Florida Water Management

Florida is one of the top 12 U.S. states for irrigated agricultural land acreage (Schaible and Aillery Reference Schaible and Aillery2012), it is fourth among U.S. states in water withdrawals for public supply (Dieter et al. Reference Dieter, Maupin, Caldwell, Harris, Ivahnenko, Lovelace and Linsey2018; Marella Reference Marella2014), and it has a significant tourism industry dependent on water resources. The Floridan aquifer underlying the state supplies drinking water for nearly 10 million people and supports agricultural and recreational needs (Marella and Berndt Reference Marella and Berndt2005). However, due to climate change, population growth, and land development, there is increasing pressure on water supplies. The total withdrawal of water in Florida accounted for 15,319 million gallons per day (Mgal/d) in 2015; of which 5,721 Mgal/d were freshwater (Marella and Dixon Reference Marella and Dixon2018). In 2015, 46 percent of the total freshwater withdrawals were used for public supply, including tourism, landscape irrigation of lawns and golf courses, and recreation, followed closely by self-supplied water for agricultural production (40%) (Bellino Reference Bellino2017; Marella and Dixon Reference Marella and Dixon2018). Interestingly, these sectors are also the most affected by reductions in water levels and water quality impairments, given their dependency on freshwater resources. Given a finite supply of water, there is direct competition between extractive water users, such as agricultural producers and households, and in-stream water uses, such as aquatic ecosystem support and tourism and recreation.

In Florida, agricultural lands total about 8 million acres, of which 24 percent are irrigated for crops. Forecasts through 2040 suggest a four percent increase in irrigated acreage associated with a 14 percent increase in water use statewide (FDACS 2017). In addition, the plethora of surface water resources and vast freshwater aquifers of the state are imperiled by nutrient pollution (from both agricultural and non-agricultural sources) and increasing water withdrawals through the state by a variety of water users. As a result, large areas of the state are classified as “water resource caution areas” (FDEP 2011). The forecast of water use in these areas suggests that they will undergo severe water supply problems within the next 20 years (FDEP 2011).

The state of Florida utilizes BMAPs to set goals by industry that, if met, would enable the basins to meet established Total Maximum Daily Loads (TMDLs) which set pollution loading limits for specific bodies of water. BMAPs are the “blueprints” for pollution reduction to improve water quality. They include a “set of strategies such as permit limits on wastewater facilities, urban and agricultural best management practices, conservation programs, financial assistance and revenue generating activities” (FDEP 2011). Basins that include significant agricultural production set target BMP adoption rates as part of their BMAPs (FDEP 2018). BMPs include specific practices designed to limit the amount of fertilizers, pesticides, animal waste, and other pollutants entering water resources from agricultural production and practices in order to reduce total water used per unit of production.

Previous Literature

Relevant literature includes studies using choice experiments to study the adoption of various agricultural or land management practices, studies analyzing the adoption of agricultural best management practices in general, and most relevant for our work, studies analyzing best management practices pertaining to water management.

Many previous studies have used choice experiments to analyze willingness to pay for environmental attributes or willingness to accept and adopt sustainable practices. Studies vary from valuing wetland attributes for recreational purposes to improved quality for domestic water demand (Alcon et al. Reference Alcon, Tapsuwan, Brouwer and Miguel2014; Barton and Bergland Reference Barton and Bergland2010). Choice experiments have also been used to analyze the adoption of sustainable practices in forestry and risk management in agriculture. Horne (Reference Horne2006) investigates factors that affect forest owners’ willingness to accept a contract for biodiversity conservation in Finland. She uses five attributes: the initiator of the conservation contract; the restrictions on forest use; the compensation per hectare annually; the duration of the contract; and the cancellation policy. Her findings suggest that respondents prefer to be the initiator of the contracts rather than having them imposed on them by a forest organization, environmental organization, or conservation trust. They also prefer to conserve small patches of forest under short contracts with the flexibility to opt out from the contract at their will. Looking into the Chinese forestry sector, Qin, Carlsson, and Xu (Reference Qin, Carlsson and Xu2009) find that shorter contract lengths, shorter wait times before harvest, and lower risk of contract termination increase the farmers’ perceived value of a forestland contract. With regard to risk management in agriculture, Mercadé et al. (Reference Mercadé, Gil, Kallas and Serra2009) investigate farmers’ preferences for crop insurance in Catalonia. Lower insurance cost, increased types of risk covered, and lower minimum thresholds for payment were the main factors associated with increased probability of adoption.

