Grantee Research Project Results
Final Report: Agent-Based Modeling of Industrial Ecosystems
EPA Grant Number: R829688Title: Agent-Based Modeling of Industrial Ecosystems
Investigators: Andrews, Clinton J. , Axtell, Robert
Institution: Rutgers University - New Brunswick , Brookings Institution
EPA Project Officer: Hahn, Intaek
Project Period: July 1, 2002 through June 30, 2005 (Extended to June 30, 2006)
Project Amount: $334,146
RFA: Corporate Environmental Behavior: Examining the Effectiveness of Government Interventions and Voluntary Initiatives (2001) RFA Text | Recipients Lists
Research Category: Environmental Justice
Objective:
Multiagent simulation (MAS) is an increasingly popular bottom-up computer modeling technique in which purposive agents (e.g., employees) interact so that aggregate performance (e.g., of a firm) is an emergent property of the system. The objectives of this research project are to: (1) investigate behavioral and organizational questions associated with environmental regulation of firms, and (2) test specifically whether a MAS approach that highlights principal-agent problems offers new insights and empirical validity.
Summary/Accomplishments (Outputs/Outcomes):
We created MAS models of firms and grounded them empirically in case studies and aggregate statistical evidence. During Year 4 of this project, we focused chiefly on writing up our results and calibrating the models to replicate specific historical circumstances.
PolyModel is a highly detailed MAS model of the interactions among employees, production technology, and the external environment within a polymer processing firm. It was adapted to accept historical time-series data on input costs and product prices as exogenous inputs. The programmer also completed minor debugging and documentation of the model.
Two new papers have been written based on PolyModel. The first examines the environmental efficacy of alternative hiring and supervisory practices when employees are either under- or oversocialized. Oversocialized employees are not very responsive to economic signals, whereas undersocialized employees are not very responsive to social signals. Insights based on modeling runs are tested against corroborating case study evidence.
The paper illuminates how the degree of socialization can influence the efficacy of both hiring and supervisory practices. If the plant manager selectively recruits employees with certain characteristics that affect behavior (say, environmentalist beliefs), they can visibly influence the firm’s performance, the more so if they are oversocialized. Supervisors adopting a tolerant and mentoring attitude toward employees see higher absenteeism, less cliquishness, and better employee attitudes under both the over- and undersocialization scenarios. Supervisors adopting an unforgiving attitude with no mentoring see the formation of cliques separating employees with good versus bad attitudes, and they also see high employee turnover with associated quality control and training issues. Corroborating case study evidence confirms that informal networks play important roles in firms, but that dramatic under- and oversocialization are rare conditions, with modest levels of both market rationality and social embedding better characterizing typical employees of real firms.
The second PolyModel paper compares technological and human resource management strategies for reducing pollution. The paper first demonstrates that the firm approximately replicates the historical trajectory of a case study firm when driven by historical values of input costs and product prices. It then considers the relative profitability and pollution levels of the firm under four management scenarios. The first scenario considered how the firm performs if environmentalism becomes widespread as a social movement and employees change their behavior accordingly. The second scenario examines what happens if a champion, the plant manager, selectively hires environmentalists. The third scenario considers what would happen if management imposed a strict quality control regime instead. The final scenario allows the plant manager to replace employees with automation as it becomes cost-effective.
The results suggest that human resource tools of hiring, training, supervision, and firing are most effective when coordinated to achieve clear objectives. The underlying labor pool bounds the firm’s potential. A rogue manager who preferentially hires environmentalists (or cronies or people from a particular ethnic group) can affect the overall performance of the firm. The most profitable strategies emphasize quality control or automation, and they provide minor environmental benefits given the characteristics of this injection molding technology. For other technologies, quality management may pay higher environmental dividends. It is noteworthy that the behavioral/human resource fixes provided benefits that were of similar magnitude to the technical fix (automation). This supports previous claims that organizational practices deserve attention from environmental regulators.
TeamPollution was the second model built. It is a multiagent system representing a highly stylized firm and it sought to abstract as much as possible from PolyModel in pursuit of general insights. The draft paper prepared in Year 3 was revised and submitted to an economics journal, and the coprincipal investigator continues to present the results and they have generated great interest.
EcoNiche was the third model built. It is a multiagent system representing a highly stylized product market in which a green niche product competes with a conventional product. This year EcoNiche was adapted to the automobile market, modeling the entry of hybrid cars as a green technology that displaces conventional technology. This required gathering empirical data on costs, quantities, and characteristics of hybrid and conventional automobiles over time. It also required the development of plausible public policy scenarios (laissez faire, technology ban, tax, information provision, etc.) and code to represent them in the computer model. It also required segmentation of consumers by search strategy, income level, and green preferences. This work now is complete and a paper summarizing the findings is being written this summer.
