2. Conservation Agriculture Practices in the U.S.

Conservation agriculture practices promote soil health, defined as the continued capacity of a soil to function as a vital, living ecosystem that sustains plants, animals, and humans, by way of minimizing soil disturbance, and maximizing soil cover and biodiversity (NRCS, 2019b). Conservation tillage, a generic term implying all tillage methods that reduce runoff and soil erosion in comparison with plow-based tillage, is known to increase soil carbon sequestration (Lal and Kimble, Reference Lal and Kimble1997). There is strong evidence on the environmental benefits of conservation tillage in the U.S. that includes the protection of soils from erosion and compaction, conservation of soil moisture, improvement in soil structure and stability, and increased presence of soil biota that help to combat pests and diseases (Claassen, Duquette, and Smith, Reference Claassen, Duquette and Smith2018; Fuglie, Reference Fuglie1999; Holland, Reference Holland2004). The economic benefits from conservation tillage include a reduction in production costs through reduced use of machinery, fuel, and labor. However, evidence of the economic benefits is mixed and varies based on crop mix, crop yields, commodity prices, fuel prices, labor availability, and labor costs (Bewick et al., Reference Bewick, Young, Alldredge and Young2008; Fuglie, Reference Fuglie1999; Jolly, Edwards, and Erbach, Reference Jolly, Edwards and Erbach1983; Juergens et al., Reference Juergens, Young, Schillinger and Hinman2004; Osei et al., Reference Osei, Moriasi, Steiner, Starks and Saleh2012; Varner, Epplin, and Stricklan, Reference Varner, Epplin and Stricklan2011; Zentner et al., Reference Zentner, McConkey, Campbell, Dyck and Selles1996). Technological innovations such as genetically engineered crops that are resistant to herbicides and the increased availability of weedicides for post-emergent weed control reduced the producer’s reliance on tillage for weed control. Meanwhile, innovations in the farm machinery sector such as planters that can cut through crop residue made planting in residue-covered fields easier.

Cover crops are grasses, legumes, and other forbs that are planted for erosion control, improving soil structure, moisture, and nutrient content, increasing beneficial soil biota, suppressing weeds, providing habitat for beneficial predatory insects, facilitating crop pollinators, providing wildlife habitat, and as forage for farm animals (NRCS, 2019a). Due to the long-term environmental benefits of cover crops, they are becoming an integral part of soil health management systems among U.S. producers (CTIC, 2017; Delgado et al., Reference Delgado, Dillon, Sparks and Essah2007; Kaspar and Singer, Reference Kaspar and Singer2011). Further, cover crops are promoted in the U.S. to address water quality issues in the intensive agricultural production areas (Bergtold et al., Reference Bergtold, Ramsey, Maddy and Williams2017; Singer, Nusser, and Alf, Reference Singer, Nusser and Alf2007). Economic benefits of growing cover crops include energy savings both by adding nitrogen to the soil and making more soil nutrients available, thereby reducing the need to apply fertilizer; yield increases for the following cash crop; and forage for livestock for producers with integrated livestock systems (Bergtold et al., Reference Bergtold, Ramsey, Maddy and Williams2017; CTIC, 2017; Kaspar and Singer, Reference Kaspar and Singer2011). However, profitability from the use of cover crops depends on factors that are under the control of producers such as current production system, selection of cover crops, establishment and termination strategies, and factors that are not under the control of producers such as soil type, weather, etc. (Bergtold et al., Reference Bergtold, Ramsey, Maddy and Williams2017). Because of the risks associated with the profitability, cost-share and subsidy programs are available at the federal and state levels to incentivize the adoption of cover crops among U.S. farmers.

