Introduction
Plant pests and diseases impose a significant constraint on agricultural productivity. Rossman (Reference Rossman2009) estimated that invasive plant pathogens cause an estimated USD 21 billion in crop losses each year in the United States alone, while Bradshaw et al. (Reference Bradshaw, Leroy, Bellard, Roiz, Albert, Fournier, Barbet-Massin, Salles, Simard and Courchamp2016) estimated invasive insects to cost USD 70 billion annually worldwide. Therefore, pesticide applications are key for farmers to mitigate the impact of pests and diseases affecting their crops. Without the application of pesticides, the amount of harvested crops would be smaller and, ceteris paribus, prices would be higher. However, farmers do not face all of the costs associated with their pesticide input choices, they take private costs into consideration while ignoring the social costs. Therefore, farmers’ pest and disease control decisions have consequences beyond their own farms.
The social costs that arise from pesticide use include adverse impact on human health of workers and consumers, the development of pesticide resistance, and potential ecological harm. The annual health costs derived from pesticide exposure in the U.S. are estimated at USD 1.5 billion (Bourguet and Guillemaud Reference Bourguet and Guillemaud2016). While the social cost generated by the development of pesticide resistance is on the rise due to the increasing reliance of agricultural producers worldwide on individual pesticide applications, global-scale cost estimates are lacking. But Carpenter and Gianessi (Reference Carpenter, Gianessi and Nandula2010) estimate that increased chemical costs due to glyphosate resistance in the U.S are USD 10 billion annually.
As a consequence of the sizable externalities created by farmers’ pesticide use, there is a strong case for government intervention because, as economic theory suggests, pesticides can be overused relative to the level preferred by society (Sexton, Lei, and Zilberman Reference Sexton, Lei and Zilberman2007). But using a command-and-control approach to regulate pesticides, which includes recommended dosages per application and maximum rates per season, leaves much to be desired (Zilberman and Millock Reference Zilberman and Millock1997a). This is so because – other than bans, which would be unfeasible to implement on the grounds that farmers would not be able to control pests and diseases affecting their crops – direct controls such as restrictions on use (i.e., when, where, and how to use pesticides) or limits on the total quantity of chemical used do not provide any incentives for reducing pesticide use. Instead, when designed properly, market-based instruments are more cost-effective relative to command-and-control (OECD 2017).
The optimal level of pesticide use is obtained when the value of its marginal product is equal to the sum of the private cost and the marginal external costs. Therefore, the first-best policy would be that which achieves no deadweight loss. The most intuitive policy to try to capture the marginal external cost would then be the introduction of a Pigouvian tax that captures the cost of the externalities. In addition, an advantage of introducing a tax is that no monitoring is necessary because decision-making is decentralized at the farmer level. Importantly, however, the externalities generated from pesticide use are not directly observable, so taxing individuals based on the generation of those externalities is not feasible. Moreover, due to the complex nature of agricultural production, marginal external costs may be dependent on location, time, and application technology, making it virtually impossible to compute the tax level that leads to equating the social marginal benefits and costs associated with the use of pesticides; even when pesticides’ externalities do not depend on location and use, it can be very challenging to determine such a tax (Sexton, Lei and Zilberman Reference Sexton, Lei and Zilberman2007). To complicate things further, typically, little data are available to accurately assess all the social costs of pesticide use.
Given the above-described challenges for computing the marginal external costs of pesticides, financial incentives such as taxes need to be applied to externality-causing inputs, whose use is observable, unlike externalities which are not. Since the relationship between the use of pesticides and the externalities they generate is not a direct one, the goal of a policy regulating the use of pesticides should be to minimize the deadweight loss derived from the market failure by inducing farmers to undertake levels of pesticide use that are closer to the social optimum (Waterfield and Zilberman Reference Waterfield and Zilberman2012).Footnote 1 Thus, a second-best approach would consist of having policy makers choose a target level of pesticide use reduction, which can be attained introducing a pesticide tax (Baumol and Oates Reference Baumol and Oates1988).
In practice, however, most environmentally related tax rates are not only significantly below marginal external costs but are also below levels that achieve significant behavioral changes (OECD 2017). Finger et al. (Reference Finger, Möhring, Dalhaus and Böcker2017) argue that, in the context of pesticides, a tax should not be ad-valorem or per unit because pesticides vary widely with respect to their properties. Instead, the authors argue that the tax should be based on the risks each pesticide poses to reflect the social marginal costs associated with their use. A tax based on such a criterion would incentivize farmers to modify their pest management decisions by substituting the use of pesticides with higher external costs for those that have lower external costs. But, Skevas, Stefanou, and Lansink (Reference Skevas, Stefanou and Lansink2013) assessed the effectiveness of different fiscal measures in encouraging farmers to reduce pesticide use and their externalities and found that taxes based on the toxicity of pesticides did not result in the substitution of high- with low-toxicity pesticides. However, their study has some limitations. For example, the data the authors use on pesticides was based on expenditures and, therefore, they used a price index to infer quantities. Importantly, the authors only considered two tax categories in their analysis, one for high and one for low toxicity, which limited substitution possibilities. In fact, they mentioned that the absence of low-toxicity alternatives may explain their result.
In this study, we assess the effectiveness of imposing a tax on pesticide use based on two different criteria by using simulations at the farm-level in the context of the spread of citrus greening or Huanglongbing (HLB) disease on Florida’s citrus industry. This is a highly relevant case study because the significant production challenge that such an invasive species has posed to citrus farmers in Florida encouraged them to adopt the use of significantly more toxic insecticides to try to control the vector of the disease and slow down the infection rate. Given that the current pesticide regulation in the U.S. – based on direct controls – does not penalize the decision of farmers to use chemicals that are more toxic in any way, it is a rational economic choice for farmers to adopt their use. Despite farmers’ widespread adoption of such insecticides, the disease still decimated the state’s citrus industry, eventually causing production to decrease by more than 90%. Such a devastating impact is evidence that the switch to more toxic chemicals by individual farmers may be futile,Footnote 2 calling into question the usefulness of the current regulation that allows them to apply pesticides without having to pay any of the social costs derived from their decisions.
