37.42 11.33 39.62 7.93 1.82* Population Density 129.66 192.76 524.78 655.43 6.90*** > Degree (%) 15.73 9.24 20.33 7.42 4.45*** >High School Degree (%) 78.15 6.91 79.91 8.52 1.86* Environmental Hardiness Zone Index 2.35 0.47 4.34 0.64 29.20*** Prime Acres 30,902.06 52,562.03 25.10 138.02 6.54*** Local Importance Acres 34,956.47 58,233.53 6,240.97 24,363.42 5.12*** Unique Importance Acres 4,491.53 17,917.56 190,549.23 183,697.78 12.09*** Total Acres 463,761.11 151,487.39 569,141.94 272,789.23 3.98*** Notes: *, **, and *** indicate significance at the 10%, 5%, and 1% level, respectively.
45 Figure 4 1 . ARS Hardiness Zone Map
46 CHAPTER 5 RESULTS For both the FM and SF models, all 5 spatial models were estimated. In cases where the spatial parameters were insignificant or the MLE estimation failed to converge, the spatial specification was discarded. Estimations were conducted using both the random and fixed effects specifications for the SAR, SDM, and SEM models. The Hausman test was then run to determine which model was more appropriate. The GSPRE model was only estimated using the random effects specification and the SAC formulation was only esti mated using the fixed effects specification. model at the statewide level, all possible spatial models were estimated. Time interaction variables where included to check for structural changes over time. E ach year of interaction variables were tested for joint significance, and years for which the interaction terms were insignificant were removed from the model. Only the linear Spatial Autoregressive (SAR) m odel and the linear Generalized Spatial Random eff ects (GSPRE) m odel correctly modeled the spatial relationships in the data . The spatial parameters in the other models were insignificant. While it would have been optimal to estimate a model that allows for nonzero values of , , and in a combined m odel, the xsmle subroutine does not allow for such a model. 1 Table 5 1 presents the results of the two spatial models. Results are robust across both spatial models. The variables pertaining to the number of farms are of 1 The fixed effects speci fication for SAR did not converge to a solution using maximum likelihood estimation (MLE) and GSPRE is by definition a random effects specification. Consequently, random effects models were used for both spatial models.
47 primary interest to this analysis . Interestingly, the coefficients on all three variables representing the count of farms are statistically insignificant before considering the year interactions . While the number of farms in each size category is insignificant, total Farmed Acres are posi tively correlated with the number of markets. This supports the earlier hypothesis that having a good supply of farm market products nearby would likely encourage the establishment of market outlet s for those products and is consistent with the finding s of both Morgan and Berning (Morgan, 2001 and Berning, 2003) . A one standard deviation change in the number of Farmed Acres has an A mong the other variables, Population size and the percentage of residents with a bachelor degree ( > Degree ) were positively correlated and significant demographic factors at the 10% level . These results are also consistent with the findings of Morgan and Berning (Morgan, 2001 and Berning, 2003). For both factors, one standard deviation change has the equivalent impact of approximately 1/3 of a , which agrees with Gallons ( 1997) . While the base effect of Farms <50 Acres and Farms 50 to 180 Acres are insignificant, in 2012, the effect of Farms <50 Acres on number of markets become s positive while the effect of Farms 50 to 180 Acres becomes negative. Berning (2013) identified these same relationships in his research. It appears that farms less then 50 acres in size might be attrac ting to their area with a one standard deviation change in the number of Farms <50 Acres increasing the number of markets in a county by a factor of 1.5 . At the same time, somewhat larger farms, farms that are less likely to target scale effects have a negative effect with a
48 one standard deviation change in the number of Farms 50 to 180 Acres decreasing the number of markets in a county by a factor of 1.2. This is likely due to constraints on agricultural land . More intermediate sized farms reduce the amount of land remaining D ummy variables for 2002, 2007, and 2012 were included in the model to identify any time trends. Both 200 2 and 2012 were positive and significant indicating a natural The variable 2007 was not significant. This was likely due to the onset of the recession that began in 2007, which would have had a negative i s are more susceptible to poor economic conditions because they do n o t typically compete on price. In terms of other structural changes, the interactions with the 2002 dummy variable were jointly insignificant so we re excluded from the model. The 2007 and 2012 interactions were jointly statistically significant, indicating that some structural changes have occurred over time. While the base effect of Population Density is insignificant, in 2007 Population Density has a positive effect on The spatial parameters in both models are statistically significant at the 0.01 level . The strongly significant and negative indicate s markets in surrounding counties has a negative effect on the number of markets in a county. This is likely due to the effects of competition. T he existence of established markets in neighboring counties makes it more difficult to open a new market and draw a sustainable number of both shoppers and farm ers. The negative indicates a negative spatial correlation across error term s and the positive indicates positive
49 spatial correlation in the county level unobserved effects . This latter correlation may be the result of cultural effects that increase demand for markets and are likely correlated across neighboring counties. The recent effect of small farms positively influencing the location pattern of is robust across specifications at the state level and similarly observed in South Florida when estimating the models by region . The analysis was run by state regions using the same models and methodology . T he se regional FM results are summarized in Table s 5 2 and 5 3 . The key outcome of the regional analysis was an indication that the southern region of the state appears to be driving the statewide results. Unlike the results of North Florida, become positively corre lated with small farms in recent years. 2 Overall, the South results were similar to those statewide ; b oth and were significant and negative. However, u nlike at the state level, is in significant for South Florida indicating no spatial correlation in the county level unobserved effects . The other results of the South Florida model resembled the state level analysis, only with approximately twice the impact. The base effect of Farms <50 Acres and Farms 50 to 180 Acres are insignificant, a nd Farmed Acres is positively correlated with the A one standard deviation change in the number of Farmed Acres in South Florida has an equivalent impact of more than 2/3 of a In 2012, the effect of Farms <50 Acres on number of markets become s positively Farms 2 It should be noted that by dividi ng the state into regions, each regional model utilizes about half of the data as the model at the state level.
