Uncovering the economic value of natural enemies and true costs of chemical insecticides to cotton farmers in China

The NCP is a densely populated agricultural area in China, covering an area of approximately 400 000 km² and accounting for about two-thirds of China's cotton acreage in 2011 [17]. Other major crops in the NCP are wheat, maize, vegetables and fruit. Insecticides are intensively used in crop production in China, and usage in cotton is among the highest on a per hectare basis [18], despite the extensive adoption of Bt-cotton since the late 1990s [19]. Broad-spectrum pyrethroid and organophosphate insecticides represent more than 85% of all insecticide use in cotton in the NCP, and both insecticide groups have similarly deleterious effects on natural enemies in cotton [20].

The study area is located about 100 km southeast of Beijing, and includes 10 villages from Langfang prefecture in Hebei and 10 villages from Wuqing district in Tianjin [21]. The household survey included 311 cotton-farming households, randomly selected from the 20 villages (15–16 households per village). In this area, cotton is grown in predominantly small farms, with an average farm area of 0.5 hectare per household, and Bt cotton has been nearly fully adopted by farmers since early 2000s. Detailed data on household characteristics, cotton cultivation and yield, decision making and practices with respect to insecticide use were collected for each household via face-to-face interview during the 2011 cotton growing season. To ensure the accuracy of recall production input data collected, we conducted four rounds of interviews: late June, late July, late August and mid-November 2011. For each household, we selected one focal cotton field to monitor cotton pests and natural enemies of these pests. Observations on the densities of cotton pests and natural enemies were made in each field in late June, late July and late August 2011 (details in online supplementary information, section 1, available at stacks.iop.org/ERL/13/064027/mmedia).

When an ecosystem service can substitute for an existing marketed input or contributes to a measurable marketed output, economic value of changes in the level of the service can be readily inferred [22, 23]. A widespread application of this factor input valuation method is fertilizer replacement value to measure the value of biological nutrient cycling, for example in cereal-legume systems [24]. Zhang and Swinton [25] used a dynamic optimal pest control model to estimate the economic value of natural pest regulation service for US soybean. Other attempts at placing economic values on pest regulation services estimated the total cost of averted pest damage due to all pest control practices and then attributed a fraction of the total to natural enemies [5, 26]. The approach adopted in our study is based on empirically estimated relationships between insect densities, insecticide quantity, labor requirement, and yield (figure 1).

Zoom In Zoom Out Reset image size

2.1.1. Marginal value of ladybeetles

The value of natural enemies (ladybeetles), MV(B), is estimated at the margin and consists of three benefits: (i) reducing insecticide use due to their effect on pest density; (ii) reducing labor use for spraying insecticides due to their effect on insecticide use; and (iii) reducing cotton yield loss (or increasing cotton yield) due to their effect on pest density (figure 1). The marginal value of ladybeetles implies the value of an additional unit of ladybeetle under current production environment where there is insecticide disruption and Bt cotton has been almost fully adopted by farmers, and is estimated as follows:

$$\begin{equation} {\rm MV(B)} = {\rm MV}_{1}(B) + {\rm MV}_{2}(B) + {\rm MV}_{3}(B) \end{equation} \tag{ 1 }$$

where MV₁(B), MV₂(B) and MV₃(B) are marginal value of ladybeetles from reducing insecticide use, reducing labor use, and reducing cotton yield loss, respectively.

To estimate MV(B), three econometric models are developed. They are farmers' per hectare insecticide use (Z, kg ha⁻¹), labor used in insecticide application (L, hour ha⁻¹), and cotton yield (Y, kg ha⁻¹) models:

$$\begin{equation} Z = a_{0} + a_{1} B + a_{2} P_{z} + a_{3} S + a_{4} A + a_{5} E + \epsilon _{z} \end{equation} \tag{ 2 }$$

$$\begin{equation} L_{z} = d_{0} + d_{1} B + d_{2} Z + d_{3} S + d_{4} A + d_{5} E + \epsilon _{L} \end{equation} \tag{ 3 }$$

