Determinants of wind and solar energy system adoption by U.S. farms:A multilevel modeling approach$

Allison M. Borchers a,n, Irene Xiarchos b, Jayson Beckman a

a Economic Research Service, USDA, USAb Office of Energy Policy and New Uses, Office of the Chief Economist, USDA, USA1

H I G H L I G H T S

� This is the first national examination of wind and solar energy adoption on U.S. farms.� Controlling for state policies distinguishes this study from past research of technology adoption.� We find net metering and interconnection policies increase the likelihood of farm adoption.� Results suggest that the design of renewable energy policies may limit their impact on farms.

a r t i c l e i n f o

Article history:Received 11 November 2013Received in revised form10 February 2014Accepted 11 February 2014Available online 15 March 2014

Keywords:AgricultureRenewable energyTechnology adoptionNet meteringInterconnection

a b s t r a c t

This article offers the first national examination of the determinants of adoption of wind and solar energygeneration on U.S. farming operations. The inclusion of state policies and characteristics in a multilevelmodeling approach distinguishes this study from past research utilizing logit models of technologyadoption which focus only on the characteristics of the farm operation. Results suggest the propensity toadopt is higher for livestock operations, larger farms, operators with internet access, organic operations,and newer farmers. The results find state characteristics such as solar resources, per capita income levels,and predominantly democratic voting increasing the odds of farm adoption. This research suggests therelevance of state policy variables in explaining farm level outcomes is limited, although in combinationbest practice net metering and interconnection policies—policies designed to encourage the develop-ment of small scale distributed applications—are shown to increase the likelihood of farm solar and windadoption. The prevalence of electric cooperatives—which are often not subject to state renewable energypolicies and often service farms—is negatively related with the propensity to adopt and suggests thatpolicy design may be a factor.

Published by Elsevier Ltd.

1. Introduction

Renewable energy (RE) is increasingly being recognized for itsability to off-set rising, and volatile energy prices, decrease carbonemissions through reducing fossil-fuel consumption, and decreasereliance on foreign fuel sources. However the costs of energy fromrenewable fuel sources is often higher than that from conventional

sources, and the institutions and systems which deliver energymay impede development of alternative sources. For preciselythese reasons, federal and state polices are in place to promote theuse of alternative energy for electricity generation by reducinginstitutional barriers and offering financial incentives. The use ofalternative fuels for electricity has been increasing in the lastdecade both for utility scale electricity production and smaller on-site consumer applications, often referred to as distributed gen-eration. Notably, wind generation in the electric power sectorincreased an average 33 percent year-on-year since 2000, whilesolar has seen a recent surge with an average 27 percent annualincrease since 2007 (U.S. Energy Information Administration,2011a). Photovoltaic solar installations outside of the electricpower sector (also referred to as customer sited installations)doubled in capacity from 2008 to 2009 and increased again by 62percent from 2009 to 2010 (Sherwood, 2010, 2009). Small wind

Contents lists available at ScienceDirect

journal homepage: www.elsevier.com/locate/enpol

Energy Policy

http://dx.doi.org/10.1016/j.enpol.2014.02.0140301-4215 Published by Elsevier Ltd.

☆The views expressed here are those of the authors and may not be attributed tothe Economic Research Service, Office of Energy Policy and New Uses, Office of theChief Economist, or the US Department of Agriculture.

1 Some of the work was performed while on detail at the Economic ResearchService, USDA.

n Correspondence to: USDA/ERS/RRED/FEB, 355 E Street, SW, Washington, DC20024-3221, USA. Tel.: þ1 202 694 5548.

E-mail address: aborchers@ers.usda.gov (A.M. Borchers).

Energy Policy 69 (2014) 106–115

systems—turbines with capacity ratings less than or equal to100 kW that are suitable for residential, farm or other on-sitegeneration—has had equally impressive increases over the lastdecade showing an average increase in capacity of 41 percent year-on-year since 2001 (American Wind Energy Association, 2011).

Farming operations are a natural fit for small scale alternativefueled generation technologies. Wind and solar applications, likeother RE options, can help farming operations stabilize electricityand energy expenditures, decrease carbon emissions, and increaseagricultural production (for example, a solar powered water pumpcould make irrigation possible where extending the electric gridmay be prohibitively expensive). Indeed, the American Wind EnergyAssociation notes the industry considers the farm sector a largemarket opportunity (). However, the occurrence of RE applicationsremains rare on US (American Wind Energy Association, 2011)farming operations. Little is known about the characteristics of thefarming operations employing alternative fuel sources for electricityat the national level, and the impact of existing state policies inpromoting RE on farms has only recently started being investigated(Xiarchos and Lazarus, 2013). The objective of this article is toestimate determinants of adoption of wind and solar technologieson US farms while accounting for the influence of state policies onadoption rates.

The inclusion of state policies and characteristics in a multilevelmodeling approach distinguishes this study from past researchutilizing logit models of technology adoption which focus on thecharacteristics of the farm operation only. Using a multilevelmodeling approach (also known as mixed effects or hierarchicalmodeling) allows for estimation of farm-level and state-levelfactors which influence adoption of renewable technologies aswell as empirically identifying the relative importance of farm andstate-level factors in explaining the model variance.

This article analyzes data from the first national survey of farmoperators about renewable energy—including solar and wind—pro-duction on farming operations. Analysis is based on data from the2007 Census of Agriculture and the 2009 follow on On-farm Renew-able Energy Production Survey (OFREPS) (National AgriculturalStatistics Service, 2009b, 2011). Until recently, no nationally repre-sentative sample of solar and wind technologies on U.S. farms wasavailable, and therefore this article presents the most comprehensivepicture of farm operation adoption of renewable technologies to date.

This research provides insights into factors that are correlatedwith adoption of solar and wind technologies on U.S. farms andexamines the relative importance of state characteristics andpolicies on farm-level decisions. In agreement with other studiesexploring the adoption of new technologies on farms we find farmsize, internet availability, organic practices and farming as aprimary occupation positively influence the probability a farmoperation adopting RE technologies. Livestock operations are morelikely to adopt wind or solar technologies than grain and oilseedfarms, while the number of years an operator has been farmingnegatively influences the probability of adoption. Evidence isfound that in combination best practice net metering and inter-connection polices can increase the likelihood of farm adoption.However, no evidence is found that other state policies orincentives significantly impact adoption.

1.1. Literature review

There exists extensive literature examining farm adoption ofnew technologies and practices (for example, Banerjee et al., 2009;Bergtold et al., 2012; D'Souza et al., 1993; Kutter et al., 2009; Lewiset al., 2011; Prokopy et al., 2008). For example, Daberkow andMcBride's (2003) estimation of adoption of precision agriculturetechnologies on US farms finds education, computer literacy, full-time farming and farm size, type and location all positively

influenced the likelihood of adoption. Soule et al. (2000) find farmsize, farmer education, and highly erodible land increased thelikelihood of conservation tillage being employed on a field. Foltzand Chang (2002) find larger farms with younger more educatedoperators are more likely to adopt rBST—a hormone found toincrease milk production per cow.

