Journal on Policy & Complex Systems Volume 1, Number 1, Spring 2014 | Page 75

Enhancing ABM into an Inevitable Tool for Policy Analysis
For policy analysis , not only artificial agents should be able to recognize a policy change in the system , they need to identify the global behaviors and adjust their behavior accordingly ( i . e ., adaptability ). For example , if inefficient light bulbs are banned in one country , an agent may start buying LED lamps because they are the only ones available in the market . Another consequence of such a policy may be the formation of black markets . The rich cognitive agent must be able to recognize this emergent phenomenon and make decision on whether to continue buying LEDs or buy banned light bulbs from the emergent black market .
The emergence and evolution of social structures is studied both in the social sciences ( Axelrod 1986 ; Janssen 2005 ; Smajgl , Izquierdo , and Huigen . 2010 ) and computer science ( Holland 2001 ; Smajgl , Izquierdo , and Huigen 2008 ). This line of research can be further enhanced to explore the effect of imposing policies on the emergence of social structures .
Enhancements for Decision Support for Policy Selection
ABM is a bottom-up approach which models individuals rather than top-down principles ( Epstein 2006 ). Policy analysis by definition is about imposing guiding principles into a social system . Therefore , ABM for policy analysis should be a combination of bottom-up and top-down model development to facilitate explicit and elaborate policy comparison and evaluation . Using System Dynamics ( SD ) techniques is one way of facilitating this combination ( Scholl 2001 ; Castella et al . 2007 ). Furthermore , participatory decision making is a new line of research in ABM where enhancements can be highly instrumental for selecting policies ( Barreteau , Bousquet , and Attonaty , 2001 ).
Finally , advances in verification and validation of agent-based models are required to be able to trust the results of comparison between policy alternatives using ABM . Results from agent-based simulations are difficult to interpret due to their size and complexity . Even more , as there is usually no test using empirical data , most evaluations do not normally go beyond a proof of concept ( Janssen and Ostrom 2006 ).
Enhancements for Monitoring Implemented Policies
Improving cognitivity in agents can provide more insights into why people give certain reactions toward an implemented policy . Furthermore , the artificial intelligence literature which provides means of implementing cognitive agents needs to be more accessible to social scientists . Currently , this line of research provides sophisticated tools that are difficult to comprehend and use by social scientists and policy analysts who are less familiar with computational sciences .
6 . Conclusion

The goal of this research was to explore the potential of ABM as a tool for policy analysis . To perform this research , we presented a systematic overview of the policy analysis cycle to identify the requirements it puts forward . We then compared various tools that are used for policy analysis , including ABM , to identify the benefits and drawbacks of each .

The comparison between the different tools provides our hypothesis that ABM can indeed be considered as an inevitable tool for policy analysis under the condition that some enhancements are made . Therefore , by using the results of our compari-
73