Policy and Complex Systems
from benchmarking and historical analysis ( Scharpf 1997 ) to computational simulations ( Gilbert 2004 ) are used .
Policies are implemented as topdown decisions but their acceptance is — at least partly — a bottom-up process . This calls for a system understanding at the microlevel in order to find out which of the alternative policies are most effective ( Scharpf 1997 ). Microlevel analysis contributes an insight into the individual ’ s unanticipated adaptive behavior , decision making , and interactions , facilitating the improvement of conditions for effective policy solutions .
The need for micro-level analysis has a good fit with what agent-based modeling ( ABM ) offers . ABM , as a bottom-up simulation approach , builds artificial societies from individual agents and their interaction , giving insight into how people may react toward different situations ( Banks et al . 2000 ). Compared to other computational approaches such as differential equations and statistical modeling , ABM imposes less assumptions on linearity , homogeneity , normality , and stationarity ( Banks et al . 2000 ). In addition , agent-based models have the power to demonstrate emergent phenomena at system level . This is especially instrumental for policy problems where the influence of individual behavior on system properties is under study ( Conte et al . 2001 ).
However , to use ABM for policy analysis there are additional requirements . For example , for evaluating policy alternatives , the policy analyst also needs means of imposing policies ( or rules ) to the simulated system in order to study individual reaction and adoption to these impulsions . Therefore , building a system purely from bottom-up may not be entirely instrumental for policy analysis . Furthermore , since the subjects of policy problems are societies with real people , the reliability of an agent simulation and the results it provides are a sensitive issue that require careful evaluation .
The goal of this paper is to show the suitability of ABM as an approach to analyze policy problems and how it can be enhanced to address even more requirements for policy analysis . To achieve this goal , we first explain the policy analysis cycle and introduce the various steps and requirements in the analysis process in Sections 2 and 3 . We then introduce the computational tools that are commonly used for policy analysis and reflect on the benefits and drawbacks of each , in Section 4 . We explain how ABM can be used as a comprehensive approach for policy analysis and discuss areas for further enhancement in Section 5 . Finally , we conclude our findings in Section 6 .
2 . The Process of Policy Analysis
A policy is a set of principles or rules to guide a social system toward those actions that are most likely to achieve a desired outcome . Policies can be implemented as social norms ( e . g ., switching off lights when leaving a place which must be internalized by people , for example , through advertisement campaigns ), legal impositions ( e . g ., subsidies and taxes on the different consumer products such as milk , LED light bulbs , etc .) or technological artifacts ( e . g ., electronic gates at stations ).
The practical activities of policymaking and implementation are distinguished from the more reflective activities of policy analysis which aim at determining which alternative policies may most likely achieve desired goals and outcomes . Policy analysis is specifically complex because the consequences of implementing a policy “ are the outcomes under external constraints of intentional action ” ( Scharpf 1997 ). In other words , human actors are driven by a com-
62