October 2002 Meeting Agenda
FIFRA SCIENTIFIC ADVISORY PANEL (SAP)
OPEN MEETING
TUESDAY, OCTOBER 1, 2002
FIFRA SAP WEB SITE https://www.epa.gov/scipoly/sap/
OPP Docket Telephone: (703)305-5805
Sheraton Crystal City Hotel
1800 Jefferson Davis Highway
Arlington, VA 22202
(703) 486-1111
Projecting Domestic Percent Crop Treated with Pesticides for Dietary Risk Assessment
- 8:30 AM Introduction of Panel Members - Ronald J. Kendall, Ph.D.
(FIFRA SAP Session Chair)
- 8:35 AM Administrative Procedures by Designated Federal Official -
Mr. Steven Knott
- 8:40 AM Welcome - Mr. Joseph Merenda, Jr. (Director, Office of Science Coordination and
- Policy, Office of Prevention, Pesticides and Toxic Substances, EPA)
- 8:45 AM Opening Remarks - Ms. Marcia Mulkey (Director, Office of Pesticide
Programs, Office of Prevention, Pesticides, and Toxic Substances, EPA)
- 9:00 AM Introduction and Background - David Widawsky, Ph.D. (Biological
and Economic Analysis Division, Office of Pesticide Programs, EPA)
- 9:10 AM The Role of Use-Related Information and Percent Crop Treated
Statistics in Pesticide Risk Assessment and Risk Management - Arthur
Grube, Ph.D. (Biological and Economic Analysis Division, Office of Pesticide
Programs, EPA) 9:55 AM Proposed Methodology for Projecting Percent Crop
Treated - Mr. Philip Villanueva, (Biological and Economic Analysis Division,
Office of Pesticide Programs, EPA)
- 10:40 AM Break
- 10:55 AM Evaluation of Proposed Methodology for Projecting Percent
Crop Treated Using Case Studies - Mr. Philip Villanueva, (Biological
and Economic Analysis Division, Office of Pesticide Programs, EPA)
- 11:40 AM Public Comments
- 12:00 PM Lunch
- 1:00 PM Questions to the Panel
Purpose of SAP Presentation
The percentage of a given crop which is treated (or, more precisely, not treated) is a critical parameter in OPP's recently instituted, probabilistic human health exposure assessments because this factor determines the proportion of a crop which is assumed to have (or not have) residues. These residues, or expected residues, determine the dietary exposure to pesticides, which is a key component of dietary risk assessment.
EPA has historical data on domestic percent crop treated for a large number of crop/chemical combinations. These data can provide a basis for projecting future usage of pesticides (e.g., percent crop treated) for regulatory decisions based on dietary risk. Given EPA's mandate to assess pesticide food tolerances on a periodic basis, it is important for EPA to be able to project expected percent crop treated using these historical data. There are a number of statistical and/or mathematical methods for projecting future values from a historical record. EPA currently has methods for conducting these projections, but believes that refinements in these methods are warranted. Therefore, EPA has explored and developed methods that we believe could serve the purpose for refining projected domestic percent crop treated, and is seeking SAP input on these method developments.
Previous Method
EPA has analyzed percent crop treated data using a distributed lag, whereby one-year projections of percent crop treated were based on weighted historical data, with older data receiving lower weight and more recent data receiving higher weight. The confidence intervals associated with these projections are based on the variability of these historical data. While these methods have been shown to generate reasonable estimates, in cases where usage is trending upwards (or downwards), the uncertainty of the estimates can increase substantially. Therefore, EPA is seeking to develop a more robust method for projecting/estimating percent crop treated.
Proposed Method
EPA is proposing to use exponential smoothing as the analytical tool for projecting percent crop treated to better reflect upward or downward trends in the data. The size of existing data sets (usually 10 or fewer observations) precludes some regression techniques, and exponential smoothing has been shown to be a robust method in settings where small data sets are common. EPA is also proposing a method for estimating prediction intervals that is consistent with exponential smoothing projections. As part of the development of these techniques, EPA has been working with and testing sets of historical data and plans to present comparisons of results in addition to the theoretical discussion.
QUESTIONS FOR THE SAP:
1 - Our focus in developing a forecasting methodology for percent crop
treated (PCT) has been univariate methods, which seem most appropriate
given the available data. In the best cases, historical information on
PCT consists of no more than 15 observations. Such a limited number of
data points excludes the use of complex forecasting models, such as Box-Jenkins.
a. Keeping this in mind, please comment on the strengths and weaknesses
of exponential smoothing as a tool for forecasting PCT.
b. Please provide any comments or suggestions for enhancing the proposed
forecasting methodology that is based on exponential smoothing.
c. Please comment on OPP's assessment of other forecasting methods, including
whether there are other methods that would be suitable for producing three
to five year forecasts of PCT with the available number of data observations.
2 - Generally OPP will not perform a risk assessment for a pesticide
every year. Therefore estimates of percent crop treated (PCT) may need
to account for pesticide use for the next three to five years. For assessing
risk due to acute exposure to a pesticide, OPP relies on an estimate of
the maximum PCT. For a given forecasting model, a prediction interval
(i.e. interval forecast) could serve as an upper bound for an individual
point forecast. However, to estimate an upper bound for multiple point
forecasts, OPP proposes the calculation of simultaneous prediction intervals.
The probability that all of the future observations being forecasted will
be less than that bound is assumed to be the product of the individual
probabilities that each of the future observations will be less than that
bound (i.e. the individual probabilities are assumed to be independent).
a. Please comment on the reasonableness of the assumption of independence
between the observations.
b. What consequences would a violation of this assumption have on the
simultaneous prediction interval being calculated?
c. Please provide comments and recommendations on refinements for calculating
simultaneous prediction intervals.
- 5:30 PM Approximate Adjournment Time