$$\begin{equation} \begin{array}{l@{}l@{}l@{}} Y = {\rm exp}(b_{0}) F^{b_1} L^{b_2} O^{b_3} A^{b_4} E^{b_5}\\ S^{b_6} [1{\rm -exp}(-c_{1} Z - c_{2} B)] + \epsilon _{Y} \end{array} \end{equation} \tag{ 4 }$$

where symbols a₀ to a₅, d₀ − d₅, b₀ − b₆, and c₁ − c₂ denote parameters to be estimated. ε_z, ε_L and, ε_Y are random error terms with a standard distribution. Equation (2) models insecticide use as a linear function of ladybeetle density B (1000 ha⁻¹, measured as the average density in late July and late August), insecticide price or average unit value of insecticides P_z (CNY kg⁻¹) (measured as total value divided by total quantity of insecticides used), field size S (ha), age of household head A (year), and education of household head E (years). In this equation, ladybeetle density B is our key interest. We did not include pest density in this model because pest density is also a function of ladybeetle density. Using observed ladybeetle density implies that we estimate the model using a reduced form of ladybeetles. In addition, including both pest (e.g. aphid) and ladybeetle densities in the model induces multi-collinearity issue in the econometric analysis.

Equation (3) models labor use for insecticide application (L_z). Here Z becomes one of the explanatory variables. If Z is correlated with the error term in equation (3), the estimation of the effect of Z on L_z will be biased [27]. To address this potential endogeneity issue, we included P_z in equation (2) as an instrumental variable for Z and included the predicted value of Z from equation (2) in equation (3).

Equation (4) models farmers' cotton yield (kg ha⁻¹), Y, as a non-linear production function with two components. The first component is a Cobb–Douglas production function that accounts for the effects of standard inputs and household characteristics. Standard inputs include fertilizer use (F, the sum of elemental N, P and K, kg ha⁻¹), total labor use (L, hour ha⁻¹), and other input uses (O, the sum of input costs of seed, irrigation and control weed, CNY ha⁻¹). The second component is an exponentially decreasing damage control function [28, 29], which reflects diminishing returns for damage abatement inputs including insecticides and ladybeetles. We chose the exponential damage control function because it fits our field survey data better than the alternative Weibull specification. Similar to equation (3), the predicted values of Z in equation (1) are used in equation (4).

From equations (2–4), which are estimated with Ordinary Least Squares (OLS), we derive MV₁(B), MV₂(B), and MV₃(B), respectively. Then MV(B) is calculated as follows:

$$\begin{equation} \begin{array}{l@{}l@{}l@{}} {\rm MV}(B) = {\rm MV}_{1}(B) + {\rm MV}_{2}(B) + {\rm MV}_{3}(B)\\ = [-a_{1} P_{Z}] + [-d_{1}W] + [c_{2} {\rm exp}(b_{0}) F^{b_1} L^{b_2} O^{b_3}\\ A^{b_4} E^{b_5} Fs^{b_6} {\rm exp}(-c_{1} Z - c_{2} B) P_c] \end{array} \end{equation} \tag{ 5 }$$

where P_Z is the average price of insecticides used by farmers (54.13 CNY kg⁻¹); W is the implicit wage of family farming labor. For the study area, the rate was estimated at 5 CNY hour⁻¹ in 2011, about two-thirds of the hourly wage of migrant labor working in the urban area during the same year (estimated at 80 CNY per day and 10 hours per day). P_c is the average price of cotton output (seed cotton, 8.33 CNY kg⁻¹ in the study area in 2011). The marginal values of ladybeetles, equation (5) can be calculated for different values of B based on the mean values of all other variables, using only those estimated parameters that are statistical significant in equations (2)–(3). A summary of statistics of all variables used in regressions is presented in the online supplementary information (table S1).

