210 likes | 385 Views
School of Geography FACULTY OF ENVIRONMENT. Modelling Individual Consumer Behaviour. Kirk Harland Alison Heppenstall. Email: a.j.heppenstall@leeds.ac.uk ; k.harland98@leeds.ac.uk 6 th May 2009. School of Geography FACULTY OF ENVIRONMENT. Presentation Motivation Project overview
E N D
School of Geography FACULTY OF ENVIRONMENT Modelling Individual Consumer Behaviour Kirk Harland Alison Heppenstall Email: a.j.heppenstall@leeds.ac.uk; k.harland98@leeds.ac.uk 6th May 2009
School of Geography FACULTY OF ENVIRONMENT Presentation • Motivation • Project overview • Population generation • Flexible modelling framework • Future Research - Adding behaviour
School of Geography FACULTY OF ENVIRONMENT • Motivation • Retail models: mature but limited • aggregate populations • limited shopping behaviours (internet, multi-purpose) • increased data and computational power • Recent work: ABM of retail market (demand side) • But: • consumers still aggregated • transport network simplistic (Heppenstall, et. al 2005, 2006, 2009)
School of Geography FACULTY OF ENVIRONMENT ESRC First Time Grant Oct 2008 – Oct 2010 Assess different methods of generating population Create population using commercial marketing survey dataset Generate agent rule-sets Replace aggregate demand-side of model with ‘individual’ agents Optional extra… incorporate realistic road network for agents to traverse
School of Geography FACULTY OF ENVIRONMENT Model Overview
School of Geography FACULTY OF ENVIRONMENT • Phase 1: Population Generation • Require a method of generating robust populations tailored to specific issues flexibly and quickly. • Different population required for a health study to that for a retail market etc… • Compared three synthesis methods to create a population and then assessed the results using a variety of statistics: • Deterministic Reweighting (Smith et al. 2009) • Conditional Probabilities (Birkin & Clarke 1988;1989) • Combinatorial Optimisation – Simulated Annealing (Press et al. 1992)
School of Geography FACULTY OF ENVIRONMENT • Population Generation – Deterministic Reweighting • Deterministic reweighting operates in two stages • Calculates a weight representing the likelihood that each individual record from the sample population appears in current zone • The weights are proportionally fitted to the known population • Very quick to execute • Tailor model using the order of constraints to incorporate geographical inconsistencies (Economically active person in city centre less likely to own a car than in suburbs?) • Determining constraint order requires more pre-run model setup time
School of Geography FACULTY OF ENVIRONMENT • Population Generation – Conditional Probabilities • Calculates the probability of a person or household appearing in a geographical area with respect to the accumulative probability distribution derived from the constraints. • Each probability is conditional on the results from the preceding constraint, hence the name conditional probabilities… • This algorithm is slower than the deterministic reweighting approach • Is not dependent on having a sample population to select from • If using a sample population and not all combinations of people available this method can have problems selecting individuals
School of Geography FACULTY OF ENVIRONMENT Population Generation – Simulated Annealing • Executes in two loops using random selection to generate and improve the created population • Changes are accepted or rejected using the Metropolis algorithm (avoids problems of hill climbing effects) • Huge number of calculations are required (creating a population for Leeds required almost 250,000 iterations of the outer loop) • Application of indexing strategy speeds up model execution • Little pre-model run setup required
School of Geography FACULTY OF ENVIRONMENT • Population Generation – Data • Experiment data consists of 715,402 individuals residing in the Leeds area. • Extract from the Sample of Annonymised Records, Small Area Microdata file is used as the sample population • 2001 Census of Population data used for constraints (univariate constraints applied)
School of Geography FACULTY OF ENVIRONMENT • Population Generation – Data • 2001 Census of Population data used for evaluation • All data cleaned to limit the impact of the disclosure control measure applied in the 2001 Census, Small Cell Adjustment Method (SCAM) (Stillwell, 2007) (both cross tabulated and univariate)
School of Geography FACULTY OF ENVIRONMENT • Population Generation – Initial Results • Voas and Williamson (2000) state that constraint tables should be well matched by all methods, however, only the simulated annealing methods produces a ‘perfect fit’
School of Geography FACULTY OF ENVIRONMENT • Population Generation – Initial Results • Even though only univariate constraints were applied for all simulations the inter-relationships between the different constraint attributes are well captured
School of Geography FACULTY OF ENVIRONMENT • Population Generation – Initial Results • As expected, the fit of attributes outside of the variables used to constrain the model are less robust.
