STATISTICAL%20ASPECTS%20OF%20COLLECTIONS%20OF%20BEES%20TO%20STUDY%20PESTICIDES

1
STATISTICAL ASPECTS OFCOLLECTIONS OF BEES TO
STUDY PESTICIDES

• N. SCOTT URQUHART
• SENIOR RESEARCH SCIENTIST
• DEPARTMENT OF STATISTICS
• COLORADO STATE UNIVERSITY

• EMAP Affiliate
• SPACE-TIME AQUATIC RESOURCEMODELING and ANALYSIS
  PROGRAM (STARMAP)

2
STARMAP FUNDINGSpace-Time Aquatic Resources
Modeling and Analysis Program

• The work reported here today was developed under
  the STAR Research Assistance Agreement CR-829095
  awarded by the U.S. Environmental Protection
  Agency (EPA) to Colorado State University. This
  presentation has not been formally reviewed by
  EPA.  The views expressed here are solely those
  of presenter and STARMAP, the Program he
  represents. EPA does not endorse any products or
  commercial services mentioned in these
  presentation.

3
PATH for TODAY

• CONTEXT Environmental Monitoring and Assessment
  Program (EMAP) Academic
• TOPICS TO CONSIDER
• What to Measure Indicators
• Other speakers will address this
• Important things to consider in designing a
  survey
• PLAN! , PLAN! , PLAN!
• A National or Regional Survey is a Substantial
  Undertaking

4
IMPORTANT THINGS TO CONSIDER IN DESIGNING A SURVEY

• 1. Probability Surveys vs Judgment Collections
• 2. Population Definition
• 3. Evaluation Units hives (colonies) or bees
• 4. Sampling Frames
• 5. Selecting the Sample Sites
• 6. Training
• 7. Collecting the Bees
• 8. Handling the Collected Bees
• 9. Quality Assurance
• 10. Data Management
• 11. Data Analysis

5
1. PROBABILITY SURVEYS versus JUDGMENT
COLLECTIONS

• Specialists Usually Know a Tremendous Amount
  About Limited Specific Situations
• This is the way science accumulates knowledge.
• But frequently specialists know a lot less
  about the overall situation than they think they
  do!
• An illustration follows
• Selection of stream segments for spawning studies
  by Oregon Department Fisheries and Wildlife

6
SELECTION OF STREAM SEGMENTS FOR SPAWNING STUDIES
(OREGON DEPARTMENT FISHERIES AND WILDLIFE)

• OBJECTIVE Estimate Number Of Coho Salmon
  Spawning in Streams of Oregons Coast Range
• Stream Segments Were Stratified As Being
• Low, Moderate, Or High, relative to
  quality of spawning habitat
• Low was not sampled high was sampled at three
  times the rate of moderate
• Quality of spawning habitat was evaluated for
  each selected segment

7
SELECTION OF STREAM SEGMENTS FOR SPAWNING STUDIES
(OREGON DEPARTMENT FISHERIES AND
WILDLIFE)(continued)
8
SELECTION OF STREAM SEGMENTS FOR SPAWNING
STUDIES(OREGON DEPARTMENT FISHERIES AND
WILDLIFE) continued

• EXAMPLE of Sampling Where Investigators Think
  Most of the Large Responses Are.
• Bad idea if knowledge isnt quite right
• Even 10 error rate can make this a very
  inefficient sampling approach
• ODFW Classification Was Off LOTS Further Than
  10.
• Many other such examples exist.

9
2. POPULATION DEFINITION

• A Population is the Set of Objects of Interest
  in a Survey
• Commercial hives
• Of cooperating beekeepers
• All hives
• All hives within 500m of a secondary road
• Species
• All
• Two species of primary interest

10
POPULATION DEFINITIONcontinued

• So What?!
• Major distinction
• Target population what you want to talk about
• Sampled population what you can talk about
• You probably dont want to talk about this sort
  of population
• All commercial hives owned by cooperating
  beekeepers within 100 miles of an EPA Regional
  Office, and within 500m of a paved secondary road
  in June, 2005.
• Where you go to collect bees does make a
  difference!

11
CONCLUSIONS ABOUT JUDGMENT SELECTED SITES

• Ecologists Typical Sites Probably Are Much
  More Homogeneous Than the Larger Context of
  Interest
• Nonprobability Samples Can Be Rather Biased for
  No Apparent Reason
• Typicalness for One Set of Responses Says Nothing
  About Typicalness for Any Other Response, i.e.
  Any Response Not Used in Determining Typicalness

12
EVALUATION UNITS HIVES (COLONIES) OR BEES?

• So what?
• If Hives (or colonies) Are Your Evaluation Units,
  You Must
• Select hives in the sampling process
• Have a response which can be attached to a
  selected hive
• Give final answers in terms hives
• Ex Proportion of hives (colonies) with yy gt xx

13
4. SAMPLING FRAMES

• A Sample Frame Provides a Means to Identify or
  Locate the Individual Units in the Population
• May be a list
• The basis for most of the older sampling theory
• Often is imperfect! Sometimes, badly so!
• Many living things must be selected by their
  location

