Survey design

The data for this study were obtained from a larger project aimed at promoting and incentivizing the adoption of conservation practices among agricultural producers in South Carolina (SC) who produce the major agricultural commodities in the state. Farmers were invited to participate in the project through email, phone, online platforms, social media, extension meetings (interactions) and through the Clemson University and South Carolina State University extension network during 2023 and 2024.Footnote ¹ Given the nature of recruitment efforts, it is unknown how many farmers and farm managers were contacted to enroll in the program during this period, for reasons such as email lists might have duplicated addresses, a lack of knowledge of who participated in the extension meetings, and uncertainty on how many people accessed the project website or social media pages.

In the first year of the project, 304 farmers applied to participate, of whom 286 met the eligibility criteria. Of these, 255 signed participation agreements. Completion of a questionnaire was required to receive incentive payments. The data used in this study were collected from survey instruments administered to participants in the second year of the program. Of the 255 initial program participants, 215 continued into year two. Surveys were conducted between November 2024 and February 2025. The final sample consisted of 177 responses and does not represent all farmers enrolled in the second year of the program, as some farmers had not completed the survey when the program was terminated in April 2025.

The survey instrument employed a blocked experimental design to control potential informational framing effects. Respondents were randomly assigned to one of two treatment groups: the “Basic Information” block and the “Additional Information” block. Respondents in the “Basic Information” block received only a definition of agrivoltaics. Respondents in the “Additional Info” block received a definition of agrivoltaics, along with information on typical project sizes and the potential benefits of leasing land for agrivoltaics projects, including a range of payments and potential impacts on agriculture.

A total of 80 respondents were assigned to the basic group and 97 respondents to the additional information group. Except for the additional provision of information as described below, the survey remained the same for both groups. The basic information statement was as follows:

To build the additional information treatment, we draw from several sources, including academic and industry outlets (Ong et al. Reference Ong, Campbell, Denholm, Margolis and Heath2013; Walley-Stoll Reference Walley-Stoll2022). The additional information block was:

After receiving the informational treatment, survey respondents were asked: “If approached by a solar energy company or government utility, how likely are you to consider signing a lease agreement allowing them to build solar panels on your land in exchange for a lease payment?” Respondents indicated their willingness to lease on a five-point Likert scale with options ranging from Very Unlikely to Very Likely. Those who selected Unlikely or Very Unlikely were further asked to provide a reason, with response options including insufficient lease payment, physical limitations of the land, concerns about potential land damage, and reluctance to enter a contract.

In another block of the survey, questions were added to capture information about the agricultural operation, such as acreage. Sociodemographic characteristics were also gathered for the respondents (i.e. gender, education, years of farming experience). Finally, respondents were asked questions to assess their agreement with environmental statements and identify their risk preferences.

Empirical strategy

Considering the nature of our variable of interest (willingness to lease land for agrivoltaics development (y _i )) reflects ordered levels of potential future participation ranging from weak to strong, we employed an ordered logit model for the analysis. This logit model and the probit model has been used in several studies on adoption of technologies (Berg et al. Reference Berg, Thayer, Silva, Vassalos, Wang and Smith2025) and willingness-to-pay (Behler et al. Reference Behler, Silva, Vassalos and Ureta2024; Umberger et al. Reference Umberger, Thilmany McFadden and Smith2009). This modelFootnote ³ can be specified as:

(1) $y_i^* = \boldsymbol{X}_i^{\prime}\beta + u_i$

where i represents the respondents (i = 1, 2, ⋯, 177), y _i * is the latent variable capturing producer i’s willingness to lease, X _i is a vector containing the independent variables, including the information treatment, farm, and farmer characteristics. β is a vector of parameters to be estimated, excluding the intercept, and u _i is the stochastic error that follows a logit distribution. The relationship between the variable of interest (y _i ) and the latent variable (y _i *) is given by:

(2) ${y_i} = j\ if{\mu _{j - 1}} \lt y_i^* \lt {\mu _j}$

where μ _j are the threshold parameters known as “cut-points” and estimated jointly with the β for j alternatives.

Figure 1 presents the distribution of the dependent variable (y_i ) by information treatment (basic versus additional). Overall, most respondents reported being very unlikely to lease their land for agrivoltaics, with only about 25% indicating that they would be likely or very likely to lease. However, differences emerge when responses are disaggregated by treatment group. Consistent with expectations, respondents who received additional information were less likely to select very unlikely (28% compared to 38% under the basic information treatment) or unlikely (16% compared to 19%). At the same time, the proportion choosing neutral (29% versus 20%) or likely (14% versus 8%) increased. Table 1 provides descriptive statistics for the key variables.

Figure 1.

Distribution of willingness to lease land for agrivoltaics development by information type – basic (n = 80) versus additional information (n = 97).

Note: Percentages for willingness to lease were calculated for each information group separately. For additional information, the percentages were 28%, 16%, 29%, 14% and 12%, roughly adding to 100%.

Table 1.

Descriptive statistics of variables used in the regression model

Note: (base) refers to the reference base in the regression analysis.