Literature specifically addressing farm adoption of solar andwind technologies is more limited. Beckman and Xiarchos (2013)examine the determinants of renewable energy adoption andsystem size on California farms and find the probability ofadoption increases with non-farm income sources, electricityprices, and organic practice; it decreases with the number of yearsfarming. Xiarchos and Lazarus (2013) find state-level adoptionrates are positively correlated with higher rates of organic farms,internet access and land ownership. Other studies examininghousehold adoption of RE technologies include a study by Williset al. (2011) which found household adoption of these renewabletechnologies to be significantly determined by the age of thehousehold, where older households are less likely to adopt renew-able energy technologies. A study in Germany finds new residen-tial structures are more likely to adopt solar hot water heatingsystems, but limited evidence of household characteristics asdeterminants of adoption (Mills and Schleich, 2009).

However, it is not farm and household characteristics inisolation which should be expected to determine technologyadoption. National and state enacted policies are designed todirectly influence the decision parameters of adoption throughprice incentives and institutional changes. These policies vary fromstate to state, and therefore some states or policies may be moreeffective at encouraging RE adoption. State policies, such as aRenewable Portfolio Standards (RPS), have been shown to beeffective in increasing utility scale capacity of renewable genera-tion (Menz and Vachon, 2006; Shrimali and Kniefel, 2011; Yin andPowers, 2010). Xiarchos and Lazarus (2013) examination of policyinfluences find that higher RPS goals increase state-level adoptionrates of solar and wind, while increases in electric cooperativeservice shares in the state decrease adoption rates. The inclusion ofstate-level characteristics distinguish this study from past researchof technology adoption which generally focuses on characteristicsof the farming operation. In addition, looking at how state-levelpolicies are correlated with customer-level adoption distinguishesthis study from other research examining the effectiveness of stateRE policy measures.

The remainder of the paper first offers an overview of therenewable energy technologies studied in this paper (solar andwind), their on-farm applications, and state policies designed toencourage customer sited renewable energy adoption. The nextsection outlines the methodology and then the model specifica-tion. Finally, the empirical results are presented with a discussionof the policy implications.

2. Materials and methods

2.1. On-farm renewable energy production survey

Electricity represents around 18 percent of total energy con-sumed on-farm and 2.5 percent of average farm expenditures(USDA-NASS, 2009). Generating electricity using renewable fuelssuch as solar and wind on-farm could reduce these electricityexpenditures as well as expenditures on other fuels if renewableenergy is used to replace them. On-farm2 applications include

2 On-farm renewable energy applications are distinguished from commercial(utility scale) applications such as wind turbines located on farm operations underwind rights lease agreements. This is consistent with the definition in the survey

A.M. Borchers et al. / Energy Policy 69 (2014) 106–115 107

grid-connected installations such as wind turbines and solarpanels for electricity production to offset power purchases fromthe electric utility. On-farm generation would also insulate theproducer from energy price fluctuations. On-farm generation alsomay offer farm operations energy reliability in areas where electricoutages are more common. Further, farms often desire an energysupply in instances where access to the electric grid is difficult orimpossible. Agricultural producers have been early adopters ofsome renewable fuel powered technology due partly to its con-venient application for small and remote power needs. Theseapplications might include electric fencing, water pumping andirrigation. Renewable fuel options could provide the producer asubstitute for remote fossil fuel needs on the farm, reducingtransportation and maintenance costs as well as reducing envir-onmental concerns (Xiarchos and Vick, 2011). On-farm renewableenergy applications of solar and wind power are increasing(National Agricultural Statistics Service, 2011), but little concreteevidence is known about the characteristics of farms where it isoccurring.

The 2009 On-Farm Renewable Energy Production Survey(OFREPS) is the first national survey of farm operators to obtaininformation on renewable energy production on farming opera-tions (National Agricultural Statistics Service, 2011). The OFREPSwas conducted as an add-on survey for operations who indicatedon the 2007 Census of Agriculture that they had produced someform of renewable energy on the farming operation. The surveyprovides a picture of the national prevalence of the use of renew-able technologies for energy, and allows this analysis of farmoperation characteristics which increase the probability of adopt-ing renewable energy. In this analysis we focus on wind and solarelectric applications, not methane gas which is also included in thesurvey, as methane gas technologies are specific to large livestockoperations and have received considerable attention in the litera-ture (for example, Gloy and Dressler, 2010; Key and Sneeringer,2011; Leuer et al., 2008). It is also the least prevalent form of on-farm renewable energy. Based on the OFREPS, the estimatednumber of farms that produce energy from solar, wind or methanedigesters was 8569 in 2009 (National Agricultural StatisticsService, 2011). Methane digesters were reported on 121 farmsacross the country, and wind turbines were reported in 1420farming operations. Solar energy production is the most prevalentform of renewable energy with an estimated 7968 farms, or 93percent of all farms with renewable energy generation (Fig. 1).

The prominence of solar technology as a renewable energysource on farms is not surprising. Solar technologies can offermany applications convenient for agricultural operations—espe-cially in remote locations where commercially-supplied electricityis prohibitively expensive. One use of off-grid electric power isirrigation pumps. The 2008 Farm and Ranch Irrigation Surveyestimates 1405 irrigation pumps are powered by solar and otherrenewable energy sources (National Agricultural Statistics Service,2009). Other studies have shown consumer preference for solarenergy over other sources of RE (Borchers et al., 2007).

The majority of on-farm solar applications were located inCalifornia, Texas, and Colorado. Solar electricity production capa-city differs substantially across farms. The average installed gen-erating capacity in North Dakota, Oklahoma and Kansas was under500 watts, compared with an average capacity of more than11,000 W in California, New Jersey, and Delaware. In CaliforniaBeckman and Xiarchos (2013) show that the size of the installedrenewable energy systems depends on farm operation

characteristics such as value of production, value of land owned,acreage, primary occupation, and the presence of a hired manager.

Wind is the second most prevalent renewable fuel source, 17percent of farms reporting renewable energy generation havewind generating capacity (Fig. 1). Interestingly, of the farms withwind generation capacity it is estimated that 66 percent of thesefarms also have solar panels. The states with the largest amount ofon-farm wind production are California, Texas, Colorado, andMinnesota. The average installed generating capacity of smallwind installations—5.7 kW per farm—is greater than the averagefarm installed solar capacity—4.5 kW per farm (NationalAgricultural Statistics Service, 2011). This is likely a function bothof the relative scalability of the technologies as well as the cost.

2.2. State policies for on-farm renewable energy development

Federal and state policies have played an important role in thegrowth of renewable energy production over the last decade.Federal policies have been largely financial incentives in the formof tax credits. These have been important to the growth of utility-scale renewable generation capacity in the USA. Most notably,annual wind capacity additions have followed a boom and bustcycle commonly recognized as a result of the short-term renewaland expiration of the federal production tax credit legislation(Barradale, 2010; Wiser et al., 2007).

However, federal policies are uniformly available across thenation and therefore do not explain the variation in renewableenergy technology adoption between states. State policies promot-ing renewable energy development vary from state to state, andtherefore some states or policies may be more effective atencouraging RE adoption than others.