School of Geography FACULTY OF ENVIRONMENT • Population Generation – Increasing Constraints • Number of constraints increased from 6 to 15, including provision of care, tenure, health and economic activity • Simulated Annealing algorithm used to create population
School of Geography FACULTY OF ENVIRONMENT Model Structure – Flexible Modelling Framework
School of Geography FACULTY OF ENVIRONMENT Model Structure – Flexible Modelling Framework
School of Geography FACULTY OF ENVIRONMENT Model Structure – Flexible Modelling Framework
School of Geography FACULTY OF ENVIRONMENT • Human Behaviour • Consumer behaviour BUT only important aspects • Beliefs Desires Intentions (BDI) • Practical reasoning: decide goals and how to achieve them • Beliefs: knowledge about the world • Desires: all goals the agent is trying to achieve • Intentions: most important goals • Unrealistic - creates completely rational agents • PECS (Schmidt, 2000; Urban 2000) • Physical Condition • Emotional States • Cognitive Capabilities • Social Status
School of Geography FACULTY OF ENVIRONMENT • PECS behavioural model http://crimesim.blogspot.com Malleson, N.S., Heppenstall, A.J., and See, L.M., Simulating Burglary with an Agent-Based Model. Computers, Environment and Urban Systems
School of Geography FACULTY OF ENVIRONMENT References Birkin, M., and Clarke, M. 1988, SYNTHESIS-a synthetic spatial information system for urban and regional analysis: methods and examples, Environment and Planning A, 20, pp 1645-1671 Birkin, M., and Clarke, M. 1989, The Generation of Individual and Household Incomes at the Small Area Level using Synthesis, Regional Studies, 23:6, pp 535-548 Heppenstall, A.J., Evans, A.J. and Birkin, M.H. (2005), A Hybrid Multi-Agent/Spatial Interaction Model System for Petrol Price Setting. Transactions in GIS 9(1): 35 - 51. Heppenstall, A.J., Evans, A.J. and Birkin, M.H., (2006) Application of Multi-Agent Systems to Modelling a Dynamic, Locally Interacting Retail Market. JASSS. vol 9(3). Heppenstall, A.J., Harland, K. and Ross, A.N. Application of an Agent-Based Model for Simulating Spatial Dynamics within a Retail Petrol Market. Journal of Business Research. Press, W. H., Teukolsky, S. A., Vetterling, W. T., and Flannery, B. P. 1992, Numerical Recipes in C: The Art of Scientific Computing. 2nd Edition. Cambridge University Press. Cambridge, England Schmidt, B. (2000). The Modelling of Human Behaviour. SCS Publications, Erlangen, Germany. Smith D., Clarke G., and Harland K. (2009) Improving the synthetic data generation process in spatial microsimulation models. Environment and Planning A. 41. pp 1251-1268 Stillwell, J. C. H., Duke-Williams, O. (2007) Understanding the 2001 UK census migration and commuting data: the effect of small cell adjustment and problems of comparison with 1991, Journal of the Royal Statistical Society 170: pp 1–21. Urban, C, (2000). PECS: A reference model for the simulation of multi-agent systems. In Ramzi Suleiman, Klaus G. Troitzsch, and Nigel Gilbert, editors, Tools and Techniques for Social Science Simulation, chapter 6, pages 83–114. Physica-Verlag.