Barriers to cover crop adoption: Evidence from parallel surveys in Maryland and Ohio

Materials and Methods

In the sections that follow we use the term “barriers” rather than “perceived barriers” for conciseness, while acknowledging that the analysis solely addresses barriers as perceived by farmers. Of course, there may be a perception-reality divergence in modeling CC barriers. A farmer’s perceptions about barriers might not match those perceived by agronomic experts. This paper focuses on farmer perceptions of these barriers, under the recognition that behavior is likely to be motivated by these perceptions, whether or not these perceptions are consistent with reality. The current data cannot (and is not intended to) evaluate the extent to which respondents’ perceptions match objective reality. However, to the extent that a divergence exists, educational programming might close any perception-reality gaps. It is also possible that better, targeted education might be as or more effective than cost-share in overcoming barriers. The goal of the paper is to inform these and other programs that explicitly target perceived barriers to enhance CC adoption. To preface the analysis that follows, the following section reviews the institutional backgrounds on CC programming in the two states studied. Then, the survey design, data collection, and variables are described.

Cover Crop Programs in Ohio and Maryland. CC adoption barriers are potentially affected by a variety of variables reflecting relevant government policies, available cost-share programs, education and extension activities, the CC input subsector, equipment availability, and the behavior of other farmers. Here, we compare Ohio and Maryland, which differ substantially in terms of programs available to support CC, agricultural production, and the context in which adoption is promoted. Primary cash crops in both states include corn (Zea mays L.), soybean (Glycine max [L.] Merr.), and small grains. Maryland has among the highest CC adoption rates among US states at 20% to 25% of total cropland (327,689 ac [132,611 ha]) (Weil and Kremen 2007; USDA NASS 2017). In contrast, CC acreage in Ohio is estimated at 357,292 ac (144,591 ha), which is approximately 3.3% of total cropland (USDA NASS 2017). Adoption of CC is promoted in the two states in part to reduce nutrient loads to the Chesapeake Bay (Maryland) and western Lake Erie basin (Ohio); both waterbodies have total maximum daily loads established. With respect to available CC programs, both states provide federal cost-share through the Environmental Quality Incentives Program (EQIP) and Conservation Stewardship Program (CSP). However, enrollment in federal programs and the scale of state and local programs differ considerably between the two states. Meanwhile, programs in the two states cannot be compared directly since the EQIP and CSP in Ohio cover all BMPs while the dominant program in Maryland incentivizes CC adoption and use.

In Ohio, the nonfederal local/state/watershed-based programs have relatively low budgets such that federal programs incentivize more planting. According to the data from USDA NRCS (2019b), EQIP funding in the fiscal year 2019 in all of Ohio totaled US$36.124 million, an increase of 43% compared to the US$25.338 million in 2010. About 70% of the total EQIP funding went to program participants as payments for implementing conservation practices, and the remainder was used for technical assistance or reimbursement funds. In 2019 there were 1,496 active contracts in EQIP, with 223,842 contracted acres (90,586 ha). CSP also provided US$11.5 million for Ohio in fiscal year 2019 (USDA NRCS 2019a). Of this total, US$9.02 million, or 78%, of CSP funding in Ohio was paid directly to the participants for maintaining and improving existing conservation practices. There were 111 active contracts in 2019, with 27,544 total acres (11,147 ha).

The presented analysis relies on data collected in eight contiguous counties in northwest Ohio, where farmers have several state and local conservation CC programs available. For the sample discussed in this paper, 5.6% of respondents reported planting CC in EQIP in 2017, 2.2% in CSP, 0.8% in other federal programs, and 8.2% in state/local programs such as the Great Lakes Restoration Initiative (GLRI), Ottawa River watershed project, Miller City Cuttoff & Pike Run Watershed Grant, and Central Fulton Conservation Project.