Renewable Portfolio Standards (RPS) impose a minimumamount of renewable energy sales, or generating capacity thatelectric utilities must provide according to a specified schedule. ByDecember 2009, 29 states and the District of Columbia hadestablished an RPS, though the specified target amount and dateto meet the requirements varied by state.3 A “set-aside,” alsocalled a “carve-out,” is a provision within an RPS that requiresutilities to use a specific renewable resource to meet a certainpercentage of their RPS. While RPS are designed to encourageutility scale investments, these set-aside provisions provide incen-tives for distributive generation such as, solar and small wind.

Fig. 1. Number of farms with on-farm renewable energy production, by fuel, 2009.Source: (National Agricultural Statistics Service, 2011).

(footnote continued)data used here (the On-Farm Renewable Energy Production Survey), and likelyconsistent with the different incentives that drive installation decisions of com-mercial applications versus on-site, distributed, applications.

3 Texas and Iowa had a renewable energy capacity requirement instead of asales requirement as found in the other RPS. Five additional states set a non-binding renewable energy goal.

A.M. Borchers et al. / Energy Policy 69 (2014) 106–115108

Sixteen states and the District of Columbia have such set-asidesimplemented (Database of State Incentives for Renewables &Efficiency, 2011).

Net-metering policies are specifically designed to encouragethe development of small scale distributed applications of renew-able fuel generation. These policies allow retail electric customersto be compensated by their electric utility for electricity generatedin excess of what they consume. Forty-one states and the Districtof Columbia had net metering policies in 2008. The specific rulesvary in design and economic return from state to state. Addition-ally, 14 states excluded electric cooperatives—the electric utilitiesthat most often service farmers and ranchers—from the netmetering legislation (Xiarchos and Vick, 2011).

Interconnection standards establish clear and uniform techni-cal requirements for small distributed generation systems con-necting to the electric grid, reducing uncertainty and time delaysin customer–utility interaction. Rules again vary by state with 37states and the District of Columbia having implemented inter-connection standards (Database of State Incentives for Renewables& Efficiency, 2011). Additionally, electric cooperatives are notsubject to the interconnection standards in 15 states (Xiarchosand Vick, 2011).

Tax incentives, rebates, and grants are examples of cost basedincentives offered by state and utility programs to encouragecustomer generated renewable energy. Rebates and grants offera payment or discount that reduces the cost of RE installations by acertain amount. For example, in some states residential solarelectric and wind projects may receive up to a certain percentageof total system costs from the state rebate program. Installationtax credits are corporate and personal (income) tax credits thatcan be applied for RE installation expenses. For example, a statetax credit may be a portion of a project's total costs. On the otherhand, production incentives are performance based incentives thatpay installations a per kilowatt hour (kWh) rate based on theelectricity generated. In 2008, eleven states with tax credits,nineteen states with grant and rebate programs and eight stateswith production incentives were identified for small scale dis-tributed generation (Xiarchos and Lazarus, 2013).

2.3. Multilevel logit model

The objective of this study is to analyze the determinants ofsolar and wind technology adoption on US farms examiningfarmer, farm operation, and state characteristics impact on theprobability of on-farm renewable energy generation. While tradi-tional models of technology adoption focus on the characteristicsof the farm operation (for example, Daberkow and McBride, 2003;Lambert et al., 2007; Soule et al., 2000) state policies are designedto directly influence the decision parameters of the operators, andpolicies vary from state to state. We employ a multilevel mixedeffects modeling approach to model the characteristics of farmsand farm operators—level 1 effects—which increase the likelihoodof adopting solar or wind generation technologies, while alsoaccounting for the influence of state characteristics—level 2 effects.A random state effect is included to account for state-levelclustering of farm adoption that accounts for the possibility thatfarm probability of adoption is statistically dependent on the statethe farms are located in.

We assume there is an unobserved or latent variable, Adoptn,which represents the expected benefits to a farm of adopting windor solar energy, that generates the observed variable Adopt,identifying the decision to adopt wind or solar energy generatingsystem. The following is assumed to represent the ith farm in thejth state decision to adopt:

Adoptnij ¼ Xijβþmjþεi; ð1Þ

where Xij is a vector of independent variables including farm,operator, and state characteristics that are hypothesized to influ-ence the decision to adopt, mj is a random state-level effect, andεi is a random error term.

Adoptij ¼ 1ðwind or solar energy or both are used to generateelectricity on�farmÞ;if Adoptnij40

¼ 0; otherwise: ð2ÞFarm ij adopts RE technology if Adoptnij40. The probability thatAdoptij¼1 is,

Prob½Adoptij ¼ 1� ¼ Prob½Adoptnij40�¼ Prob½Xijβþmjþεi40�¼ 1–Fð�Xijβ�mjÞ¼ FðXijβþmjÞ; ð3Þ

where F ( � ) is the cumulative distribution function. If the randomerror term εi is assumed to be distributed as a logistic randomvariable then the logit model is used.

The main motivation for employing this modeling approach isthe inherent grouping of farming operations by states. Multilevellogistic regression allows the farm probability to be correlatedwithin groups or areas—in this application, states. For example,two farms within a state are likely not independent, but havesimilar state-level policies, culture and climate impacting theirdecisions and this correlation is important to model. Indeed,evidence exists that adoption of new technologies is encouragedby neighboring adopters (Lewis et al., 2011). This approach allowsfor regression estimates to be corrected for this correlation, butalso allows for the estimation of additional information, such asthe relative importance of state characteristics in explaining thevariation in outcomes. Such models are common in the education(for example, Bowers and Urick, 2011) and health care literature(for example, Leyland, 2005; Merlo et al., 2006), and are increas-ingly being found in other disciplines (for example, Ballas andTranmer, 2012). In the model above the random state-levelintercept, mj, is assumed to be normally distributed, with meanzero and variance s2, which characterizes the variation betweenstates. The use of a random intercept in this specification has theadded advantage of statistical efficiency over the alternative ofspecifying 47 state specific dummy variables (Alaska and Hawaiiare excluded from this analysis).

This estimation method offers information as to the importanceof state level factors, including renewable energy policies relativeto individual farm-level characteristics on the propensity of farmsto adopt. Several measures of these effects are discussed in theresults.

The occurrence of RE adoption on U.S. farming operations israre—less than one percent of farms. While unusual in manyapplications, modeling rare events is common in some literaturesuch as political science (King and Zeng, 2001), medical fields(Andrews, 2001) and even natural disasters (Van Den Eeckhautet al., 2006). Two main concerns arise when working with rareevent data. One, gathering data using a traditional random sampleor population can be disproportionately expensive and timeconsuming when the actual event of interest occurs infrequently.In this application, data gathering is not a limitation as the Censusof Agriculture provides data on the entire population of U.S. farmseliminating the need for any sampling or modeling adjustments toprovide data gathering efficiencies. Second, the predictive powerof logistic models when the outcome variable is rare is known tobe bias downward (King and Zeng, 2001). We avoid this short-coming by presenting statistics and results—for example, oddsratios and median odds ratios—which are not dependent on theproportion of outcomes (Merlo et al., 2006).

A.M. Borchers et al. / Energy Policy 69 (2014) 106–115 109

2.4. Model specification

The on-farm adoption of solar and wind energy technologies ismodeled from the population of all U.S. farms identified by the2007 Census of Agriculture. The dependent variable of the multi-level logit model, Adopt, was specified as binary, equal to 1 if theoperation had on-farm solar or wind or both electric generationtechnologies as reported from the OFREPS, and 0 otherwise. Allindependent variables are summarized in Table 1.

