Using e-Research Tools for Micro-Level Simulation


Published on

A showcase of some of the microsimulation work that is going on as part of the NeISS project.

Published in: Technology
  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide
  • Computationally expensive - Single model run reveals a theorem, but no information about robustness (Axtell, 2000) - Sensitivity analysis and many runs required - NGS can helpSmall errors can be replicated in many agents - problem with scaleablility, 1000s of agents might have small error - small error in decision making or agent interaction replicated“Methodological individualism” (O’Sullivan and Haklay, 2000) - one-way notion of emergence. - agents can influence society which emerges from their interactions but are not themselves influenced by society. - impossible for society to make a conscious effort to affect individualsModelling “soft” human factors: - complex psychology, seemingly irrational behaviour - very difficult to account for in a computer program
  • Recent shift in criminology from motivation of offender to social / environmental contexts in which crimes occur. - instead of focusing on underlying reasons that offender has propensity to commit crime, focus on social / environmental factors that make crime attractive - routine activities theory: convergence of victim + offender – capable guardian - rational choice perspective: offenders make a cost/benefit analysis of surroundings - crime pattern theory: which opportunities an offender is aware ofTo model modern opportunities effectively, must account for complex, micro-interactions of offenders and their victims / environment – essential to determine whether or not a crime will occur. - e.g. environment: visibility of property to neighbours / passers-by, burglar alarm, possible rear entry - e.g. offenders: feel “safe” in area, know about property (awareness space)Other models: - Groff: street-level models of robbery. Interesting results but simple offender/victim behaviour - Dan Birks(?) - Mastermind: agent-based model, aimed at collaborative crime theory development
  • The analysis illustrates the impact of measures to cut government expenditure within the current austerity programme.  The specific policy changes to be considered are targeted towards reduction in housing benefit. Figure 1 (bottom left) shows the current distribution of housing benefit in Leeds.  Displays a strong concentration in the deprived areas towards the centre of the city.  Data is provided by Leeds City Council for more than 80,000 benefits claimants over a four year period,Figure 2 (bottom right) shows the spatial effect of four major changes in the rules on housing benefit; i. reduce rents from 50th centile of market prices to 30th centile (yellow); ii. clawback savings from people living in cheap rented accommodation (blue); iii. reduced entitlement for young people in family households (increased room sharing) (red); iv. reduced entitlement for  large families (green)Figure 3 is similar to Figure 2 but shows total loss by small area indicating that worst affected areas are the most deprived (which is not entirely obvious intuitively, as many family areas and high rental areas at periphery)Figure 4 is exploring the effects on the providers of accommodation because the market can no longer sustain the higher prices (using a simple ‘agent’ model of landlord behaviour).  Second order effect is therefore a transfer of housing away from the rented sector to other uses, and therefore increased homelessness, squatting and other social problems
  • The slide illustrates our recent work to connect social simulation, crowd-sourced behavioural data and policy analysis.Data is captured from the public using web-based surveys (CASA’s MapTube or SurveyMapper technology). In this example, residents in the North-West were asked to state their response to a new congestion charge in the centre of Manchester. The survey was promoted through BBC North-West and attracted more than 15,000 responses.The spatial distribution of responses is shown in the second illustration. Although views were expressed from as far away as Aberdeen and Cornwall, most respondents are concentrated in the Greater Manchester area!We designed a simulation model which predicts transport behaviour in relation to the distribution of homes; the financial and non-monetary costs of trip-making; the socio-economic status of residents, and the ‘attractiveness’ of workplaces and retail outlets. An instance of this model can be executed in a few seconds on a serial machine, so that any given parameter set has an associated behaviour profile. In order to match model behaviour with the stated behaviour of the survey population, we executed many thousands of simulations, using a genetic algorithm to optimise the fit between the model and the survey. This procedure was implemented via the National Grid Service. The chart shows progress improvement in the goodness-of-fit of the model across 100 iterations (in each of which the model is executed 100 times).From the resulting model, we can evaluate the impact of various alternative transport policy scenarios on the traffic patterns of the city of Manchester.This work is currently in press: Birkin M, Malleson N, Hudson-Smith A, Gray S, Milton R (2011) Calibration of a Spatial Interaction Model with Volunteered Geographical Information, International Journal of Geographical Information Science, xxx.
