2.1. General CRP

Enrollment in General CRP involves several steps (Figure 1). First, the FSA announces a specific application submission date, which usually occurs annually. Once the application period opens, producers submit an offer/bid for a particular parcel of their environmentally sensitive agricultural lands with a proposed conservation practice. Annual payment rates are capped by a parcel’s soil rental rate (SRR). The SRR is calculated using parcel-specific soil productivity measures, county-level average estimates of non-irrigated cropland rental rates, and professional expertise of field office staff (Hellerstein, Reference Hellerstein2017). After submission, FSA ranks the submitted offers based on the Environmental Benefits Index (EBI), which considers wildlife habitat potential (10–100 points), water quality benefits (0–100 points), erosion reduction (0–100 points), post-CRP environmental benefits (0–50 points), air quality benefits (3–45 points) and annual payments (USDA, 2021a). The FSA then sets a threshold for EBI after the distribution of offers is known, to accept only those offers that meet or exceed the threshold. In addition, to be eligible for General CRP, the land must have a crop history (the land must have been planted to an agricultural commodity in four of the six crop years before the year of the most recent Farm Bill) or be expiring CRP acres. Furthermore, the land must have considerable erosion potential or be in a national or state CRP conservation priority area (USDA, 2024a). The number of acres enrolled is subject to both a National Cap mandated by the most recent Farm Bill and a county limit of 25% cropland (Hellerstein, Reference Hellerstein2017). Finally, producers enroll their environmentally sensitive lands in General CRP for 10–15 years and, upon the contract expiration, either seek re-enrollment or exit the program.

Figure 1. General Conservation Reserve Program enrollment procedure.

2.2. Study area

Our survey population included agricultural producers in six southeastern states (Alabama, Florida, Georgia, Mississippi, North Carolina, and South Carolina) either currently (as of 2023) or previously enrolled in the program (Figure 2). One reason for choosing this study area is the rapid pace of land-use change in the Southeast region as a result of economic and population growth and urban expansion (Napton et al., Reference Napton, Auch, Headley and Taylor2010; Zhao et al., Reference Zhao, Liu, Sohl, Young and Werner2013). Additionally, this region is significant from a conservation point of view. According the National Environment Education Foundation (2024), the Southeast includes one of the world’s oldest and best-preserved freshwater systems. The United States Fish and Wildlife Service (2019) National Wetlands Inventory reported that stormwater runoff significantly contributes to water quality problems in North Carolina, and wetlands are the most essential filters of agricultural runoff in the coastal region. Moreover, the region is also a part of the North American Biodiversity Hotspot (Walls, Reference Walls2014), and is, therefore, a critical target for conservation efforts. Finally, CRP enrollment fell roughly 50% between 2015 and 2023 to 752 thousand acres in the study area (2023c, USDA, 2015).

Figure 2. Land use map of the six southeastern states selected for the study (USDA, 2021d).

2.3. Sampling method and data collection

We used the USDA FSA administrative data as a sampling frame to select farmers enrolled in CRP (General and Continuous CRP). We followed a stratified random sampling strategy to select CRP participants from six southern states who were previously or currently enrolled in the CRP as of 2023. The survey was sent to a total of 5,000 randomly selected CRP participants; the percentage randomly selected from each state was chosen to be proportional to the percentage of CRP participants in each state from a total of six states (USDA, 2023c, 2023d). The Institutional Review Board (#00007239) approved the mail survey at the University of Georgia, and the mail packet included a survey questionnaire and a pre-addressed stamped return envelope to return the completed questionnaire. The survey included questions related to socio-demographics (gender, age, educational status, annual income, employment status, etc.), farm characteristics (farm tenure, total owned and operated land, and distance of home from land), information regarding producers’ awareness of, sentiments about, and participation in CRP.