50 50 to 180 Acres becomes negatively correlated. A one standard deviation change in the number of Far ms <50 Acres increases the number of markets in a county by a factor of 2.8, while a one standard deviation change in the number of Farms 50 to 180 Acres decreases the number of markets in a county by a factor of 2.4. Estimation of the model for the North Florida region revealed no spatial significance and exhibited little explanatory value among In terms of base effects, o ( > Degree ) was significant , indicating a positive effect of a more educated population on the number of markets . A one standard deviation change in the percentage of the population with a degree increase s the number of markets in the county by a factor of 0.6. While t he North Florida regional had less overall explanatory power, the model did identify some structural change in 2012 with regards to the effects of Population , Farmed Acres , and Population Density . The changes indicate an increased positive effect of education. As with the South region, observations were limited in the analysis and may lack the degrees of freedom to precisely fit the estimates. Small Farms Model For the Small Farms (SF) model at the statewide level, the f arms of less than fifty a cres definition of small far ms was considered first. Again, all spatial models were estimated and those models without statistically significant spatial parameters were discarded. As was the case for FM, the Spatial Autoregressive (SAR) and the Generalized Spatial Random effects (GSP RE) models best specify the spatial relationships. However , this time a log model was preferred to the linear model, based
51 on comparison of R 2 values and the ease of variable interpretation . 3 Both of these random effects models exhibited high R 2 values. R esults are presented in Table 5 4. The primary variable of interest is , and it was not statistically significant in terms of its base value or when interacted with the 2007 or 2012 year dummy variables. This suggests that markets do not encourage the development of more small farms in the surrounding area. Not surprisingly, the number of intermediate and large farms in a county are positively correlated with Farms <50 Acres . The presence of intermediate and large farms leads to var ious kinds of infrastructure , which are necessary for the viability of small farms. A one percent increase in the number of Farms 50 to 180 Acres would increase the number of Farms <50 Acres by 0.5 percent; this would translate to a 44.0 percent increase i n Farms<50 for a one standard deviation increase in Farms 50 to 180 Acres or approximately 55 small farms. A one percent increase in the number of Farms >180 Acres would increase the number of Farms <50 Acres by 0. 4 percent; this would translate to a 32.1 percent increase for a one standard deviation increase in Farms 50 to 180 Acres or approximately 25 small farms. Other significant explanatory variables include Population Density , county size ( Total Acres ) , and Farmed Acres . These indicate that counties with a high er population density , more total acres, and fewer agricultural acres would likely have more farms of less than 50 acres. Population Density and fewer Farmed Acres indicate more urbanized areas so the relationships found likely capture relationships with hobby farms. 3 The MLE of the SAR fixed effects specification converged to a result, but showed little significance and failed to meet the asymptotic assumpt ions of the Hausman test. The model was discarded in favor of the random effects specification.