In contrast, the Maryland Agricultural Water Quality Cost-Share (MACS) Program is a large state initiative with comparatively high budgets that allow nearly 100% of applicants to obtain some type of cost-share. In 2019 to 2020, the total funding for this program was US$22.5 million, and farmers could receive up to US$90 cost-share ac^–1 (US$222.40 cost-share ha^–1) of CC planted. There were 1,687 farms that participated in the program with 558,976 ac (226,210 ha) during 2016 to 2017 (Maryland Department of Agriculture 2017). Maryland farmers are also eligible for federal programs. EQIP funding in Maryland in 2019 was US$18.56 million, or roughly half of the funding in Ohio (USDA NRCS 2019b). The total active contracts and acreage for EQIP in Maryland are 341 and 14,045 ac (5,684 ha), which are about 22.8% of total active contracts and 6.3% of the enrolled acreage in Ohio. For the sample discussed in this paper, 0.9% of respondents planted CC in EQIP during 2017, 1.3% in CSP, 1.1% in other federal programs, and 66.0% in MACS. Hence, there are considerable differences in CC program use across the two states.

Barriers to Cover Crop Adoption and Survey Variables. There is an extensive CC adoption literature (Lee and McCann 2019), summaries of which are provided elsewhere (Carlisle 2016; Miller et al. 2012). Rather than repeating the general insights in prior reviews, the following section highlights more recent findings of this literature as related to the variables included in the empirical analysis. In this way, we summarize the primary ways in which existing work motivated our research design. In supplementary materials, we develop an updated version of the summary table in Miller et al. (2012), focusing solely on barriers to adoption (our table S1 lists 13 barriers from prior studies, which we reduce to 7 key barriers and an “other” category). The 7 dependent variables in our empirical analysis take the form of binary indicators identifying whether farmers perceive certain types of barriers as affecting CC use on their farm.

The survey was developed over a two-year process, including (1) an initial literature review, (2) six structured interviews with CC program personnel and other experts in both states and the federal government, and (3) focus groups with farmers in each state. The purpose of the structured interviews was to understand relevant state and federal CC programs, patterns in enrollment, perceived challenges in maintaining and increasing CC adoption/enrollments, etc. Information from these interviews was supplemented with the results of four focus groups conducted with 25 farmers in Ohio and Maryland (held in December of 2016 and January of 2017). These focus groups explored the way farmers and local CC program personnel thought about CC, available CC programs, and different types of barriers to adoption. This insight was used to develop the set of barriers for which information was elicited by the survey instrument and that were ultimately included as dependent variables in the quantitative models. Focus groups also provided information on the factors hypothesized to cause systematic variations in these barriers across different farms and farmers—these comprise the explanatory variables. A final goal of focus groups was to refine and pretest the language in the survey instrument, following survey design guidelines in Johnston et al. (2017).

Barriers were identified for each farmer who responded to the survey via the following question: “What do you think are the barriers to using cover crops on your farm? (…whether that means planting a cover crop for the first time or planting more than you currently plant) Please check all that apply.” This was followed by a list of seven statements (detailed in supplementary table S2), along with an “other” category. Those who checked the “other” category were able to describe this barrier in a follow-up text box.

Most literature cites economic or financial challenges as one of the primary barriers to adoption (Carlisle 2016; SARE 2017; Daryanto et al. 2019). CC have direct costs associated with the purchase of seed, labor, and equipment used in planting and termination (Lichtenberg 2004; Arbuckle and Ferrell 2012; Bergtold et al. 2012). Studies show that the absence of cost-share programs or insufficient cost-share rates in some states contribute to farmers’ financial concerns (Maryland Department of Agriculture 2005; Singer et al. 2007; Baumgart-Getz et al. 2012; Dunn et al. 2016). The overall concern that CC do not pay for themselves was captured by the variable ROI for return on investment, or the financial sustainability of planting CC. This variable takes a value of 1 if the survey respondent identified the following as a barrier within the survey: “I am not confident that cover crops are financially sustainable.” The use of the term “sustainable” implies a focus on long-term returns. A qualification is that some respondents may have conflated their other financial concerns with CC, being more pessimistic about CC ROI than in some other years. This might be expected because 2016 and 2017 were 2 of the 12 years between 2000 and 2019 when net farm income was below the average (USDA ERS 2021). Indeed 2015 to 2019 was an extended low period for this measure. Although the frequency with which this barrier is reported might be higher than in some years, we do not anticipate a systematic bias in the effect of the explanatory variables on ROI in the presented empirical analysis.