2.4. Conceptual framework

Producers with eligible agricultural lands can choose to enroll in General CRP. According to the random utility theory, they anticipate the costs and direct and indirect benefits when deciding whether to enroll or not (Ochieng et al., Reference Ochieng, Afari-Sefa, Muthoni, Kansiime, Hoeschle-Zeledon, Bekunda and Thomas2022; Pham et al., Reference Pham, Chuah and Feeny2022). A rational producer compares the expected utility received during the contract period (between 10 and 15 years after enrollment) with the expected utility received during the same period of non-enrollment. In complex and uncertain situations involving risk, attitudes and social norms can guide decision-making more effectively than relying solely on calculated expected utility. Regarding the benefits of non-enrollment in CRP, producers presumably continue to earn income from crop production or other land uses. Accordingly, under the random utility framework, the expected utility for the i ^th producer in any enrollment year (t) from the enrollment in CRP can be represented as (π _eit ), which is compared to the expected utility with non-enrollment (π _neit ). The i ^th producer will choose to participate in CRP in any given year (t) if the expected utility (benefit) from enrolling in CRP is greater than the utility (benefit) of not enrolling in CRP (Plantinga et al., Reference Plantinga, Alig and Cheng2001). The expected net utility (U _i *) by the producer in any given enrollment year (r _t ) can be represented as:

(1) $$U_{i}^{*}=\left[\pi _{eit}-\pi _{neit}\right]\gt 0$$

Enrollment in General CRP is a stepwise (i.e., multi-stage) and dynamic process (i.e., the enrollment process can change based on eligibility criteria, changes in program rules, changes in environmental priorities, etc.) affected by policy changes in which producers, based on their knowledge and experiences, form an attitude towards the program (i.e., expected net benefits, suitability with existing agricultural conditions, and CRP attributes) over time and then decide to enroll or not enroll at a particular time. However, the decision to enroll in General CRP also varies with individual producers, which might be related to their household socio-demographics, farm characteristics, perspectives on CRP enrollment process, and perceptions and attitudes. Furthermore, external factors such as agriculture and energy policies could also influence CRP enrollment. We used a discrete-time duration analysis model to examine the key determinants influencing how soon enrollment in General CRP occurs. This model assumes that producers who have higher expected returns from enrolling in General CRP enroll sooner. The early enrollees are less likely to be influenced by time-varying factors, as they have already committed to the program and are receiving its benefits.

2.5. Econometric model specification

Duration analysis deals with the simultaneous adoption and diffusion of practice or technology within a dynamic process that helps in identifying barriers to adoption and strategies to promote faster and more widespread diffusion (Alcon et al., Reference Alcon, de Miguel and Burton2011; Genius et al., Reference Genius, Koundouri, Nauges and Tzouvelekas2013). More recently, duration analysis has been widely used in the adoption of happy seeder technology (Kaur et al., Reference Kaur, Singh, Anand and Singh2023), laser land leveler technology (Sheikh et al., Reference Sheikh, Mugera, Pandit, Burton and Davies2022), conservation tillage (D’Emden et al., Reference D’Emden, Llewellyn and Burton2006; Gao et al., Reference Gao, Ren, Lu and Feng2023), sustainable agricultural practices (Dadi et al., Reference Dadi, Burton and Ozanne2004; Yigezu et al., Reference Yigezu, Mugera, El-Shater, Aw-Hassan, Piggin, Haddad, Khalil and Loss2018), conservation agriculture (Khataza et al., Reference Khataza, Doole, Kragt and Hailu2018), and organic farming (Burton et al., Reference Burton, Rigby and Young2003; Kallas et al., Reference Kallas, Serra and Gil2010). Our study used duration analysis to examine the factors influencing the length of a spell, i.e., the length of time taken to enroll in General CRP. The spell starts when CRP enrollment is initiated for the first time (in 1985) or the year in which the producer began managing/operating the farm, if that is later; the spell ends when a producer enrolls the farm in the General CRP. Producers who enrolled before the end of the survey are called uncensored producers. We only included uncensored producers because, due to our sample frame, our sample does not include those who have never participated in the program. In cross-sectional surveys, the researchers are dependent on recall data on the first enrollment date of General CRP or the year in which a producer began managing/operating the farm, if that is later; these dates are usually reported on grouped or interval-censored scales, such as months or years (An and Butler, Reference An and Butler2012; Burton et al., Reference Burton, Rigby and Young2003; Khataza et al., Reference Khataza, Doole, Kragt and Hailu2018). It is reasonable to apply a discrete-time duration analysis model over the continuous-time duration analysis approach for interval-censored (discrete time) data (Jenkins, Reference Jenkins1995; Khataza et al., Reference Khataza, Doole, Kragt and Hailu2018; Tiller et al., Reference Tiller, Feleke and Starnes2010). Thus, we applied the discrete-time duration model, and the model specification for it is as follows (Jenkins, Reference Jenkins1995; Khataza et al., Reference Khataza, Doole, Kragt and Hailu2018):