The OFREPS was merged with the 2007 Census of Agricultureto supplement the data with farm and operator characteristics.Independent farm-level variables included farm size and typevariables, operator characteristics, and operation managementattributes. Farm size was measured by the acreage of the operation(in 100s of acres), as well as the total value of production. A basichypothesis regarding technology adoption on farms is that largerfarms will tend to adopt new technologies sooner than smallerfarms given the uncertainty and fixed transaction costs associatedwith these new technologies (Daberkow and McBride, 2003). Totalvalue of production was included as indicator variables fordifferent levels of production due to the wide distribution of farmsizes and this specification allows the effect of farm size to varynonlinearly with adoption. The omitted category is farms with lessthan $1000 in production value. It is hypothesized that increasingvalue of production makes financing RE applications increasinglyfeasible. In addition, the percentage of the acres operated that areowned by the operator is included. It is expected that, all elseequal, increasing the amount of acres owned will increase thelikelihood of adopting RE. One hypothesis is that land owners aremore likely to adopt new technologies. However, since RE applica-tions are not directly tied to the land acreage like other agricul-tural innovations (conservation tillage, for example), it is not clearthe expected impact of this tenure variable.

Farm operation management attributes such as whether theoperation is organic are included. It is expected that the likelihoodof adopting RE technologies is increasing in these variables. Wealso include whether the operation has more than one operator,and the number of days the primary operator works off farm,which may capture whether the operator has the time available togather knowledge, install and manage the new technology(Fernandez-Cornejo et al., 2005).

Farm household income is controlled for with an indicatorvariable for incomes above $50,000. This is the highest incomecategory reported in the Census of Agriculture. Given the highercost of RE applications it may be that higher income householdscan more easily finance the new technology. Finally, whether theoperator has internet access is included. Use of the internet iswidely regarded as a tool to gather information and providesexposure to new ideas and should help facilitate the adoption ofnew technologies.

The operator characteristics, such as primary residence on farmand whether the operator's primary occupation is farming areincluded as indicator variables. The number of years the operatorhas been farming is included.

Farm type heterogeneity is captured with indicator variablesusing the North American Industry Classification System (NAICS)code to categorize farm specialty into (1) cash grains or oilseedsoperations, (2) livestock operations, and (3) vegetables, fruit, nutsand other crops, which is the omitted category (Office ofManagement and Budget, 2005).

State-level factors are hypothesized to play a significant role inthe decision to adopt renewable energy technologies. State-levelpolicy variables (compiled by the Office of Energy Policy and NewUses, USDA), were used here to reflect the prices and policies in2008 to capture the conditions at the time the OFREPS wasadministered.

Net metering allows retail customers to receive utility bill priceadjustments for excess energy production if their RE installation isgrid connected. Interconnection standards help to reduce barriersfor retail customers to connect their RE installations to the grid.Therefore, while both these policies are expected to have a positiveinfluence on the likelihood of adoption the effect of these policiesis expected to be complimentary. Variables are included to capturewhether states have net metering, interconnection or both poli-cies. However, since states' enacted net metering and interconnec-tion legislation varies in efficacy, the occurrence of these policieswas augmented using a report that grades the use of legislationbest practices to limit impediments and encourage customerparticipation (Rose, 2008). This follows the approach used inDoris et al. (2009). A variable is included to indicate only stateswhich have both statewide net metering and interconnection rulesand where both legislations received a grade of ‘A’ or ‘B’ or ‘C’ by thisreport. For the remainder of this paper we refer to these as ‘bestpractice’, following the language from the report which compiledthe grades (Rose, 2008). Similarly, an indicator variable for statesthat had both net metering and interconnection but either one, orboth received a grade of ‘D’ or ‘F’ by the report is included. Finally,indicators for states where only net metering policies or onlyinterconnection policies exist is included. While a study empiricallytesting a more complete characterization of these two policies iscertainly of interest there is a limited number of parameters whichcan be included given the degrees of freedom available from only 48data points (e.g. states). Given the 12–15 individual policy pointsaddressed in the grading metric, using the grades of this report is aneffective way to capture the variation in policies.

A state-level indicator variable is included for rebates, taxcredits, grants, and other cost based financial incentives for solaror wind installations in effect prior to 2008. Similarly, an indicatorvariable captures if a state had production based incentives ineffect prior to 2008 and if the state's RPS has a provision for solaror distributed generation. Because the included policy variableswere all enacted with the intention of promoting RE applications,the probability of farms adopting RE is expected to be increasingwith all policy variables.

Historically, electric cooperatives were formed to service ruralareas and as a result many farms are served by electric coopera-tives. Electric cooperatives may receive special consideration orexceptions in policy. For example, state policies—such as RPS or

Table 1Correlation between policy variables.

Price Net M Inter Cost Prod Co-op Set aside

Price Average retail elec. price 1.000Net M Net metering 0.250 1.000Inter Interconnection rules 0.155 0.646 1.000Cost Cost incentives 0.296 0.150 0.241 1.000Prod Production incentives 0.065 0.050 0.129 �0.037 1.000Co-op Pct elec. co-ops �0.483 �0.202 �0.296 �0.367 �0.387 1.000Set aside DG set-aside in RPS 0.345 0.331 0.428 0.157 0.136 �0.239 1.000

A.M. Borchers et al. / Energy Policy 69 (2014) 106–115110

net metering—may set lower goals and requirements or excludecooperatives from the standards entirely. While the electric utilityfor each observation in the sample is not known, the percentage ofelectric customers in a state served by an electric cooperative isincluded based on data available from the U.S. Energy InformationAdministration (U.S. Energy Information Administration, 2011b).The probability of farms adopting RE is expected to decrease withhigher electric cooperative customer shares in the state.

Several USDA subsidy programs promote increased energyefficiency of rural businesses, farms, and homes. The RenewableEnergy Systems and Energy Efficiency Improvement Program,renamed Rural Energy for America (REAP) in the 2008 Farm Billhas provided financial support to solar and small wind installa-tions. Most of the awards however have been for energy efficiency;for example 74 percent in 2008. From 2001 to 2009 USDA's RuralDevelopment funded 550 solar and small wind projects with atotal of over $17.5 million in funds. We include a variable to controlfor the total dollars spent on REAP projects awards in each state.

The state average retail electric price is included to approx-imate a farm's level of avoided costs through on-farm generationof electricity. While some farm RE applications may be moredirectly avoiding the use of other fuels, such as diesel, stateaverage diesel prices are highly correlated with state average retailelectricity prices (Pearson¼0.73) and only one is used in analysis.

The solar resource potential is included as the state annualaverage daily total solar resource. This was calculated in ArcGISfrom the low resolution data (surface cells of approximately 40 kmby 40 km in size) developed by the NREL's Climatological SolarRadiation Model (National Renewable Energy Laboratory, 2009).Arizona is found to have the highest average state annual solarresource potential at 6.2 kWh/m2/day, and Michigan has the low-est state average potential at 4.2 kWh/m2/day.

