Background and literature review

Citrus production in Florida and Texas faces several challenges related to pests, soil quality, and weather fluctuations. While damages from freezes and water shortage are more common in Texas, pest population and soil quality are the major challenges for citrus producers in Florida. One of the biggest challenges to citrus production is HLB, or the citrus greening disease, that was first detected in Florida in 2005. HLB is an incurable disease caused by the Candidatus Liberibacter asiaticus (CLas) bacterium and is vectored by the Asian citrus psyllid. HLB has also impacted the citrus industry in Texas, although to a much lesser extent than in Florida (Graham et al. Reference Graham, Gottwald and Setamou2020).

HLB is responsible for widespread losses to the citrus industry in Florida and has also caused production costs to increase each year since 2005 (Moss et al. Reference Moss, Grogan, Farnsworth and VanBruggen2015). Fruits from HLB-infected trees are sour and bitter making them unsuitable for consumption. Between Florida’s 2006–2007 and 2010–2011 growing seasons, HLB-related losses were estimated at $4.51 billion in GDP and $8.62 billion in revenue (Hodges and Spreen Reference Hodges and Spreen2012) or nearly $1 billion per year (Farnsworth et al. Reference Farnsworth, Grogan, van Bruggen and Moss2014). Thus, reducing the impact of HLB is crucial for long-term sustainability of the citrus industry. Given that there is no natural cure for HLB, management practices to mitigate the effects of the disease are diverse with varying effectiveness. While there is no consistent management practice that could solve the HLB issue, management practices that improve soil quality, such as using cover crops, can lead to improved soil quality and tree health and allow growers to continue producing in HLB endemic conditions.

A general theme for fighting a disease in tree crop production is cultivating stronger trees that are more capable of fighting infection. Cultivar selection and breeding to develop a tree that is tolerant or resistant to the bacteria causing the disease is an ongoing goal for citrus researchers (Dutt et al. Reference Dutt, Barthe, Irey and Grosser2015). Until these cultivars are available, strategies to improve tree health are continually being studied and researchers are looking for solutions to a range of issues targeting each step in the citrus production process. Conversely, growers’ actions have been motivated by short- and long-term costs, farm location and size, and trust in the effectiveness of various solutions. Thus, disease management in citrus, particularly for HLB, is characterized by heterogeneous practices and inconsistent outcomes, which could deviate from the ideal management practices recommended by researchers.

Determining the best management practices or bundle of practices currently available for growers is critical for maintaining the citrus industry. The cost of implementing multiple strategies may be high. However, when compared to the economic impacts of citrus greening disease, the benefits may outweigh the risk (Alvarez et al. Reference Alvarez, Rohrig, Solís and Thomas2016). Cover crops could be a viable addition to the management practices that aim to improve the growing conditions of citrus trees and ameliorate the impacts of HLB and other pests in citrus groves.

Survey design and data collection

We develop and use a survey to gather information from citrus growers and use a double-bounded dichotomous choice model to measure WTP for cover crops. Our survey starts with questions on growers’ demographics, including whether and what type of citrus fruits they grow. The survey next asks whether the respondents used cover crops and elicits respondents’ levels of agreement or disagreement on benefits and drawbacks of using cover crops in citrus. Respondents also ranked benefits associated with using cover crops according to the inherent value that they provide to the grower.

For questions related to growers’ valuation of cover crops, respondents were asked if they would be willing to implement cover crops at no cost to them. If they answered “yes,” they moved on to the question representing the next part of the double-bound model, which adds an additional interval to explain a skew in data from respondents that are unwilling to implement at any price. The respondents were next asked whether they would be willing to implement cover crops at a “middle” price – a price that was semi-randomly generated. A follow-up question, with a “higher” or “lower” price (semi-randomly generated), was asked if the response to the first question was “yes” or “no,” respectively. There was one respondent that answered “no” to implementing cover crops at no cost but indicated that they had already adopted cover crops. We assigned our estimated price of cover crop adoption ($850/acre per year) to this respondent. In our survey design, respondents that answered no to adopting cover crops at no cost and had not adopted cover crops had their survey terminated.

