Literature Review

Analysis of chemical usage, specifically insecticides, is complex given the number of factors that influence usage, including advances in seed varieties, introduction of new chemical compounds and fluctuating pest infestation levels.

Recent studies have focused on various trends in pesticide use. For example, as new ground is broken for growing a crop there is a tendency for pesticide use to increase. The literature indicates that this occurred in the early 1980s. Osteen and LivingstonReference Osteen and Livingston⁷ reported that there was an increase in planted acres and also an increase in pesticide application rates. The literature also discusses a recent trend toward mono-cropping or reduced use of rotations to meet market demands. This trend is in contrast to integrated pest management practices that reduce insect pest populations, and monocultures may even intensify pest pressureReference Lewis, van Lenteren, Phatak and Tumilnson⁸. Indeed, biodiversity and species evenness are both critical aspects of agroecosystems’ natural ability to resist pest pressure, and this ability is reduced or eliminated by monoculture cropping practicesReference Crowder, Northfield, Strand and Snyder^9, Reference Lundgren and Fergen¹⁰. Insecticide use is thus necessary to control the resulting increasing pest populations in monocultures.

The presence of damaging insect pests is an important determinant of insecticide use. Intuitively one can assume that insect pest infestation levels are closely related to the amount of insecticide applied. That is to say, ceteris paribus, the more insects that are present the more a producer will spray. These infestation levels, however, can fluctuate from year to year, are often difficult to predict and can spread rapidly. Previous studies indicate that the inability to predict pest infestations can lead producers to spray insecticide prophylacticallyReference Lazarus and Swanson^11, Reference Foster, Tollefson, Nyrop and Hein¹².

Insect pest infestation data have been used in previous studies to investigate the issue of producers making a simultaneous decision of insecticide use and GM adoptionReference Fernandez-Cornejo and Li^13, Reference Vialou, Nebring, Fernandez-Cornejo and Grube¹⁴. Perceived pest pressure likely affects whether a producer adopts GM seed and also how much they spray. By omitting pest pressure data there is likely a correlation between the error term and the dependent variable. Two-stage regressions are often used to model this type of decision-making process.

The most common empirical methodology used to investigate insecticide usage on agricultural crops is to evaluate changes in total pounds of insecticide appliedReference Vialou, Nebring, Fernandez-Cornejo and Grube^14–Reference Fernandez-Cornejo and McBride¹⁷. In the literature, however, an evaluation of chemical usage (such as ‘volume applied’) ignores important chemical attributes like strength and application practicesReference Osteen and Livingston⁷. As a result, problems arise when trying to assess changes in insecticide application rates over time.

One alternative method that accounts for changing chemical attributes is a hedonic pricing model. A hedonic pricing model can provide a proxy for changes in chemical potency and breadthReference Vialou, Nebring, Fernandez-Cornejo and Grube¹⁴. Another alternative empirical method proposed by Alston et al.Reference Alston, Hyde, Marra and Mitchell¹⁸ estimates chemical usage by multiplying the total acreage by the percentage treated for corn rootworm specifically. We have adopted a similar approach in this study. Our empirical methodology focuses on estimating the changes in chemical usage as a percentage of crop acres treated with insecticides in South Dakota.

There is a general consensus in the literature that total pesticide (herbicide plus insecticide) usage (total pounds applied) has increased due to increased GM acreage plantedReference Benbrook¹⁹, largely driven by increased glyphosate usage associated with glyphosate-tolerant GM crops. The general consensus is GM adoption has decreased insecticide per acre usage on US corn acreage planted over the past decadeReference Fernandez-Cornejo and Caswell²⁰. However, BenbrookReference Benbrook¹⁹ argues that only a small proportion of corn acres were treated with insecticides prior to the introduction of Bt varieties. Therefore, the reduction in per acre insecticide usage associated with Bt adoption may be overestimated in the recent literature.