Our contributions to the literature can be summarized as follows. First, we quantify the significant increase in toxicity derived from Florida orange farmers’ decisions to adopt an intensive insecticide program to control the insect vector of HLB. Such an estimate should be useful to increase policymakers’ awareness on the consequences of having no policies in place for restricting the use of pesticides that are highly toxic and, therefore, pose significant risks for human health and the environment. Second, we model the profit-maximizing problem of a representative farmer by specifying a functional form for the damage function that incorporates the biological impact of the pathogen-vector system on yield as well as the abating impact of different insecticides on the insect vector population based on previous entomological work.
Importantly, we also use estimates of the marginal benefit and toxicity for each pesticide based on multiple and detailed toxicity criteria. This allows us to compute a differentiated tax rate for each chemical, which arguably provides a better opportunity for substitution patterns among them. In fact, our results show in detail how the substitution of high- with low-toxicity pesticides occurs under the different alternative policies we consider. To our knowledge, we are the first to model a functional form for the damage function of HLB to Florida citrus farmers that considers the underlying biological process that determines the relationship of the insect vector and the yield damage that the disease causes. By running simulations of such a farm-level model, we are able to quantify the differential impact of choosing a pesticide indicator that accounts for the heterogeneous properties of chemicals versus a quantity-based tax. The model is valuable because little empirical research has been conducted regarding the implementation of financial incentives in pesticide policy and their impact on pesticide use, farm income, and externalities at the farm level (exceptions include Falconer and Hodge (Reference Falconer and Hodge2001) and Skevas, Stefanou and Lansik (Reference Skevas, Stefanou and Lansink2013)). Third, this case study analysis helps renew attention to the significant inefficiency associated with the current command-and-control pesticide policy in the United States and makes a case for the need to complement it with an incentive system that would minimize the associated deadweight loss.
The rest of this article is organized as follows. In the next section, we provide an overview on the current pesticide policies in the U.S. and in Europe. We also discuss the impact of HLB in Florida and its relevance as a case study for the present analysis. In the following section, we present the theoretical model and the specification for tax policies based on two possible criteria. In the subsequent section, we show how we calibrate the model in terms of yield, prices, and the correlation between those two variables to run simulations of the model applied to the Florida citrus industry. In addition, we also specify a functional form to the damage function. We then summarize and discuss the results, and, in the last section, we offer some conclusions and make recommendations for the design of a more efficient pesticide policy in the United States.
Background
Pesticide policies in the U.S. and in Europe
The experience in OECD countries regarding the introduction of environmental taxes on agricultural production inputs, such as pesticides, is limited because only a small number of countries have implemented them (Nielsen et al. Reference Nielsen, Hedegaard Konrad, Pedersen and Gyldenkærne2023). Instead, many governments have opted for a command-and-control approach to regulate pesticide use. Such a policy choice can largely be attributed to the challenges of assessing the damages derived from pesticide use and their variability caused by the spatial and temporal heterogeneity in applications.
In the United States, for example, the Environmental Protection Agency (EPA) regulates pesticides through their registration and labeling under the premise that their use “will not generally cause unreasonable adverse effects on the environment.” By such a statement, the EPA means any unreasonable risk to man – including dietary risk from residues on foods that result from use of a pesticide – or the environment (EPA 2024a). Thus, the EPA requires that manufacturers test chemicals and mixtures for toxic effects prior to their approval, and monitors risks and side effects (EPA 2024b). Should a pesticide be found to have harmful effects on the environment or human health, they get banned. But there are no financial incentives for farmers to reduce the use of pesticides other than following the recommended application rates (Zilberman and Millock Reference Zilberman and Millock1997b). Moreover, the command-and-control approach can sometimes be too lenient. The Environmental Protection Agency (EPA) is developing new guidelines for the use of pesticides to ensure improved compliance with the Endangered Species Act (ESA), which were brought about by lawsuits claiming that the EPA had failed to comply with the ESA when registering pesticides (EPA 2024c).
In Europe, there is a greater emphasis on financial incentives relative to the United States. National action plans are mandatory for all European Union member states since 2012. Pesticide policies in the different national action plans include subsidies for upgrading equipment, educating farmers, and increasing awareness regarding the introduction of pesticide taxes (Böcker and Finger Reference Böcker and Finger2016). The main stated goal of pesticide policies is to reduce the “risks and impacts of pesticide use on human health and the environment” (Remáč Reference Remáč2018, p.14). Taxes on pesticide use can contribute to achieve such a goal and a few countries have adopted such market-based policy based on different criteria.
Sweden was the first country to introduce a flat quantity tax on pesticides based on the volume sold in 1984 (Böcker and Finger Reference Böcker and Finger2016). Norway, on the other hand, introduced a tax on pesticides in 1988, which was designed as an ad-valorem tax but changed in 1999; it currently consists of a base rate per area plus an additional rate that is based on which of the seven possible categories the pesticide belongs to according to the risks it presents for human health and the environment. In 2009, France replaced a volume tax on pesticides with a 3-category tax by which products are charged a different fee based on the non-point agricultural pollution category to which they belong. Thus, mineral-based pesticide products are charged the lowest fee, products considered to be dangerous to the environment are charged approximately double the fee, and products that are carcinogenic or hazardous are charged the highest fee, which is approximately five times that of the lowest fee. Starting in 2013, Denmark implemented a pesticide tax by which each pesticide product receives a differentiated tax rate based on the environmental and health impacts of the product. Such a pesticide policy is considered to be the most sophisticated currently in place (OECD 2017) and makes the country a pioneer in implementing such an approach to curb the externalities associated with the application of pesticides in agriculture (Böcker and Finger Reference Böcker and Finger2016).
Citrus greening disease in Florida as a case study
Florida has historically been the largest orange- and citrus-producing state in the U.S. and a top global producer (USDA-NASS 2022). Orange production in Florida typically accounts for 90 percent of the state’s annual citrus production.Footnote 3 Therefore, in our analysis we focus on such a crop. Since 2005, Florida citrus farmers have been facing the impact of HLB, a devastating bacterial disease transmitted by an insect vector (the Asian citrus psyllid) that has decimated the state’s citrus industry (Singerman and Rogers Reference Singerman and Rogers2020). As illustrated in Figure 1, orange production in Florida has decreased by more than 90% since the disease was first found. As a consequence, Florida has lost its status as the top citrus and orange-producing state in the United States (USDA-NASS 2022).