2.1.2. Marginal cost and benefit of insecticide use

The costs and benefits of insecticide use are also evaluated at the margin and interpreted as the marginal value of an additional unit of insecticide use under current conditions. The estimation consists of three components (figure 1): (a) MC₁(Z), marginal cost of insecticide product (or unit price of insecticide, P_c); (b) MC₂(Z), marginal cost of labor input for insecticide application; and (c) MB(Z), marginal benefit of insecticide use from reducing cotton yield loss (or increasing cotton yield). A fourth component, MC₃(Z), which is the marginal cost of insecticide use due to ladybeetle mortality, is also discussed and presented in section 3 of SI but is not included in the final estimate as we opt for a conservative valuation method in light of the uncertainty about the causal directions (explained in section 3.2). The analysis also does not cover the costs to farmer's health and local environment. The marginal value of insecticide use, MV(Z), is given by:

$$\begin{equation} {\rm MV}(Z) = {\rm MB}(Z) - {\rm MC}_{1}(Z) - {\rm MC}_{2}(Z). \end{equation} \tag{ 6 }$$

From equations (3) and (4), we can derive MB(Z) and MC₂(Z), respectively, as follows:

$$\begin{equation} \begin{array}{l@{}l@{}l@{}} {\rm MB}(Z) = [c_{1} {\rm exp} (b_{0}) F^{b_1} L^{b_2} O^{b_3} A^{b_4} E^{b_5} F^{b_6}\\ {\rm exp}(-c_{1} Z - c_{2} B)] P_c \end{array} \end{equation} \tag{ 7a }$$

$$\begin{equation} {\rm MC}_{2}(Z) = -d_{2} W \end{equation} \tag{ 7b }$$

where variables and parameters are the same as defied above.

Then MV(Z) can be estimated as:

$$\begin{equation} \begin{array}{l@{}l@{}l@{}} {\rm MV}(Z) = [c_{1} {\rm exp}(b_0) F^{b_1} L^{b_2} O^{b_3} A^{b_4} E^{b_5} F^{b_6}\\ {\rm exp}(-c_{1} Z - c_{2} B)] P_{c} - P_{z} - [d_{2} W]. \end{array} \end{equation} \tag{ 8 }$$

To explore how the marginal value of insecticide use, MV(Z), varies under different circumstances, we conducted simulations, using mean values of all other variables in equations (3) and (4). The parameters are estimated from equations (3) and (4), again, we used only those estimated parameters that are statistically significant.

Farmers frequently applied insecticides for the control of cotton pests. The surveyed cotton plots were treated 1–14 times over the growing season, 8.2 times on average (table 1). Our survey confirmed that most insecticides used are broad-spectrum, particular pyrethroid and organophosphate insecticides [20, 28, 30]. The average amount of formulated insecticide product applied was 22.4 kg ha⁻¹, at an average cost of 1091.7 CNY ha⁻¹ (approximately USD 169)⁸. In addition, farmers put in 173.2 hours of labor per hectare, on average, to carry out the insecticide applications. For a household that planted 0.5 hectare of cotton in our study area, the estimated total cost of insecticide application was equivalent to 3.7% of household average income or 10% of the average crop income of rural households in Hebei in 2011 [17].

Cotton aphids (Aphis gossypii) were the only abundant cotton pest (figure 2). Whiteflies (Bemisia tabaci) were present in July and August, but much less abundant than aphids. Densities of spider mites (Tetranychus cinnabarinus; <0.2/plant), mirid bugs (mainly Apolygus lucorum; <0.05/plant), and bollworm (Helicoverpa armigera; <0.03/plant) were negligible. Ladybeetles (mainly Harmonia axyridis, Propylea japonica, and Coccinella septempunctata) were dominant among the natural enemies (figure 3). Their numbers were very low (1.6/100 plants) in June—none were found in 250 of the 311 focal fields and those appeared in 61 fields were predominately larvae. Ladybeetle numbers were much greater in July and August (figure 3). Most individuals observed in July and August were in the adult stage, while larvae were mostly absent, indicating that the individuals found may originate from habitats other than the focal fields. Given that aphids are the primary pest in cotton and that ladybeetles are the main natural enemy group that control cotton aphid in our study villages and the NCP [20, 21], our analysis focuses on the cotton aphid–ladybeetle relationship.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We interviewed farmers about their insecticide application decisions for each major pest (table 2). The vast majority based their insecticide use decisions on the presence of aphids in the cotton field (34.6% + 56.6%), while the other 9% used prophylactic (preventative) spraying to prevent cotton aphid. The practice of prophylactic treatment was more widespread for mirid bugs and cotton bollworm, at rates of 27% and 35%, respectively. Only one of the 311 interviewed farmers indicated awareness of aphid natural enemies (e.g. ladybeetles) in his cotton field but he did not take them into consideration when making insecticide treatment decisions. This finding reveals that farmers' decision-making on pest management did not factor in the potential of biological pest control by natural enemies.