State per capita income is included to control for between statewealth effects. A political variable is included to capture policypreferences. This is specified as whether the state voted democrator republican in the 2008 president election, where blue is equal toone if the majority of the state voted democrat.

3. Results and discussion

Table 2 presents summary statistics of the explanatory variables,and compares the means of these characteristics for farmingoperations which had solar or wind generation and those that didnot. A t-test of means across the groups finds that operator and farmcharacteristics differed significantly between farms which hadadopted these technologies and those that did not for a majorityof the characteristics studied. Producers generating renewableenergy own considerably more land per farm than the average U.S. farming operation. Among farm types, cattle farms were the mostcommon farm types to have wind or solar technologies accountingfor 29% of all farms with these technologies. Operations usingorganic methods were more common in the adoption group thanamong all U.S. farming operations, and farms adopting RE hadoperators that, on average, had farmed fewer years.

Based on this statistical test of group means, operationsgenerating wind or solar energy were, on average, in states withhigher electric prices, effective net metering and interconnectionpolicies, and states offering rebates, tax incentives, or other costincentives. These statistics are illustrative, but regression resultsprovide a more controlled test of these effects.

3.1. Multilevel logistic regression results

SAS version 9.3 software was used to estimate the multilevelmixed effects logit regression using the GLIMMIX procedure.

Table 3 presents the results of the regression and Table 4 presentssupporting model statistics. Model 1 shown is an empty modelconsisting of only intercept terms used to illustrate changes inresults as state and individual covariates are added to the model.Using only the intercept terms the proportion of residual variationdue to farm-level variation and the proportion due to state-levelvariation can be estimated. This is termed the intraclass correla-tion coefficient (ICC) or variance partition coefficient (VPC). Tocalculate the ICC—the proportion of the total residual variation dueto unobserved state-level variation—we employ the simulationmethod outlined in Goldstein et al. (2002) for non-linear models.The between-state variance as measured by the ICC is small(Table 3). The percentage of the residual variance attributable tostate-level variation is 1.22 percent, while the remainder of thevariance is attributed to differences in farming operations. Thisresult suggests that adoption decisions are subject to the influ-ences of state factors, but interpreted alone implies that individualfarm characteristics are more important in understanding farmadoption of RE. However, this measure may not be the preferredmeasure for non-linear models, and especially in applications,such as this one, modeling rare events. The ICC is a function of theproportion of events and when the proportion is very low—as inthis application—or very high, the ICC tends to be low (see Merloet al., 2006).

Because of the limitations in using the ICC in nonlinear models,it may be more informative in this application to quantify thevariation between states using the median odds ratio (MOR).Using the odds ratio scale is a sound approach in logistic regres-sion applications, and is often preferable in the interpretation ofdata with low prevalence of event outcomes (Merlo et al., 2006).The MOR is the median of the distribution of odds ratios betweenfarms with the same covariates, but from different states. The MORis easily calculated and directly comparable to fixed–effects oddsratios. This is calculated following Merlo et al. (2006).4 Model 1,the empty model, has a MOR of 3.03 suggesting significantheterogeneity exists between states. This MOR can be comparedto other odds ratios calculated in Models 2 and 3.

3.1.1. Adding state-level covariatesTo explain this state-level heterogeneity Model 2 includes

state-level covariates to capture some of the between-state varia-tion found in Model 1. This specification is presented to illustratethat the inclusion of these variables reduce both the ICC and theMOR compared to Model 1 suggesting that the variables includedare successful at explaining some of the state-level variation foundin the data. Using the empty model as a reference level, theinclusion of the state-level covariates explained 43 percent of thestate variance, as shown in Table 3.

The state control variables for solar resource potential, percapita income and political majority are all significant with theexpected signs. Solar resource potential is significant at thepo0.01 level suggesting higher levels of solar resources increasethe propensity of adoption. Increasing per capita income increasesthe odds of RE adoption and farms in democratic leaning stateshave a higher probability of adoption. State average electric priceis not found to be a statistically significant determinant of farmlevel adoption. This is in contrast to Xiarchos and Lazarus's study(2013) that found electric prices were a statistically significantdeterminant of state-level adoption rates of renewable electricity.

4 Following Merlo et al. (2006) the MOR is calculated as MOR¼exp(√(2 n s2) nФ�1(0.75)), where Ф( � ) is the cumulative distribution function of the normaldistribution with mean of zero and variance of one, and exp( � ) is the exponentialfunction.

A.M. Borchers et al. / Energy Policy 69 (2014) 106–115 111

Farms in states with both best practice net metering and inter-connection policies are significantly more likely to adopt RE technol-ogies. While no evidence is found of impact of these policies whenimplemented alone or when they do not both follow best practices asdefined by the grading completed by Rose, 2008. This finding suggeststhat these policies are complimentary and that following best prac-tices leads to effective policies. Previous studies do not find this result,however we note that past studies have different dependent variablesand that the interaction effects included in this study allow for morenuanced effects (Xiarchos and Lazarus, 2013; Yin and Powers, 2010).Doris et al. (2009) also finds policy interactions are significant in thegrowth of RE development.

The percentage of state electric customers served by an electriccooperative is negative and significant (po0.05), suggestinghigher levels of electric cooperative service territory reduces theprobability of RE adoption. This is in agreement with the findingsreported in Xiarchos and Lazarus (2013) where state solar andwind adoption rates are negatively related to state cooperativeservice shares. This finding is in line with the fact that electriccooperatives are often not subject to state policies that promoterenewable energy and suggests that incorporating electric coop-eratives into RE policy would increase the efficacy of policyincentives on the farming sector.

The measure of USDA program dollars spent on RE projects isnot found to be statistically significant. It may be that the projectspecific awards do not have any spillover effects on adoption of REapplication at other farms. In addition, through 2009 these awardsdid not focus on smaller systems (Xiarchos and Vick, 2011).

Consequently adoption of small solar and wind as modeled in thisstudy may not be correlated with REAP awards. Program changesafter 2009, if the new emphasis and program funding continueunchanged, may make them a more influential factor for smallsystems (Xiarchos and Vick, 2011).

Contrary to existing literature, no evidence is found that RPSset aside increases the likelihood of on-farm RE adoption. We alsotried modeling the relative stringency of the RPS set-aside using acontinuous variable and find similar results. However, we note ourmodel differs from existing literature in that we are modelingfarm-level adoption rather than aggregate state utility develop-ment (Yin and Powers, 2010) or adoption rates (Xiarchos andLazarus, 2013) therefore the difference in results is less surprising.

To understand the size of the effects of covariates in non-linearmodels odds-ratios are often reported. In this multi-levelingmodel since there is no variation for within state comparison,for state-level variables the calculation of odds ratios involvesfarms with different random effects, and requires a slightly morecomplicated measure than the traditional odds ratio—the intervalodds ratio (IOR). The IOR is an interval—here the interval betweenthe 10th and 90th percentiles—containing the distribution of oddsratios of farms differing in state-level covariates, but similar farm-level covariates. The calculation again follows Merlo et al. (2006).5

Table 2Description and comparison of model variable means across groups.