Orange production in Florida from 1955 to 2022, with shaded period (2005–2022) denoting citrus greening disease outbreak.
To try to mitigate the impact of the disease, Florida orange farmers have adopted several different practices, including an intensive insecticide program to control the propagation of the insect that transmits the disease (Shen et al. Reference Shen, Cevallos-Cevallos, Nunes da Rocha, Arevalo, Stansly, Roberts and van Bruggen2013).Footnote 4 Thus, farmers substituted the traditional insecticides they had been using (i.e., mainly petroleum or mineral oil) with significantly more toxic chemicals such as neonicotinoids. Figure 2 shows the toxicity measured in load (see appendix for a detailed explanation of how the load is computed) and quantity per acre of insecticide use for growing oranges in Florida from 2005 to 2017. As illustrated by Figure 2, there is little difference, approximately 3%, in the quantity of insecticide used per acre from 2005 to 2017. However, such a figure also shows that the toxicity of insecticides used in orange production in Florida increased by 472% during that same period.
Insecticide use toxicity and quantity per acre for producing oranges in Florida from 2005 to 2017.
While the estimated increase in toxicity is very significant, to the extent that it only includes insecticides, it represents a lower bound on the total increase in toxicity of chemicals used by Florida citrus farmers to try to manage the impact of HLB. For example, the estimate does not include the toxicity and risks associated with citrus farmers’ adoption of antibiotics as foliar sprays – following a decision of Florida policy makers to approve them in 2016.
The questionable approval of the use of antibiotics as foliar sprays (see Singerman and Lence Reference Singerman and Lence2023) is highly relevant because they can contribute to the development of antibiotic resistance with its consequent impact on public health. Given that the use of antibiotic foliar sprays was eventually found to be ineffective to enhance the health of HLB-infected trees (Li et al., Reference Li, Wu, Duan, Singerman and Guan2020), their use was short lived. However, the decision of farmers to substitute the use of insecticide sprays with antibiotics induced greater variability in the use of insecticides, which contributed not only to an increase in the population of the insect that transmits the disease, but also to an increase in resistance to insecticides (Chen et al., Reference Chen, Gill, Ashfaq, Pelz-Stelinski and Stelinski2018). The development of pesticide resistance can cause a further increase in their use to offset the lower susceptibility of the pest, which exacerbates the negative externalities generated (Dover and Croft Reference Dover and Croft1986).
Given the drastic change in the toxicity of insecticides used by Florida orange farmers as a response to the exceptional challenge imposed by HLB, in this study, we use such a case to illustrate the change in outcomes derived from possible market-based interventions to capture, at least partially (as explained in the introduction), the external costs of pesticide use. In particular, we provide a conceptual and empirical framework to examine the change in farm-level outcomes (i.e., insecticide toxicity, profit, and yield per acre) from implementing a tax on insecticide use based on different pesticide indicators. The social cost mitigation achieved by lowering the use of pesticides through the implementation of a tax would make society better off (by decreasing the magnitude of the externalities and, thus, of the deadweight loss). In addition, given that producer surplus can be defined as revenue minus variable costs and, equivalently, as profit plus fixed costs, a tax based on the pesticide indicator that results in the greatest toxicity reduction at the lowest profit loss for producers will generate an outcome that is closer to the social optimum.
Model
We base our model on the damage control framework developed by Lichtenberg and Zilberman (Reference Lichtenberg and Zilberman1986), in which the farmer’s use of pesticides is viewed as a damage control input. The premise of the model is that farmers apply pesticides to try to reduce the damage from pests and diseases and, thus, increase the proportion of potential output they can realize. The starting point of our model is the profit function for a representative farmer, which is given by
${\rm{\Pi }} = TR - TC$
, where
${\rm{\Pi }}$
,
$TR,{\rm{\;and}}\;TC$
denote profit, total revenue, and total private cost, respectively; all three terms expressed in dollars per acre. For a representative Florida orange farmer,
$TR = p \cdot q$
, where
$p$
is the price of oranges per box and
$q$
is the quantity of oranges boxes produced per acre. Following Lichtenberg and Zilberman (Reference Lichtenberg and Zilberman1986), in our model,
$q = y \cdot \left( {1 - D\left( {{N_1}} \right)} \right)$
, where
$y$
is the maximum potential yield per acre in the absence of disease, and
$\;D\left( {{N_1}} \right)$
denotes the damage to yield caused by the density of the vector population,
${N_1}$
, that spreads the disease. Therefore,
$D\left( {{N_1}} \right)$
is between 0 and 1, and
${N_1}$
is a function, k, of the initial pest population
${N_0}$
and the quantity of insecticide
$I$
used per acre during the season, so that
${N_1} = k\left( {{N_0},I} \right)$
.
Economic theory offers no guidance regarding the functional form that should be used to specify the damage function, so studies typically consider a variety of functional forms (Sexton, Lei and Zilberman Reference Sexton, Lei and Zilberman2007). Importantly, and in contrast to such an approach, in our model we use the findings of an entomological study on HLB to specify a functional form for the damage function that incorporates the underlying biological processes of the insect vector and the pathogen of the disease on yield. Thus, we model the damage to orange yield based on the study of Monzo and Stansly (Reference Monzo and Stansly2017). In their work, the authors used non-linear least-squares to model the relationship between yield losses and the cumulative number of psyllids in experiments they conducted in commercial groves of mature orange trees in Florida. They then fitted the data to a rectangular hyperbolic yield loss model by using the Newton-Raphson iterative estimation procedure. Thus, we define the yield damage function as follows:
$$D\left( {{N_1}} \right) = \left( {{{\theta \cdot k\left( {{N_0},\;I} \right)}}\over{{1 + {{{\theta \cdot k\left( {{N_0},\;I} \right)}}\over{\eta }}}}} \right) \cdot {100^{ - 1}}$$
where
$\theta $
denotes the slope of the curve at the origin – that is the rate of yield loss at low pest density – and
$\eta $
denotes the horizontal asymptote of the function, which denotes the maximum yield loss. Furthermore, based on feedback from entomologists at University of Florida Citrus Research and Education Center, we model the function
$k\left( \cdot \right)$
in equation (1) as follows:
$$k\left( {{N_0},\;I} \right) = {N_0} \cdot \mathop \prod \nolimits_{i = 1}^n \left( {1 - {x_i} \cdot {d_i} \cdot {I_i}} \right)$$
where
${{\rm{x}}_{\rm{i}}}$
is the rate of reduction of the insect vector population from applying insecticide i,
${{\rm{d}}_{\rm{i}}}$
is the duration of the effect of insecticide i in days expressed as a proportion of a year,
${I_{\rm{i}}}\;$
is the application rate of insecticide i, and n is the number of insecticides used.