We arrived at estimates of the semi-randomly generated prices from two sources. First, we obtained citrus budget items and figures for existing groves from Singerman (Reference Singerman2021).Footnote ¹ The budgets served as benchmarks for typical costs of citrus production. Second, we obtained the total per acre costs of implementing cover crops from four anonymous growers who are using cover crops in their groves. The growers were early adopters of cover crops and had been using them for multiple years in their groves. They indicated that it took them several seasons before finding the exact mix of cover crop species, the optimum method of planting the seeds, management costs, and the termination method. This was especially true for Florida since the citrus growing regions in the state are primarily sandy. Thus, the growers tried and tested various mixes of legume and non-legume seeds before choosing the right mix for increasing both SOM and nitrogen levels in the soil. We took a 4-year average of the costs of using cover crops for the four growers. We found that it cost an average of $500/acre for cover crop seeds for the growers. Costs varied by the type of cover crop species used.

The growers indicated that for terminating cover crops, they mowed them and left the residues to decompose and mix with the soil. For mowing costs, we spoke to the growers and referred to the “Central Florida (Ridge) and Indian River-Southwest Florida Citrus Custom Rate Charges” report that is published in Singerman and Aiya (Reference Singerman and Aiya2022). It costs on average $14/acre to mow an 8–12 feet wide row middle in 2021–2022. This translated to $28/acre for two applications required per year. In addition to mowing, management costs included costs of renting a seed broadcaster or a no-till planter, an ATV, and scouting for pests. Thus, the total management and cover crop termination cost was $100/acre per year. Additional labor and supervision cost an additional $250/acre per year. Including additional labor costs and costs of cover crop management and termination, we found that the total initial cost of implementing cover crops was $850 per acre per year. For the survey, we set the “middle” price to be randomly generated within the $600–$800 interval, while “low” and “high” prices were randomly generated from the $100–$500 and $900–$1200 price intervals. These prices were randomly generated for each respondent by Qualtrics.

After measuring the respondent’s WTP, the survey asks additional demographic, practice, and behavioral questions. These questions captured management practices, involvement in conservation programs, and overall farm diversity. Responses pertaining to experience in citrus farming, availability of credit, percentage of land rented, and farm size were also requested. Finally, to eliminate duplication and allow for follow-up questions, farmers were asked for contact information, with promised confidentiality.

Survey respondents were recruited through various channels beginning in early 2021: Qualtrics, University of Florida Citrus Extension Agents’ email listservs, cold calling groves and nurseries, the Indian River Citrus League email listserv, the California Citrus Research Board e-newsletter, and in-person interviews at the 2021 Citrus Show in Ft. Pierce, Florida. Data collection ended on November 2, 2021. The response rate to our survey was 35%, and in total, we received responses from 359 growers from Florida, Texas, California, and Georgia, of which 173 identified themselves as citrus growers. Samples from California and Georgia were not representative of commercial citrus producers and we excluded them from our study. After removing inconsistent and incomplete observations, we had 79 observations with 59 from Florida and 20 from Texas.

Estimation strategy

Using cover crops in citrus is not common, and to our knowledge, no study has tried to elicit WTP for cover crops in citrus in the USA. Given the direct and indirect benefits from using cover crops and the lack of published estimates for costs of adopting them, we used the CV method to estimate WTP for growers. In this approach, the estimates of the respondents’ “valuation” of an environmental good or a conservation practice is “contingent” upon the hypothetical market or the scenarios that we present to them (Hanemann Reference Hanemann1994; Carson and Hanemann Reference Carson and Hanemann2005). To elicit respondents’ WTP, we used the double-bounded dichotomous choice approach. This approach starts with a simple dichotomous choice question, which is followed by another set of questions that is contingent upon the information provided in the earlier question. The double-bounded CV approach yields more efficient estimates of WTP than the single dichotomous choice model, in which only a single question is asked (Hanemann et al. Reference Hanemann, Loomis and Kanninen1991).