14
PLAUSIBLE SPATIAL SAMPLING FRAMES(Courtesy of
Tony Olsen, EMAP, US EPA)

• Use 6th Field HUCs as Spatial Units. 
• Select sample of HUCs incorporation landcover/use
  into probability of selection.  Then find
  beekeepers within HUC.  Sample locations where
  hives are set up.
• Same as Above, Except Ignore Beekeepers. 
• Go out an trap any bees at selected points within
  HUC - possibly use landcover again within HUC as
  selection probability.
• Use NRI Sample Points as Frame and Subsample
  Them.
• Use NASS Spatial Frame Sample Points and
  Subsample Them.
• Use NLCD (8million pixels). 
• Select GRTS sample of pixels based on landcover
  class.  Either trap bees or use that the identify
  if bee hives are present (in some way).  Have to
  do oversample if expect most pixels to not have
  hives.....

15
PLAUSIBLE SPATIAL SAMPLING FRAMES(Courtesy of
Tony Olsen, EMAP, US EPA)

• JARGON!!! - means what?
• HUC Hydrologic Unit Code
• NRI National Resources Inventory oriented
  toward soil erosion (Iowa State U)
• NASS National Agricultural Statistical Survey
• NLCD National Land Cover Data
• GRTS Generalized Randomized Tessellation
  Stratified
• VERY promising approach provides easy and
  defensible way to accommodate access denials, etc

16
PLAUSIBLE SPATIAL SAMPLING FRAMES(Courtesy of
Tony Olsen, EMAP, US EPA)Where to Find Info

• JARGON!!! - where to find out more about the
  content the jargon represents
• HUC http//water.usgs.gov/GIS/huc.html
• NRI http//www.nrcs.usda.gov/technical/NRI/
• NASS http//www.usda.gov/nass/
• NLCD http//www.epa.gov/mrlc/nlcd.html
• GRTS httporegonstate.edu/dept/statistics
  epa_program/docs/ spatial_balance_imperfect_fr
  ame.pdf

17
A PLAUSIBLE SPATIAL SAMPLING FRAMEHydrologic
Units

• Level 1 Two digit
• 21 major geographic areas, or regions
• Level 2 Four Digit
• divides the 21 regions into 222 subregions
• Level 3 Six Digit
• 352 hydrologic accounting units
• Level 4 Eight Digit
• There are 2150 Cataloging Units in the Nation

18
5. SELECTING THE SAMPLE SITES

• There are Lots of Ways to Select Collection Sites
  Depending On
• Objectives
• Sampling Frame
• Units chosen (hives or bees)
• Possible stratification factors

19
SELECTING THE SAMPLE SITEScontinued

• One Which Has Come Out of the EMAP Experience
• Generalized Randomized Tessellation Stratified
  (GRTS) Sampling
• It allows
• Spatially distributed sites
• Variable sampling rates depending factors of
  interest
• A well-defined means for adding sites to
  accommodate problems like access denial
• Implemented in several computational contexts
• Using GIS, or
• Statistical software

20
6. TRAINING

• Data Cannot Be Combined Across Areas, etc Unless
  It is Comparable Across Those Same Features
• IMPLICATION Good Training is Critical to Assure
  Consistent Procedures
• Various plausible contingencies must be
  identified in advance, and
• Plans made for how they should be dealt with

21
7. COLLECTING THE BEES

• Make Sure Field Crews Follow the Collection
  Protocols
• Be sure collection times dont collide with fair
  labor laws
• Does a federal employee need to be a member of
  each field crew?
• For safety purposes, crews may need to have at
  least two members
• Collect the Bees As Planned

22
8. HANDLING THE COLLECTED BEES

• Ship the Collected Material to the Appropriate
  Labs, According to Specified Protocols
• Need ice?
• Consider crew logistics, like
• housing, transportation, permits, location of
  shipping point, availability of shipping point by
  day of the week
• Plan for custody of the collected material

23
9. QUALITY ASSURANCE

• EPA has Stringent Quality Assurance (QA)
  Processes
• Approval of a QA plan may take several months
• Plan for that
• Implication Indicator(s) needs to be chosen
  early in the process

24
10. DATA MANAGEMENT

• This Will Be a Much Larger Effort Than You May
  Expect
• This has a QA component, too
• 20 30 of resources! Not 5!
• The collected information becomes part of the
  public record.
• You need to plan to make it available to various
  interested parties!

25
11. DATA ANALYSIS

• Plan Intended Summaries from the Beginning
• Record and keep track of all of the design
  information,
• Like the rate at which various kinds of sites
  were selected
• Consider making estimated cumulative distribution
  functions (cdf) a major part of the survey
  summary

26
STUDY CONTEXT FOR ILLUSTRATION OF CDFs
27
ESTIMATED CUMULATIVE DISTRIBUTION FUNCTION (cdf)
OF SECCHI DEPTH, EMAP AND DIP-IN

• Use cdfs tails often are of interest
• Confidence bounds
• Misinformation from convenience data

28
END OF PREPARED TALK

• QUESTIONS ARE WELCOME

29
BACK
30
HYDROLOGIC UNITS
BACK