agent-based modeling in marketing– modeling of online auctions (wolfgang jenk) – dynamic web...
TRANSCRIPT
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Agent-Based Modeling in Marketing
Roland T. Rust and William RandCenter for Complexity in Business
Robert H. Smith School of BusinessUniversity of Maryland
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Computational Methods to Model Complexity
Computationally Intensive methods
Non-intensive methods
Old New
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
The Center for Complexity in Business
• Decreasing cost of computation• Complexity of business phenomena• Rising literature on complexity in
business• Chris Dellarocas & Bill Rand• Goal: the focal point for complexity
research in business
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Agents in Marketing
• Consumers• Companies• Governments• Individual managers
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Issues with Acceptance of Agent-Based Models
• Not traditional methodology• No critical mass of ABM researchers in
substantive fields• No generally accepted standards of
rigor for agent-based models
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Standards of Rigor
• Tradition of frameworks for application of methodology
• e.g., Churchill (1979) – development of scales in marketing research
• e.g., Anderson & Gerbing (1988) – development of structural equation models
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Goals of Our Presentation
• Overview ABM in marketing• Suggest conditions under which ABM
may be appropriate• Propose guidelines for establishing rigor
in ABM• Give examples of applying ABM in
marketing
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Previous Work in ABM and Marketing
• Diffusion of Innovations– Watts and Dodds (2007), Watts (2002)– Goldenberg et al. (2009)– Rahmandad and Sterman (2004)– Toubia, Goldenberg, and Garcia (2008)– Saikh, Rangaswamy and Balakrishnan (2005)
• Firm Positioning– Wilkinson and Young (2002)– Tay and Lusch (2002, 2005), Lusch and Tay
(2004)
• Adaptive Firms– Midgely, Marks and Cooper (1997), Marks,
Midgley, and Cooper (1998, 2001, 2006)– Marks et al. (1999), Midgley, Marks and
Kumchamwar (2007)
• Evolution of Cooperation in Marketing
– Hill and Watkins (2007), Watkins and Hill (2009)
• Cellular Automata– Garber et al. (2004), Goldenberg et al. (2000,
2004)– Goldenberg, Libai and Muller (2001a, 2001b)– Libai, Muller, and Peres (2005, 2009)– Moldovan and Goldenberg (2003)– Frels et al. (2006)
• Network Theory and Marketing– Goldenberg et al. (2007), Goldenberg, Libai,
and Muller (2002)
• Conjoint Analysis and ABM– Garcia, Rummel, and Hauser (forthcoming)
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
When is ABM Appropriate?
• Medium Numbers• Complex but Local Interactions• Heterogeneity• Rich Environments• Temporal Aspects• Adaptation
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Guidelines for Developing a Model
1. Design the Model
2. Construct the Model
3. Analyze the Model
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Designing an ABM1. Scope of the Model2. Agents3. Properties4. Behaviors5. Environment6. Input and Output7. Time Step
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Guidelines for Model Rigor
• Verification – Conceptual Model matches Implemented Model
• Validation – Implemented Model corresponds to the Real World
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Rigor in Verification1. Documentation – Conceptual design and the implemented model
should be documented.2. Programmatic Testing – Testing of the code of the model.
– Unit Testing – Each unit of functional code is separately tested.– Code Walkthroughs – The code is examined in a group setting.– Debugging Walkthroughs – Execution of the code is stepped through.– Formal Testing – Proof of verification using formal logic.
3. Test Cases and Scenarios – Without using data, model functions are examined to see if they operate according to the conceptual model.
– Corner Cases – Extreme values are examined to make sure the model operates as expected.
– Sampled Cases – A subset of parameter inputs are examined to discover any aberrant behavior.
– Specific Scenarios – Specific inputs for which the outputs are already known.– Relative Value Testing – Examining the relationship between inputs and outputs.
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Rigor in Validation1. Micro-Face Validation – Showing that the elements of the
implemented model correspond to real world elements.2. Macro-Face Validation – Showing that the processes and
patterns of the implemented model correspond “on face” to real world processes and patterns.
3. Empirical Input Validation – Showing that the data used as inputs to the model corresponds to real world data and facts.
4. Empirical Output Validation – Showing that the output of the model corresponds to real world data and facts.– Stylized Facts / Subject Matter Experts – Generally known patterns of
behavior that are important to reproduce in the model.– Real World Data – Recreating real world results using the model.– Cross-Validation – Comparing the new model to a previous model that has
already been shown to be valid.
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Marketing and Diffusion• Bass Model
– h(t) = p + qF(t)• h(t) = rate of adoption, given not adopted so far• F(t) = cumulative fraction of adopters
• Parameters– p = coefficient of innovation; tendency to adopt
independent of social contagion– q = coefficient of imitation; tendency to adopt
due to social contagion– m = market size
Bass, 1969“A New Product Growth Model for Consumer Durables”
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Agent-Based Version
• Setup m agents• Each time step agents determine if they
will innovate on the basis of:– pa – innovation– qa – imitation on the basis of other
adopters• qa (nt /m)
• Observe number of adopters
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Reproducing the Original Bass Experiment
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Network Based Model
• Small Change– Rather than q being modified by the whole
population of adopters, it is only modified by the local neighborhood
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Preferential Attachment Peak Adoption
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Future Work on Bass ABM
• Look at other network structures– More fine-grained resolution– Small World Networks– Empirical network structures
• Cross-validate with other extensions• Examine the role of influentials
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
The Future of ABM in Marketing
• Many other applications– Adaptive Advertising, modeling both consumers and firms– Web 2.0, sponsored search, website navigation, viral
marketing
• GIS and ABM– Understanding consumer behavior in spatial circumstances
(Heppenstall, Evans, and Birkin 2006; Brown et al. 2005)
• Machine Learning and ABM– Extracting agent rules from large datasets
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
Improving the Methodology of ABM
• Difficulties in Verification– Emergence vs. Bugs vs.
Miscommunication• Difficulties in Validation
– Standards for Statistics Reporting– Examination of Path Dependent Results
• Development of Reporting Protocols (Grimm et al. 2006; Polhill et al. 2008; Parker et al. 2008)
© 2008 Robert H. Smith School of BusinessUniversity of Maryland
CCB Projects• Diffusion of Innovation
– Extracting Social Networks from Aggregate Data (Michael Trusov)– Adaptive Agents in Innovation Adoption
• Spatial Modeling– Spatial Demography and Live Event Ticket Sales (Wendy Moe, Peggy Tseng)– Design of Servicescapes (PK Kannan)
• Web 2.0– Content Creation Networks and Media (Chris Dellarocas)– Modeling of Online Auctions (Wolfgang Jenk)– Dynamic Web Metrics
• Combining Analytical Methods with ABM– Game Theory (Yogesh Joshi)– Statistical and Bayesian Analysis (Wolfgang Jenk, Michael Trusov)
• Social Networks– Pedagogy of Social Network Visualization (Ben Shneiderman, Derek Hansen)