Using Adaptive Agent-Based Simulation Models to Assist Planners in Policy Development: The Case of Rent Control

Computer simulation modeling for policy development in planning has had difficulty gaining a consistent foothold. Reasons for this include bad experiences with large-scale, comprehensive models (e.g., Forrester, 1969) and the lack of theory that one can quantify (Batty, 1994). Batty (1994) has suggested that new types of computational models, based on the tenets of complexity theory (Bernard, under revision), may prove useful. One type of complexity theory model is an "adaptive agent based model" in which the actions, interactions, and adaptations of many autonomous, heterogeneous "agents" (households, firms, etc.) produce emergent, system-wide behavior. One can examine this emergent behavior using commonly employed metrics, but one can also garner a richer, more intuitive understanding of how the individual behavior of the agents self-organize to produce the entire system. Using this type of modeling for small-scale planning problems can both inform planning theorists and improve planning practice by providing rich understanding that standard quantitative models do not. In this paper, I will present an agent-based model of rent control. Household agents (with different income levels) rented apartments from landlord agents--these apartments were situated on a lattice. Landlord agents continually adapted to the conditions of the marketplace (apartment demand, type of rent control in place, and so on), raising and lowering their prices as they saw fit. I varied conditions of rent decontrol and measured various metrics, such as vacancy rate, apartment quality, tenant income, and average rent paid. I found that a market with rent control typically has tenants with lower incomes than a nonrent-controlled market, even substantially after the market has been suddenly decontrolled. In addition, I found that there were lower vacancy rates in regimes of rent control. As these results are not based on actual data, they are merely presented as suggestive. In fact, the point of abstract computational models such as the one presented here is not as the ultimate predictors of policy decisions, but as tools to inform and provoke discussion among policy makers. Thus, I will conclude by speculating on the use of adaptive agent-based models for assisting in policy formulation.