A respondent i’s WTP for cover crops can be assumed to be related to their demographic as well as grove information through the relationship in equation 1, where ${z_i}$ is the vector of explanatory variables, and ${u_i}$ is the error term that is assumed to be normally distributed. We do not directly observe $W{P_i}$ and instead know whether they fall in the specific price intervals created in the survey.

(1) $$WT{P_i} = {z_i}{\rm{\beta }} + {\rm{\;}}{u_i}$$

A respondent first responds either “yes” or “no” to whether they would adopt cover crops at no cost. If they answered affirmatively, they are presented with a pair of dichotomous choice questions, the latter of which is dependent on the response to the earlier question. These dichotomous choice questions require either “yes” or “no” responses to hypothetical costs of implementing cover crops. Given that a respondent can only progress to the double-bounded dichotomous questions if they answer “yes” to implementing cover crops at no cost, the four possible outcomes for the pair of dichotomous questions are (yes, yes), (yes, no), (no, yes), or (no, no). Let $y_i^1$ represent the response to the first question in the double-bounded model (randomly generated middle price), $y_i^2$ represent the response to the second question in the double-bounded model (which is either high or low price based on the response to the previous question), ${t_1}$ represent the first price, which is the middle price, and ${t_2}$ represent the second price, which could be either high or low price depending on the response to the first part. Then, for a respondent whose WTP lies in the interval that is greater than or equal to the middle price or less than the high price in the double-bounded dichotomous choice model, the following is the probability of a potential outcome from the double-bounded discrete choice model (the equations are adopted from Lopez-Feldman (Reference Lopez-Feldman2012)):

$$y_i^1 = 1\;and\;y_i^2 = 0$$

$$ = \gt \Pr \left( {y,n} \right) = r\left( {{t^1} \le WTP \lt {t^2}} \right)$$

$$ = \gt \Pr \left( {y,} \right) = \Pr \left( {{t^1} \le z_i^{\prime}\beta + {u_i} \lt {t^2}} \right)$$

(2) $$= \gt \Pr \left( {y,n} \right) = {\rm{\Phi }}\left( {z_i^{\prime}{\beta \over \sigma } - {{{t^1}} \over \sigma }\;} \right) - {\rm{\Phi }}\left( {z_i^{\prime}{\beta \over \sigma } - {{{t^2}} \over \sigma }} \right)$$

Similarly, when the WTP is greater than or equal to the high price in the double-bounded dichotomous choice model, the probability of the outcome is shown in equation 3. The responses thus comprise the right-censored data.

$$y_i^1 = 1\;and\;y_i^2 = 1$$

$$ = \gt \Pr \left( {y,y} \right) = Pr\left( {WTP \gt {t^1},\;WTP\; \ge {t^2}} \right)$$

$$ = \gt \Pr \left( {y,y} \right) = \Pr \left( {z_i^{\prime}\beta + {u_i} \gt {t^1},z_i^{\prime}\beta + {u_i} \ge {t^2}} \right)$$

(3) $$ =\gt \Pr \left( {y,y} \right) = {\rm{\Phi }}\left( {z_i^{\prime}{\beta \over \sigma } - {{{t^2}} \over \sigma }} \right)$$

When the WTP lies in an interval that is less than the middle price but greater than or equal to the lower price, equation 4 represents the probability of the outcomes.

$$y_i^1 = 0\;and\;y_i^2 = 1$$

$$= \gt \Pr \left( {n,y} \right) = Pr\left( {{t^2} \le WTP \lt {t^1}} \right)$$

$$ = \gt \Pr \left( {n,y} \right) = \Pr \left( {{t^2} \le z_i^{\prime}\beta + {u_i} \lt {t^1}} \right)$$