Diverse crop rotation refers to growing three or more crops in a rotation, particularly among row crop producers. Careful selection of a crop rotation system offers the possibility of reducing the trade-off between farm profitability and environmental impact by internal nutrient recycling, maintaining the long-term productivity of the land, and by breaking weed and disease cycles (Gebermedhin and Schwab, Reference Gebermedhin and Schwab1998). Evidence on the environmental benefits of diverse crop rotation includes a reduction in soil erosion, improvement in soil fertility, promotion of ecosystem services and ecosystem health, and improved air and water quality (Altieri, Reference Altieri1999; Datu Research, 2014; Davis et al., Reference Davis, Hill, Chase, Johanns and Liebman2012; Ikerd, Reference Ikerd1991; Fausti, Reference Fausti2015; Fausti et al., Reference Fausti, Tia, Lundgren, Li, Keating and Catangui2012). Available evidence from agricultural experiment station research shows that diverse crop rotation has the potential to provide substantial economic returns to producers, particularly in areas affected by herbicide-resistant giant ragweed infestations in the midwestern U.S. (Golpen et al., Reference Golpen, Coulter, Sheaffer, Becker, Breitenbach, Behnken and Gunsolus2018). The total energy use in 3- or 4-year rotations (1.41 and 1.50 BTU, respectively) is substantially lower than the typical 2-year rotation of corn and soybean (3.53 BTU) (Johanns, Chase, and Liebman, Reference Johanns, Chase and Liebman2012). Although not conclusive, there is some evidence on the yield-enhancing effect of crop rotations (Mulik, Reference Mulik2015). Despite these economic and environmental benefits, a combination of technological innovations in agriculture such as the introduction of genetically modified crops, federal agricultural policies including commodity support programs focusing mainly on five crops (corn, soybean, wheat, rice, and cotton) and crop insurance and market assistance programs focusing on limited commodities, biofuel policies supporting the production of a limited number of crops, and high market prices for select commodities have contributed to a steep decline in the prevalence of diverse crop production systems particularly during the last two decades (Benbrook, Reference Benbrook2012; Brookes and Barfoot, Reference Brookes and Barfoot2013; EPA 2018; Fausti, Reference Fausti2015; Fausti et al., Reference Fausti, Tia, Lundgren, Li, Keating and Catangui2012).

As per a recent USDA study, conservation tillage was used in 67% of the wheat area in 2017, 65% of the corn area in 2016, and 70% of the soybean area in 2012. The adoption of conservation tillage varies by region, with a higher rate of adoption in drier and warmer regions which include Northern Great Plains, Prairie Gateway, and the South (Claassen et al., Reference Claassen, Bowman, McFadden, Smith and Wallander2018). Only about 4% of U.S. farmers adopted cover crops on some portion of their fields in 2010–2011, and only about 1.7% adopted cover crops on cropland (Wade, Claassen, and Wallander, Reference Wade, Claassen and Wallander2015). Due to financial incentives, cover crops acres in the U.S. increased 50% from 10.3 million acres in 2012 to 15.4 million acres in 2017. Across the 12 states of the U.S. Corn Belt, about 70% of the planted crop area is under corn–soybean rotations, leaving only 30% for diverse crop rotation (Mulik, Reference Mulik2015).

As per the 2017 NRCS cropping system inventory for South Dakota, 45% of crop area is under no-till, followed by 22% in mulch tillage, 17% in reduced tillage, and 16% in conventional till (NRCS, 2017). Between 2012 and 2017, there was an 89% increase in planted acres under cover crops; from 149,383 acres in 2012 to 281,649 acres in 2017 (LaRose and Meyers, Reference LaRose and Meyers2017). As per the 2017 Census of Agriculture data, harvested crop acreage of corn and soybean in South Dakota increased 13%, from 57% in 2002 to 70% in 2017 (NASS, 2017).

3. Empirical Framework

Regional variations in adoption rates of conservation practices suggest the importance of spatial effects, independent of social, economic, and institutional factors in farmers’ adoption decisions. Spatial effects can be conceptualized as two types of impacts that influence farmers’ adoption decisions through different processes. The first type of spatial effect relates to geographic factors such as soil, climate, and topography. The observed distributional pattern of farmers’ adoption decisions can be explained partially by the variations in these factors. We define these exogenous/contextual characteristics commonly shared by individuals within a group as spatial heterogeneity (Sampson and Perry, Reference Sampson and Perry2019). In the case of spatial heterogeneity, the relationship between the spatial factor and a farmer’s adoption decision is constant for a particular spatial factor (e.g., cropping district), but varies between spatial factors (e.g., between cropping districts). There is substantial variation among the adoption rates of conservation practices within South Dakota (see Figure 1). For example, the adoption of no-till and crop diversity is the highest in the central part of the state while the heaviest use of cover crops is in the central and northeastern part of the state (NRCS, 2017). Figure 1 shows that the adoption of conservation practices was observed as higher in northern areas than in the south, and higher in districts located on the west side than on the east side.

Figure 1. Crop-district wise adoption of conservation practices in eastern South Dakota.