For simplicity, in our model, we define the total private cost per acre as the sum of purchasing and applying inputs so that
$TC\left( {I,Z} \right) = C\left( I \right) + \;W\left( Z \right)$
, where
$C\left( I \right)$
represents the private cost for pest control inputs I and
$W\left( Z \right)$
represents the cost of all other grove caretaking inputs Z. Since we are particularly interested in the farmer’s insecticide decisions, we assume that all other inputs are constant and given by
$\overline {W\left( Z \right)} $
.
Given that Florida citrus farmers typically spray insecticides mixed along with other chemicals, we assume that there is no additional cost of applying insecticides. Thus, the private cost of pest control inputs I is then given by the cost of the materials used. The total cost for a representative Florida orange farmer in the baseline scenario, which represents the current regulation, is then given by the following expression.
$$TC\left( {I,Z} \right) = \mathop \sum \limits_{i = 1}^n \left( {{r_{i\;}} \cdot {I_{i\;}}} \right) + \overline {W\left( Z \right)} $$
where
${\rm{\;}}{r_i}$
is the price of insecticide i and
${I_i}\;$
is the annual application rate per acre of insecticide i. Importantly, insecticide use must be constrained so that each chemical application does not exceed its maximum recommended use
$\overline {{I_i}} $
, so that
${I_{i\;}} \le \overline {{I_i}} $
. Therefore, the problem for the representative farmer whose objective is to maximize profit by choosing the level of each insecticide used can then be formalized as follows:
$$ma{x_{{I_i}}}\;E\left\{ {p \cdot y\left[ {1\; - \left( {{{\theta \cdot {N_0} \cdot \mathop \prod \nolimits_{i = 1}^n \left( {1 - {x_i} \cdot {d_i} \cdot {I_i}} \right)\;}}\over{{1 + {{{\theta \cdot {N_0} \cdot \mathop \prod \nolimits_{i = 1}^n \left( {1 - {x_i} \cdot {d_i} \cdot {I_i}} \right)\;}}\over{\eta }}}}} \right) \cdot {{100}^{ - 1}}} \right]} \right\} - \;\mathop \sum \limits_{i = 1}^n \left( {{r_{i\;}} \cdot {I_{i\;}}} \right) - \overline {W\left( Z \right)} $$
$$s.t.{\rm{\;\;\;}}$$
$$\;{I_{i\;}} \le \overline {{I_i}} $$
Social planner and taxing
The setup in (3) can be modified to represent that of a social planner whose objective is to maximize social welfare, by subtracting an additional term denoting the net social costs of all insecticides used:
$SC(\mathop \sum \limits_{i = 1}^n {I_{i\;}})$
. Thus, we assume such term represents all social benefits and costs of each insecticide used and that the latter are larger than the former. Such a setup, and in particular the estimate of the last term, would allow implementing a first-best policy that achieves no deadweight loss because all social costs are taken into account. However, as mentioned in the introduction, it would be virtually impossible to capture all social costs associated with pesticide use. Therefore, a second-best approach would consist of implementing of a tax policy that would reduce the size of the deadweight loss resulting from the market failure derived from farmers not internalizing the social costs of their input choices. A first tax policy we consider is based on adopting quantity indicator of insecticides used, whereas a second policy we consider is based on an indicator that accounts for the toxicity of insecticides.
If policy makers were to implement a tax on the quantity of insecticides used, the total cost to an orange farmer would be given by:
$$TC\left( {I,Z} \right) = \mathop \sum \limits_{i = 1}^n \left( {\left( {{r_{i\;}} + \;{T_q}} \right) \cdot {I_{i\;}}} \right) + \overline {W\left( Z \right)} $$
where
${T_q}$
represents the rate for the quantity tax (in dollars per acre). If instead, policy makers were to implement a tax on the toxicity of insecticides (i.e., a load tax), the total cost to an orange farmer would be given by:
$$TC\left( {I,Z} \right) = \mathop \sum \limits_{i = 1}^n \left( {\left( {{r_{i\;}} + \;{L_i} \cdot {T_L}\;} \right) \cdot {I_{i\;}}} \right) + \overline {W\left( Z \right)} $$
where
${L_i}$
represents the load of insecticide i and
${T_L}$
represents the rate for the load tax (in dollars per load).
Depending on whether a tax on quantity or on the toxicity of insecticides is implemented, the variable cost term in expression (3) needs to be modified according to equation (4) or (5), respectively. Thus, the simulations allow us to compare the outcomes in terms of yield, profit, and aggregate toxicity of insecticides in the baseline scenario to those obtained when implementing a tax to curb the social cost of insecticides.Footnote 5
Data and calibration
In this section, we describe the data we have available on insecticide use by Florida orange farmers, including active ingredients, cost, application rate, and cost-effectiveness. Then, we describe how we calibrate key variables (i.e., yield, prices, and the correlation between those two) for a representative Florida orange farmer to run our simulations. In the appendix we provide a detailed explanation of the procedure we followed to estimate the toxicity of each insecticide.