On each actual insecticide application, we also asked farmers to record the primary targeted pest(s) (PTPs) that was aimed to control and non-PTPs that could also be controlled at the same time (table 3). The results show that: (1) aphids, among the singularly targeted pests, received the highest number of insecticide applications; and (2) farmers also often use insecticides to control more than one pest at a time (rows 4–7); and 3) on the average, 9% of 8.2 insecticide applications (or 0.74) also helped controlling non-PTPs.

Greater ladybeetle population densities are associated with lower insecticide use (figure 4). There are multiple effects and causal pathways underlying this: (1) ladybeetles reduce pest level, which in turn reduces the volume of insecticides used; (2) a greater proportion of ladybeetles are killed in fields that receive more insecticides; (3) fields with fewer aphids will accumulate fewer ladybeetles as the residence time of ladybeetles depends on prey density [31]. Dynamically reciprocal causality in the ladybeetle-insecticide use relationship, if present, would introduce self-reinforcing feedbacks in the pest control system. This implies a potential for either a vicious cycle (i.e. the pesticide treadmill; [11]) or a virtuous cycle, when re-establishment of biocontrol services reduces the need for insecticide, further strengthening biocontrol service potential. Such dynamic relationship could in theory be modeled but this requires additional assumptions that are difficult to justify based on our survey data. Here, we identify the relationship between ladybeetle density, insecticide use, and farm economic outcomes based on econometric modelling from a farmer's behavioral point of view. The econometric model was estimated with insect density data collected in late July and late August when ladybeetle individuals were predominantly adults in our study area which are less susceptible to insecticide poisoning than larvae. Because other factors might also affect farmers' insecticide use (kg ha⁻¹) and associated labor input (hour ha⁻¹), we ran multivariate regressions to control for the confounding factors.

Zoom In Zoom Out Reset image size

Ladybeetle density is significantly and negatively associated with per hectare insecticide use (table 4). An increase of 1000 ladybeetles per hectare, equivalent to nearly 2 ladybeetles per 100 cotton plants at a density of 54 000 plants ha⁻¹ in the study villages (details in SI, section 2), is associated with a reduction of insecticide use of 0.69 kg ha⁻¹ (P < 0.01). Furthermore, an increase of 1000 ladybeetles per hectare is associated with a reduction in labor use in insecticide application of 1.95 hours ha⁻¹ (P < 0.10). Expressed per ladybeetle individual, the corresponding savings are estimated at 0.69 g formulated insecticide and 7 seconds labor time per ladybeetle. Given the huge potential densities of ladybeetles per hectare, these are major potential benefits.

The estimated coefficient for ladybeetle density in the cotton yield equation was positive and statistically significant (c₂ in table 5), indicating that an increase in ladybeetle density reduced cotton yield loss from pest attack. Based on the fitted equation, a 50% reduction of yield loss to pests is reached at a ladybeetle density of 17 000 individuals ha⁻¹, holding everything else constant at the sample mean. Likewise, the fitted model indicates a positive impact of insecticide use on cotton yield (c₁ in table 5), corresponding to a 50% reduction of yield loss to pests at an insecticide input of 2.24 kg ha⁻¹, approximately a factor 10 smaller than the actual usage of insecticides.

Other factors also contributed to cotton yield. These include total labor input, education level of household head, and size of cotton field. The latter confirms the economy of scale effect for cotton cultivation in the study area. Fertilizer use was not significantly correlated with yield, suggesting over-use of fertilizers by farmers. To ensure high yield, overuse of fertilizers in crop production has been common in China [32]. Farmers in our sample applied 192 kilograms of fertilizer per hectare. Insignificant coefficient of fertilizer use implies that marginal impact of fertilizers is about zero, which is consistent with previous findings [32, 28].