The other type of spatial effects refers to the impact of neighbors’ adoption decisions on farmers’ adoption decisions. Farmers may observe and learn from their neighbors about technologies and practices through social interactions that are strongly conditioned by the geographic distance between individuals (Bollinger and Gillingham, Reference Bollinger and Gillingham2012; Festinger, Schachter, and Back, Reference Festinger, Schachter and Back1950; Gonzalez, Hidalgo, and Barabasi, Reference Gonzalez, Hidalgo and Barabasi2008; Haynes, Reference Haynes1974; Muller and Rode, Reference Muller and Rode2013; Sampson and Perry, Reference Sampson and Perry2019). Therefore, the diffusion of information, ideas, and technology may occur within a certain geographic distance. We define this type of spatial effects as spatially mediated peer effects or peer effects in short. Farmers’ decisions to adopt knowledge-intensive conservation agriculture practices are primarily driven by economic motives (Dessart, Barreiro-Hurlé, and Bavel, Reference Dessart, Barreiro-Hurlé and van Bavel2019). The economic success/profitability of a specific practice will depend greatly on factors that are under the control of farmers in a specific production context. A decrease in the learning costs associated with switching to the new practice will likely increase the profitability of the practice and thus the likelihood of adoption of the practice.

In this paper, we focus on the spatially mediated peer effects on farmers’ adoption decisions. Previous work has observed regional variations in the adoption of conservation practices in the U.S. and within South Dakota (NRCS, 2017; Wade, Claassen, and Wallander, Reference Wade, Claassen and Wallander2015). Hence, we control for spatial heterogeneity to avoid biased peer effect estimates. We include agricultural statistics district (ASD)/cropping district dummies to capture spatial heterogeneity effects related to geography, climate, and cropping practices. Cropping districts are defined groupings of counties in each state, by geography (soil type, terrain, and elevation), climate (mean temperature, annual precipitation, and length of the growing season), and cropping practices (crops grown and use of irrigation).

Defining a peer group is a major challenge in studies focusing on spatial effects (Sampson and Perry, Reference Sampson and Perry2019). In sparsely populated rural areas, we expect that the peer effects may not emerge at the same level of geography as it would be in an urban or more populated setting. In this study, we use two different peer-group definitions: (i) neighbors in a 15-mile radius and (ii) neighbors in a 30-mile radius. The 15-mile radius more or less corresponds to the average school district boundary in South Dakota. Given the large farm sizes in South Dakota, the inclusion of a smaller geographic boundary (less than 15 miles) will not provide us enough information to generate a meaningful variable that captures spatial peer effects. Additionally, sample data may not give good representative statistics for the peer effects if the size of the neighborhood is much reduced. The inclusion of a larger boundary (e.g., 50-mile radius) will conceptually capture at least part of the effects (if any) of the crop district dummies. As mentioned earlier, the focus of this study is to isolate the effect of spatial peer effects after controlling for spatial heterogeneity and other confounding factors, hence selecting a larger geographical boundary for defining the peer group may not be effective, conceptually. To test the sensitivity of peer effects to the specification of peer-group definition, we have conducted analyses using a 50-mile radius to define the peer group and the results are presented in Table S2.

To compute the measurement of spatially mediated peer effects, we first create a radius of a given distance (15, 30, and 50 miles, respectively) for each farmer. Within the radius for each farmer, we then count the total number of adopters. If the focal farmer is not an adopter, then the number of adopters is the number of neighboring adopters. If the focal farmer is an adopter, the number of neighboring adopters is calculated by subtracting 1 from the number of adopters. The adoption rate is then calculated by dividing the number of neighboring adopters by the number of neighboring farmers.

Although peers’ adoption in previous years might be influential in a focal farmer’s adoption decision, some studies use contemporaneous adoption by peers to capture the peer effects. Manski (Reference Manski1993)’s discussion of the contemporaneous effects through peer influence pointed out that although it would be more realistic to assume the lag in the transmission of the effects, the dynamic process may have a “unique stable temporal equilibrium of the form.” Brock and Durlauf (Reference Brock and Durlauf2002) and Durlauf (Reference Durlauf, Vernon Henderson and Thisse2004) have a detailed discussion of the contemporaneous effects as one form of neighborhood effects by which individual behaviors are influenced by interaction with peers. Previous studies on peer effects on a wide range of social behaviors have used similar approaches. Gaviria and Raphael (Reference Gaviria and Raphael2001)’s analysis of school-based peer effects used the proportion of students in the school engaged in the defined activity after excluding the individual. Baerenklau (Reference Baerenklau2005) used the number of members in the defined peer group as the measure of peer-group influence to examine whether a producer will consider adopting an anaerobic digester. More recently, Cowley and Brorsen (Reference Cowley and Wade Brorsen2018) used the number of in-state neighbors as the measure of peer group influence on the adoption of anaerobic digesters. We used a distance-based approach to define who would be included in the peer group of the focal farmer. We consider this is part of the geographic network of the focal farmer. This also addresses the identification problem raised by Manski (Reference Manski1993).