Insecticide data
USDA-NASS has collected data on the quantity of insecticide used for growing oranges in Florida from 2005 to 2017 by surveying farmers every odd year during that period, excluding 2013. According to the USDA-NASS data, eight different active ingredients – out of a total of 30 – account for over 90% of all insecticide use by Florida orange farmers. Our simulations and analysis are then based on the cost and traits of the most popular commercial products in the market that contain those active ingredients. To obtain estimates of the cost of insecticides, we collected quotes from chemical retailers and used their average. In addition, for each insecticide, we also gathered data regarding the rate of application per acre, the duration and magnitude of adult psyllid population reduction, and the proportion of active ingredients per product from the entomological literature available (Qureshi, Kostyk and Stansly Reference Qureshi, Kostyk and Stansly2014). Table 1 summarizes the data just described for each insecticide active ingredient and corresponding commercial product along with other relevant variables that we computed for each of them. Thus, the information in Table 1 is used for the variables defined in equation 2. Table 1 also includes the cost-effectiveness of each insecticide, which we obtained by using the prices and the duration of adult insect vector reduction listed in such table. The cost-effectiveness of insecticide i in dollars per acre per day is computed by multiplying the price of insecticide
$i$
by the application rate and dividing it by the duration of adult insect vector reduction in days multiplied by the magnitude of the insect vector reduction.
List of insecticide active ingredients (which account for over 90% of all insecticide use by Florida orange farmers) and corresponding most popular commercial product used to grow processed oranges in Florida along with their prices, application rates, proportion of active ingredient per product, duration, and magnitude of psyllids reduction, effective days of coverage, and cost-effectiveness
1 Source: Qureshi, Kostyk and Stansly (Reference Qureshi, Kostyk and Stansly2014). Note: In their study, the authors had two different application rates for each insecticide: low and high. Given the urgency of farmers to deal with the disease, we assume they applied the maximum rate possible (i.e., high).
2 Source: Environmental Protection Agency (EPA) product labels.
3 Source: Authors’ calculations.
The main indicators used as a basis for taxing are the Quantity of Active Ingredients (QA) and the Load Index (LI), which measures the toxicity of a given pesticide on organisms other than the targeted one. While the QA indicator is simpler and, thus, often used in policy making, the LI indicator accounts not only for differences in dosage of pesticides but also for their environmental and health impacts at the product level (Kudsk, Jørgensen, and Ørum, Reference Kudsk, Jørgensen and Ørum2018). Möhring et al. (Reference Möhring, Gaba and Finger2019) found that quantity-based pesticide indicators fail to identify important qualitative characteristics of pesticides, particularly for applications that have the highest values of the LI indicator.
Pesticides load
The load of a pesticide provides a measure of the toxicity from the use of such a chemical on human health and the environment. To estimate the load of insecticides used by Florida citrus farmers, we used data from the International Union of Pure and Applied Chemistry database regarding the properties of pesticides and bio-pesticides and follow the procedure established in the literature (Samsøe-Petersen et al., Reference Samsøe-Petersen, Bøje, Hansen, Kjølholt, Jørgensen, Kudsk and Ørum2012). The load of a pesticide has three components: human health load, environmental fate load, and environmental toxicity load. We explain how we computed each of those components in detail in the appendix.Footnote 6
Calibration
To calibrate our simulations, we use publicly available data from the United States Department of Agriculture’s (USDA) National Agricultural Statistics Service (NASS) regarding oranges’ historical yield and prices in Florida. For cost estimates, we use publicly available data from the University of Florida. We calibrate our model to represent Valencia oranges because it is the variety with the highest proportion of planted acreage in the state, currently accounting for approximately 60% of the area. Importantly, the management practices and cost of production for non-Valencia oranges are identical to those of Valencia oranges.
To calibrate yield, we first estimate the trend for Valencia oranges yield data in Florida from 1955 to 2004, which we then use for obtaining two other estimates as explained next. Given that yield is a random variable in our model, we assume it follows a normal distribution. Using the estimated historical yield trend of Valencia oranges, we can obtain a sensible estimate for the mean of the yield distribution by computing the trend yield for year 2005. Thus, such an estimate represents the maximum potential yield of Valencia oranges per acre in the absence of HLB, which is reasonable because 2005 was the year in which HLB was first found in the state. In addition, we also use the trend to compute the de-trended yield for each year in our series so as to reflect 2005 technology but historical yield variability due to weather (i.e., weather draws). Thus, by estimating the variance of the resulting series, we obtain a sensible estimate for the variance of the yield distribution that we use as a basis for our simulations.
Regarding prices, we assume they follow a log normal distribution to avoid obtaining negative prices in our simulations. Given that the HLB outbreak was in 2005, we use the average of real prices in Florida for Valencia oranges from 1990 to 2004 as the mean of the distribution. The reason for choosing such a period is because in the 1980s there were several widespread freezes in Florida that caused orange trees and, therefore, orange supply to decrease and prices to increase significantly as a consequence. We obtained estimates of real prices by adjusting the nominal prices published by the USDA-NASS for inflation based on the annual Producer Price Index (PPI) for all U.S. citrus fruit (USDA-ERS 2019).
The final step in the calibration consisted of including an estimate of the negative correlation between prices and yield. Therefore, we computed the historical correlation between the two variables for Valencia oranges grown in Florida from 1955 to 2004, which we estimated at –0.33. We then used the Iman and Conover (Reference Iman and Conover1982) methodology – which only re-sorts the original yield and price draws – to impose such an observed correlation in our simulations.
In equation (1), we set parameters θ = 0.97 and η = 50.91 following the work of Monzo and Stansly (Reference Monzo and Stansly2017), who obtained those estimates by pooling the data in the experiments they conducted in commercial groves of mature orange trees in Florida. Regarding the cost of inputs and applications for a representative orange farmer of processed oranges in Florida, we use farmer survey estimates provided by Singerman (Reference Singerman2020), excluding the costs associated with the management of the insect vector of HLB (which we estimate in our simulations). Thus, our estimate for the cost of all other grove caretaking inputs denoted by
$\overline {W\left( Z \right)} $
in expression (3) is $1,600 per acre.
Results
We run simulations of the profit maximization problem for a representative Florida orange farmer denoted by expression (3) for the baseline scenario. We show results for population levels of the psyllid insect vector (
${N_0}$
) from 0 to 80 because before 2012, when the significant increase in toxicity depicted in Figure 2 occurred, the average level of psyllids per orange block was less than 10, but it increased to a peak of 27 per orange block in 2017 (Singerman Reference Singerman2017). Thus, a level of psyllids that almost triples that count allows us to consider the likely further increase in psyllid population levels in following years.Footnote 7
We then run simulations of the profit maximization problem for the cases in which a quantity-based tax and a load-based tax are implemented, which are denoted by the combination of expressions (3) and (4) for the former case and by the combination of expressions (3) and (5) for the latter case. The objective is to estimate the magnitude of the change in load, yield, and profit per acre when implementing each tax policy relative to the baseline.