Let ′T′ be a random variable, which represents the discrete-time (T ≥ 0) before enrollment in General CRP (in years); f(t) represents a continuous probability density function of a random variable (T), where ′t′ is the time duration from initiation of General CRP or the year in which a producer began managing/operating the farm, if that is later, until enrollment in General CRP; F(t) refers to the corresponding cumulative density function of a random variable (T) (describing the probability that a producer will have enrolled in General CRP before the time ′t′), which is given by:

(2) $$ F\left(t\right)=\int _{0}^{t}f\left(s\right)\,{\rm d}s= {\rm Pr} \; \left(T\leq t\right),\quad t\geq 0 $$

The survival function (S(t)) refers to the probability of surviving (i.e., not enrolling in General CRP) until time ′t′, which can be expressed as:

(3) $$ S\left(t\right)=1-F\left(t\right)={\rm Pr}\; \left(T\gt t\right) $$

The hazard function (h(t)) estimates the probability of enrolling in General CRP at the time ′t′, given that the producer has not enrolled before the time ′t′, which can be specified as follows.

(4) $$ h\left(t\right)=\lim _{{\rm d}t\rightarrow 0} \left[{\Pr \left(t\leq T\lt t+{\rm d}t|T\geq t\right) \over {\rm d}t}\right]=\lim _{{\rm d}t\rightarrow 0} \left[{{\rm F}\left({\rm t}+{\rm d}t\right)-{\rm F}\left({\rm t}\right) \over {\rm d}t\left(1-F\left(t\right)\right)}\right]={f\left(t\right) \over S\left(t\right)} $$

The hazard function depends on the elapsed time length and explanatory variables/covariates (e.g., household socio-demographics, farm characteristics, CRP attributes, institutional factors) on the time duration until enrollment. A proportional hazard specification examines the effects of both the baseline hazard and explanatory variables on a hazard function (Jenkins, Reference Jenkins1995; Khataza et al., Reference Khataza, Doole, Kragt and Hailu2018). The baseline hazard function is only a function of time but not any covariates, and it is common for all producers, i.e., enrollment is associated with time only. In contrast, the relative hazard is contingent on explanatory variables or covariates, which vary with each producer (Yigezu et al., Reference Yigezu, Mugera, El-Shater, Aw-Hassan, Piggin, Haddad, Khalil and Loss2018). For the estimation of the proportional hazard model, parametric duration models (Weibull, exponential, logistic, log-normal, log-logistic, and Gompertz) were used for continuous-time duration data (Kiefer, Reference Kiefer1988), while logistic and complementary log–log specifications were used for discrete-time duration data (Khataza et al., Reference Khataza, Doole, Kragt and Hailu2018). The generalized representation of the proportional hazard model is specified as follows:

(5) $$ h\left(t,X\right)= h_{0}\left(t\right)\exp \left(\beta^\prime X\right) $$

where h(t, X) is the hazard rate at the time ′t′; h ₀(t) is the baseline hazard function, which depends on ′t′ (but not on any covariates ′X′); exp (β′X) is a producer-specific non-negative function of covariates ′X′, which scales the baseline hazard function.