Variable name Description RE producing farms Non-RE Farms

Farm-level variablesLand operated Acres in operation, in 100 s 1.74n 0.41Pct owned Percent of acres owned by operator 1.49 1.59TVP 1K Total value of production 4$1k, o10 K (0,1) 0.29n 0.31TVP 10K Total value of production 4$10k, o100 K (0,1) 0.29 0.29TVP 100K Total value of production 4$100k, o350 K (0,1) 0.28n 0.24TVP 350K Total value of production 4$350k, o700 K (0,1) 0.09 0.09Organic Farm operation is organic (0,1) 0.11n 0.01Grain NAICS code is cash grains or oilseeds operations (0,1) 0.04n 0.15Livestock NAICS code is livestock operation (0,1) 0.57n 0.52Operators More than one farm operator (0,1) 0.60n 0.42Days off farm Number of days operator works off-farm (0,1) 3.09 3.07On farm residence Operator resides on farm (0,1) 0.86n 0.77Farming occupation Farming is primary occupation (0,1) 0.51n 0.45Internet Operator has internet access (0,1) 0.76n 0.56Income 50K Income category 5 (0,1) 0.56n 0.50Years farming Years operator has been farming 17.53n 21.64

State-level variablesAvg elec. price State average retail electric price (cents/kWh) 9.93n 8.70PV resource State average solar potential (kWh/m2/day) 5.23n 5.00Best Practice Netm & Interconnect Both policies following best policy practices (0,1) 0.50n 0.14Not Best Practice NetM & Interconnect One or both not following best policy practices (0,1) 0.38n 0.58NetM only Not effective net metering policy (0,1) 0.05n 0.10Interconnection only Not effective interconnection policy (0,1) 0.01n 0.03Cost incentives Grants, rebates, tax credits (0,1) 0.52n 0.32Production incentives Incentives paid based on production (0,1) 0.36n 0.16DG RPS set aside Distributive generation Set Aside in RPS (0,1) 0.36 0.39Pct elec. co-op Percentage customers served by electric cooperative 11.54n 17.85USDA dollars Dollars distributed to RE projects ($1,000) 534.97n 441.66Per capita income State per Capita Income ($1,000) 40.16n 37.75Blue state Voted democratic in 2008 presidential election (0,1) 0.54n 0.33Number of farms 6525 1,512,027

Note:Results are weighted based on NASS supplied weights.

n Indicates mean values are statistically different across groups at the p40.01 level.

5 Following Merlo et al. (2006) the IOR is calculated as IORlower¼exp(β(x1�x2)þ√(2s2)Ф�1(0.10)), and IORupper¼exp(β(x1�x2)þ√(2s2)Ф�1(0.90)),where Ф( � ) is the cumulative distribution function of the normal distributionwith mean zero and variance of one, and exp( � ) is the exponential function.

A.M. Borchers et al. / Energy Policy 69 (2014) 106–115112

The IOR is not a confidence interval; it follows a log-normaldistribution, therefore the probability the IOR is greater than 1 isreported in Table 4. The interval is narrow if there is little variationin the variable between states and large if the variation betweenstates is large. The IOR of each of the state-level variables is widesuggesting large between-state variation. In addition, most IORscontain the value one, implying the effect of that variable is not aslarge as the remaining heterogeneity. The magnitude of the IORalso provides insight into the relative importance of the statecharacteristics and policy variables and confirms the combinedimportance of net metering and interconnection on adoption.

3.1.2. Adding farm-level and state-level covariatesThe full model with both state and farm-level fixed effects is

shown as Model 3 with a number of significant operator char-acteristics, farm characteristics, and management attributes, aswell as the state-level variables introduced in the specification ofModel 2.6 As shown in Table 2, all farm-level covariates meet

conventional standards of statistical significance, except the tenurevariable. As expected operator characteristics such as living on thefarming operation, internet access and higher household incomeincrease the likelihood of adopting solar or wind generation, whilethe number of years farming decreases the likelihood of adoption.These findings are consistent with much of the technology adop-tion literature (Daberkow and McBride, 2003; Foltz and Chang,2002; Soule et al., 2000) and Beckman and Xiarchos' (2013)examination of on farm renewable electricity adoption in Califor-nia. The number of acres operated is positively correlated with theadoption of RE technologies, and increasing value of productionshows positive, but nonlinear impact on the odds of adoption.

For farm-level variables, the usual odds ratios (OR) apply.7 Forexample, for farms within the same state, the odds ratio of theorganic effect is interpreted as the ratio of odds of two farmswithin the same state, with the same covariates, except for orga-nic certification. The results suggest that, all else equal, an organicoperation is around five times more likely than a non-organicoperation to adopt RE technologies.

Table 3Estimates and standard errors from MLM regression of farm adoption of solar and/or wind energy systems.

Dependent variable: farm adopted RE energy (0,1)

(1) (2) (3)

Farm-level variablesIntercept �5.970 (0.172)nnn �12.822 (2.181)nnn �13.866 (1.935)nnn

Land operated – – 0.001 (0.000)nnn

Pct owned – – 0.000 (0.000)TVP 1K – – 0.440 (0.072)nnn

TVP 10K – – 0.472 (0.069)nnn

TVP 100K – – 0.672 (0.066)nnn

TVP 350K – – 0.626 (0.073)nnn

Organic – – 1.664 (0.053)nnn

Grain – – �0.837 (0.082)nnn

Livestock – – 0.192 (0.034)nnn

Operators – – 0.414 (0.031)nnn

Days off farm – – �0.020 (0.010)nn

On farm residence – – 0.440 (0.044)nn

Farming occupation – – 0.192 (0.036)nnn

Internet – – 0.575 (0.037)nnn

Income 50K – – 0.097 (0.032)nn

Years farming – – �0.012 (0.001)nnn

State-level variablesAvg elec. price – �0.044 (0.072) �0.066 (0.064)PV resource – 1.031 (0.321)nnn 0.995 (0.284)nnn

Best Practice NetM & Interconnection – 1.427 (0.563)nn 1.109 (0.449)nn

Not Best Practice NetM & Interconnection – 0.307 (0.432) 0.210 (0.384)NetM only – 0.627 (0.561) 0.560 (0.499)Interconnection only – �0.627 (1.079) �0.453 (0.956)Cost incentives – �0.160 (0.307) �0.085 (0.272)Production incentives – �0.885 (0.466)n �0.730 (0.413)n

DG RPS set aside – �0.450 (0.343) �0.282 (0.304)Pct elec. co-op – �0.032 (0.014)nn �0.029 (0.012)nn

USDA dollars – 0.000 (0.000) �0.294 (0.304)Per capita income – 0.059 (0.034)n 0.059 (0.030)n

Blue state – 0.924 (0.469)n 0.818 (0.415)n

Random effects (s42)State 1.377 (0.291) 0.789 (0.199) 0.614 (0.157)State random effects Yes Yes YesFarm-level covariates No No YesState-level covariates No Yes Yes

n Indicates statistical significance at the p40.10 levels respectively.nn Indicates statistical significance at the p40.05 levels respectively.nnn Indicates statistical significance at the p40.01 levels respectively.

6 With the random effects approach there still may be state-level variables thatbias the parameter estimates for the farm-level variables. We test the robustness ofthe farm-level variable estimates using a state-level fixed effects model (modelavailable upon request). The results suggest the farm-level estimates are robust to

(footnote continued)these changes in specification. We thank an anonymous reviewer for this helpfulsuggestion.

