Judge Business School Sociophysics Simulations of Technology Adoption and Consumer Behavior WJ Nuttall and Tao Zhang University of Cambridge, Cambridge CB2 1AG, UK D Hamilton University of Glasgow, Glasgow, G12 8QQ, UK F. Roques CNRS-CIRED, Nogent, France Authors copyright unless otherwise attributed (all rights reserved)
Sociophysics? Builds upon techniques and models from theoretical condensed matter physics: Ising Model Potts Model 2D-XY Model etc. Then applied to natural phenomena E.g. Propagation of Forest Fires (Per Bak and colleagues circa 1990) Image Mike Creutz, BNL http://www.cmth.bnl.gov/~maslov/soc.htm
Sociophysics More recently applied to social systems - sociophysics Arguably economist Thomas Schelling got there first in 1971 with his dynamic models of segregation Image D Vinkovic & A Kirman, source IAS Princeton press release: http://vinkovic.org/projects/schelling/
Consumer Technology Adoption in Electricity Two projects reported here. First: Hamilton, Nuttall and Roques have modelled nearest neighbour interactions in a system close to physics Second: Zhang and Nuttall have modelled a small world system aiming for greater social realism
Hamilton, Nuttall and Roques Simulations Phenomenological analysis of a model where thousands of electricity consumers agents governed by very simple rules continually reassess their choice of electricity source. Choice between base-load grid supply, solar power or micro-combined Heat and Power (CHP). Crucially, this is a spatial discrete-event simulation, with consumer s neighbours influencing the decision making process.
Electricity Consumer Agents Two types of consumer agents are randomly populated throughout a city : residential and business consumers. All begin the simulation connected to the grid supply. Residential agents are influenced by their nearest neighbours; business agents are not.
Consumer Decision Rules Each consumer can switch supply every three months, based upon their satisfaction with the grid characterised by a very simple utility function. Heterogeneity of the consumer agents, in terms of thresholds for change, is implemented through standard Monte Carlo sampling methods. Change to solar power becomes more likely if nearest neighbours have solar power (but not CHP).
One Example Response Scenario A (Stable) Solar relative attractiveness weak High proportion of business consumers Stable equilibrium reached after only two years. Not particularly interesting.
Another Example Response Scenario B (Asymptotic) Neighbour effects dominate: No. Solar >> No. CHP. Continuous adoption of new technologies recreates standard S-curve behaviour. Stable equilibrium reached after five years. Very different from previous scenario.
S-curves? This is one of many examples where agent based simulation reproduces the ubiquitous Schumpeterian S-curve. We believe it is nothing more than a consequence of the shape of the probabilistic distributions underpinning the decision rules. For an introduction to S-curves in technology adoption see Everett Rogers, Diffusion of Innovations, Fourth Edition, Free Press/Simon and Schuster (1995)
Scenario C (Near-critical) Transient behaviour characterised by long periods of quasi-stability and sharp discontinuities. More than 10 years to reach equilibrium. Volatile dynamic effects are not caused by changes in model parameters Very difficult to extrapolate from any intermediate stage.
Comparison of System Evolution
Which Parameters Govern the System Behaviour? Final-state equilibrium is extremely sensitive to relative attractiveness of solar PV and CHP and the ratio of business to residential consumers. All other model parameters do not have a significant influence on equilibrium state. Use the percentage of consumers remaining on the grid at equilibrium as an equilibrium order parameter
Variation of Key Parameters towards a phase diagram? Phase diagram showing the percentage of consumers remaining on the grid at equilibrium as a function of both the relative attractiveness (A initial ) and the ratio of business to residential consumers (R).
Zhang and Nuttall s Simulations Consider the roll-out of a new electricity metering technology smart meters As previously, householder agents occupy a square lattice for a discrete-event simulation, with consumer s neighbours influencing the decision making process in a small world model Detail from a 250x250 grid with PBCs We aim to structure the agents decision rules in a simple, but more theoretically rigorous way
Icek Ajzen s Theory of Planned Behaviour 1991 Backgroud Factors Individual Personality Mood, emotion Intelligence Values, stereotypes General attitudes Experience Social Education Age, gender Income Religion Race, ethnicity Culture Information Knowledge Media Intervention Behavioral beliefs Normative beliefs Control beliefs Attitude toward the behavior Subjective norm Perceived behavioral control Intention Actual behavioral control Behavior
Promoting Smart Meter Deployment Initial Observations Monopoly Supplier Mar ket Shar e 90 80 70 60 50 40 30 20 10 0 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 Ti me St ep Random Deployment - Ent husi asm On Random Deployment - Ent husi asm Off Cent r al i zed Depl oyment - Ent husi asm Off
Towards a more realistic electricity supply market who should pay? Government financed competitive rollout Government financed monopoly rollout Electricity supply company financed competitive rollout Distribution company financed monopoly rollout Six private supply companies separated from distribution
Conclusions Spatial agent based simulations of residential electricity consumers can offer a range of insights for energy policy makers Complex effects (e.g. volatile dynamics) can occur Equilibrium states are sensitive to some few key initial conditions and these may even be parameterised as a phase diagram. Modelling can incorporate agent decision rules grounded in the social sciences. Agent based simulations can suggest preferred public policies for new technology deployment.
Acknowledgements We are most grateful to the ESRC Electricity Policy Research Group for financial assistance and for providing anonymous peer review for some of the work presented here. We are also grateful to Tobias Galla, Philip Ball and the committee of the Institute of Physics Nonlinear and Complex Physics Group for advice and assistance.