(4) $$ = \gt \Pr \left( {n,y} \right) = {\rm{\Phi }}\left( {z_i^{\prime}{\beta \over \sigma } - {{{t^2}} \over \sigma }\;} \right) - {\rm{\Phi }}\left( {z_i^{\prime}{\beta \over \sigma } - {{{t^1}} \over \sigma }} \right)$$

Finally, equation 5 represents the probability of outcomes when the observed WTP is less than the middle as well as the lower price. These responses are thus the left-censored data, that is, the responses for which the lower bound is unknown.

$$y_i^1 = 0\;and\;y_i^2 = 0$$

$$ = \gt \Pr \left( {n,n} \right) = Pr\left( {WTP \lt {t^1},\;WTP\; \le {t^2}} \right)$$

$$ = \gt \Pr \left( {n,n} \right) = \Pr \left( {z_i^{\prime}\beta + {u_i} \lt {t^1},z_i^{\prime}\beta + {u_i} \le {t^2}} \right)$$

(5) $$ = \gt \Pr \left( {n,n} \right) = 1 - {\rm{\Phi }}\left( {z_i^{\prime}{\beta \over \sigma } - {{{t^2}} \over \sigma }} \right)$$

The resulting maximum likelihood equation was estimated through the interval regression function in Stata. The interval regression function estimates the probability of a latent variable, in our case, $WT{P_i}$ in equation 1, belonging to specific intervals. Using the regression coefficients, we predicted both the mean and median WTPs (Yu et al. Reference Yu, Gao and Zeng2014). We also constructed normal distribution-based 95% confidence intervals around means and medians. We used bootstrap standard errors with 200 replications to construct the confidence intervals (Cawley Reference Cawley2008).

We ran separate regressions for Florida and Texas. This is because the citrus production systems for the two states are different: growers in Texas primarily produce organic oranges, while Florida growers are less likely to produce organic oranges.

Data

Out of the 173 respondents who identified as citrus growers, several responses were unusable because they were partially complete. We dropped responses from growers who had less than 0.5 acres of planted citrus and those that did not participate in our valuation method. Respondents were asked to provide the zip codes of their groves, which we used to identify their state. One respondent from Florida answered “yes” when asked whether they were currently using cover crops and “no” to whether they would be willing to implement cover crops at $0 cost. We assigned the benchmark price of $850 for cover crops to the upper and lower bounds for that respondent. We also dropped one response from Florida who was not currently using cover crops in citrus and did not want to implement them at $0 cost. Our final data consist of 79 observations – 59 from Florida and 20 from Texas.

Table 1 gives the summary statistics of orange acreage by state from the sample. The average acreage for a grower from the Florida sample is 3,034.2 acres, while that for Texas respondents is 155.1 acres. However, the median acreage is 80 acres for respondents from Florida and only 5 acres for those from Texas. Thus, the distribution of citrus acreage is skewed to the right for respondents for the two states. Based on total orange acreage and the number of orange groves in the Census of Agriculture, 2017, we estimated that the average orange grove size is between 187 and 199 acres for Florida. These include both commercial and non-commercial groves. The respondents to our survey were primarily commercial growers with larger groves. Moreover, if we exclude the outlier groves that have 10,000 acres or more under citrus, the average grove size is less than 400 acres for the Florida sample. For the Texas sample, the average grove size of 155.1 acres was higher than the estimated average grove size of 23.7 acres from the 2017 Census of Agriculture. However, 80% of the respondents have less than 100 acres of grove on average. Table 1 also gives the total orange acreage from the sample and the total orange acreage for each state, where the latter are USDA 2019–2020 estimates. Respondents from Florida represented a little more than 48% of the total orange acreage in the state, and respondents from Texas represented about 13% of the total orange acreage in their state. Thus, our sample is a strongly representative sample for Florida in terms of citrus acres.

Table 1. Orange acreage summary statistics (state acreage data from USDA 2019–2020)

In our sample, on average 3/4 of the area of a grove is covered with mature citrus trees, that is, by citrus trees that are 3 years or older. However, while 82% of citrus acreage was covered with mature trees in Texas, 76% of the citrus acreage were covered with mature trees in Florida, on average. In terms of tree density, most of the groves across the two states had 100–200 trees per acre, and between 30 and 35% of the groves had 200–300 trees per acre. Table 2 gives the summary statistics of the variables that we used in our WTP analyses.