7 The odds ratios are calculated in the usual manner: OR¼exp(β(x1�x2)).

A.M. Borchers et al. / Energy Policy 69 (2014) 106–115 113

Results suggest livestock operations are more likely to adoptrenewable energy generation systems than vegetable, fruit nutsand other crop producers, while cash grain and oilseed operationsare less likely to adopt these technologies.

4. Conclusion and policy implications

This study offers the first national look at determinants ofadoption of wind and solar energy generation on U.S. farmingoperations. The modeling approach differs from traditional modelsof technology adoption through its incorporation of state policyeffects as determinants of renewable energy adoption, and theability to empirically measure the relevance of between statedifferences in understanding on-farm solar and wind adoption.This study also expands the examination of state policy impacts oncustomer adoption of distributed renewable energy generation.

Results suggest that farm characteristics such as livestockoperations, owned acreage, operators with internet access, organicoperations, and newer farmers increase the propensity to adoptsolar and wind generation. The results also find state character-istics such as solar resources, per capita income levels, andpredominantly democratic voting increase the odds of adoption.Parties interested in expanding RE use on farms could targetefforts on types of farms identified here as more likely to adoptin order to increase the impact of their outreach. Alternatively,given the evidence presented from these results, additional

research could better identify why barriers to adoption might varyacross farm types.

Renewable energy is an explicit goal in USDA's rural develop-ment initiatives, as evidenced by the 2014 Farm Bill whichprovides for over $800 million in continued funding for USDA'srural renewable energy and biofuels programs, and $50 million perannum for REAP. Within this environment of continued supportfor rural renewable energy, this paper can provide direction onhow to plan for effective targeting and policy design.

In terms of state policy implications, in combination netmetering and interconnection policies which follow best practices(designated in this analysis through the use of Rose's, 2008grading system) are shown to increase the likelihood of farm REadoption. This result recognizes the importance of following bestpractices in the design of policy, and further highlights synergiesof successful policies. It is logical that these policies are mosteffective when enacted in combination—net metering providescost incentives for adoption when connected to the utility grid,while interconnection standards reduce institutional and infra-structure barriers to grid installed applications. Disappointingly,results suggest that the relevance of other policy variables like costincentives and RPS in explaining renewable energy adoptiondecisions at the farm level is limited. But this may not besurprising as net metering and interconnection are specificallydesigned to encourage the development of small scale distributedapplications of renewable fuel generation in direct contrast toother RE policies such as RPS which target utility scale invest-ments. These results suggest that further work should focus on the

Table 4Odds ratios for adopting solar and/or wind energy systems.

Dependent variable: farm adopted RE energy (0,1)

(1) (2) (3)

Farm-level fixed effects Odds ratioLand operated – – 1.001Pct owned – – 1.000TVP 1K – – 1.553TVP 10K – – 1.603TVP 100K – – 1.958TVP 350K – – 1.870Organic – – 5.280Grain – – 0.433Livestock – – 1.212Operators – – 1.513Days off farm – – 0.980On farm residence – – 1.553Farming occupation – – 1.212Internet – – 1.777Income 50K – – 1.102Years farming – – 0.988

State-level fixed effectsInterval odds ratio %OR41 Interval odds ratio %OR41

Avg elec. price – 0.191 4.784 0.49 0.226 3.874 0.48PV resource – 0.561 14.023 0.79 0.654 11.192 0.82Best practice NetM & interconnection – 0.833 20.824 0.87 0.733 12.544 0.84Not best practice NetM & interconnection – 0.272 6.794 0.60 0.298 5.105 0.58NetM only – 0.374 9.363 0.69 0.423 7.244 0.69Interconnection only – 0.107 2.670 0.31 0.154 2.631 0.34Cost incentives – 0.170 4.261 0.45 0.222 3.801 0.47Production incentives – 0.083 2.065 0.24 0.116 1.994 0.26DG RPS set aside – 0.128 3.190 0.36 0.182 3.121 0.40Pct elec. co-op – 0.194 4.841 0.49 0.235 4.020 0.49USDA dollars – 0.200 4.998 0.50 0.180 3.084 0.40Per capita income – 0.212 5.305 0.52 0.256 4.389 0.52Blue state – 0.504 12.600 0.77 0.548 9.377 0.77

Random effects Median odds ratio

State 3.05 2.32 2.11Intraclass correlation coefficient 1.22% 0.36% 0.25%

A.M. Borchers et al. / Energy Policy 69 (2014) 106–115114

synergies and importance of grouping well designed policies forincreasing effectiveness. Guidance does exist for developing suc-cessful and effective policies. For example, one source of guidanceon best practices for net metering and interconnection has beenprovided by Freeing the Grid since 2007 (Rose, 2008).

Also, specific to farming operations, the significance of thevariable measuring the prevalence of cooperatives in the state(pct_coop) suggests that treatment of cooperatives could affectfarm level renewable energy decisions. As mentioned above, insome states electric cooperatives, which have historically servedrural areas and farm customers, are excluded from the full impactof renewable energy legislation. Although data limitations pre-clude a more detailed analysis of farms served by electric co-opscompared to those that are not, the results may suggest thatunequal application of renewable energy policies may impact thefarm sector. Additionally future research could evaluate whetherthere is any heterogeneity on policy influence by farm-levelcharacteristics, since some types of farms may be more sensitiveto policy instruments.

References

American Wind Energy Association, 2011. 2010 U.S. Small Wind Turbine Market Report.Available at: ohttp://awea.files.cms-plus.com/FileDownloads/pdfs/AWEA_SmallWind_GMS2011Report_Final.pdf4 (assessed 3.14.2014).

Andrews, N.J., 2001. Statistical assessment of the association between vaccinationand rare adverse events post-licensure. Vaccine 20 (Supplement 1(0)),S49–S53.

Ballas, D., Tranmer, M., 2012. Happy people or happy places? A multilevel modelingapproach to the analysis of happiness and well-being. Int. Reg. Sci. Rev. 35 (1),70–102.

Banerjee, S., Martin, S.M., Roberts, R.K., Larson, J., A., Hogan, R.J.J., Johnson, J.L.,Paxton, K.W., Jeanne, M. Reeves, 2009. Adoption of conservation-tillagepractices and herbicide-resistant seed in cotton production. AgBioForum 12(3 and 4), 258–268.

Barradale, M.J., 2010. Impact of public policy uncertainty on renewable energyinvestment: wind power and the production tax credit. Energy Policy 38 (12),7698–7709.

Beckman, J., Xiarchos, I.M., 2013. Why are Californian farmers adopting more (andlarger) renewable energy operations? Renew. Energy 55, 322–330.

Bergtold, J.S., Duffy, P.A., Hite, D., Raper, R.L., 2012. Demographic and managementfactors affecting the adoption and perceived yield benefit of winter cover cropsin the southeast. J. Agric. Appl. Econ. 44 (1), 99–116.

Borchers, A.M., Duke, J.M., Parsons, G.R., 2007. Does willingness to pay for greenenergy differ by source? Energy Policy 35 (6), 3327–3334.