We used a complementary log–log specification for the proportional hazard model because it is the discrete-time equivalent of the continuous-time proportional hazard model (Bontemps et al., Reference Bontemps, Bouamra-Mechemache and Simioni2013; Jenkins, Reference Jenkins1995; Khataza et al., Reference Khataza, Doole, Kragt and Hailu2018), which can be represented as follows:

(6) $$\ln \left(-\ln \left[1-h\left(t,X\right)\right]\right)=h_{0}\left(t\right)\exp \left[X\left(t\right)\beta +e_{i}\right]=\alpha \gamma _{t}+\beta^\prime X+e_{i} $$

where h(t, X) is the hazard rate for the t ^th time interval; h ₀(t) is the baseline hazard function; γ _t is a time variable capturing the pattern of duration dependence (baseline hazard); α and β are the unknown parameters to be estimated; X is a vector of covariates; e _i is an error term representing producer-specific characteristics, unobserved variables (e.g., market conditions, policy changes, motivation, risk tolerance, life-events, etc.). All coefficients for the discrete-time duration analysis model are reported in terms of hazard ratios (HRs), which are reported in the exponential form (e ^β′). The coefficient of HR (the null hypothesis of HR is not different from one) is interpreted as a proportional shift in the hazard function to a per unit change in the covariates (Beyene and Kassie, Reference Beyene and Kassie2015). For instance, an HR of two for a particular covariate implies a doubling of the hazard rate with a one-unit increase in the covariate (Khataza et al., Reference Khataza, Doole, Kragt and Hailu2018). This model also assumes a higher hazard rate for early enrollees than for late enrollees in General CRP. In this complementary log–log model, explanatory variables or covariates can enter the model in two ways: time-invariant covariates (e.g., gender) enter as fixed predictors across all time intervals, while time-varying covariates (e.g., policy changes) are allowed to change at each discrete time point. These covariates influence the hazard or probability of the event (i.e., enrolling in General CRP) occurring at each time point, conditional on survival up to that point.

2.6. Data and variable description

The dependent variable in our model is the time duration between the year 1985, when CRP was first initiated or the year a producer begins to manage/operate the farm, if that is later, until the moment the producer enrolls in General CRP. The explanatory variables in the model include socio-demographic factors (gender, age, employment status, annual income, educational status, and race), farm characteristics (farm tenure type and total area of owned and operated land), and awareness about CRP incentives and perspective with CRP attributes (sign-up ranking process, enrollment process (office and site visit), rules and regulations, grass establishment, and tree establishment). A description of all model variables appears in Table 1.

Table 1. Description of dependent and explanatory variables used in the discrete-time duration analysis model

Note: *Total land area: We rescaled the total land area by dividing by 100. As the original variable’s value decreases by 100, its original regression coefficient (0.00016) increases by 100 to balance this change in the original variable’s value. However, we reported the hazard ratio in our study; it changes the original hazard ratio (1.00016) to a new hazard ratio (1.016) after rescaling.

** CRP awareness: CRP awareness means knowledge about CRP objectives, eligibility criteria for enrollment, knowledge of the application process for enrollment, knowledge of enrollment types, knowledge about program benefits.

For the estimation of a discrete-time duration analysis model, the original data structure was reorganized from one producer with one observation format to one producer with multiple observation formats (Jenkins, Reference Jenkins1995). In other words, a unique producer can have as many data rows of observations as the total number of time intervals (i.e., years) with a risk of the event (i.e., enrolling in CRP) occurring. A unique producer consists of data rows of zeros (since the inception of General CRP Sign-up in 1985 or the year in which the producer began managing/operating the farm, if that is later) and is replaced with a value of 1 when the producer had enrolled in General CRP in time t ≤ T. Reorganizing our data in this way also helps to incorporate time-varying variables in the model and easy estimation of log-likelihood for the model (Jenkins, Reference Jenkins1995).