These findings reported by Benbrook suggest that the observed increase in acres treated with insecticide in South Dakota may indicate a reversal in the downward trend for insecticide usage by producers in states with high GM adoption rates. This trend reversal may be reinforced by unintended consequences associated with very high Bt adoption rates. For example, the introduction of Bt corn has allowed US producers to practice mono-cropping or reduced use of rotations to increase corn production. Insect populations facing greater exposure to a single environmental barrier may genetically adapt, thus reducing the effectiveness of the GM trait. Meihls et al.Reference Meihls, Higdon, Siegfried, Miller, Sappington, Ellersieck, Spencer and Hibbard²¹ reports western corn rootworm has the ability to adapt to Bt toxins if refuge acre policy recommendations are not followed, and Bt-resistant corn rootworms were recently discovered in Iowa cornfieldsReference Gassmann, Petzold-Maxwell, Keweshan and Dunbar²². Our goal is to make a contribution to the GM/pesticide debate by providing insight on increased insecticide usage in a state having a high GM adoption rate targeting corn insect pests.

Methodological Issues

Corn, soybean, hay and sunflowers are important crops for South Dakota agriculture and each has significant pest problems. Corn has several pests that cause crop damage every year, including the European corn borer (Ostrinia nubilalis) and the western and northern corn rootworms (Diabrotica virgifera and D. barberi, respectively). Soybeans in South Dakota have recently (2001) been invaded by the soybean aphid (Aphis glycines Matsamura) which is present in all soybean-producing regions of South Dakota (J.G. Lundgren, pers. comm., November 13, 2009). In South Dakota, 46 counties produced soybean in 2007. Additional information on soybean aphid infestation levels at the county level can be found at http://www.ncipmc.org/index.cfm. In 2002, the soybean aphid was present in 20 of 66 counties (30%) in South Dakota countiesReference Catangui²³. By 2007, the soybean aphid had spread to 46 counties in South Dakota (≈70%). Alfalfa weevil (Hypera postica) affects alfalfa and is present in every county in South DakotaReference Catangui²⁴. Sunflowers, because they are a native species and harbor an endemic insect fauna, face a substantial pest threat from 15 different insectsReference Knodel, Charlet and Gavloski²⁵. Infestations can be highly variable even within a single sunflower field. Over time these pests have undoubtedly affected insecticide use. As this study does not directly link GM adoption to insecticide use, the inclusion of pest infestation data is not critical. However, we included a proxy variable for soybean aphid infestation at the county level in 2007. This proxy is potentially important because the scope of infestation and insecticide-use responses have changed dramatically over the time-frame of this study.

To estimate the effect of multiple explanatory variables on insecticide use for all counties in South Dakota, a fixed-effect model was employed. A fixed-effect model is often used on non-experimental data, where a scientific control group is not available or possible, treating each observation as its own controlReference Allison²⁶. This model also accounts for time-invariant unobserved effects that are not captured with available dataReference Wooldridge²⁷.

A typical ordinary least-squares (OLS) model often suffers from omitted variable bias. In the case of insecticide usage, the omitted variable is frequently the magnitude of the pest population. In South Dakota there are limited data on this particular issue, making it challenging to describe the relationship between insecticide use and other variables. Using the fixed-effect model we can assume there are two types of unobserved effects, those that vary over time and those that are constant, denoted as U _it and a _i, respectivelyReference Wooldridge²⁷. The unobserved effect, a _i, accounts for unobserved effects that do not change over time (i.e., latitude, longitude, soil type and cultural norms). The time-varying error term U _it is subject to many assumptions that require some discussion. Examples include precipitation and technology adoption.

Accounting for pest populations is of particular importance. Some pest populations are unpredictable and can vary greatly from year to year. Also, pest communities have experienced dramatic change over the past 30 years. To account for the potential effect of soybean aphid pest populations at the county level we include a variable that identifies soybean aphid infestation presence in South Dakota in 2007. Other changes in pest populations are contained within the time-varying error term. We also make the assumption that pest populations are not correlated with crop choice as we believe there are other factors that have greater influence on crop selection, including market demands and prices as well as protective measures if pest infestations occur. If correlation between crop proportion and pest population does exist it would violate several of the assumptions necessary to ensure unbiased estimationReference Wooldridge²⁷. Furthermore, we are not interested in how market factors are affecting cropping patterns. We consider this an issue for future research. Given the econometric caveats, our goal is to investigate how changes in cropping patterns may have affected the proportion of crop acreage applied with insecticide. Of special interest is corn acreage treated with insecticide because it is the only major crop in South Dakota that has a GM solution for insect pests.