A large literature considers adoption of best management practices. Two meta-analyses summarize this extensive literature, with one focusing on studies prior to 2008 (Prokopy et al. Reference Prokopy, Floress, Klotthor-Weinkauf and Baumgart-Getz2008) and the second focusing on studies published between 2008 to 2017 (Liu, Bruins, and Heberling Reference Liu, Bruins and Heberling2018). Common factors associated with adoption include farmers’ characteristics (income, social networks, demographics, knowledge, risk preferences, and environmental awareness), farm characteristics (land tenure, farm size, fertility, slope, altitude, proximity to urban areas), the characteristics of the best management practices (cost-effectiveness, time requirement, ease of use, flexibility, observability, potential for spatial and temporal spillover effects) (Prokopy et al. Reference Prokopy, Floress, Klotthor-Weinkauf and Baumgart-Getz2008, Liu, Bruins, and Heberling Reference Liu, Bruins and Heberling2018). Among other points, the authors suggest the understanding of the decision-making process through the analysis of farmers’ preferences is an area in need of more research. Therefore, one goal of this article is to fill this void by investigating the farmers’ preferred cost-share program attributes for water management BMPs in Florida.

Previous work utilizing choice experiments for water management options are most pertinent for our study. Alcon et al. (Reference Alcon, Tapsuwan, Brouwer and Miguel2014) explore farmers’ willingness to pay a premium for irrigation water in Spain. Their results suggest that a guaranteed quantity of water supplied increases a farmer's willingness to pay. Interestingly, willingness to pay based on water source varies by region. In regions with ample water supply, retaining their current water source is associated with higher willingness to pay, whereas in regions with water shortages, utilizing new water sources increases willingness to pay. In South Africa, Saldias et al. (Reference Saldias, Speelman, Van Huylenbroeck and Vink2016) attempt to understand farmers’ preferences for wastewater reuse in agricultural irrigation. Their findings show that respondents are willing to adopt reclaimed wastewater if (i) the water quality is guaranteed, (ii) there are low restrictions on water usage for irrigation, and (iii) the water service providers are privately managed.

Despite the vast literature on adoption of environmentally friendly practices, to the best of our knowledge, no work has considered the effects of the implementing agency on probability of adoption. Within most states, implementation of an agri-environmental program could be done through either the department of agriculture or the department of environmental protection. Our results suggest the former will lead to increased adoption relative to the latter. Similarly, allowing states to implement programs as opposed to federal implementation could increase adoption.

Theorical Framework and Empirical Model

We model producer decisions using random utility theory (McFadden Reference McFadden and Zarembka1974), which implies that every producer is a rational decision-maker i maximizing utility, U _i, based on his or her choice set of available practices. Mathematically, utility received from choosing practice j will be represented as:

$$U_{ij} = V_{ij} + \varepsilon _{ij}$$

where V _ij represents the deterministic component of expected utility, while ɛ_ij is the random component of utility. In an agricultural context, utility is largely determined by profit, but as our results will show, growers also have preferences for things like agencies with whom they prefer to work, making a utility a more all-encompassing metric to use.

The functional form for the utility can be specified as:

$$U_{ij} = {\beta }^{\prime}X_{ij} + {\delta }^{\prime}Z_i + \varepsilon _{ij}$$

where X _ij is a vector of observed alternative-specific attributes, and Z _i is a vector of producer and farm characteristics. Based on previous literature, Z _i includes demographic characteristics of the producer (age, experience, gender, education, income), farm attributes (acreage, crop type, organic certification status), experience with previous cost-share programs, and perceptions of alternative practices. β and δ are assumed to be constant across respondents, while the error term is assumed to be an independently and identically distributed Gumbel distribution.

Producer i will choose an alternative j over alternative k if U _ij > U _ik. Considering alternatives j and k, and assuming the random component in the utility function follows the Type 1 extreme value distribution (i.e., the Gumbel maximum distribution), the probability of choosing alternative j for individual i is given by:

$$P_{ij} = \displaystyle{{e^{V_{ij}}} \over {e^{V_{ij}} + e^{V_{ik}}}}$$

We use the conditional logit to estimate parameters β and δ. This model assumes homogeneous preferences across respondents and the independence from irrelevant alternatives (IIA) property, which suggests that the relative probabilities of two options being chosen are unaffected by introduction or removal of other alternatives. If the IIA property is violated, the CL estimates will be biased, and we must use a discrete choice model that does not require the IIA property, like the random parameter logit (RPL)—also known as the Mixed Logit (ML) model. We test for the IIA property of the CL model using the test of Hausman and McFadden.

Data

Choice experiment design

The survey included a choice experiment with a fractional factorial design with a D-efficiency of 97.57. This design included 18 choice sets blocked into three groups of six. Each respondent received one block. A design with full efficiency would have required 36 choice sets. This option was discarded out of concern for survey length and sample size. For each set, the respondent was asked to choose which of two plans (A or B) in which they would hypothetically enroll to cover a portion of the costs of adopting soil moisture sensors, annual crop and soil tissue testing, and use of UF/IFAS nutrient recommendations. The respondent also had the option of rejecting both plans (Figure 1).

Figure 1.