52 The interactions of the time dummy variables with the explanatory variables indicate that there has been some structural change in relationship s during the last decade. A test for joint significance of interactions by year indicated that only the 2007 and 2012 sets of interactions were statistically significantly different than zero. Some of these structural changes indicate a change in location of small farms to areas wi th locally important acreage and away from areas with land designated as Uniquely Important Acres . As of 2012, small farm operators appear to be moving south to the warmer areas of the state as indicated by the positive coefficient on the Hardiness Zone In dex . For this model, two of the spatial correlation parameters were significant. A positive indicate s that the presence of small farms in the larger region encourages small farms in the observed county. This is likely due to the effects of infrastructur e, knowledge, and a shared sense of community. Having the retail outlets, agricultural skills and services and the local knowledge of how to farm encourages more small farms. The negative indicates spatially correlated error term s. For this model, wa s found to be insignificant, indicating no spatial correlation across the county level unobservable terms . Finally, the Small Farms (SF) m odel at the statewide level using a definition of small farms as Farms <180 Acres was analyzed. Again, all spatial mod els were estimated and those models without statistically significant spatial parameters were discarded. As was the case for FM, the Spatial Autoregressive (SAR) and the Generalized Spatial Random effects (GSPRE) models best specify the spatial relationships. A Hausman test was conducted on the estimates generated under both
53 the random and fixed effects specification for the SAR model and the fixed effects formulation was rejected. Both model specifications exhibi ted high R 2 values. R e sults are presented in Table 5 5 . As was the case when small farms were defined as less than 50 acres, when defined as less than 180 acres, the coefficient on is statistically insignificant, both in terms of its base value and for the interaction terms. Under the new small farm designation of less than 180 acres, the presence of large farms is still positively correlated with the number of small farms. A one percent increase in the number of Farms >180 Acres would inc rease the number of Farms < 18 0 Acres by 0.3 percent; this would translate to a 22.1 percent increase in Farms <180 Acres for a one standard deviation increase in Farms >180 Acres or approximately 17 small farms for a county with the average number of Farms >180 Acres . Other positively correlated explanatory agricultural variables include Prime Acres and county size ( Total Acres ) . Like the previous small farm scenario of less than fifty acres, small farms are also positively correlated with Population Density . Unexpected ly, the percentage of high school graduates in a county ( >High School Degree ) i s negative ly correlat ed with small farms. The interactions of the time dummy variables with the explanatory variables indicate that there has been some struct ural change in relationships during the last decade. A test for joint significance of interactions by year indicated that only the 2007 and 2012 sets of interactions were statistically significantly different than zero. Again, small farms seem to be moving to warmer regi ons of the state and making more use of Local Important Acres over time . Small farms also appear to be locating in less
54 population dense regions and using less land designated as Uniquely Important Acres in 2012 relative to previous years . T he statistically significant and positive indicate s that the presence of other similarly sized farms in the larger region encourages small farms in the observed county. This again is likely due to the effect of infrastructure and knowledge in the region small farms. is not significant under this small farm specification indicating no spatial correlation in the error terms. However, is significant and positive. This implies that the county level unobserved effects are spatially correlated . Due to the limited number of observations available for the small farms models, the regional analysis was not included here. Both small farm cate gories require a large number of variables to estimate the SF log model. For Farms <50 Acres , the model requires 39 variables and has only 93 observations available for the South region. This means the model may not have enough degrees of freedom to accura tely fit the estimates. In summary, the results of this research aligned well with the literature in that it This study found support for Morgan markets and income, population size, and farmed acreage. Berning (2013) determined that a combination of market size, education levels, farming activities, and decreased farm size were all The research here supp orts those findings as well . Like Berning, this study also determined a negative relationship between larger
55 Table 5 1 . SAR Model GSPRE Model Variable Coeff SE Coeff SE 2002 0.4294** 0.1872 0.2292** 0.1066 2007 0.2008 2.6089 0.3791 2.6344 2012 6.0898** 2.5882 5.7939** 2.6033 Population (1,000s) 0.0009* 0.0005 0.0007* 0.0004 Farms >180 Acres 0.0021 0.0033 0.0013 0.0032 Farms 50 to 180 Acres 0.0011 0.0021 0.0007 0.0021 Farms <50 Acres 0.0001 0.0004 0.0002 0.0004 Medium Income (1,000s) 0.0154 0.0175 0.0153 0.0168 Farmed Acres 0.0025*** 0.0010 0.0043*** 0.0010 Population Density 0.0000 0.0003 0.0002 0.0003 >High School Degree (%) 0.0202 0.0264 0.0219 0.0248 > Degree (%) 0.0382* 0.0223 0.0476** 0.0212 2007x Population 0.0009 0.0007 0.0008 0.0006 2007x Farms >180 Acres 0.0006 0.0054 0.0013 0.0055 2007x Farms 50 to 180 Acres 0.0004 0.0034 0.0002 0.0032 2007x Farms <50 Acres 0.0003 0.0005 0.0001 0.0005 2007x Medium Income 0.0046 0.0217 0.0063 0.0223 2007x Farmed Acres 0.0010 0.0014 0.0010 0.0014 2007x Population Density 0.0013*** 0.0004 0.0013*** 0.0004 2007x >High School Degree 0.0046 0.0392 0.0019 0.0395 2007x > Degree 0.0217 0.0324 0.0178 0.0339 2012x Population 0.0065*** 0.0008 0.0072*** 0.0007 2012x Farms > 180 Acres 0.0081 0.0054 0.0071 0.0056 2012x Farms 50 to 180 Acres 0.0095** 0.0038 0.0075** 0.0036 2012x Farms <50 Acres 0.0027*** 0.0006 0.0021*** 0.0006 2012x Medium Income 0.0021 0.0215 0.0107 0.0224 2012x Farmed Acres 0.0061*** 0.0016 0.0066*** 0.0015 2012x Population Density 0.0004 0.0004 0.0000 0.0004 2012x >High School Degree 0.0515 0.0381 0.0694* 0.0380 2012x > Degree 0.0028 0.0321 0.0098 0.0330 Constant 1.4406 1.7085 2.2008 1.6094 Spatial rho 1401.9031*** 479.5713 lambda 3234.1602** 1599.8225 phi 5165.7276*** 344.3522 N 268.00 268.00 R 2 0.8638 0.8432 Notes: *, **, and *** indicate significance at the 10%, 5%, and 1% level, respectively.
