The PLS Agent – Agent Behavior Validation by Partial Least Squares

Winter Simulation Conference, 2009

Agent-based modeling and simulation (ABMS) is a new approach to modeling systems comprised of autonomous, interacting agents. Computational advances have made possible a growing number of agent-based models across a variety of application domains. Applications range from modeling agent behavior in the stock market, supply chains, and consumer markets, to predicting the spread of epidemics, mitigating the threat of bio-warfare, and understanding the factors that may be responsible for the fall of ancient civilizations. Such progress suggests the potential of ABMS to have far-reaching effects on the way that businesses use computers to support decision-making and researchers use agent-based models as electronic laboratories. Some contend that ABMS "is a third way of doing science" and could augment traditional deductive and inductive reasoning as discovery methods. This brief tutorial introduces agent-based modeling by describing the foundations of ABMS, discussing some illustrative applications, and addressing toolkits and methods for developing agent-based models.

2011

This paper describes the use of validation methods in model building. We address issues associated with the increasing complexity of models that is in part a response to the growing popularity of Agent-Based Models (ABM), commonly used to study cognitive, natural, and social phenomena. The first section of this manuscript discusses model categories and attributes. The second section discusses the stages of validating a simulation model: verification, validity, and sensitivity analysis. The third section presents specific validation approaches, with an emphasis on six specific tests that are described in detail. The final section summarizes the goals of model validation and modeling. Keywords-Agent-Based Models, Validation, Verification, Infectious Disease Models.

Results from agent-based or multi-agent simulation (AB/MAS) modelling can provide relevant information for policy makers, scientists and stakeholders about the boundary conditions of rural development and the uncertainties involved in land-use/cover change (LUCC). However, the process of model validation that can build trust in the outcomes for new parameter conditions and in future scenarios is not a trivial problem. Apparently, no common measure of the degree of confounding between parameterization and validation data sets exists. The current lack of success and the effort necessary for validating the models can be traced to the weak theoretical representation of human decision making in current models. Thus, this paper reviews various ways to represent land-use decision making using AB/MAS models. It briefly describes process-based decision making as an alternative approach to address the problem of weak theoretical representation of human decision making, and presents a case study of an agent decision-making model applying an empirical validation technique.

2006

ABSTRACT. There is an increasing drive to combine agent-based models with empirical methods. An overview is provided of the various empirical methods that are used for different kinds of questions. Four categories of empirical approaches are identified in which agent-based models have been empirically tested: case studies, stylized facts, role-playing games, and laboratory experiments. We discuss how these different types of empirical studies can be combined.

Winter Simulation Conference, 2011

Agent-based simulation (ABS) is an approach to modeling systems comprised of individual, autonomous, interacting "agents." Agent-based modeling offers ways to more easily model individual behaviors and how behaviors affect others in ways that have not been available before. There is much interest in developing agent-based models for many application problem domains. Applications range from modeling agent behavior in supply chains and the stock market, to predicting the success of marketing campaigns and the spread of epidemics, to projecting the future needs of the healthcare system. Progress in the area suggests that ABS promises to have far-reaching effects on the way that businesses use computers to support decision-making and researchers use agent-based models as electronic laboratories to aid in discovery. This brief tutorial introduces agent-based modeling and simulation by describing the basic ideas of ABS, discussing some applications, and addressing methods for developing agent-based models.

2007

Perspectives on Science

Agent-based models (ABM) derive the behavior of artificial socio-economic entities computationally from the actions of a large number of agents. One objection is that highly idealized ABMs fail to represent the real world in any reasonable sense. Another objection is that they at best show how observed patterns may have come about, because simulations are easy to produce and there is no evidence that this is really what happens. Moreover, different models may well yield the same result. I will rebut these objections by focusing on an often neglected, but crucial function of ABMs. Building on Gelfert’s (2016) account of the exploratory uses of scientific models I show that, in the absence of an accepted underlying theory, successful ABMs lend inductive support to assumptions concerning certain structural feutures of the behavioral rules employed. One core step towards this goal is what I call multiple-model robustness analysis.

Oxford Bibliographies in Sociology, 2017

Agent-Based Modeling is a research method that represents theories of social behavior as computer programs of a particular kind, rather than narratives (as ethnography does) or equations (as regression models do). Like existing research methods in sociology (both qualitative and quantitative) it can be applied throughout the discipline and offers advantages for certain research questions. In particular, the approach is referred to as agent-based because the computer program unambiguously represents interactions between heterogeneous social actors while also explicitly determining their aggregate simulated consequences. This distinguishes Agent-Based Modeling from existing quantitative approaches in sociology where the relationship between aggregate associations and individual agency is often unclear. It also distinguishes the method from existing qualitative approaches that, while investigating individuals and their interactions, have no systematic techniques for establishing their aggregate consequences. Given this capability, the methodology of Agent-Based Modeling has a distinctive logic. Agent-Based Models are calibrated using data on individual behavior (for example using ethnography or laboratory experiments) and then the computer program generates simulated aggregate data. This can then be compared with equivalent real data for validation. It is the independence of these two activities that provides Agent-Based Modeling with its distinctive claim to explanatory power. The explicitly represented link between individual and aggregate respects the complexity of social systems, the phenomenon in which individuals and their simple interactions may produce surprisingly counter-intuitive aggregate outcomes. Agent-Based Models are thus particularly suitable for investigating sociological issues involving heterogeneous actors, diverse cognitive processes and social systems mediated by entities operating between the level of the individual and the aggregate (like schools and churches).

Article history: Available online xxxx Keywords: Agent-based Modelling Model development Research process Modelling Cycle Methodology

2006

This paper addresses the problem of finding the appropriate method for conducting empirical validation in agent-based (AB) models, which is often regarded as the Achilles' heel of the AB approach to economic modelling. The paper has two objectives. First, to identify key issues facing AB economists engaged in empirical validation. Second, to critically appraise the extent to which alternative approaches deal with these issues. We identify a first set of issues that are common to both AB and neoclassical modellers and a second set of issues which are specific to AB modellers. This second set of issues is captured in a novel taxonomy, which takes into consideration the nature of the object under study, the goal of the analysis, the nature of the modelling assumptions, and the methodology of the analysis. Having identified the nature and causes of heterogeneity in empirical validation, we examine three important approaches to validation that have been developed in AB economics: indirect calibration, the Werker-Brenner approach, and the history-friendly approach. We also discuss a set of open questions within empirical validation. These include the trade-off between empirical support and tractability of findings, the issue of over-parameterisation, unconditional objects, counterfactuals, and the non-neutrality of data.