Bowers, A.J., Urick, A., 2011. Does high school facility quality affect studentachievement? A two-level hierarchical linear model. J. Educ. Finance 37 (1),72–94.

D'Souza, G., Cyphers, D., Phipps, T., 1993. Factors affecting the adoption ofsustainable agricultural practices. Agric. Resour. Econ. Rev. 22 (3), 159–165.

Daberkow, S.G., McBride, W.D., 2003. Farm and operator characteristics affectingthe awareness and adoption of precision agriculture technologies in the US.Precis. Agric. 4 (2), 163–177.

Database of State Incentives for Renewables & Efficiency, 2011. Available at: ⟨www.dsireusa.org⟩ (accessed 11.4.11).

Doris, E., Busche, S., Hockett, S., McLaren, J., 2009. The role of state policy inrenewable energy development. Paper Presented at American Society ofMechanical Engineers Third International Conference on Energy Sustainability.San Francisco, California.

Fernandez-Cornejo, J., Hendricks, C., Mishra, A., 2005. Technology adoption and off-farm household income: the case of herbicide-tolerant soybeans. J. Agric. Appl.Econ. 37 (3), 549–563.

Foltz, J.D., Chang, H.-H., 2002. The adoption and profitability of rbST on Connecticutdairy farms. Am. J. Agric. Econ. 84 (4), 1021–1032.

Gloy, B.A., Dressler, J.B., 2010. Financial barriers to the adoption of anaerobicdigestion on US livestock operations. Agric. Finance Rev. 70 (2), 157–168.

Goldstein, H., Browne, W., Rasbash, J., 2002. Partitioning variation in multilevelmodels. Underst. Stat. 1 (4), 223–231.

Key, N., Sneeringer, S., 2011. Climate Change Policies and the Adoption of MethaneDigesters on Livestock Operations (Rep. ERR-111). Economic Research Service,U.S. Department of Agriculture, Washington, DC.

King, G., Zeng, L., 2001. Explaining rare events in international relations. Int. Organ.55 (3), 693–715.

King, G., Zeng, L., 2001. Logistic regression in rare events data. Political Anal. 9 (2),137–163.

Kutter, T., Tiemann, S., Siebert, R., Fountas, S., 2009. The role of communication andco-operation in the adoption of precision farming. Precis. Agric., 1–16

Lambert, D.M., Sullivan, P., Claassen, R., Foreman, L., 2007. Profiles of US farmhouseholds adopting conservation-compatible practices. Land Use Policy 24(1), 72–88.

Leuer, E.R., Hyde, J., Richard, T.L., 2008. Investing in methane digesters onPennsylvania dairy farms: implications of scale economies and environemtnalprograms. Agric. Resourc. Econ. Rev. 37 (2), 188–203.

Lewis, D.J., Barham, B.L., Robinson, B., 2011. Are there spatial spillovers in theadoption of clean technology? the case of organic dairy farming. Land Econ. 87(2), 250–267.

Leyland, A., 2005. Socioeconomic gradients in the prevalence of cariovasculardisease in Scotland: the roles of compositon and context. J. Epidemiol.Community Health 59, 799–803.

Menz, F.C., Vachon, S., 2006. The effectiveness of different policy regimes forpromoting wind power: experiences from the states. Energy Policy 34 (14),1786–1796.

Merlo, J., Chaix, B., Ohlsson, H., Beckman, A., Johnell, K., Hjerpe, P., 2006. A briefconceptual tutorial on multilevel analysis in social epidemiology: usingmeasures of clustering in multilevel logistic regression to investigate contextualphenomena. J. Epidemiol. Community Health 59, 290–297.

Mills, B.F., Schleich, J., 2009. Profits or preferences? Assessing the adoption ofresidential solar thermal technologies. Energy Policy 37 (10), 4145–4154.

National Agricultural Statistics Service, 2009a. 2007 Census of Agriculture, vol. 2,Subject Series. U.S. Department of Agriculture, Washington, DC.

National Agricultural Statistics Service, 2009b. 2008 Farm and Ranch Land Irriga-tion Survey. U.S. Department of Agriculture, Washington, DC

National Agricultural Statistics Service, 2011. On-Farm Renewable Energy Produc-tion Survey (2009). U.S. Department of Agriculture, Washington, DC

National Renewable Energy Laboratory, 2009. Lower 48 PV Low Resolution. U.S.Department of Energy. Available at: ⟨http://www.nrel.gov/gis/data_solar.html⟩.

Office of Management and Budget, 2005. North American Industry ClassificationSystem-Update for 2007, Notice. Federal Register 70 (47), pp. 12390.

Prokopy, L.S., Floress, K., Klotthor-Weinkauf, D., Baumgart-Getz, A., 2008. Determi-nants of agricultural best management practice adoption: evidence from theliterature. J. Soil Water Conserv. 63 (5), 300–311.

Rose, J. Freeing the Grid: the best and worst practices in state net metering policiesand interconnection procedures. Network for New Energy Choices.

Sherwood, L., 2009. U.S. Solar Market Trends. Interstate Renewable Energy Council.Sherwood, L., 2010. U.S. Solar Market Trends. Interstate Renewable Energy Council.Shrimali, G., Kniefel, J., 2011. Are government policies effective in promoting

deployment of renewable electricity resources? Energy Policy 39 (9),4726–4741.

Soule, M.J., Tegene, A., Wiebe, K.D., 2000. Land tenure and the adoption ofconservation practices. Am. J. Agric. Econ. 82 (4), 993–1005.

U.S. Energy Information Administration, 2011a. Annual Energy Review 2010.Washington, DC.

U.S. Energy Information Administration, 2011b. Form EIA-861 Data Files. Availableat: ⟨http://www.eia.gov/electricity/data/eia861/index.html⟩.

USDA-NASS, 2009. 2007 Census of Agriculture. Volume 1, Geographic Area Series.U.S. Department of Agriculture, Washington, DC

Van Den Eeckhaut, M., Vanwalleghem, T., Poesen, J., Govers, G., Verstraeten, G.,Vandekerckhove, L., 2006. Prediction of landslide susceptibility using rareevents logistic regression: a case-study in the Flemish Ardennes (Belgium).Geomorphology 76 (3–4), 392–410.

Willis, K., Scarpa, R., Gilroy, R., Hamza, N., 2011. Renewable energy adoption in anageing population: heterogeneity in preferences for micro-generation technol-ogy adoption. Energy Policy 30 (10), 6021–6029.

Wiser, R., Bolinger, M., Barbose, G., 2007. Using the federal production tax credit tobuild a durable market for wind power in the United States. Electr. J. 20 (9),77–88.

Xiarchos, I., Lazarus W.F., 2013. Factors Affecting the Adoption of Wind and Solar-Power Generating Systems on U.S. Farms: Experiences at the State Level.Washington, DC.

Xiarchos, I., Vick, B., 2011. Solar Energy Use in U.S. Agriculture: Overview and PolicyIssues. U.S. Department of Agriculture, Office of the Chief Economist, Washing-ton, DC

Yin, H., Powers, N., 2010. Do state renewable portfolio standards promote in-staterenewable generation? Energy Policy 38 (2), 1140–1149.

A.M. Borchers et al. / Energy Policy 69 (2014) 106–115 115