56 Table 5 2 . SAR Model GSPRE Model Variable Coeff SE Coeff SE 2002 0.9664** 0.3894 0.6684* 0.3578 2007 3.5991 5.2875 6.1390 5.8107 2012 4.5626 4.8665 0.5911 5.2112 Population (1,000s) 0.0003 0.0006 0.0004 0.0006 Farms >180 Acres 0.0035 0.0047 0.0062 0.0051 Farms 50 to 180 Acres 0.0002 0.0038 0.0010 0.0043 Farms <50 Acres 0.0003 0.0007 0.0004 0.0008 Medium Income (1,000s) 0.1104 0.0766 0.1501* 0.0884 Farmed Acres 0.0049*** 0.0011 0.0051*** 0.0013 Population Density 0.0003 0.0003 0.0001 0.0003 >High School Degree (%) 0.0207 0.0327 0.0253 0.0360 > Degree (%) 0.1245 0.0810 0.1572* 0.0953 2007x Population 0.0010 0.0009 0.0005 0.0010 2007x Farms >180 Acres 0.0092 0.0110 0.0011 0.0119 2007x Farms 50 to 180 Acres 0.0060 0.0078 0.0003 0.0083 2007x Farms <50 Acres 0.0007 0.0010 0.0006 0.0012 2007x Medium Income 0.1220 0.1125 0.1629 0.1299 2007x Farmed Acres 0.0010 0.0021 0.0001 0.0024 2007x Population Density 0.0018*** 0.0005 0.0017*** 0.0006 2007x >High School Degree 0.0333 0.0599 0.0416 0.0648 2007x > Degree 0.1211 0.1345 0.1666 0.1559 2012x Population 0.0057*** 0.0011 0.0056*** 0.0013 2012x Farms >180 Acres 0.0109 0.0079 0.0196** 0.0090 2012x Farms 50 to 180 Acres 0.0198*** 0.0075 0.0296*** 0.0082 2012x Farms <50 Acres 0.0043*** 0.0012 0.0053*** 0.0014 2012x Medium Income 0.3663*** 0.1129 0.3319** 0.1305 2012x Farmed Acres 0.0088*** 0.0021 0.0109*** 0.0024 2012x Population Density 0.0009* 0.0005 0.0009 0.0006 2012x >High School Degree 0.0442 0.0536 0.0303 0.0576 2012x > Degree 0.3948*** 0.1213 0.4105*** 0.1384 Constant 0.2370 2.8561 0.3424 3.2201 Spatial rho 6448.0440*** 1417.661 lambda 6596.855*** 2528.4159 phi 12788.2282 9.724e+08 N 124.00 124.00 R 2 0.9182 0.9111 Notes: *, **, and *** indicate significance at the 10%, 5%, and 1% level, respectively.
57 Table 5 3 . SAR Model GSPRE Model Variable Coeff SE Coeff SE 2002 0.2108 0.1605 0.3272*** 0.1049 2007 0.1311 3.0243 0.1352 3.0417 2012 2.7188 2.9457 2.6668 2.9842 Population (1,000s) 0.0003 0.005 0.0021 0.0051 Farms >180 Acres 0.002 0.0047 0.0028 0.0049 Farms 50 to 180 Acres 0.00 0.002 0.00 0.0019 Farms <50 Acres 0.00 0.0006 0.0003 0.0006 Medium Income (1,000s) 0.0076 0.0129 0.0119 0.0135 Farmed Acres 0.0005 0.0046 0.0018 0.0049 Population Density 0.0005 0.0037 0.0022 0.0037 >High School Degree (%) 0.0118 0.0363 0.006 0.0365 > Degree (%) 0.0617*** 0.0212 0.0554** 0.022 2007x Population 0.0066 0.0055 0.0084 0.0054 2007x Farms >180 Acres 0.0089 0.007 0.0094 0.0073 2007x Farms 50 to 180 Acres 0.0013 0.0026 0.0016 0.0026 2007x Farms <50 Acres 0.001 0.0007 0.0011 0.0007 2007x Medium Income 0.0041 0.0152 0.0032 0.0155 2007x Farmed Acres 0.0103 0.007 0.0109 0.0073 2007x Population Density 0.0045 0.0041 0.0058 0.004 2007x >High School Degree 0.004 0.0463 0.0032 0.0469 2007x > Degree 0.0600** 0.026 0.0515* 0.027 2012x Population 0.0161*** 0.0056 0.0121** 0.0055 2012x Farms >180 Acres 0.0068 0.0069 0.0058 0.0071 2012x Farms 50 to 180 Acres 0.0015 0.0038 0.0004 0.0038 2012x Farms <50 Acres 0.0001 0.0007 0.0004 0.0007 2012x Medium Income 0.0124 0.0145 0.008 0.0152 2012x Farmed Acres 0.0127* 0.007 0.0099 0.0073 2012x Population Density 0.0072* 0.0042 0.0041 0.0041 2012x >High School Degree 0.046 0.0436 0.0498 0.0444 2012x > Degree 0.0009 0.0256 0.0156 0.0276 Constant 0.2547 2.3078 0.1018 2.3002 Spatial rho 998.8436 739.3718 lambda 2260.8127 2416.8602 phi 4680.0374 6048.0262 N 144.00 144.00 R 2 0.7642 0.7529 Notes: *, **, and *** indicate significance at the 10%, 5%, and 1% level, respectively.
