Probabilistic Evaluation of Pesticide Impact Metrics

The ratings of 1, 3, and 5 in the EIQ method are surrogates for low, medium, and high risk or impact or toxicity or persistence, depending on the factor of interest. For demonstration purposes only, we show how converting the ratings to estimates of risk probabilities for only four of the factors limits the value of the EIQ method. The EIQ factors, “long-term health effects”, “leaching potential”, and “surface runoff potential”, and ratings of “little-none”, “possible”, “definite”, “small”, “medium”, and “large” imply that they are risks. Therefore, they have a probability of occurrence rather than an absolute certainty of occurring. Similarly, the factor “beneficial arthropod toxicity” has ratings of “low impact”, “moderate impact”, and “severe impact”. Degrees of impact also have associated uncertainty.
Because the ratings of 1, 3, and 5 are surrogates for risk, they can be converted to risk intervals that incorporate the underlying probabilities. Therefore, the simplest, yet coarse, way to do this is to assume the ratings of 1, 3, and 5 span the range of risk from 0 to 1 (or 0 to 100%). A rating of 1, when mapped onto an interval of risks would be 0 to 0.32. A score of 3 would be 0.33 to 0.66 and a score of 5 would be 0.67 to 1. Consequently, if a pesticide has a “surface runoff potential” factor that has a score of 3, it is at medium risk of runoff. However, a discrete score of 3 does not capture the probabilistic nature of risk, yet the score of 3 is intended to represent medium risk. Therefore, the score needs to be mapped to an estimate of risk. This can be done most simply by assuming a uniform probability density function of risk values from 0.32 to 0.66 for medium risk. Medium risk implies uncertainty and probability, but a score of 3 does not accommodate that risk estimate. An interval of 0.33 to 0.66, however crudely, accommodates the probability of occurrence.
To demonstrate the consequences of mapping discrete risk ratings to probabilities, we calculated adjusted EIQs for a group of 20 actual insecticide active ingredients with unadjusted EIQs ranging from 22.1 (methiocarb) to 44 (diazinon). The insecticides evaluated were chosen randomly from lists of active ingredients in Yu (2008) , who provides a relatively complete list of currently registered insecticides. Five insecticides each were chosen randomly from four chemical classes: carbamates, neonicotinoids, organophosphates, and pyrethroids. The unadjusted EIQs and ratings were obtained from the New York State Integrated Pest Management Program, Cornell University (www.nysipm.cornell.edu/publications/eiq/). The four factors discussed above were converted to probability ranges of risk and all other factors were held constant at their respective deterministic scores. To align those deterministic scores with the probability ranges mapped for the four factors, the ratings were converted to static probabilities proportional to the value of the scores. For example, a score of 3 for fish toxicity was converted to 0.5.
Using Monte Carlo simulation (Oracle Crystal Ball^® 11.2, Denver, CO), we calculated adjusted EIQs under different hypothetical scenarios by incorporating the probability ranges associated with the four factors (Fig. 1). Probabilities of occurrence of adjusted EIQ values were determined by incorporating sampling from the statistical probability density function of each input variable used to calculate the EIQ. Each of the four input variables was sampled 20,000 times. Then, the variability for each input was propagated into the output of the model so that the output reflected the probability of values that could occur.