Many studies find that the costs of seed, labor, and equipment can also prevent broader CC adoption, apart from more general ROI concerns. The survey therefore asks separate questions about each of these costs: COSTS-LABOR, COSTS-SEED, and EQUIPMENT (we report the survey questions and summary statistics by state in supplementary table S2). For example, seed costs are variable (they depend on whether CC are planted) while equipment costs are fixed in the short run (they occur regardless of whether CC are planted, and the survey question focuses on buying equipment rather than operating and maintaining it). Labor costs might be fixed or variable. As a result of these and other differences, these costs might present different types of barriers to different types of farming operations, for example as a function of factors such as financial position, experience, level of capitalization, and size of operation. ROI addresses respondents’ overall concern with the financial sustainability over a long term—which considers both revenues and costs—while COSTS-LABOR, COSTS-SEED, and EQUIPMENT emphasize specific sources of cost.

CC might introduce perceived risks to a farming operation, captured by the variable RISK and linked to the statement “I believe there are risks to my farm introduced by planting cover crops.” One possible risk is that CC may interfere with the fall harvest or spring planting of the cash crop (Snapp et al. 2005; Dunn et al. 2016; Roesch-McNally et al. 2017). Other risks may be farm- or farmer-specific. For instance, during focus groups in Ohio, respondents expressed concern that CC might provide winter food for nuisance animals, thereby increasing pest impacts in future years. Farmers may also fear that planting CC might increase some diseases and/or that CC could become a weed (Snapp et al. 2005). Farmers might also be concerned about financial risks and impacts on yield variability.

Lack of farmer experience (or related information and education challenges) may further discourage the use of CC. Some farmers lack sufficient information, such as information on the benefits of planting CC (Singer et al. 2007; Arbuckle and Ferrell 2012; Miller et al. 2012) or related to selection of CC species within a crop rotation (Miller et al. 2012; SARE 2017). Prokopy et al. (2019) review 35 years of adoption literature, and one conclusion is that future research should measure the impact of different “avenues of reaching farmers.” Here, we distinguish between two potential barriers related to experience and knowledge. The first reflects basic knowledge about CC (KNOWLEDGE), whereas the second reflects a more general disagreement that CC are important (IMPORTANCE). The former is linked to the statement “I don’t know enough about cover crops” while the latter is linked to the statement “I don’t think of cover crops as being an important part of my farming.” Both of these barriers may be addressed with continuing education about the private benefits and costs of CC. Research also can test whether specific avenues of educational programming are most effective at reaching farmers who were unpersuaded by previous efforts to promote adoption. Some novel approaches may even be effective, such as peer mentoring programs (we thank an anonymous referee for this suggestion).

The “other” option with an open-ended follow-up question allows respondents to discuss other types of barriers that are not explicitly addressed by other survey questions. The survey did not ask farmers to consider an exhaustive list of all possible CC adoption barriers. This was done intentionally. First, listing all possible adoption barriers would have led to an exceptionally long list that would risk frustrating survey respondents. Second, and most importantly, many types of barriers do not vary systematically across farmers and hence cannot readily be addressed by farm-level interventions. For example, there may be macro-scale barriers to CC adoption related to factors such as the structure of the state or regional farm economy. Our study does not address barriers of this type because they apply similarly to all farmers in a region.