3.1. Empirical Model

Suppose there are $N$ farmers in the region and consider that farmer $i$ will adopt a conservation practice if the expected utility of adoption which includes stochastic monetary profit exceeds the utility of non-adoption. The stochastic monetary profit depends on factors such as costs of production, yield, market price, government subsidies, weather effects, and farm and farmer characteristics. As mentioned earlier, the derived utility from adopting a conservation practice will also depend on the learning costs associated with adopting the new practice.

Let Y _i = 1 denote the decision of producer i to adopt a conservation practice and let Y _i = 0 denote the decision to not adopt the conservation practice. Let the perceived profit associated with adoption decisions be denoted by π _i ^Yi . The relative net profit from adopting the practice is defined as:

(1) $$\Delta {\pi _i}\, = \,{\pi _i}^1 - {\pi _i}^0$$

Let U _i ^Yi denote the utility for producer i from decision Y _i . Adoption occurs when

(2) $$E\left( {U\left\{ {1,{\pi _i}^1,L{C_i}\left[ {{I^1}\left( {{a_j}} \right)} \right],X} \right\}}\right){\rm{ }} \gt E\left[ {U\left( {0,{\pi _i}^0,X} \right)} \right],$$

where LC _i is the learning cost associated with adopting the practice which depends on conservation practice-specific information available from the neighboring farmers’ adoption decisions I ¹ (a _j ) and X, a vector of observable covariates.

The producer’s utility function U(Y _i , π _i ^Yi ; X) is unknown to us, and the deterministic part of the utility function is V(Y _i , π _i ^Yi ; X). So the inequality in equation (2) can be written as:

(3) $$V\left\{ {1,\pi _i^1,L{C_i}\left[ {{I^1}({a_j})} \right],X} \right\} + {U_1} > V\left( {0,\pi _i^0,X} \right) + {U_0}$$

where U ₁ and U ₀ are independent and identically distributed random disturbances with zero means and unit variances.

Equation (3) implies that a reduction in the learning costs (spatially mediated peer effects) will increase the expected utility of adopting a conservation practice. We expect that as the percentage of adopters in the neighborhood increases it might lead to a reduction in the learning costs associated with the practice and thus the adoption costs. However, due to the idiosyncrasies associated with various conservation practices described earlier, the role of learning costs might vary between practices.

The conceptual model described above can be represented as the following latent equation:

(4) $$Y_i^* = \;\beta '{X_i} + \alpha '{l_i} + \emptyset {Y_j} + { \in _i}$$

where $Y_i^*$ is the latent variable, ${X_i}$ is the observable farm and farmer characteristics, $\;{l_i}$ is the location-specific controls to account for spatial heterogeneity, and ${Y_j}$ captures spatially mediated peer effects expressed by the percentage of adopters in the defined peer group (a proxy for the reduction in learning costs).

However, we observe only the binary outcome ${Y_{i\;}}$ (whether farmer i has adopted the practice or not) and equation (4) can be empirically estimated as equation (5) using a univariate probit model that uses maximum likelihood estimation:

(5) $${Y_{i{\rm{\;}}}}={\rm{\;}}\beta '{X_i} + \alpha {\rm{'}}{l_i} + \emptyset {Y_j} + { \in _i}$$

where $\alpha$ and $\emptyset $ are parameters to be estimated.

In our study, we are focusing on three different conservation practices: conservation tillage, diverse crop rotation, and cover crops. Previous studies have shown that farmers’ adoption decisions on conservation agricultural practices could be correlated (Arbuckle and Roesch-McNally, Reference Arbuckle and Roesch-McNally2015). This implies error terms across the three independent adoption equations are correlated. So equation (5) becomes the M-equation multivariate probit model where M = 3:

(6) $$Y_{im}^*=\;\beta _m'{X_{im}} + \alpha '{l_i} + \emptyset {Y_{jm}} + { \in _{im\;}},\;\;\;\;\;m = CT,CC,{\rm{\;\;and\;\;}}DCR$$

where $Y_{im}^*$ is the latent variable,

(7) $${Y_{im}}=1\;if\;Y_{im}^* \gt 0\;{\rm{and\;}}0{\rm{\;otherwise}},$$

where ${Y_{im}}$ is the observed outcome.