58 Table 5 4 . Statewide Small Farms Log Model Results ( Farms <50 Acres ) SAR Model GSPRE Model Variable Coeff SE Coeff SE 2012 3.0691 2.9678 2.0982 2.9033 2007 0.5225 2.8256 0.2898 2.7353 lnFarms >180 0.3950*** 0.0887 0.3366*** 0.0894 lnFarms 50 to 180 0.4762*** 0.0783 0.5914*** 0.0743 lnFarmers' Markets 0.0505 0.0963 0.0151 0.0912 lnHardiness Zone 1.0240*** 0.2901 1.1047*** 0.2831 lnPrime Acres 0.0109 0.0189 0.0135 0.0195 lnLocal Importance 0.0009 0.0159 0.0122 0.0157 lnUnique Importance 0.0285* 0.0146 0.0189 0.0153 lnTotal Acres 0.4113*** 0.1329 0.3613*** 0.1357 lnMedium Income 0.0958 0.2781 0.1772 0.2724 lnFarmed Acres 0.1965** 0.0884 0.2069** 0.0858 lnPopulation Density 0.3733*** 0.0696 0.3471*** 0.0706 ln>High School Degree 1.6299 1.1131 0.7104 1.0958 ln> Degree 0.0923 0.2355 0.0241 0.2326 2007xln Farms >180 0.0919 0.0959 0.0778 0.0898 2007xln Farms 50 to 180 0.0357 0.0786 0.0988 0.0709 2007xln Farmers' Markets 0.0736 0.0922 0.1599* 0.0891 2007xln Hardiness Zone 0.2484 0.1879 0.1625 0.1543 2007xln Prime Acres 0.0087 0.0115 0.0005 0.0115 2007xln Local Importance 0.0212** 0.0089 0.0191** 0.0092 2007xln Unique Importance 0.0158 0.0097 0.0257*** 0.0089 2007xln Total Acres 0.0435 0.0858 0.0466 0.0830 2007xln Medium Income 0.0185 0.1876 0.0871 0.1742 2007xln Farmed Acres 0.0656 0.0732 0.1104 0.0694 2007xln Population Density 0.0576 0.0539 0.0493 0.0548 2007xln >High School Degree 0.5194 0.6818 0.3461 0.6498 2007xln > Degree 0.2302 0.1549 0.3048** 0.1536 2012xln Farms >180 0.1023 0.1015 0.0712 0.0961 2012xln Farms 50 to 180 0.0384 0.0880 0.1303 0.0846 2012xln Farmers' Markets 0.0384 0.0929 0.0770 0.0872 2012xln Hardiness Zone 0.5220*** 0.1788 0.4809*** 0.1472 2012xln Prime Acres 0.0090 0.0116 0.0021 0.0120 2012xln Local Importance 0.0035 0.0090 0.0035 0.0092 2012xln Unique Importance 0.0258** 0.0105 0.0441*** 0.0105 2012xln Total Acres 0.0832 0.0857 0.1383 0.0843 2012xln Medium Income 0.0964 0.1915 0.0506 0.1857 2012xln Farmed Acres 0.1207* 0.0731 0.2021*** 0.0704 2012xln Population Density 0.0765 0.0572 0.0689 0.0570 2012xln >High School Degree 1.2080* 0.7309 1.0450 0.7034
59 Table 5 4 . Continued SAR Model GSPRE Model Variable Coeff SE Coeff SE 2012xln > Degree 0.2073 0.1679 0.2081 0.1681 Constant 0.1938 4.5470 1.5849 4.5765 Spatial rho 301.1818* 155.9988 lambda 9302.308*** 2492.6669 phi 118.2120 2338.0989 N 201.00 201.00 R 2 0.8863 0.8831 Notes: *, **, and *** indicate significance at the 10%, 5%, and 1% level, respectively.