Based on estimated coefficients in tables 4 and 5, we computed the marginal value of ladybeetles to cotton farmers, accounting for (1) the reduced cost of insecticides; (2) the reduced labor cost associated with insecticide application; and (3) the yield benefit estimated from the damage control production function.

Ladybeetle density in our study fields averaged 13 500 ha⁻¹ in July and August 2011. An increase of ladybeetle density by 1000 ha⁻¹ (7.4% of the current density level) in the NCP is worth 47.74 CNY (equivalent to USD 7.39), given the current cotton production practices (table 6). Lower insecticide use is associated with higher densities of ladybeetles and savings of about 38 CNY ha⁻¹ (or USD 5.8) on costs of insecticides and 10 CNY ha⁻¹ (or USD 1.5) on costs of labor. The marginal impacts on cotton yield and income are small (0.047 kg ha⁻¹ on yield and 0.39 CNY ha⁻¹ on income). This is partly due to the fact that farmers in the NCP have in general applied excessive amounts of insecticides to control pests (22.35 kg ha⁻¹ on average), substantially curbing the efficiency and marginal value of both insecticides and natural pest regulation.

The marginal value of ladybeetles declines as their density rises, regardless of the level of insecticide use (figure 5). Moreover, the marginal value of ladybeetles decreases with the volume of insecticide used, especially at lower ladybeetle densities. For example, at the density of 1000 individuals ha⁻¹, the marginal value of ladybeetles is 47.74, 67.65 and 163.27 CNY ha⁻¹ when evaluated at the mean, one-half of the mean, and one-fourth of the mean insecticide use volumes, respectively. This considerable driving effect of insecticide use on the marginal value of ladybeetles is mainly due to higher marginal value of ladybeetles at lower insecticide use. At the current average density of ladybeetles (13 500 individuals ha⁻¹), the marginal value of the ladybeetles would rise from 48 CNY ha⁻¹ at the current insecticide use (22.35 kg ha⁻¹) to 118 CNY ha⁻¹ at one fourth of the current insecticide use.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

High insecticide use contributes negatively to farmers' income. At the current average insecticide use level, for each additional kilogram of insecticides applied in cotton fields, farmers not only pay for the purchase price of insecticides (54.13 CNY kg⁻¹ on average), but also incur labor cost for spraying (12 CNY ha⁻¹) (table 6), while gaining a mere 0.36 kg ha⁻¹ yield saving valued at 3.02 CNY ha⁻¹. As a result, one additional kilogram of insecticide per hectare from the current usage level would reduce farmers' income by 63.14 CNY ha⁻¹, indicating that the 'true' marginal value of insecticides is in actuality negative. Given the uncertainty about the causal directions in the system as discussed above, the actual marginal cost of insecticide use can be greater or smaller. For example, an attempt to explicitly account for the loss of biological control provided by ladybeetles (estimated at 2.71 CNY ha⁻¹) would increase the marginal cost of insecticide use to 65.18 CNY ha⁻¹ (63.14 CNY ha⁻¹ + 2.71 CNY ha⁻¹). Despite the uncertainties and limitations in the data and the interpretation, there is convincing evidence that ladybeetles are an important driver in the system as the estimated coefficients of ladybeetle density in both insecticide use and cotton yield equations are statistically significant at 1% level (tables 4 and 5). It is plausible that, by suppressing aphid levels, ladybeetles are likely to drive down insecticide use, offering an economic value to cotton farmers. We put the upper bracket of the marginal cost of insecticides at 63.14 CNY ha⁻¹, as opposed to 65.18 CNY ha⁻¹, as it is deemed a more conservative estimate.

Figure 6 reports the marginal value of insecticides evaluated at different levels of insecticide use, showing clearly the extent of excessive use by farmers in our study area. At the current average price of insecticides and level of ladybeetle density, the optimal level of insecticide use, at which the marginal cost of insecticides is equal to the marginal value product of insecticides, is estimated to be 10.39 kg ha⁻¹. This value places an upper bracket on the optimal insecticide use because the negative social and environmental costs associated with insecticides are not incorporated in the analysis. The actual use in our sample is more than two times the calculated upper bracket for the optimum.