${ \in _{im\;}}$ , $m$ = $\;CT,CC,{\rm{\;and}}\;DCR$ are error terms distributed as multivariate normal, each with a mean of zero, and variance–covariance matrix V, where V has values of 1 on the leading diagonal and correlations ${\rho _{jk\;}} = \;{\rho _{kj\;}}$ as off-diagonal elements. Following equation (7), we estimate equation (8) below:

(8) $${Y_{im\;}}=\beta _m'{X_{im}} + {\alpha ^{'{l_i}}} + \emptyset {Y_{jm}} + \;{ \in _{im}}$$

The log-likelihood function for a sample of N independent observations when M = 3 is given by,

(9) $$L = \mathop \sum \limits_{i = 1}^N {W_i}log{\emptyset _3}\left( {{\mu _i};\;\Omega } \right)$$

where ${W_i}$ is an optional weight for observations and ${\emptyset _3}$ is the trivariate standard normal distribution with arguments ${\mu _i}$ and $\Omega $ (Cappellari and Jenkins, Reference Cappellari and Jenkins2003). Equation (9) is estimated using Geweke–Hajivassiliou–Keane (GHK) smooth recursive conditioning simulator where trivariate normal distribution function is expressed as the product of sequentially conditioned univariate normal distribution functions.

4. Data

The data used in this study were collected from a farm-level survey conducted in eastern South Dakota during the spring of 2018. We used the survey to collect information on the perceptions of benefits and challenges of farm management practices, particularly conservation tillage, cover crops, and diverse crop rotation; years of adoption of these practices; farmer demographics; and farm characteristics. The list of eligible survey participants in the state was obtained from the Farm Service Agency (FSA) using a Freedom of Information Act (FOIA) request. Using FSA as the source for the participants’ list is reasonable as most of the farm operations in South Dakota work with FSA programs. We employed proportionate stratified-random sampling to select a representative sample of 3,000 farm operators in the eastern part of the state where most of the corn and soybean production occurs. We used four rounds to contact survey participants in 2-week intervals: (i) an invitation letter that described the survey with a link to answer the survey online was mailed to all operations selected (including a $2 bill incentive in half of the letters to test for the effects on response rates), (ii) a hard copy of the survey with return envelopes were mailed to those who did not respond to the survey online; (iii) a reminder postcard was mailed to those who did not respond in round 2; and (iv) a hard copy of the survey with return envelopes were mailed to remaining non-responders. We received 708 completed survey responses. Excluding operations that stopped farming or rented out all of their lands, we had a 30% response rate. However, 190 of the returned survey responses had P.O. boxes as the postal address making it impossible to geocode them for categorizing them into peer groups. Hence, there were 518 completed surveys left for the analysis.

We use farmers’ age and cropland acres operated to evaluate the representativeness of respondents in our sample. As per the U.S. Department of Agriculture 2017 Census of Agriculture data, the average age of producers in South Dakota was 56.1 years. Similar to the Census data, the average age of respondents in our sample is 56.7 years. Based on the 2017 Census data, farm size averaged 1,443 acres in South Dakota. On average, respondents in our sample had an average farm size (total acreage of farmland operated in 2017 planting season) of 1,170 acres. Overall, the key demographics of respondents in our sample is comparable to the state-level demographics reported in the 2017 Census data.

It is evident from Table 1 and Figure 1 that the adoption rate of diverse crop rotation is low (24%) among our respondents compared to the other conservation practices such as conservation tillage (66%), and cover crops (47%). Similar to previous studies (Arbuckle and Roesch-McNally, Reference Arbuckle and Roesch-McNally2015; Singer, Nusser, and Alf, Reference Singer, Nusser and Alf2007), analysis of the pair-wise correlation between adoption decisions show a statistically significant correlation between conservation practices: conservation tillage and diverse crop rotation (r = 0.12, P value = 0.00); conservation tillage and cover crops (r = 0.22, P value = 0.00); and diverse crop rotation and cover crops (r = 0.19, P value = 0.00).

Table 1. Summary statistics of key variables

***, ** indicate mean values are statistically significantly different between adopters and non-adopters at 1% and 5% levels, respectively.