Integrating Microsimulation, Mathematics, and Network Models Using ABM– prospects and issues
Upcoming SlideShare
Loading in...5
×
 

Like this? Share it with your network

Share

Integrating Microsimulation, Mathematics, and Network Models Using ABM – prospects and issues

on

  • 220 views

A talk at the workshio on "Microsimulationof chronoc disease: current methods & future directions", 27th Feb 2014, LSHTM, Tavistock place, London

A talk at the workshio on "Microsimulationof chronoc disease: current methods & future directions", 27th Feb 2014, LSHTM, Tavistock place, London

Statistics

Views

Total Views
220
Views on SlideShare
220
Embed Views
0

Actions

Likes
0
Downloads
5
Comments
0

0 Embeds 0

No embeds

Accessibility

Categories

Upload Details

Uploaded via as Microsoft PowerPoint

Usage Rights

CC Attribution-NonCommercial-ShareAlike LicenseCC Attribution-NonCommercial-ShareAlike LicenseCC Attribution-NonCommercial-ShareAlike License

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment

Integrating Microsimulation, Mathematics, and Network Models Using ABM – prospects and issues Presentation Transcript

  • 1. Integrating Microsimulation, Mathematics, and Network Models Using ABM – prospects and issues Bruce Edmonds Centre for Policy Modelling Manchester Metropolitan University Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 1
  • 2. The Modelling Background We use many kinds of model in the development and expression of knowledge, including: • data, equations, logic/rules, networks, NL descriptions, pictures and computer programs These capture what we observe, our ideas and how our ideas and observations relate at different levels of: • abstraction, granularity and generality They also vary according to their intended purpose or use to which we attempt to put them, including: • prediction, explanation, illustration, storage, description, communication, detecting patterns, understanding ideas, simplifying Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 2
  • 3. The “No Free Lunch” Theorems • These are a set of theorems from the field of Machine Learning (e.g. Wolpert 1996) that say: – There is no technique that will automatically succeed in prediction, search, pattern detection across all kinds of problem and kinds of data • That is, you have to choose the technique that works best for your goals, the nature of the data and the nature of what is being investigated • In other words, for good prediction etc. one has to apply knowledge about the situation to get better results out from any technique • Thus I will start with a review of pros and cons of the various techniques I am discussing Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 3
  • 4. Mathematical Equations • Represents complex relationships between a set of variables in a formal way • Is global to the “system” it is applied to (but that system can be at many degrees of granularity) • Holds out the possibility of general form solutions, but only if the equations are simple enough, otherwise using (numerical) simulation • Is good at representing dynamics over time • Is poor at distributed systems requiring hundreds of separate but linked equations (since this effectively reduces one to simulation anyway) • Tends to be theory-driven and global Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 4
  • 5. MicroSimulation Models (MSM) In the most abstract terms: – Divides the data into chunks (e.g. geographically) – Then applies a model to each chunk (maybe fuzzily) – Aggregates or displays the results from all chunks Thus, in practice, tends to: – Have a great many implicit free parameters, hence can flexibly fit a broad variety of patterns – Be more data-driven than theory-driven – Fits patterns local to the chunks, thus can be contextsensitive (relative to the way the data has been divided) Can be seen as a kind of data-mining technique that uses knowledge in terms of how to segment the data and what models are applied to each segment Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 5
  • 6. Social Network Models (SNM) These use a particular kind of abstraction step: – The representation of interactions between agents as a link between them • • • • • • This can be data-driven, but always given the assumptions implicit in the abstraction to links and the assumptions in their analysis Capture elements of structure well Links are essentially static (each link representing a series of interaction over a period) Lots of mathematical results, but these difficult to know if these are applicable to any particular network Are very hard to validate, but are suggestive Tend to be explanatory rather than predictive Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 6
  • 7. Agent-Based Models Divides the system up into parts, then represents the interactions between these parts in terms of messages between parts of a computer simulation • Bridges the micro- and macro-levels • Good at revealing complex dynamics in systems • Is very flexible in terms of structure and rules, in particular in terms of heterogeneity and context-specificity • Can be very abstract and divorced from data… • …but can also be very complex and specific to particular sets of evidence • Needs a lot of data to validate well • Are always somewhat theory driven, but the “theory” can be mundane and informed by evidence • Tend not to be predictive in any narrow sense, but can be useful for an informative but possibilistic “risk analysis” Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 7
  • 8. What happens in ABSS • Entities in simulation are decided on • Behavioural Rules for each agent specified (e.g. sets of rules like: if this has happened locally then do this) • Repeatedly evaluated in parallel to see what happens • Outcomes are aggregated, inspected, graphed, pictured, measured and interpreted in different ways Specification (incl. rules) Representations of Outcomes Simulation Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 8
  • 9. Some modelling trade-offs Use of existing knowledge Network Models MSM ABM Contextspecificity Global Statistical Models Macro predictive goal Abstract Mathematics Capturing complex dynamics Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 9
