RTI ppt template

1
Creating a Synthesized U.S. Agent Database for
Agent-Based Modeling ISDS Conference October 12,
2007 Bill Wheaton
RTI International is a trade name of Research
Triangle Institute
3040 Cornwallis Road P.O. Box 12194
Research Triangle Park, North Carolina, USA
27709
Phone [phone]
e-mail wdw_at_rti.org
Fax [phone]
2
Acknowledgments

• This work funded under the Models of Infectious
  Disease Agent Study (MIDAS) for the National
  Institute of General Medical Sciences (NIGMS)
• RTI wishes to thank Irene Eckstrand and the MIDAS
  Steering Committee for funding and support.
• Prior Research and Techniques
• Beckman, Richard J., Baggerly, Keith A., McKay,
  Michael D., Creating Synthetic Baseline
  Populations, Transportation Research, Vol. 30,
  No. 6, pp. 415-429, 1996.
• Norman, Paul, Putting Iterative Proportional
  Fitting on the Researchers Desk. Working Paper
  99/03, School of Geography, University of Leeds.
• TranSims Transportation Analysis Simulation
  System.
• http//transims.tsasa.lanl.gov

3
Microsimulation/Agent-based Models

• Microsimulation methodologies aim at building
  large-scale data sets on the attributes of
  individuals or householdsand at analyzing policy
  impacts on these micro-units through the
  simulation of economic, demographic and social
  processes.
• If we do not have a micro data base on
  individuals and households then there is a
  necessity to simulate one
• -- Ballas, D., Clarke, G., Turton, I. Exploring
  Microsimulation Methodologies for the Estimation
  of Household Attributes, paper presented at the
  4th International conference on GeoComputation,
  Mary Washington College, VA., 25-28 July 1999.
• Thus the Idea Produce a national,
  geospatially-explicit synthetic population for
  the United States.

4
Micro (Individual) vs. Macro (Aggregate) Data

• Macro/Aggregate Data
• Census counts by geographic area
• State, County, Census Tract, Block Group
• Does not provide information on household
  structure
• Micro/Individual Data
• Individual or Household-level data
• Household structure maintained

5
Creating a Synthetic Population Data Inputs and
Techniques

• Block-group Level Demographics
• SF3 (2000 decennial census)
• Public Use Microdata (PUMS)
• Actual Census long-form records (from U.S. Bureau
  of the Census, 2000)
• Household and individual level data
• Family structure maintained
• 5 Sample within Public Use Microdata Areas
  (PUMAs)
• PUMAs contain about 100,000 persons
• Household Locations
• Randomly generated w/in block groups
• Iterative Proportional Fitting (IPF)
• Uses conditional probabilities to fill out a
  synthetic population that matches SF3 counts
  based on PUMS microdata samples.

6
Geographical Context

• Counties
• Census Tracts
• Block Groups
• Public Use Microdata Areas (PUMAs)
• Households
• Clone particular records of the 5 PUMS sample
  (red outlines) to match census counts at block
  group level (black outlines)

7
Transims Population Generator

• Transims A transportation modeling package
  developed at Los Alamos National Lab
• Became the basis for EpiSims infectious disease
  modeling software
• Included development of code that uses IPF to
  generate a synthetic population
• Details in Beckman, Richard J., Creating
  Synthetic Baseline Populations, Transportation
  Research, Vol 30, No.6, pp 415-429, 1996

8
IPF Attributes for MIDAS

• Works on HOUSEHOLD Attributes
• The MIDAS Synthetic Population Uses
• Persons
• Population lt 18
• Workers in Family
• Vehicles Available
• Household Income
• Other household attributes could be used in
  future

9
Example Household and Persons
Randomly Selected Synthetic Household
Household Attributes
Persons
10
Results

• Households
• 105,480,101 generated vs. 104,926,825 in
  census
• X,Y locations
• Household attributes
• Persons
• 273,624,650 generated vs. 281,421,906 (Group
  Quarters persons subsequently added)
• Individual attributes (age, sex, etc.)
• Family Structures Maintained
• Closely Matches Census Counts

11
Schools, Workplace Assignments

• Assign school-aged children to schools
• We have locations of schools by grade and
  capacity for U.S.
• Developed method of assigning school-aged
  children in synthetic population to schools
• Assignments are generated for particular
  geographic area when needed
• Workers assigned to workplaces based on STP64
  commuting patterns

12
Group Quarters

• Persons in Group Quarters accounted for 2.8 of
  U.S. population in 2000
• Group Quarters
• Institutional
• Non-Institutional
• For synthesized agent database, created Group
  Quarters locations (nursing homes, prisons,
  military bases) and synthesized agents to occupy
  them

13
Limitations Gaps

• Ethnicity, Race and Other Personal
  Characteristics Not Used
• Group Quarters
• In the works, currently not included
• Spatial Locations Could be Enhanced
• Geographic Limitations
• Counties, Tracts, and BGs with few people are
  less accurate

14
Potential Usefulness to Syndromic Surveillance

• An aid in modeling/predicting effects of
  epidemics identified by syndromic surveillance
• Assign synthesized persons an infected code
  based on demographic characteristics and
  geographic location of actual patients (seeds)
• Use those agents as the seeds in agent-based
  modeling
• Run different mitigation scenarios through the
  model to predict different outcomes and to inform
  policy decisions

Seeds
15
Acknowledgments

• Production Team Bernadette Chasteen, Justine
  Allpress, Michael Bacon, Jamie Cajka
• Support Team Doug Roberts, Gana, Diglio Simoni,
  Phil Cooley, Diane Wagener

16
Conclusion

• Thank You for Attending!