Table 2. Summary statistics of key variables

Owning a no-till drill/planter could be an indicator of a grower’s inclination toward adopting a technology for conservation practice. Moreover, the ownership of a no-till drill is inherently exogenous to WTP for cover crops and is also included in our analyses to understand growers’ experience with specialized machinery. We hypothesize that ownership of a no-till drill would have a significant positive effect on growers’ WTP for cover crops. In our sample, 36% of growers in Florida own a no-till drill compared to 60% in Texas.

We include awareness of using cover crops in citrus in our analyses to check whether a grower who is aware of cover crops has significantly different WTP for cover crops than one who is not aware. On average, 79–80% of the respondents were aware of cover crops across both states. To maximize response rates, the survey did not ask questions on personal information. Instead, we use years of experience in citrus farming as a proxy for age as well as knowledge of citrus farming, and farm size instead of income in our analyses. On average, Florida and Texas growers had similar years of experience growing citrus. Farm size is strongly correlated with citrus acreage. Therefore, we used the latter since our survey is concerned about cover crop usage in citrus.

We include the number of crop types that a grower produces as an indicator of the amount of diversity within a grove. This could represent growers’ adaptability and stake in one crop. A grower from Florida on average had greater crop diversity than one from Texas. The primary benefits from using cover crops, such as increased SOM and nutrients, and consequent savings from reduced fertilizer use, take time to accrue. We included the percentage of a grower’s citrus grove that is planted on rented land as an indicator of how much an average grower may be willing to personally gain from long-term investments that enhance soil and crop quality. Roughly 16% of the sample citrus acreage in Florida is on rented land. This figure is much lower than Texas’ 34.2%.

Out of the 59 respondents in Florida, 24 respondents or roughly 40% of the sample were using cover crops in citrus, while 50% of the respondents in Texas were using cover crops in their groves (Table 3). Using cover crops in citrus is more common among growers that also grow organic citrus, with 12 out of 16 organic producers in Florida and 10 out of 14 organic producers in Texas using cover crops. Growing organic citrus is prevalent among growers in Texas, where 70% of the growers produced organic citrus as well. However, in Florida, only 27% of the growers produced organic citrus. Overall, 30 out of 79 respondents also produced organic citrus. Among growers that were not growing organic citrus, there was no cover crop usage for those in Texas, while 28% of such growers in Florida used cover crops.

Table 3. Cover crop use by respondents

Growers that grew organic citrus were also more likely to own a no-till planter than those who did not grow organic citrus (Table 4). This relationship is significant for Florida since we reject the null hypothesis of independence between the two variables using a chi-square test [chi² test statistic (1) = 10.72, pr. = 0.001]. The relationship is however not significant for respondents from Texas. Therefore, we did not include the status of organic production in the regression specifications for Florida.

Table 4. No-till planter ownership and organic citrus production

Respondents were asked how they ranked potential benefits from using cover crops. Table 5 shows the state-wide distribution of benefits that were ranked 1 by the respondents. Most of the respondents – 36% from Florida and 30% from Texas – ranked nutrient retention in soil as the most important benefit that cover crops would provide. Also, 25% of the respondents from Texas indicated that maintaining soil moisture was the most important benefit of cover crops, while 20% of respondents from Florida ranked pest control as the most important benefit from using cover crops.