  • Figure 1 (top left) illustrates the processes in the dynamic model.  Figure 2 (top right) shows some trends from this model for Leeds.   One of the major difficulties is modelling highly transient populations of students and young people, shown in the blue areas of figure 3 (bottom left).  To address this we adopt an agent-based representation.  In this way we can also represent morbidity histories of individuals – this allows us to simulate the impact of factors such as family influences and previous residences (e.g. Person moving from the coalfields to the coast does not suddenly absorb the life expectancy profiles of a new location).
  • The potential for applications in domains such as economics and business studies can be illustrated by spatial pricing models. In Figure 3.1, an agent-based model of the behaviour of petrol station managers was constructed as a basis for understanding petrol price variations within and between urban areas (Heppenstall et al, 2006). The model is able to capture micro-level variations between different types of sites, as shown in Figure 3.2. This work has also been extended to include other effects such as network planning and competitive bidding for land (Birkin and Heppenstall, 2011). As in all of these examples, our concern in this proposal is to develop these applications by engaging with disciplinary partners about ‘their’ data and ‘their’ problems and to move towards instances of our social simulation technology which are most appropriate to these contexts.
  • Using e-Research Tools for Micro-Level Simulation

    1. 1. School of Geography<br />FACULTY OF ENVIRONMENT<br />Using e-Research Tools for Micro-Level Social Simulation<br />Nick Malleson<br />University of Leeds<br />
    2. 2. Outline<br />Introduction to “micro-level” social simulation<br /><ul><li>Complexity / emergence
    3. 3. Microsimulation
    4. 4. Agent-based modelling</li></ul>Population reconstruction and projection<br />The NeISS project<br />Crime simulation<br />Other work<br />Conclusion<br />
    5. 5. Prediction: Emergence<br />The Need for Individual-Level Models (ILMs)<br />Individual-level models provide a natural description of the system<br />Emergence<br />Patterns at one level arising from lower level effects<br />Simple rules -> complex patterns<br />Not intended by the individuals<br />Examples from human systems?<br />
    6. 6. Individual Level Models (ILMs)<br />Model from the “bottom-up”<br /><ul><li>“Natural description” of the system under study</li></ul>Range of ILM methodologies<br />Microsimulation<br /><ul><li>Rule based systems
    7. 7. Retain diversity / uniqueness</li></ul>Agent-Based Models (ABMs)<br /><ul><li>“Agents” are unique entities in the model – capable of self control and decision making
    8. 8. Strong emphasis on individual behaviour / psychology</li></li></ul><li>Data Disaggregation<br />Can also use microsimulation to disaggregate data<br />For each area, heuristically find the optimal combination of individuals that match the aggregate demographics<br />These data are often used as input to other individual-level models<br />Example: disaggregating the census<br />Synthetic individuals<br />Aggregate (census table)<br />
    9. 9. Agent-based Modelling (ABM)<br />Autonomous, interacting agents<br />Focus on behaviour, psychology, autonomy<br />Represent individuals or groups<br />
    10. 10. Advantages of Individual-Level Models<br />More “natural” for social systems than statistical approaches<br />Dynamic history of system<br />Can include physical space / social processes in models<br />Designed at abstract level: easy to change scale<br />Bridge between verbal theories and mathematical models<br />
    11. 11. Disadvantages of Individual-Level Models<br />Computationally expensive<br /><ul><li>Lots of agents, lots of iterations, lots of computation!