For the load-based tax model, we run different tax rates per load (
${T_L}$
) and found that, as the number of psyllids increases, a tax rate of $2.50 makes the total load be asymptotic at a level of 29, which is the level of load in the baseline scenario when the number of psyllids is 20 (roughly representing a middle point in the spike of psyllid population between 2012 and 2017 described earlier). Therefore, we run the simulations using such a value as a “target load” because it arguably reflects a sensible maximum tax rate that would make sense implementing to reduce the toxicity of insecticides (i.e., higher tax rates would be futile in achieving a greater reduction in load).
In panel A of Table 2, we show the results of the simulations in levels for the baseline scenario, as well as for the scenario in which a load-based tax rate of $2.50 per load is implemented. For comparison purposes, we also show the results for the scenario in which a quantity-based tax is implemented using the same rate. However, to make the comparison more meaningful between tax policies, in Table 2 we also show the results of a scenario for a quantity-based tax with a rate of $12 because we found it yields a comparable target load level of 29 when the number of psyllids is 50. However, that quantity tax rate does not prevent toxicity from further increasing as the number of psyllids increases (as the load tax rate does). Hence, we also show the results for a quantity tax rate of $20 that achieves a slightly lower but similar load to the load-based tax rate of $2.50 when the psyllid number is 80.
Profit maximization simulation results showing change in load, yield, and profit per acre for the baseline, for a load-based tax of $2.50, as well as for a quantity-based tax of $2.50, $12, and $20 for different psyllid levels when the correlation between yield and prices is –0.33
In panel B of Table 2, we show the results of the simulations in panel A in percentage change relative to the baseline. The load-based tax rate of $2.50 induces a significant reduction (74%) in the load of insecticides when the level of psyllids per acre is 10. The underlying substitution in insecticides causes a reduction in yield of 3% and of 16% in profit relative to the baseline. As expected, as the number of psyllids increases, both the percent reduction in load and profit increases significantly.Footnote 8 When the psyllids level is 60, the load is reduced by 93% relative to the baseline, but the profit is also reduced significantly – to 86% – because the tax prevents farmers from using more toxic insecticides profitably. For higher levels of psyllid population, the profit under a load tax of $2.50 becomes negative.
The results of implementing a quantity-based tax model using a rate of $2.50 in Table 2 show that using such a tax type and rate initially induces a reduction in the load of insecticides. But as the number of psyllids increases, a farmer would find it profitable to use more toxic insecticides despite the tax and, therefore, the load increases. Thus, for levels of psyllids greater than 30, the percent change in load relative to the baseline starts decreasing (in absolute value). When the psyllids level is 60, the load is reduced by only 22% (versus 93% with the load-based tax). The percent reduction in profit is increasing due to the effect of the tax combined with the increase in the number of psyllids. Even though a quantity tax of $12 is more effective at reducing the load relative to the rate of $2.50, the pattern in load reduction is similar; it starts decreasing for levels of psyllids greater than 50. Only when the quantity-based tax is $20, the percent reduction in load remains steady at more than 90% when the number of psyllids is 40 or more. Importantly, however, the percent reduction in profit using the $20 quantity-based tax rate roughly doubles that obtained when using a load-based tax of $2.50 for a number of psyllids of 40 or more.
Table 3 shows the prevalence and pattern of insecticide use in each of the scenarios included in Table 2 and for each level of psyllid population. The table illustrates, for example, that in the baseline scenario the criterion used to choose the level of use of each insecticide is based on the cost-effectiveness (which is shown in Table 1). In addition, Table 3 also shows the pattern of substitution that takes place when a tax on load is implemented; the most toxic insecticides such as Naled and Chlorpyrifos are no longer used because they are taxed more heavily. Interestingly, when the quantity tax is $20, Naled is still used (albeit at a very low level) when the psyllid level is 80.
Prevalence of insecticide use (as percentage of maximum possible application) by scenario for each psyllid population level
Sensitivity analysis
Our model hinges on the correlation imposed between prices and yield of processed Valencia oranges that we estimated based on historical values. However, such a value may no longer hold under current market conditions. Thus, we conduct a sensitivity analysis for different correlation values to examine whether and to what extent the simulation results change. We run the simulations for correlation values between the two variables of –0.50 and –0.75. When the (absolute) value of the correlation between prices and yield increases, yield draws that have lower values are matched more often with the price draws that have higher values to reach the target correlation (as described in the calibration section when referring to the Iman and Conover (Reference Iman and Conover1982) methodology).
The simulation results for the correlation values of –0.50 and –0.75 are shown in Tables 4 and 5, respectively. The results in Tables 2, 4, and 5 show that as the (absolute) value of the correlation increases, profit in all scenarios decreases for all psyllids levels. This is because a higher correlation value implies that a change in yield is offset more often by a change in prices in the opposite direction. As the absolute value of the correlation between yield and prices increases, the choice of insecticides used changes slightly in the baseline and quantity tax-based scenarios as denoted by the change in load. Nevertheless, for the load-based tax case the load and yield change only marginally. Thus, an additional advantage to implementing a load-based policy to curb the social costs of insecticide use is that such a tax would be more stable should the correlation between price and yield change.
Profit maximization simulation results showing change in load, yield, and profit per acre for the baseline, for a load-based tax of $2.50, as well as for a quantity-based tax of $2.50, $12, and $20 for different psyllid levels when the correlation between yield and prices is –0.50
Profit maximization simulation results showing change in load, yield, and profit per acre for the baseline, for a load-based tax of $2.50, as well as for a quantity-based tax of $2.50, $12, and $20 for different psyllid levels when the correlation between yield and prices is –0.75
Conclusions and policy implications
Applications of pesticides are critical for farmers to manage pests and diseases affecting their crops. But when farmers apply pesticides, they ignore the social costs associated with their use. In the United States, pesticides are regulated only through their registration and labeling; there are no financial incentives for farmers to reduce the use of pesticides other than following the recommended application rates. Given the sizable externalities generated by pesticide use, there is a rationale for further government intervention. As argued by Zilberman and Millock (Reference Zilberman and Millock1997a), the role of economists in assessing pesticide policies is to expose the inefficiency associated with existing regulations and propose alternatives. Thus, using citrus greening disease in Florida as a case study – where farmers adopted the use of highly toxic insecticides to try to control the spread of the insect that transmits the disease – we use simulations to compare the outcomes of implementing a tax on insecticide use based on different criteria to curb the social costs associated with their use.