Table 5. State-wide distribution of benefits that were ranked 1 by growers. Percentages given in parentheses

Most of the respondents from Florida that were using cover crops ranked nutrient retention as the number one benefit of using cover crops (11 out of 24). Out of the 35 respondents in Florida that were not using cover crops, 10 ranked nutrient retention and 8 ranked pest control as the most beneficial outcome of using cover crops (Table 6). Although the distribution of the number 1 ranked benefits varies by status of cover crop adoption, we did not find any significant relationship between the two for either Florida or Texas (Fisher’s exact = 0.592 and 0.505 for FL and TX, respectively). In our regression specifications for Florida, we included the highest ranked perceived benefit of cover crops for each respondent. However, we did not include the ranked 1 perceived benefits in the regression specifications for Texas because the number of observations in the Texas sample is only 20 and adding the perceived benefits would require estimating coefficients for five additional variables.

Table 6. Distribution of growers by their status of cover crop use and their highest ranked perceived benefits of cover crops

In addition to variables obtained from our survey, we also included a soil quality indicator and an index of market access in the regression models. The market access index and the index of soil quality are geospatial data sets that are obtained in the form of gridcells. For market access, we used Verburg et al. (Reference Verburg, Ellis and Letourneau2011)’s market accessibility index. The index is estimated by combining travel times to major cities, ports, and towns and accounts for a region’s topography and other physical characteristics. The extracted index ranges from 0 to 1, with 1 being the closest to markets. The data are in 5 arcminute resolution which roughly translate to an area of 74.49 sq. km. in Florida. Since the market accessibility index data are in the form of gridcells, we took the set of gridcells for Florida and Texas and mapped the centroids of the growers’ grove zip codes, which we obtained from our survey, to the gridcells. For soil quality, we used the National Commodity Cropland Productivity Index, NCCPI as the indicator. To be precise, we used cropland weighted NCCPI, in which we weighted NCCPI by the proportion of cropland obtained from USDA’s Cropland Data Layer (CDL). We obtained the NCCPI data from USDA-NRCS’ Gridded Soil Survey Geographic (gSSURGO) database. The index divides soil quality based on several attributes such as moisture, soil carbon content, weather, and crop type and classifies them according to the unique combinations of the attributes. The index ranges from 0 to 1 indicating low soil to high soil productivity. NCCPI is a geospatial data product at 10 m resolution provided by gSSURGO. We aggregated the 10 m × 10 m pixels to 30 m × 30 m pixels, which is the resolution at which CDL data are available and multiplied them by the cropland proportion in the 30 m × 30 m CDL pixels. We combined 2019 gSSURGO data with cropland proportions from 2017 CDL data. Next, we aggregated the combined 30 m × 30 m gridcells to 5 arcminute-level gridcells, which we used in our analysis.

Grower reported zip codes for groves varied in size. Moreover, the distribution of respondent zip codes is not uniform across the citrus growing regions of the two states. Therefore, it is possible that some of the zip code centroids were located at the border of two or more 5 arcminute gridcells. For such respondents, we took the mean of the accessibility index and the soil quality index. It is also possible that there are multiple respondent zip codes within a particular gridcell. For such observations, the cropland weighted NCCPI and the market accessibility index had equal values. Overall, we were able to match 45 respondents from Florida and all 20 respondents from Texas to the 5 arcminute gridcells used in our study. To keep the locations and identities of our respondents confidential, we do not show a distribution of respondents by zip codes, nor do we explicitly show them on a map. In total, there were 51 unique gridcells in which 65 zip codes of the respondents were located. Table 7 gives the state distribution of unique gridcells by the number of respondents in them. Florida has 1 gridcell with 4 respondents in it, and 2 gridcells with 3 respondents in each of them, and 4 gridcells with 2 respondents in each gridcell. Texas has 3 gridcells with 2 respondents in each of them. In total, there were 41 respondents that are mapped to unique gridcells.

Table 7. State-wide number of gridcells by number of respondents located in them

The average NCCPI for growers in Florida is 0.02 and for Texas growers it is 0.1 (Table 2). However, the market accessibility index shows a reverse trend, with growers in Florida having higher index on average at 0.65 followed by those in Texas at 0.58. There is possible negative correlation among the two variables, which we discuss in the results and discussion section.

Our survey included additional questions such as access to credit, soil health, and agricultural practices that had either a low response rate or were correlated with other variables that we use in our model.