Data

Data from several sources were used in this project. Data were collected from the National Agriculture Statistics Service (NASS) as well as the US Census of Agriculture. Data regarding total acres planted and crop-specific acres planted at the county level were retrieved from NASS's Quick ‘Stats All States Data-Crops’²⁸. The census data on chemical usage were collected from ‘Agriculture Chemicals Used, specifically: sprays, dusts, granules, fumigants, etc. to control insects on hay and other crops’. The census reports all acres treated at the county level, but does not report crop-specific acres treated. The empirical hurdle we overcome is the lack of data on which crops are being treated, and by extension providing insight on why there has been an increase in acres treated as a proportion of acres planted at the county level. Soybean aphid infestation information by county is provided by the Plant Science Department at South Dakota State University and the Agricultural Research Service Station in Brookings South Dakota.

Empirical Model

The data set contains observations for each county over the past seven Census reporting periods (1978–2007). A fixed-effect modeling approach was selected to analyze the constructed panel data set. The data are a balanced panel, meaning that there are no missing observations for any counties over the observed years. Given the econometric issues discussed earlier, we used a ‘cluster robust standard error’ approach to account for both the intra-cluster correlation and between-cluster heteroscedasticity. Intra-cluster correlation can be caused by unobserved omitted variables (e.g., insect populations: see equation 21–13 in Cameron and TrivediReference Cameron and Trivedi²⁹ for more details). A complete listing of the variables used in the model, and their definitions, is located in Table 1. Summary statistics for the variables are provided in Table 2.

Table 1. Variables used in panel regression.

Table 2. County-level summary statistics: percentage of acres planted: N = 462.

The analysis uses South Dakota county-level data on acres planted with corn, soybean, sunflower and hay. In aggregate, these four crops accounted for 77% of all acres planted in 2007. The empirical analysis employs two regression models. The first regression model (Eqn 1) used all the crop variables as well as two dummy variables: (a) shift variable for 2007 and (b) a dummy variable to account for county-level aphid infestation in 2007. The second regression model (Eqn 2) includes interaction terms to capture the influence of each crop variable on insecticide usage in 2007.

The decision to include only the 2007 dummy variables (i.e., year, aphid) to capture changes in insecticide use over time is based on data indicating that insecticide use in South Dakota remained fairly stable from 1978 to 2002, with 8–10% of county-level planted acres being treated with insecticides. In 2007, South Dakota counties experienced an increase in acres treated with insecticides, on average; roughly 20% of the planted acres were treated with insecticides. This increase is depicted in Figure 2.

(1)$$\eqalign{Y_{1\,{it}} = \,& {\rm \delta} _1 + {\rm \delta} _2 {\rm YR}07_{2\,{it}} + {\rm \delta} _3 {\rm Aphid}_{3\,{it}} + {\rm \beta} _1 {\rm Corn}_{1\,{it}} \cr & + {\rm \beta} _2 {\rm Soybean}_{2\,{it}} + {\rm \beta} _3 {\rm Sunflower}_{3\,{it}} \cr &+ {\rm \beta} _4 {\rm Hay}_{4\,{it}} + a_i + U_it} $$

(2)$$\eqalign{Y_{2\,{it}} =\, & {\rm \delta} _1 + {\rm \beta} _1 {\rm Corn}_{1\,{it}} + {\rm \beta} _2 {\rm Soybean}_{2\,{it}} \cr &+ {\rm \beta} _3 {\rm Sunflower}_{3\,{it}} + {\rm \beta} _4 {\rm Hay}_{4\,{it}} + {\rm \beta} _5 {\rm Corn}{^*}07_{5i} \cr &+ {\rm \beta} _6 {\rm Soybean}{^*}07_{6i} + {\rm \beta} _7 {\rm Sunflower}{^*}07_{7i} \cr &+ {\rm \beta} _8 {\rm Hay}{^*}07_{8i} + {\rm \alpha} _i + U_{it}, \cr &\ \ \ \ {\rm for}\,i\,{\rm county}\,{\rm in}\,{\rm year}\,t.}$$

The dependent variable in both regressions is a county-level variable and is defined as: Y _it ≡ total acres treated with insecticide_it/total acres planted_it. By using the proportion of acres treated with insecticide as the dependent variable, instead of a weighted measurement of pesticides applied, we can avoid the problem of aggregating across a heterogeneous group of pesticidesReference Burrows³⁰.