    12. 12. Models often stochastic so single model run reveals a theorem, but no information about robustness
    13. 13. Sensitivity analysis and many runs required
    14. 14. (High performance computing – e.g. National Grid Service – can help)</li></ul>Small errors can be replicated in many agents<br />“Methodological individualism”<br />Modelling “soft” human factors<br />
    15. 15. The National e-Infrastructure for Social Simulation (NeISS)<br />Census Data Service<br /><ul><li>Extract required census data for a given area (e.g. Leeds)</li></ul>Population Reconstruction Model (PRM)<br /><ul><li>Disaggregate the census to create a simulated population of individuals</li></ul>Dynamic Population Projection Model<br /><ul><li>Simulate the population change over time (e.g. Leeds 2010)</li></ul>Mapper service<br /><ul><li>Use MapTube to map results (CASA)</li></ul>Simulations<br /><ul><li>Transport</li></li></ul><li>Practical Application:Crime Simulation<br />Shift in criminology towards “opportunity theories”<br />Focus on social / environmental factors surrounding an individual crime<br />Routine activities theory, rational choice perspective, crime pattern theory<br />Complex micro-level interactions of individuals and environment<br /><ul><li>Individual houses: visibility of properties, burglar alarm, back door etc
    16. 16. Individual burglars: feel “safe” in neighbourhood?, aware of opportunity?, drug addiction?, “professional” or “opportunist”?</li></ul>Problems with aggregate models:<br /><ul><li>Spatio/temporal aggregation
    17. 17. Fail to capture dynamics of crime system</li></ul>Solution: ABM<br />
    18. 18. An Agent-Based Crime Model<br />Major components:<br />Virtual Burglar Agents<br />Individual level<br />Physical Virtual Environment<br />Houses, roads, bars, busses etc<br />Individual level<br /><ul><li>Social virtual environment</li></ul>“Communities”<br />Aggregate level<br />
    19. 19. ‘Victims’ in the Model<br />Houses are modelled as individual objects<br />But people (non-burglars) are estimated at Output Area level<br />Also, census data to build victims is out of date<br />Will use the NeISS infrastructure to combine synthetic individuals with the burglary model<br />
    20. 20. Policy Implications<br />Work closely with Safer Leeds (crime and disorder reduction partnership)<br /><ul><li>Involved in model development
    21. 21. Provide data</li></ul>Burglary simulation (via NeISS) part of a “crime reduction toolkit”<br /><ul><li>Predict effects of urban regeneration / crime reduction schemes</li></li></ul><li>Housing<br /> 4<br /> 3<br /> 1<br /> 2<br />Jordan, R., Birkin, M., Evans, A. (2011): ‘Agent-based Simulation Modelling of Housing Choice and Urban Regeneration Policy’. In: Bosse, T.,  Geller, A. and Jonker, C. (eds.), Multi-Agent-Based Simulation XI.  Springer, Berlin, 152-166.<br />
    22. 22. Crowd-Sourced Traffic Simulation<br /> 3<br /> 2<br /> 4<br /> 1<br />Birkin M, Malleson N, Hudson-Smith A, Gray S, Milton R (2011) Calibration of a Spatial Interaction Model with Volunteered Geographical Information, International Journal of Geographical Information Science, forthcoming.<br />
    23. 23. Demographics<br /> 1<br /> 2<br />3<br />Wu, B. M., Birkin, M. H. and Rees, P. H. (2011) A dynamic microsimulation model with agent elements for spatial demographic forecasting, Social Science Computing Review, 29(1), 145-160.<br />Wu B, Birkin M, Rees P (2008) A spatial microsimulation model with student agents, Computers Environment and Urban Systems, 32, 440-453.<br />
    24. 24. Retail<br />Heppenstall A, Evans A, Birkin M (2007) Testing and optimisation of a hybrid multi-agent model for the retail petrol market, Environment and Planning B, 34, 1051-1070.<br />Heppenstall A, Evans A, Birkin M (2006) Application of multi-agent systems to modelling a dynamic, locally interacting retail market, Journal of Artificial Societies and Social Simulation, 9, 3. <br />
    25. 25. Conclusion<br />Need for more individual-level social models!<br />(Previously) many barriers:<br /><ul><li>Data
    26. 26. Methods
    27. 27. Computation</li></ul>e-Research tools are supporting the social simulation lifecycle<br />
    28. 28. Links<br />Video introductions to NeISS service:<br /><ul><li></li></ul>NeISS demo service:<br /><ul><li></li></ul>RepastCity project<br /><ul><li></li>