60 Table 5 5 . Statewide Small Farms Log Model Results ( Farms <180 Acres ) SAR Model GSPRE Model Variable Coeff SE Coeff SE 2012 2.8050 2.5506 3.0354 2.5106 2007 1.1624 2.4142 1.2897 2.3760 lnFarms >180 0.2727*** 0.0856 0.2475*** 0.0829 lnFarmers' Markets 0.0244 0.0848 0.0100 0.0867 lnHardiness Zone 0.2329 0.3784 0.7278 0.5442 lnPrime Acres 0.0529** 0.0253 0.0430* 0.0243 lnLocal Importance 0.0020 0.0213 0.0003 0.0196 lnUnique Importance 0.0138 0.0191 0.0162 0.0212 lnTotal Acres 0.7380*** 0.1685 0.7621*** 0.1608 lnMedium Income 0.0225 0.3068 0.0082 0.3081 lnFarmed Acres 0.0207 0.0797 0.0114 0.0794 lnPopulation Density 0.5635*** 0.0780 0.4811*** 0.0802 ln>High School Degree 2.7190* 1.4640 3.0287** 1.4293 ln> Degree 0.1954 0.3075 0.0728 0.2949 2007xln Farms >180 0.0454 0.0638 0.0380 0.0628 2007xln Farmers' Markets 0.0645 0.0780 0.0767 0.0803 2007xln Hardiness Zone 0.2792* 0.1513 0.2504* 0.1493 2007xln Prime Acres 0.0098 0.0098 0.0090 0.0098 2007xln Local Importance 0.0223*** 0.0076 0.0218*** 0.0075 2007xln Unique Importance 0.0052 0.0079 0.0063 0.0079 2007xln Total Acres 0.0664 0.0722 0.0672 0.0714 2007xln Medium Income 0.1084 0.1610 0.1138 0.1607 2007xln Farmed Acres 0.0763 0.0603 0.0651 0.0596 2007xln Population Density 0.0550 0.0355 0.0551 0.0348 2007xln >High School Degree 0.5688 0.5797 0.6131 0.5697 2007xln > Degree 0.2355* 0.1338 0.2409* 0.1318 2012xln Farms >180 0.0039 0.0621 0.0002 0.0612 2012xln Farmers' Markets 0.0635 0.0792 0.0741 0.0830 2012xln Hardiness Zone 0.6547*** 0.1468 0.6145*** 0.1439 2012xln Prime Acres 0.0104 0.0099 0.0094 0.0100 2012xln Local Importance 0.0105 0.0075 0.0100 0.0074 2012xln Unique Importance 0.0234*** 0.0080 0.0245*** 0.0082 2012xln Total Acres 0.1195 0.0731 0.1171 0.0746 2012xln Medium Income 0.0060 0.1635 0.0014 0.1619 2012xln Farmed Acres 0.0043 0.0597 0.0051 0.0597 2012xln Population Density 0.1520*** 0.0390 0.1523*** 0.0396 2012xln >High School Degree 1.1717* 0.6276 1.2273** 0.6167 2012xln > Degree 0.1569 0.1494 0.1531 0.1458 Constant 2.7438 6.0899 6.5944 6.0412
61 Table 5 5 . Continued SAR Model GSPRE Model Variable Coeff SE Variable Coeff Spatial rho 627.4741*** 194.7893 lambda 218.9295 1509.5403 phi 4949.6839*** 453.5149 N 201.00 201.00 R 2 0.7765 0.5492 Notes: *, **, and *** indicate significance at the 10%, 5%, and 1% level, respectively.
62 CHAPTER 6 CONCLUSIONS The results of the FM model demonstrate that small farms, and farms in general, have not historically had an impact on the location an markets not only become positively correlated with Farms <50 Acres and negatively correlated with Farms 50 to 180 Acres , but this relationship is likely causal . The use of a lagged Farms <50 Acres variable in the FM model combined with the lack of significance of , also lagged, as an explanatory variable in the SF model demonstrates this causal relationship . From 2012 forward, the number of surrounding the county and appear to form as a result of increases in the number of farms less than 50 acres in size . The resulting impact is not small as the magnitude of number of small farms moves one standard deviation from the mean number of small farms. Another structural shift that has occurred in t he state have become more prevalent in the south ern region of the state. The northern region of the state has fewer small farms ( Farms <50 Acres ) and, on average, more intermediate sized Farms 50 to 180 Acres . In addition, the SF m odels demonstrate that small farms are also shifting south seeking a warmer climate as farmers of small farms transition from using Uniquely Important Acres to producing on Locally Important Acres . This result implies that the northern parts of Florida
63 future when it already has less than half the number of markets as the southern part of the state. Two policy implications arise from this result. First, it provides conf i rmation that there is a causal relationship b etween the location of and the location of small farms. This suggests that markets are responding to the availability of fresh local produce and are seeking to provide a market outlet for small farms in the region. Conversely, it could be t he case that small farms are actively attempting to form For communities in areas with a large number of small farms less than 50 acres , primarily in South Florida, it indicates that they should expec t an increase in the number of markets and should prepare to effectively support them. The second policy implication relates to the northern part of the state, which has less than one half the number of small farms ( Farms <50 Acres ) as compared to South Florida. North Florida also has a larger number of f arms in the 50 to 180 acre category, in the region thereby potentially restricting the availability of locally produced foods. More important may be the implications of the effectiveness of rural development policies in this area of the state. These policies tend to encourage the development of small farms to support of the local economy and, as suggested by these findings, may not be working as effect ively as would be hoped in this part of Florida. This research has also confirmed that do not have an effect on the location and number of small farms. Instead, small farms develop around farming communities with larger local farms. Also, the SF model results suggest that small farms