  • 10. ABM as a tool for integration ABM can relate to a broad range of evidence, e.g.: • Macro-level quantitative statistics • Distributions and tendencies in dynamics • Qualitative evidence or expert knowledge to inform micro-level rules • Aggregate behavior and stats at all levels of aggregation including local and meso-levels • Network data either as an input or as an abstraction of the interactions coming out of it The disadvantage is that it is so flexible, there are many ways to simulate any system, a lot of choice The advantage of this is that this can all be explicit Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 10
  • 11. Staging Abstraction (in SCID) SNA Model Analytic Model Abstract Simulation Model 1 Abstract Simulation Model 2 Data-Integration Simulation Model Micro-Evidence Macro-Data Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 11
  • 12. Chains/Clusters of Model “Chains” or “Clusters” of model allow one to combine the need for different goals, e.g.: – – – – relevance and rigour prediction and explanation connection to data and what-if analyses context-specificity and global outcomes However this is at a cost of a plurality of models, which involves more input in terms of: development, maintenance and checking… …especially in the relationship between models But can help: – – stage abstraction more carefully maintain meaningful reference of model components Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 12
  • 13. Examples of “Causal Stories” Initial party preference inherited – party preference can be linked to learning from parents. People vote out of habit – going to the polls in one election will lead to a greater likelihood of returning to the polls in a subsequent election. People vote because they care about who wins - voters are more likely to turnout if they have a stronger preference for one party or another. Voting is a social norm – civic duty is an important rationale for individual-level turnout. People share the political views of their greater networks – probability of agreement within a network depends on the distribution of political opinion within one‟s network (autoregressive networks). Electors can be mobilised to vote by family, friends and political parties – household members, friends and political parties will ask people to vote on election day. Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 13
  • 14. Overall Structure of SCID Voter Model Demographics of people in households Social network formation and maintenance (homophily) Influence via social networks • Political discussions Output Input Underlying data about population composition Voting Behaviour Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 14
  • 15. Changing personal networks over which social influence occurs A Household Class Activities Age Etc. Ethnicity Level-of-Political-Interest Composed of households of individuals initialised from detailed survey data Each agent has a rich variety of individual (heterogeneous) characteristics Memory Discuss-politics-with person-23 blue expert=false neighbour-network year=10 month=3 Lots-family-discussions year=10 month=2 Etc. Including a (fallible) memory of events and influences An Agent’s Memory of Events Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 15
  • 16. Example Quantitative Output Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 16
  • 17. Simulated Social Network at 1950 Majority: longstanding ethnicities Newer immigrants Established immigrants: Irish, WWII Polish etc. Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 17
  • 18. Simulated Social Network at 2010 Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 18
  • 19. How to integrate MicroSimulation I To condition the context-specific rules of an ABM, i.e. an input to it, staging the abstraction from data • One could cluster/segment the data according to the different strategies that actors use • Then use MSM to estimate the context-specific strengths of interactions/behaviours, e.g.: – In different communities/localities – In different classes or economic circumstances • This would allow the ABM to be better grounded in the data, not only in terms of local initialisation but also in the varying strategies of agents Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 19
  • 20. How to integrate MicroSimulation II To use a MSM along side an ABM, both models simulating the same phenomena, using the same basic segmentation of the system. • The MSM: – Being more data driven – Providing „surprise free‟ but numerical predictions • The ABM: – Adding in more interaction – Applying other features and constraints based on domain knowledge – Providing possibilistic, „what if‟ risk analyses covering some of the possible structural changes Both models could be validated against each other as well as separately against their data and outcomes Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 20
  • 21. How to integrate MicroSimulation III Interlace ABM and MSM techniques together in the same model. • This is a little hard, due to the fundamentally different natures of the two approaches (interactive vs. independent, data-driven vs. theory driven, predictive vs. explanatory etc.) But is possible in some cases, e.g.: – Some aspects of the environment of agents being determined by Microsimulation – „Fitting‟ an ABM to each data segment, allowing a weaker interaction between segments – Movement (or other action) of agents, changing the basis of the MicroSimulation analysis Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 21
  • 22. Conclusions • Integrating a variety of techniques is possible, and ABM often provide a flexible way of doing this • A shift to „packages‟ of models where the properties of each model is understood and with a clear purpose • Rather than trying to use a single model for many different purposes • I argue this is inevitable to make progress with complex phenomena (Edmonds 2013) • MSM, ABM and SNM allow for an inclusion of context-specific/local behaviours compared to analytic mathematical models (in practice) Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 22
  • 23. Thanks! Bruce Edmonds http://bruce.edmonds.name Centre for Policy Modelling http://cfpm.org I will (soon after) make these slides available at: http://www.slideshare.net/BruceEdmonds Integrating Microsimulation, Mathematics, Network Models Using ABM, Bruce Edmonds, Microsimulation of chronic disease, London, 27th Feb 201. slide 23