Our simulations’ results show that taxing farmers using a load-based tax would provide them with a strong incentive to reduce the use of pesticides that are more toxic and substitute them with less toxic chemicals. Instead, a policy using a quantity-based tax would need to have a significantly higher rate to achieve a similar effect but the percentage reduction in profit and yield would be significantly higher in such a case. Thus, a load-based tax would be more effective to help correct the market failure derived from farmers not internalizing the social costs of their input choices.
A significant proportion of the crop damage that pesticides aim to prevent is caused by invasive species such as HLB. Economics can not only be helpful in informing pesticide policy, but it can also aid in the design of policies that would minimize and mitigate the effects of invasive species (Sexton, Lei and Zilberman Reference Sexton, Lei and Zilberman2007). In this regard, since our results show that a load-based tax would significantly increase the cost of using insecticides to combat pests and diseases – which would in turn cause loss of yield and profit – policy makers could, for instance, exempt farmers from paying the tax under certain circumstances.
Invasive species like HLB pose a collective-action problem (see, for example, Perrings et al. Reference Perrings, Williamson, Barbier, Delfino, Dalmazzone, Shogren, Simmons and Watkinson2002; Florec et al. Reference Florec, Sadler, White and Dominiak2013; Singerman and Rogers Reference Singerman and Rogers2020). Hence, the ability of individual farmers to effectively control mobile pests in the long run depends critically on the actions of neighboring farmers (Lazarus and Dixon Reference Lazarus and Dixon1984), which creates a public-good dilemma. It is, therefore, key for policy makers to realize that the lack of policies restricting the toxicity of pesticides being used may not even be beneficial for farmers in the long run – as it has been the case of citrus farmers in Florida – because farmers (incorrectly) view pesticides as a guarantee to achieve fewer production losses (Deguine et al. Reference Deguine, Aubertot, Flor, Lescourret, Wyckhuys and Ratnadass2021), which promotes a false belief in the possibility of self-reliance for controlling pests that require instead collective action (Bagavathiannan et al. Reference Bagavathiannan, Graham, Ma, Barney, Coutts, Caicedo, De Clerck-Floate, West, Blank, Metcalf, Lacoste, Moreno, Evans, Burke and Beckie2019).
Policy makers could achieve multiple desirable societal objectives by designing a policy that would minimize the deadweight loss derived from farmers ignoring the social costs of their input choices and simultaneously mitigate the impact of invasive species as exemplified next. Spraying insecticides as part of a regional pest and disease control such as an Area-wide Pest Management program (AWPM) has been shown to be superior to site-specific control when pests and disease vectors are mobile (Faust Reference Faust, Koul, Cuperus and Elliott2008; Singerman, Lence and Useche Reference Singerman, Lence and Useche2017). The underlying premise for AWPM is that coordinating sprays allows neighboring farmers to control the pests and diseases at a lower (social) cost by increasing the productivity of their inputs. However, due to its voluntary character, an AWPM program can suffer from low farmer participation (Hendrichs et al., Reference Hendrichs, Kenmore, Robinson, Vreysen, Vreysen, Robinson and Hendrichs2007; Singerman and Useche, Reference Singerman and Useche2019). A policy combining a load-based tax and a potential exemption from it when neighboring farmers coordinate sprays could contribute to reduce pesticide use and therefore the externalities associated with it because farmers could choose to either apply insecticides individually (and be subject to the tax) or in coordination with neighbors (and be exempt from the tax). Thus, farmers would not necessarily be worse off by the introduction of the tax policy but they would unequivocally reduce the externalities associated with their pesticide choices.Footnote 9
Last, but not least, there is evidence that the introduction of a load-based tax targeting the externalities of pesticide use in Denmark has been effective in inducing behavioral change on farmers and has resulted in an overall reduction of pesticide toxicity by 18% (Nielsen et al. Reference Nielsen, Hedegaard Konrad, Pedersen and Gyldenkærne2023). Given that global warming will likely contribute to increase the vulnerability of crops to pests and diseases and, thus, lead farmers to increase chemical use (Lyall, Suk and Tait Reference Lyall, Suk and Tait2006), policy makers in the United States could consider complementing the current pesticide regulatory system with financial incentives, such as a load-based tax, to reduce the risks for human health and the environment derived from the toxicity of pesticide use.
Data availability statement
The authors confirm that the data supporting the findings of this study will be available within the article.
Competing interests
The author(s) declare none.
Appendix: pesticide load calculation
The first component of a pesticide load is the human health load, which measures the risk of exposure that an insecticide poses to human health when handling and applying the pesticide. The load of each pesticide is calculated based on the hazard statements that appear on the label of the product; each statement is assigned risk points, and then those points are converted to a load per pound of product by multiplying the points by a load factor. Table A1 shows the risk points and corresponding load on human health for the different hazard statements that appear on the pesticides’ labels. As an example, Table A2 summarizes the sum of risk points from the hazard statement and the Human Health Load for eight insecticide active ingredients – out of a total of 30 – that account for over 90% of all insecticide use by Florida orange farmers. The conversion from risk points to load per lb. of product involves dividing the total by a reference value of 300. We then convert the Human Health Load per lb. of product to Load per lb. of active ingredient.
Risk points and load on human health associated with the different hazard statements found on pesticides’ labels
Source: Samsøe-Petersen et al., Reference Samsøe-Petersen, Bøje, Hansen, Kjølholt, Jørgensen, Kudsk and Ørum2012, p. 17.