Explanatory Variables and Hypotheses. Two types of variables potentially influence barriers: 12 farm and farming operation characteristics and 13 information sources (table 1). LNACRE measures the log of acres of cropland for commodity crops in 2017. Studies show that farmers with larger acreage may perceive fewer barriers to use CC or conservation practices, in general, because they have economies of scale in the fixed costs of learning, equipment, etc. (Featherstone and Goodwin 1993; Upadhyay et al. 2003; Gabrielyan et al. 2010; Vitale et al. 2011). In contrast, Bergtold et al. (2010) found that larger acreage has a negative impact on CC adoption due to the lack of time and labor (Bergtold et al. 2012). Other studies find an insignificant relationship between farm size and conservation practices (Prokopy et al. 2008; Bergtold et al. 2012). Various explanations have been offered for the mixed results on the farm-size/adoption relationship. For instance, there may be different measures of farm size, either acreage managed or acreage owned (Prokopy et al. 2008). Or, there may be scale-related challenges with capital investment, risk tolerance (Carlisle 2016), and/or anticipated environmental benefits (Arbuckle and Ferrell 2012). In this paper, we expect that LNACRE should have no effect, or a negative effect, on the two barriers (EQUIPMENT and ROI) due to economies of scale. However, acreage should be positively associated with labor because of lumpiness in labor supply, or an inability to hire more labor, especially for the small farms in our sample (76.5%) with gross cash farm income (GCFI) less than US$350,000 for whom hiring new labor may not be routine.

Farming experience is measured using four binary indicators for the years of farming by the principal operator: EXPERIENCE<16, EXPERIENCE16-30, EXPERIENCE31-45, and EXPERIENCE46+. Our survey also included measures of farmer’s age and GCFI; however, these two measures were correlated with farming experience. Therefore, we use experience in the regressions alone, which also partially captures age and GCFI effects. To evaluate whether this specification choice has a significant impact on model results, we also estimated preliminary models that included all three of these types of variables, but only 1 of 32 new coefficients was significant. Because these new variables introduce multicollinearity and the existing results are robust, we proceed with the model that incudes experience alone, as a proxy for combined age, income, and experience. We hypothesize that the relationship between farm experience and adoption barriers will depend on farmers’ risk preferences (Abadi Ghadim and Pannell 1999; Bergtold et al. 2012), but also on age because younger farmers are more likely to use CC (Clay et al. 2020). That said, older farmers would tend to have more years of opportunity to experiment with CC and identify an optimal mix of inputs to enhance profits. Thus, in concept, greater experience should lead to reduced concern about CC labor requirements (COSTS-LABOR) because accumulated skills should substitute for labor quantity. However, more experienced farmers might also be resistant to change. More experience might also proxy for financial security, so more experience should lead to fewer financial barriers. Considering these factors, we hypothesize that less experienced farmers will be more likely to report barriers, while those with more experience will be less likely.

The set of explanatory variables also includes indicators for the use of conventional tillage (CONVTILL), conservation tillage (CONSTILL), and no-till (NOTILL). Within the survey, conventional tillage was defined as “less than 30% of the soil surface is covered by residue after planting,” while conservation tillage refers to “at least 30% of the soil surface is covered by residue after planting.” No-till is the case in which there is “no soil disturbance prior to planting.” Survey respondents were allowed to select multiple options in this question. Because CC management under no-till often incorporates equipment for termination, such as a “roller-crimper” (Miller et al. 2012), one possible hypothesis is that farmers who use no-till (NOTILL) will be more likely to report barriers related to equipment (EQUIPMENT). However, Bergtold et al. (2010) found that farmers who already adopted conservation tillage were more likely to adopt CC. Thus, the use of conservation tillage (CONSTILL) or no-till (NOTILL) is expected to have negative relationship with CC barriers.