Choice Experiment Instructions and Example Choice Set

The cost-shares varied along five attributes: the cost-share contracting agency, the duration of the contract, the type of costs included in the cost-share, the monitoring method used to verify the producer's practices, and the percent of costs covered by the agency (Table 1). For contracting agencies, we consider a state-level agricultural agency (Florida Department of Agriculture and Consumer Services or FDACS), a state-level environmentally focused agency (Florida Department of Environmental Protection or FDEP), and a federal-level agricultural agency (United States Department of Agriculture or USDA). We hypothesize that producers may have a preference for agriculture agencies over environmental agencies and a preference for local over non-local. The duration of the contract has three levels in our choice experiment: 5, 10, and 15 years. According to previous literature, we hypothesize that respondents will be less likely to choose longer contracts. For the verification process, we consider self-reporting by producers or on-farm visits by the contracting agency's agents. Agricultural producers tend to value privacy, so we hypothesize that self-reporting would be preferred to on-farm visits. The fourth attribute is the costs included in the program. We consider two options: covering both the initial investment (primarily soil moisture sensor installation) and the annual soil and tissue testing costs, or covering only the initial investment. In the latter case, the annual testing costs will be paid entirely by the producer. Given the high cost of initial investments, we are uncertain, ex ante, if growers will prefer subsidies for annual investment costs over just the initial investment. Finally, we consider three levels of cost-share: 30 percent, 50 percent, and 70 percent of incurred costs, intuitively hypothesizing that higher subsidies should induce higher hypothetical enrollment rates. Respondents were not given costs for practices because these would vary by crop and region. For example, the number of soil moisture sensors required depends on the heterogeneity of soil type and topography within a farmer's operation (Zotarelli, Dukes, and Paranhos Reference Zotarell, Dukes and Paranhos2013), and their cost can vary substantially depending on the degree of functionality and compatibility with other equipment like automated irrigation systems desired by the producer. This variation in cost does introduce uncertainty about costs incurred by individual respondents. As such, we avoid converting model coefficients into measures of willingness to accept for cost-share attributes.

Table 1.

Attributes and Levels of the Choice Experiment

Choice experiment data collection

The survey was conducted via mail following Dillman's method during the summer of 2018. Due to budget limitations, we did not provide incentives to participants. The mailing targeted all farmers who are currently enrolled in the FDACS BMP program. Consequently, these farmers have adopted at least one BMP, but the vast majority have not yet adopted all key practices. This sample was chosen because it was the largest mailing list of growers available. These are also growers who have demonstrated an interest in adopting BMPs and working with FDACS to do so. These growers are the most likely to enroll in future cost-share programs. A total of 4,279 surveys were distributed. Five hundred forty-one completed surveys were received, resulting in a response rate of 12.6%. This falls within the reported range (8.1% to 15%) of response rates for Florida agricultural producers found in the literature across a range of topics, including both environmental and production-oriented surveys (Kreye, Pienaar, and Adams Reference Kreye, Pienaar and Adams2017; Milleson, Shwiff, and Avery Reference Milleson, Shwiff and Avery2006; Paxton et al. Reference Paxton, Mishra, Chintawar, Roland, Larson, English, Lambert, Marra, Larkin, Reeves and Martin2011; van Dijl, Grogan, and Borisova Reference van Dijl, Grogan and Borisova2015).

The summary statistics of the sample are presented in Table 2. Most of the respondents are male (84%) with an average age of 62 years old. While the average age is similar to that found in the USDA Census of Agriculture (2017) for Florida, our respondents include a smaller proportion of females than the state overall. Collectively, the individuals surveyed have substantial experience in the agriculture industry, which is reflected by an average of 27 years of experience among the respondents. Comparison data are not directly available, but about 73 percent of Florida farmers report at least 11 years of experience in the Census of Agriculture (USDA 2017). The average respondent farm size is about 666 acres, relative to an average size of 201 acres reported in the Census (2017). Our sample includes several large cattle ranches that skew our acreage distribution. Our results may be less applicable to smaller farms. However, from an environmental perspective, large farms will have a larger impact, so this sample selection bias is not as worrisome as a bias towards small farms would have been.

Table 2.

Summary Statistics

^a Comparison data obtained from USDA, 2017.

^b Comparison data obtained from USDA, 2018.

Our respondents primarily raise/grow cattle (33.6% of the respondents), citrus (12.5%), ornamentals plants (9.8%), and hay (8.9%). The most used irrigation type is micro-irrigation (28.5%), followed by overhead sprinklers (26%). Almost half of the respondents (47%) do not use any irrigation system. The Census of Agriculture reports that 76.4% of Florida operations do not use irrigation (USDA 2017). This suggests that many non-irrigating producers chose not to complete the survey. As with farm size, the direction of this selection bias is less troubling than the reverse because we are most interested in water conservation behaviors, which is most relevant for farms with irrigation systems.

Over 60 percent of farmers plan to pass on their farms to a family member for continued agricultural use upon retirement. Fifty-one percent of the sample report an annual income above $100,000. Table 3 presents the participation rate in cost-share programs for the respondents. Except for the Environmental Quality Incentives Program, local or state agency programs have higher participation than federal agency programs.

Table 3.

Participation in Cost-Share Programs (N= 541)