Figure 2. Proportion of planted acres treated with insecticide by county in South Dakota.

The outliers within the data identified in Figures 2 and 3 are worth discussion. Figure 2 shows the level and year of outlier occurrence, whereas Figure 3 provides the geographical location of outlier counties where an unusually high percentage of acres treated with insecticides occurred. The outliers tend to form a cluster in the southeast region of the state, which is also a large corn and soybean production area. One 2007 outlier, Shannon county in western South Dakota, represents a county with very few planted acres, thus easily achieving a high percentage of their acres treated with insecticides. Using a robust model accounts for the outliers and therefore these observations were left in the dataReference Bramati and Croux³¹.

Figure 3. Map of South Dakota county outliers by year.

The crop variables are defined as a percentage of county-level total acres planted that are devoted to each specific crop for a specific county. County-level crop production patterns from 1978 to 2007 are provided in Figure 4. A change in a crop's acreage share at the county level is hypothesized to influence insecticide usage at the county level. Corn has recently experienced a large increase in acres planted at the county level, increasing from 23.5% in 2002 to 28.5% of total acres planted in 2007. Soybean acres planted have also dramatically increased over the span of the study, increasing from just 2.4% in 1978 to 21% in 2002 before dropping to 18% in 2007. Hay acres harvested and planted (1978–2007) ranged between 25 and 30% of acres planted at the county level. A majority of the hay acres harvested are located in western South Dakota. From 1978 to 2007, the proportion of acres planted with sunflowers was variable, ranging from 0.5 to 4% of county-level total acres planted.

Figure 4. County-level crop acres planted in South Dakota.

The rapid expansion of the soybean aphid from 20 counties in 2001 to 46 counties in 2007 (all soybean-producing counties in South Dakota) has resulted in it becoming targeted as a major pest. To capture the aphid effect on insecticide usage a binary variable is included, set to one for counties when the soybean aphid is present in 2007 and zero otherwise. This variable is a proxy for aphid pest populations and it is hypothesized that the presence of the soybean aphid affected insecticide usage in South Dakota.

Bt Corn and Insecticide Usage in South Dakota Counties

The regression estimates (Table 4) indicate that there is a statistical relationship between the proportion of corn acres planted and the proportion of acres treated. Holding all other variables constant the prediction equation is

(3)$$E(Y_{2\,{\rm corn}} ) = 0.2212{^*}{\rm Corn} + 0.2309{^*}{\rm Corn}{^*}07.$$

Equation 3 includes the corn*2007 interaction term. For all Census years except for 2007, the effect of the proportion of corn acres related to the proportion of acres treated with insecticide is given by the first term only. Thus, a 1% increase in acres planted with corn results in a 0.22% increase in the proportion of acres treated, ceteris paribus.

In 2002, on average for South Dakota counties, the proportion of corn acres to total acres planted is estimated at 0.2347. By substituting the average proportion of corn acres planted for 2002 into Equation 3, the predicted overall contribution of corn acres planted to the proportion of acres treated is estimated to be 0.0518 (0.2211*0.2347). The summary statistics for 2002 indicate that, on average, 7.1% of planted acres were treated with insecticide. Corn acres planted in 2002, on average, accounted for 73% (0.0518/0.071) of total acres treated and 24.6% (0.0518/0.2347) of corn acres planted in 2002 were treated with insecticide. In 2002, 43% of corn acres planted were Bt or stacked gene corn acres (state average). Therefore, Bt corn accounted for 0.1001 of the 0.2347 share of corn acres (0.43*0.2347) to total acres planted. Thus, the proportion of non-Bt corn acres planted accounted for 0.1346 of total acres planted. Assume Bt corn acres were not treated with insecticide (this assumption is made to in order to determine the upper bound of non-Bt corn acres treated). The implication is that in 2002, 38.5% (0.0518/0.1346) of non-Bt corn acres planted in South Dakota were treated with insecticide. It should be noted that the above estimates are based on the assumption that the percentage of acres planted with Bt seed at the state level is consistent with Bt plantings at the county level. County-level data on Bt usage is not available.