The variables CORN, SOYBEANS, and WHEAT indicate whether the respondent planted the crop. The survey also includes a single variable, OTHERCROP, which includes those who reported “any other crops” and the few respondents who (perhaps erroneously) selected “no crops.” Participants selected all that apply (and table 2 shows that >80% of respondents planted corn and soybeans). The use of any irrigation (IRRIGATION) was an indicator variable. Studies show that farmers in areas with a dominant corn–soybean production system can find it challenging to integrate CC into the crop rotation because most of the agricultural infrastructure is oriented toward corn and soybeans (Arbuckle and Ferrell 2012; Roesch-McNally et al. 2017). Accordingly, there may be a positive tendency for corn–soybean farmers to report one or more of labor, seed, or equipment barriers. As irrigation should reduce crop yield risks and indicate high levels of capitalization, we anticipate that IRRIGATION will be negatively associated with ROI, COSTS-SEED, and EQUIPMENT.

Farmers can learn about CC in many ways, and we consider 10 possible information sources identified by our interviews and focus groups as being particularly relevant to farmers in at least one of the two states: the USDA Natural Resources Conservation Service (NRCS) (INFONRCS), Soil Conservation District Office (INFODISTICT), Cooperative Extension web services (INFOEXTWEB), the Departments of Agriculture in each state (INFOSTATE), extension agents (INFOEXTAGENT), seed dealers (INFODEALERS), growers’ associations (INFOGROWERS), Farm Bureau (INFOFB), and informal channels like neighboring farms (INFOFARM-NEIGH) or other farmers (INFONOFARMNEIGH). Respondents were asked which of these information sources were used and could select more than one source. The information source of the Ohio No-Till Council (INFOOHNOTILL) only exists in the Ohio survey. This survey lists 11 specific information sources and two possible answers for other sources. Although it may seem possible that our barriers and information sources could have dual causation, we ran separate regressions with and without the information sources and find no compelling evidence that the inclusion of (potentially endogenous) information sources lead to potential biases in other conclusions drawn from the model (table S6 in the supplementary files). Hence, although we lack the data to address any possible endogeneity of information sources formally, other model results are robust to the inclusion or exclusion of these variables. We proceed with models that include the information-source variables, while noting that results for these variables should be interpreted with at least some caution due to possible endogeneity.

Existing studies highlight the importance of CC information for adoption (Snapp et al. 2005; Arbuckle and Ferrell 2012; Miller et al. 2012; Carlisle 2016), and we seek insight about the relative impact of different information sources in reducing barriers. Indeed, the naïve hypotheses are that each source should reduce barriers. However, we also hypothesize that farmers who obtain information on CC from public agencies (e.g., Departments of Agriculture, USDA NRCS, and extension agents) would be less likely to report barriers of all types because these agencies provide information that is designed to help farmers overcome barriers. In contrast, it is possible that information from private networks might have some built-in biases that attenuate barrier-reducing potential; for instance, a seed company might favor a more capitalized model of agriculture and thus this information source might be associated with an increase in labor barriers. We also expect the two farmer-information sources to be the least likely to reduce barriers because their objective functions do not necessarily involve CC promotion.

Statistical Methods. CC adoption studies often have dichotomous dependent variables (adopt/not adopt) analyzed with discrete choice regression techniques such as logit or probit models. The model in this paper is similar; however, four response variables (barriers) are simultaneously modeled using a more general MVP (Greene 2012) (equation 1): 1

where the is a vector of unobserved latent variables of respondents’ perceptions about whether barriers m = 1,…, M exist (subscript m references each barrier), x_m is a vector of observable influences, and ϵ_m is a vector of unobservable influences that are treated as errors. The equation error distributions are assumed multivariate normal with zero means and a variance-covariance matrix R, where R has 1s on the leading diagonal and correlated coefficients ρ_mn between equation m and n on the off-diagonal elements.