64 are moving south in the state and transitioning from farming Uniquely Important Acres to producing on Locally Important Acres . This appears to be consistent regardless of whether you define small farms as Farms <50 Acres or as Farms <180 Acres . The faster growth of small farms in the southern part of the state is an interesting observation since the region is four times more densely populated and contains vast regions of environmentally sensitive land. This is e ven more striking when considered with the transition to production on locally important acreage, which is in relative ly short supply in the area. This would suggest impending pricing pressures on an ever shrinking land resource. Considering that this tend ency is relatively new and small farms generally obtain small margins on their production, this appears to be an unsustainable trend. The other concerning trend is that beginning in 2007 small farms have become negatively correlated with more highly educat ed populations as demonstrated by the significant and negative coefficient on >Bachelor Degree. This could indicate that as more residents obtain higher education, they are choosing careers other than farming. A long term implication of this trend could be a loss in innovation in the industry. This research is limited by several factors. First, the analysis was constrained to the state of Florida and may not be applicable to some other areas of the county. Although many of the results of this study align we New England, future research needs to assess these effects generally across all regions of the country. As the regional analysis here demonstrates, the effects of small y even within the state Florida.
65 of the two stage equilibrium outcome entry game : there have to be enough (small) farms as potential market entrants in the area . It would be helpful to better understand what it is about inhibit a In a sense, what are the thresholds that give rise to this associative relationship ? Then, how c ould community leaders identify those thresholds and make the best use of the knowledge in its application to rural development policy ? Finally, some improvements to the dataset would allow better estimates of these ef fects and should be considered for future research. Zip code or Metropolitan Statistical Area (MSA) level data would improve the understanding of geographical relationships and better estimate spatial impacts. These small spatial scales would also provide more observations in each time period, for better estimation of structural changes over time. a simple translation be dr An added benefit of using ARMS data would be a n improvement in the frequency of the agricultur al da ta from five years to annual. Increasing the frequency of observations would impr ov e the estimation ability of the models and provide a better picture of trends over time .
66 LIST OF REFERENCES Agricultural Marketing Resource Center. 2014. arkets web page. AgMRC, U.S. Department of Agriculture, Washington, D.C. Available at http://www.agmrc.org/markets__industries/food/farmers markets (accessed June 2014 ). Agricultural Marketing Service National Farmers AMS, U.S. Department of Agriculture, Washington, D.C. Agricultural Marketing Service National Directory of Farmers Markets AMS, U.S. Department of Agriculture, Washington, D.C. Agricultural Marketing Service National Directory of Farmers Markets AMS, U.S. Department of Agriculture, Washington, D.C. Agricultural Marketing Service 02 FM Directory AMS, U.S. Department of Agriculture, Washington, D.C. Agricultural Marketing Service 2006 Farmers Market Directory Databas e -Release Archive AMS, U.S. Department of Agriculture, Washington, D.C. Agricultural Marketing Service 2008 Farmers Market Directory Database -Release Archive AMS, U.S. Department of Agriculture, Washington, D.C. Agricultural Marketing Ser vice FM Database 2011 AMS, U.S. Department of Agriculture, Washington, D.C. Agricultural Marketing Service Geographic Coordinates for U.S. Farmers Markets AMS, U.S. Department of Agriculture, Washington, D.C. Agricultural Marketing Serv Farmers Market Growth AMS , U.S. Department of Agriculture, Washington, D.C. Available at http://www.ams.u sda.gov/AMSv1.0/ams.fetchTemplateData.do?template=Templ ateS&navID=WholesaleandFarmersMarkets&leftNav=WholesaleandFarmersMar kets&page=WFMFarmersMarketGrowth&description=Farmers%20Market%20Gr owth&acct=frmrdirmkt (accessed June 2014). Agricultural Research Plant Hardiness Zone Plant Hardiness Zone , U.S. Department of Agriculture, Washington, D.C. Available at http://planthardiness.ars.usda.gov (accessed February 2014 ). Anselin, L. 1988. " Spatial E Dordrecht, The Netherlands: Kluwer Academic Publishers. Belotti, F., G . Hughes, and A . P . Mortari. 2013 . XSMLE A Com mand to Estimate German Stata Users Group Meeting , Potsdam, Germany .