We also controlled for farmer characteristics, aiming not only to obtain an accurate estimate of the effect of additional information but also to shed light on which characteristics are associated with a greater willingness to lease their land for agrivoltaics. To capture the influence of farmers’ demographics, we created a dummy variable female equal 1 if the respondent was female, 0 otherwise; nonwhite equal 1 if the respondent was non-white and 0 otherwise; and bachelor or higher equal 1 if the respondent has a bachelor’s degree or higher, 0 otherwise. As shown in Table 1, most of our respondents are white males, with only 15% identifying as female and 36% responding non-white. To capture a farmer’s experience, we created two dummy variables Farmer5to10, equal to 1 if the farmer reported 5–10 years of farming experience, 0 otherwise and FarmerMore10, equal to 1 if the farmer reported 11 years or more of experience, 0 otherwise. Most respondents in our sample (74%) have more than 10 years of experience.

Following Finger et al. (Reference Finger, Wüpper and McCallum2023), respondents were asked to self-assess their financial risk preferences with respect to external financing on a scale from 0 (very unwilling to take risk) to 10 (very willing to take risk). This measure, referred to as Financial Risk, was included among the explanatory variables. The mean value was 3.68, with a median of 4 (Table 1), suggesting that respondents are generally risk averse, leaning closer to unwilling than willing to take risks. Notably, 28.8% of respondents selected 0 or 1, while only 11.3% selected 7 or higher, indicating that – with few exceptions – most participants identify as financially risk averse.

The willingness of farmers to lease their land for agrivoltaics might also be associated with their environmental beliefs, especially because energy production alternatives have been shown and are widely associated with the environment. To account for this association, a question was added to the survey that asked respondents to state their agreement to the following statement using a 5-level Likert scale: “Considering the Environment in Production Decisions Lowers Profit.” The dummy variable Environmental trade-off was created and equals 1 if the respondent has selected Agree and Strongly Agree. Approximately 20% of the respondents reported either agreeing or strongly agreeing to the statement.

Some respondents may have previously acquired knowledge about agrivoltaics through various channels, including meetings with extension agents. Respondents were asked: “How often have you interacted with an extension agent in the last 12 months?” and prompted to select ranges of interactions from none to 12 or more times in the last 12 months. To account for the potential association between such interactions and willingness to lease, we included multiple binary variables to characterize a farmer’s interactions with extension agents. A dummy variable of Extension2to3, coded as 1 if the farmer meets with extension agents 2–3 times per year, 0 otherwise was added. A second dummy variable of Extension4to5, coded as 1 if the farmer meets with extension agents 4–5 times per year, 0 otherwise, was added. A third variable of Extension 6orMore, coded as 1 if the farmer met with extension agents 6 or more times a year, 0 otherwise, was added. These thresholds were chosen to capture interactions occurring roughly twice a year, up to every other month or more, which exceeds typical meeting frequency expectations. In our sample, 17% of respondents met with extension agents six or more times over the past year, with 43% among those who received basic information and 57% among those who received additional information.

The survey asked respondents to report total sales for the operation for the last three years. Approximately 8% of the sample reported sales greater than $1,000,000. Yet, 42% of the sample reported sales of less than $25,000. Using the midpoint of the ranges, the mean was $289,000 and median sales were $125,000. Based on the USDA measure of small farms (USDA, 2021),Footnote ⁴ less than or equal to $350,000, we created our dummy variable, Sales larger than $350k. It is important to notice that even though our sample split producers into three programs (Leafy Greens, Beef Forage and Peanuts), producers in each program oftentimes report multiple production types. Our sales measure captures this production diversity. See the appendix for the question text.

As mentioned above, our sample is composed of producers enrolled in a project split into three programs – leafy greens, beef forage and peanut programs. Program assignment was selected by the producers, mostly based on their main commodity; however, farmers may produce more crops than the commodity listed in the program. For example, Leafy Greens program enrollees also reported producing other vegetables (97%) and row crops (39%), Beef Forage program enrollees also reported producing timber (41%) and other livestock (100%), and Peanuts program enrollees also reported producing other row crops (100%), timber (34%) and livestock (25%). These numbers also show that even though our sample consists of respondents participating in a project with programs that cover three different commodities, the number of crops produced is greater than just the incentivized commodity and is a reasonable representation of the broader agricultural production in the state.Footnote ⁵ For a full breakdown of production types reported by producers in each program please see Appendix C.

To control for factors and characteristics that motivated producers to enroll in one of the three programs, we created three dummy variables (Leafy Greens Program, Beef Forage Program, and Peanut Program), each equal to 1 if enrolled in that program and 0 otherwise. Note that producers could only enroll in one program. To control for differences in production across the state, we created four regional dummy variables – Upstate (10 counties, and 27 respondents), Low Country (12, and 65), Midlands (12, and 25) and Peedee (12, and 60). We also controlledFootnote ⁶ for solar exposure using solar irradiance by zip code obtained from the United States Department of Energy NREL. A geospatial data layer showing direct normal irradiance (DNI) from 1998 to 2016 for the state of South Carolina was downloaded (Sengupta et al. Reference Sengupta, Xie, Lopez, Habte, Maclaurin and Shelby2018), and an average was calculated based on zip code using ArcGIS Pro.