Sum of risk points and human health Load for the top eight insecticides’ active ingredients used to grow processed oranges in Florida
The second component of a pesticide load is the environmental fate load, which measures three distinct characteristics of a pesticide: degradation, bioaccumulation, and leaching. Degradation measures how long an insecticide persists in the environment and is calculated using the following expression:
$${D_i} = {{DT_{i}^{50} \cdot D_{Ref}^{LF}}}\over{{DT_{Ref}^{50}}}$$
where
${D_i}$
denotes insecticide
$i$
’s degradation in Load per pounds of active ingredient (L/lbs a.i.),
$DT_i^{50}$
is the half-life (in days) of insecticide
$i$
’s active ingredient in the soil, and
$D_{Ref}^{LF}$
and
$DT_{Ref}^{50}$
represent the values for the load factor (in L/lb a.i.) and the half-life (in days) of the active ingredient of a reference substance, respectively (Samsøe-Petersen et al. Reference Samsøe-Petersen, Bøje, Hansen, Kjølholt, Jørgensen, Kudsk and Ørum2012). The reference substance is an approved active ingredient with the worst value for load.
Bioaccumulation is the second characteristic that defines the environmental fate load of an insecticide and it refers to the progressive accumulation over time of a contaminant in an organism relative to its level in the ambient medium. The bioconcentration factor (BCF) is an indicator measuring the ability of a chemical to be assimilated by organisms from the water and it is used to calculate bioaccumulation (Wang, Reference Wang, Blasco, Chapman, Campana and Hampel2016). When the BCF of an insecticide was not available in the database, following the literature, we estimated it by using the so-called log Pow model, which correlates laboratory bioconcentration factors determined in fish and n-octanol/water partition coefficients. The relationship between the two variables was studied by Devillers et al. (Reference Devillers, Bintein and Domine1996), and the formula linking them that we used is the following:
$$BC{F_i} = \;{10^{0.85 \cdot \log (Po{w_i}) - 0.7}}$$
where
$BC{F_i}$
denotes insecticide
$i$
’s bioconcentration factor and Powi is the octanol/water partition of insecticide i. Bioaccumulation, denoted by
${B_i}$
, is then computed using the following expression:
$${B_i} = {{BC{F_i} \cdot BCF_{Ref}^{LF}}}\over{{BC{F_{Ref}}}}$$
where
$BCF_{Ref}^{LF}$
and
$BC{F_{Ref}}$
represent the coefficients for the bioconcentration factor load (in L/lb a.i.) and bioconcentration factor for a reference substance, respectively.
Leaching is the third characteristic that defines the environmental fate load of an insecticide. It measures the mobility of pesticides in soil, which indicates their potential for migrating to groundwater. This factor is calculated by using the Screening Concentration in GroundWater (SCI-GROW) model – a screening model used to estimate insecticide concentrations in groundwater (EPA, 2016). Leaching is calculated using the formula:
$$L{e_i} = {{SC{I_{GRO{W_i}}} \cdot Le_{Ref}^{LF}}}\over{{L{e_{Ref}}}}$$
where
$L{e_i}$
denotes insecticide
$i$
’s leaching in Load per pounds of active ingredient (L/lbs a.i.),
$SC{I_{GRO{W_i}}}$
is the SCI-GROW index of insecticide
$i$
, and
$Le_{Ref}^{LF}$
and
$L{e_{Ref}}$
represent the coefficients for the leaching load factor (in L/lb a.i.) and the SCI-GROW index of a reference substance, respectively. Thus, to compute the environmental fate load of each insecticide active ingredient, we simply added the values we obtained for each of its three sub-components: degradation, bioaccumulation, and leaching.
The third and last component of the load of a pesticide is the environmental toxicity load, which measures the load of an insecticide on different organisms other than the targeted one. This measure is composed of the short-term effects on birds, mammals, fish daphnia, algae, aquatic plants, earthworms, and bees, and the long-term effects on fish daphnia, and earthworms (Samsøe-Petersen et al. Reference Samsøe-Petersen, Bøje, Hansen, Kjølholt, Jørgensen, Kudsk and Ørum2012). Table A3 shows the different reference values and load factors for each component of the environmental toxicity load. Each of the components is multiplied by a reference value and divided by a reference load:
$${E_{{i_n}}} = {{R{V_n}} \over {{v_{{i_n}}}}} \cdot R{L_n}$$
where
${E_{{i_n}}}$
is insecticide
$i$
’s effect on organism
$n$
in Load per pounds of active ingredients (L/lb a.i.),
$R{V_n}$
is the value for the reference substance on organism
$n$
,
$R{L_n}$
is the reference load of the reference substance on organism
$n$
in Load per pounds of active ingredients (L/lbs a.i.), and
${v_{{i_n}}}\;$
is the effect/value of insecticide
$i$
on organism
$n$
. Short- and long-term effects are then added up to obtain the total environmental toxicity load of a pesticide.
Parameters of reference substances to be used in the environmental toxicity load calculation
Source: Samsøe-Petersen et al., Reference Samsøe-Petersen, Bøje, Hansen, Kjølholt, Jørgensen, Kudsk and Ørum2012, p. 20.
We obtain the total load for each of the eight insecticides we consider by simply summing up the values we obtained for each of its three components: the human health load, the environmental fate load, and the environmental toxicity (see Table A4). In the last column of Table A4, we also show the area load for each insecticide. Calculating the load of each insecticide per acre allows us to compare the load of insecticides used by Florida orange farmers for a standard measure of unit area. Thus, we obtain the total area load of insecticide i at the maximum yearly recommended use (
$A{L_i}$
) in L/acre, as follows:
$$A{L_i} = \left( {{H_i} + E{F_i} + E{T_i}} \right) \cdot {U_i}$$
where is
${H_i}$
represents the Human Health Load of insecticide
$i$
,
$E{F_i}$
is insecticide
$i$
’s environmental fate load,
$E{T_i}$
is the environmental toxicity load of insecticide
$i$
– all in L/lb a.i. – and
${U_i}$
is the actual use of insecticide
$i$
on Florida oranges in lb a.i./acre.
Toxicity (load) components and total load per pound and per acre for the top eight insecticides’ active ingredients that are used to grow processed oranges in Florida
In contrast to the calculations above, the Quantity of Active Ingredients index simply measures insecticide use so that
$Q{A_i} = {U_i}$
, where
$Q{A_i}$
is the quantity of active ingredients in lb a.i./acre.
Open access