In 2007, on average for South Dakota counties, the proportion of corn acres to total acres planted is estimated at 0.284. The total effect of corn acres planted in 2007 on acres treated is determined by both coefficients in Equation 3. At the county level, the predicted contribution of the proportion of corn acres planted to the proportion of acres treated with insecticide is estimated to be 0.128, i.e. (0.2211*0.284 + 0.231*0.284). The estimated predicted value of acres treated associated with acres of corn planted in 2007 indicates that approximately 12.8% of the total acres planted were treated corn acres.

The summary statistics for 2007 indicate that on average, 20.1% of total (all crops) planted acres were treated with insecticide. Corn acres planted in 2007, on average, accounted for 63.8% (0.128/0.201) of acres treated. Furthermore, it is estimated that 45% (0.128/0.284) of corn acres planted in 2007 were treated with insecticide, a substantial increase relative to 2002 (24.6%). Given that 59% of corn acres planted in South Dakota contained Bt, our estimate for corn acres treated suggests that some Bt acres were treated in 2007.

These back of the envelope findings are surprising on a number of levels. First, GM corn acreage share and insecticide corn acreage coverage both increased. Secondly, in 2007, on average at the county level, corn accounted for 28.4% of total acres planted. Bt corn account for 59% of corn acres planted. Therefore, Bt corn accounted for 16.75% of total acres planted (0.59*0.284). If we assume Bt corn acres were not treated with insecticide, then non-Bt corn acres accounted for 11.65% of total acres planted. The implication is that in 2007 109.87% of non-Bt corn acres planted in South Dakota was treated with insecticide (0.128/0.1165). We regard this estimate as an upper-bound for non-Bt acres treated and it suggests that South Dakota producers treated some Bt acres with insecticide in 2007.

This proportion is exceedingly high given: (1) that in 2002 it is estimated that 38.5% of non-Bt corn acres were treated acres; (2) traditional insect pest management patterns prior to GM introduction; and (3) the USDA refuge acreage requirement for producers who use Bt seed. Our estimate of a very large increase in corn acres treated raises some interesting questions. Has the change in cropping or seed marketing patterns forced producers to rely more heavily on chemical control? Is it the result of producers moving away from traditional rotation control methods used in the past to suppress insect pests? Are producers treating Bt corn acres in a manner they treat non-Bt acres? Additional research needs to be conducted on this issue.

Next, the coefficient estimates for the interaction terms found in Table 4 provide additional empirical evidence of increased acreage treated with insecticide being linked to changing cropping patterns. The corn interaction term coefficient estimate indicates that the slope relationship between corn acres planted and total acres treated with insecticide increased dramatically in 2007 from 0.2212 during the 1978–2002 period to 0.4521 in 2007 (0.2212 + 0.2309). This steepening of the slope indicates an increase in the marginal contribution of corn acres planted to total acres treated in 2007, relative to previous census years.

Overall, the empirical evidence suggests that corn, hay and sunflower production in South Dakota have experienced a recent intensification of insecticide use in 2007 relative to past Census reporting years. Empirical evidence indicates that the presence of the soybean aphid in South Dakota did contribute to increase the acres treated in 2007.

Broader Implications of the South Dakota Case Study

One question raised by the findings reported in this case study is whether the 2007 increase in acres treated with insecticide in South Dakota was an isolated phenomenon or whether other mid-western states experienced a similar increase. Figure 5 shows the proportion of acres treated with insecticide from 1987 to 2007. Of the 13 states that were looked at, Illinois, Indiana, Iowa, Minnesota and Kentucky seem to have experienced a marked increase in the proportion of acres treated in 2007. It is worth noting that while there appears to be a large and sudden increase, because of the limited nature of the data, it is unclear as to whether this increase was in fact gradual for the years in between 2002 and 2007.

Figure 5. Proportion of acres treated with insecticide for mid-west states.

For some states, like Illinois, the increase in the proportion of acres treated began in 1992. Other states, including Iowa, Minnesota, South Dakota, Indiana, Kentucky, had relatively low, though variable, proportions of their cropland treated with insecticides until the increase in 2007. The differences between states could be the result of varying or spreading pest populations, but more likely this is the result of a national paradigm shift in insecticide use in spite of the widespread adoption of Bt.

The 13 states listed in Figure 5 make up the Midwest region of the United States and are intensive crop-producing states. These states will provide the basis for further analysis at the county level and could provide greater insight into the reasons for the 2007 increase in the proportion of acres planted being treated with insecticide.