Key findings and lessons for the U.S. Environmental Protection Agency from the PolyModel, TeamPollution, and EcoNiche models were discussed above and in the Year 3 annual report. There also are several cross-cutting insights from the project as a whole.
Overall, this project has confirmed that agent-based modeling has utility in studying the effects of organizational behavior on environmental management. It offers insights that are complementary to econometric studies, organizational case studies, social psychology experiments, and other modes of organizational analysis. This type of simulation modeling provides a method for formal theorizing and for prospective, forward-looking analysis of organizational behavior in general and environmental management in particular.
Yet the MAS approach also poses challenges. In practical terms, it is still rare to encounter graduate students who have a sophisticated understanding of both the social sciences and of Java programming. Thus, implementing the project has depended on pairing personnel with complementary skills, and getting them to collaborate successfully. The best collaborations have involved graduate students in computer science and public policy working under close supervision of the principal investigator. Needed are more graduate programs that cross-train computational social scientists.
Validating multiagent models of organizational behavior remains difficult, because typically the only available quantitative data are aggregated at the level of the firm. Case studies of employee interactions provided some empirical anchoring for the modeling but still leave numerous degrees of freedom. Hence the models are better suited to producing: (1) generic insights about decision rules and management strategies; and (2) management training tools, and they are less well suited to predictive modeling of specific investment or public policy tradeoff decisions.
Each year, however, the modeling becomes easier and more standardized in implementation. Computing power also continues to become cheaper and more ubiquitous, so there is much room for optimism regarding this research approach.
In particular, multiagent systems provide a tractable way to bring the insights of behavioral economics to policy relevance. By building realistic, boundedly rational, purposive individual behavior into models of organizations, markets, and other relevant large-scale phenomena, troubling artifacts of simplifications like axiomatic firm-level profit-maximization disappear. Better public policy is a likely outcome. Instead of floundering in the analytical no-mans land of second-best policies or resorting to simple-minded policy prescriptions, it becomes increasingly possible to identify optimal public policies based on realistic behavioral expectations.
Vibrant research communities using agent-based techniques have appeared within economics, organization science, and industrial ecology, and this project has encouraged the growth of those communities.
Future off-budget activities include additional publications, updating the project Web site, and applying multiagent simulation techniques to new topics.
We expect to adapt EcoNiche for use in studying emerging green buildings technologies and then we will write a paper on that topic. We also expect to write a computer science-oriented paper on the Bayesian updating algorithm developed in EcoNiche. All new papers, model code, and executable versions of EcoNiche will be posted on the project Web site, supplementing the significant amount of material that is already there. Andrews has submitted a proposal to the National Science Foundation to apply multiagent techniques to simulate the energy and environmental performance of green buildings.
Axtell’s main focus will be to get his ideas on team pollution into: (1) the academic community through publication of the paper mentioned in last year’s report; and (2) the policy community, possibly through a Brookings Policy Brief. If he finds additional funding, he will have a research assistant turn his extant extensive inventory of the cost of abatement literature into a truly comprehensive one through a more systematic literature search. This would put the empirical research results on even stronger footing, although he feels they are sufficiently strong already to support the main conclusions of the paper. One additional future activity will be to put these papers and analyses on his research center’s Web site. This summer he is Director of the Trento Summer School on agent-based computational economics.
Journal Articles:
No journal articles submitted with this report: View all 21 publications for this projectSupplemental Keywords:
pollution prevention, green chemistry, life-cycle analysis, alternatives, sustainable development, clean technologies, innovative technology, renewable, waste reduction, waste minimization, environmentally conscious manufacturing, public policy, decisionmaking, psychological, preferences, sociological, social science, industry (polymer processing, plastics), NAICS Code 3261, NAICS Code 325991, NAICS Code 326199, team production, team pollution, local increasing returns, increasing returns to scale, Nash equilibrium, under-abatement, hypothetical cost savings, explicit cost savings, heterogeneous marginal costs of abatement, non-optimal firm behavior, sub-optimal firm behavior, inefficient market-based regulation,, RFA, Scientific Discipline, Sustainable Industry/Business, cleaner production/pollution prevention, Corporate Performance, Economics and Business, Social Science, environmental policy case studies, corporate decision making, cleaner production, corporate environmental policy, sustainable development, computer generated alternative synthesis, environmentally conscious manufacturing, government intervention, clean technology, computer simulation modeling, government-industry interaction, environmental behavior, agent based modeling, pollution prevention, clean manufacturing designs, corporate environmental behaviorRelevant Websites:
http://www.brookings.edu/dynamics Exit
Progress and Final Reports:
Original AbstractThe perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.