If the unobserved latent variable for a farmer’s opinion on a specific barrier () is positive, then the farmer has reported this factor (y_m) to be a barrier to CC use. These perceptions may overlap across different barrier types—for instance, the perception of a ROI barrier may be related to concerns about the equipment needed to plant CC. The MVP model allows one to test whether the observed variables affect one barrier in the same way or differently than other barriers (e.g., β_1l ≠ β_1k where l and k indicate different set of barriers).

The MVP model also tests for variation across barriers through unobserved components via the correlation coefficients (ρ). The unobserved variations may come from the potential connections between farmers’ perceptions of barriers to adopt CC. However, the omitted variables that are unmeasurable (or costly to measure) explanatory variables, such as awareness of and concern about the environmental impact and farmland quality, can also contribute to the unobserved variations. For example, some types of farmers may simply be predisposed to perceive many types of CC adoption barriers, ceteris paribus. Because the adoption barriers are binary outcomes with potential correlation in unobserved components, the MVP model is more appropriate than a univariate model. The univariate models, which ignore the unobserved and common factors and correlation among adoption barriers and estimate the models independently, can lead to inefficient and less informative estimates (Kassie et al. 2013; Lin et al. 2005; Mulwa et al. 2017). The model is estimated using maximum likelihood with GHK simulation (Cappellari and Jenkins 2003).

Survey Methods. The survey was implemented using a mixed-mode mail/internet approach from late January to March of 2018. The survey sample frame was all potential CC growers in Maryland and the eight counties of northwest Ohio (Allen, Fulton, Hancock, Henry, Lucas, Putnam, Seneca, and Wood). To obtain a mailing list for this group, we obtained a list of commodity crop (corn, soybean, and small grains) farmers from the USDA Farm Service Agency, and supplemented this list in Maryland with an additional state program participant list and the Maryland Grain Producers and Utilization Board list. The combined mailing list was cleaned manually to remove duplicates, nonspecific location addresses (like post office boxes), corporation names that suggested an individual farmer might not be answering the survey, and businesses like golf courses and cemeteries. This procedure yielded 8,774 names to which survey invitation letters were sent (5,551 in Ohio; 3,223 in Maryland).

Survey implementation followed the tailored design method for mixed-mode surveys outlined by Dillman et al. (2014). The first contact was a letter sent by postal mail that explained the purpose of the research and invited the respondents to complete an online version of the survey. This letter provided the URL and subject identification number necessary to access the survey. Paper surveys could also be requested. Three rounds of follow-up (reminder) mail contacts were sent to those who did not respond to earlier contacts, with a paper copy of the survey included with the third contact letter. This procedure yielded a total sample size of 2,804 completed surveys, with 1,845 from Ohio (65.8%) and 959 from Maryland (34.2%). Various respondents were subsequently excluded from this analysis based on survey responses, suggesting they were not potential CC growers. These included farmland owners who did not farm or had land entirely out of production, and hence were not making decisions with regard to CC planting. The overall response rate is 41.7% in Maryland and 37.4% in Ohio, but only some of the observations were usable and came from commodity crop farmers (59% in Maryland and 77% in Ohio). CC were familiar to the respondents; respectively, 95.4% and 74.5% of Maryland and Ohio respondents reported having planted CC at some point in the past.

The data used for the analysis are drawn from sections of the survey that asked about barriers to CC adoption (see above), with explanatory variables derived from questions on demographic, farming, and farm characteristics. The survey defined CC as “any crop planted for late fall, winter, or early spring seasonal vegetative cover.” After initial data screening (see above) but before accounting for missing data on individual variables used within the model (due to incomplete survey questions), the data included 1,518 observations from Ohio and 716 from Maryland. There were 2,804 observations in total from the survey, but the presented models are estimated using survey responses only from landowners who plant commodity crops. This screen left 2,277 observations, and a further 481 observations were dropped because of missing values (table S3 in the supplementary files). However, as the regression analysis below models the two states separately and controls these factors in each analysis, the impact of missing values on parameter estimation is likely small. The final usable sample size is n = 1,796, with 546 from Maryland and 1,250 from Ohio.