67 EA & CAES Joint Annual Meeting, Washington, DC. Bode, E . 2011. "Annual Educational Attainment Estimates for US Counties 1990 2005 ." Letters in Spatial and Resource Sciences 4(2): 117 127 Brown, C. American Journal of Agricultural Economics 90(5): 1296 1302. Bureau of Economic and Business Research . 2014. Demographi Florida Statistical Abstract and Florida St BEBR, University of Florida, Gainesville, Florida . Available at http://www.bebr.ufl.ed u/data/series/catalog/alphabetical/all (accessed April 2014). Conner, D.S., S. Smalley, K. Colasanti, Increasing Farmers Market Patronage: A Michigan Survey Journal of Food Distribution Research 41(2) : 26 35. Darby, K., M.T. Batte, S. Ernst American Journal of Agricultural Economics 90(2): 476 486. Farm Financial and Crop Production Practices web page. ERS , U.S. Departmen t of Agriculture, Washington, D.C. Available at http://www.ers.usda.gov/data products/arms farm financial and crop production pra ctices.aspx#.U6HK6o1dUsw (accessed May 2014). Florida Geographic Data Library . 2014. 2002 Census Counties 2002 Census Counties University of Florida's GeoPlan Center , Gainesville, FL. Available at http://www.fgdl.org/metadata/fgdc_html/cencnt2000.fgdc.htm (accessed December 2013). U.S. Department of Agriculture, Washington, D.C. Available at http://www.fns.usda.gov/ebt/what farmers market (accessed May 2014). G allons, J., U.C. Toensmeyer, J. R . Bacon, and C.L. German. 1997. s of Consumer Characteristics Concerning Direct Marketing of Fresh Produce in Journal of Food Distribution Research 28 ( 1 ): 99 106. Jarosz, L. 2008. The city in the country: growing alternative food networks in Metropolitan areas. Journal of Rural Studies 24 (3): 231 244.
68 LeSage, J.P. and R.K. Pace. 2009. Introduction to Spatial Econometrics. Boca Raton, FL: Taylor & Francis. Manski, C.F. 1993. Identification of Endogenous Social Effects: The Reflection Problem . The Review of Economic Studies 60 (3): 531 542. Marsden , T Journal of Rural Studies 14(1) : 107 117 . Martinez, S. , M. Hand, M. Da Pra, et al. Local Food Systems : Concepts, Impacts, and Is E conomic R esearch R eport 97, E conomic R esearch Service, U.S. Department of Agriculture, Washington, D.C . Monson, J. , D. Mainville, and N. Kuminoff. Analysis of Small Fruit and Specialty Product Markets in Vir Journal of Food Distribution Research 39(2): 1 11. Morgan, T. and D. Alipoe . 2001. Factors affecting the number and type of small farm direct marketing outlets in Mississippi. Journal of Food Distribution Research 32(1): 125 132. National Available at http://www.agcensus.usda.gov/ and http://quickstats.nass.usda.gov/?source_desc=CENSUS (accessed December 2013). Natural Resources Conservation Service. 2014. andbook title 430 VI National Soil Survey H andbook U.S. Department of Agriculture, Washington, D.C. Available at http://www.nrcs.usda.gov/wps/portal/nrcs/detail/soils/survey/?cid=nrcs 142p2_054 242 (accessed April 2014). Natural Resources Conservation Service. 2014. National Cooperative Soil Survey data Web Soi U.S. Department of Agriculture, Washington, D.C. Available at http://websoilsurvey.sc.egov.usda.gov/App/HomePage.htm (accessed March 2014). Ragland , E . and D . Tropp. 2009. USDA National Farmers Market Manager Survey 2006. Agricultural Marketing Service , U.S. Department of Agriculture, Washington, D.C. Renting, H., T. K. Marsden, and J. Banks. 2003 . Networks: Exploring the Role of Short Food Supply Chains in Rural D evelopment. Environment and Planning A 35 (3): 393 411.
69 Sage, J.L. 2012. raphic Exploration of the Social and Economic Sustainability of D issertation , Graduate School , W ashington S tate U niversity. Son nino, R. and T. Marsden . 2006. Relationships Between A lternati ve and Conventional Food N etworks in Europe. Journal of Economic Geography 6(2): 181 199. o. 1073, Oregon State University Extension Service, Corvallis, OR. Uematsu , H. and A . K. Mishra . 2011. Use of Direct Marketing Strategies by Farmers and Their Impact on Farm Business Income. Agricultural and Resource Economics Review 40( 1 ): 1 19 U.S. Census Bureau. 2014. American Community Survey data from Census Bureau, U.S. Department of Commerce , Washington, D.C. Available at http://fa ctfinder2.census.gov/faces/nav/jsf/pages/index.xhtml (accessed May 2014). U.S. Census Bureau. 2014. 1990 & 2000 United States Decennial Census data from USA Counties Census Bureau, U.S. Department of Commerce , Washington, D.C. Available at http://censtats.census.gov/usa/usa.shtml (accessed May 2014).
70 BIOGRAPHICAL SKETCH William Barker earned his Master of Science from the Food and Resource Economics department at the University of Florida in the summer of 2014. He previously earned a Bachelor of Science degree with a double major in economics and applied mathematics from Yale University in the spring of 1996. Born and raised in Plant City, FL, Will graduated salutatorian from Plant City High School and served six years in the United States Navy .