Policy and Complex Systems
Conclusions and Discussions
Our study is one of the first that integrates economic experiments with agent-based modeling in a nonpoint source pollution setting . The ABM extends and scales up the findings from the economic experiment by providing a spatially explicit simulation environment based on an actual watershed . Instead of assuming full rationality , the economic experiment calibrates and validates the ABM by defining human-based bounded rational decision rules for the agents . We apply a modification of a classic game theoretical model from the environmental economics literature to the ABM and the experiment as the core underlying model in both scenarios . We define the target level ( fully rational theoretical level ) by solving for unique dominant Nash strategy . Using experimental data , we first identify the number of behavioral groups using exploratory cluster analysis and then group agents into the three identified groups by multinomial logistic model ; second , we define agent decision rules by estimating adoption and production deviations from the target levels based on the information treatment , type , size , and location of each agent . The result of our simulation experiment demonstrates that both information “ nudges ” help the performance of the ambient-based policy . Individual level information induces higher policy efficiency compared to group level information , where the individual decisions tend to be anchored to the group averages , even though it may not be in their best interest . Our results show in a spatially explicit watershed setting that ambient-based policies , coupled with information “ nudges ” to provide guidance to people ’ s behavior , have the ability to induce group level compliance , and the policy efficiency is higher when individual level information is being provided . Therefore , it is important to use informational “ nudges ” to help people make better decisions , especially under complex heterogeneous scenarios .
There are a number of limitations and directions of future work based on our research . First , a more complicated hydrological model may be developed and incorporated in the ABM and the experiment . Examples of such models include the WWACShed model by Tesfatsion et al . ( 2017 ) and the SWAT model used in Ng et al . ( 2011 ). However , if one attempts to also include bounded rationality in the agent decision processes and use economic experiments to capture these irrationalities , it is crucial to ensure that the conclusions from the experiment could be safely carried over to the ABM . In our experiment , this link was built by adopting the same underlying model and therefore the same incentives around the dominant Nash strategies . If a more complicated model were in place , it would be hard to solve for a perfect rational utility maximization prediction , and therefore would be difficult to have a baseline to compare with actual human behavior . Additionally , the more complicated a model is , the more information burden is introduced to the participants and the harder for the participants to generate informed decisions , so one needs to think carefully about the tradeoff .
Second , another extension of this research is to use farmer sample instead of a sample from university students in the experiment , aiming to increase externality validity of the experiment . The majority of research comparing samples from students and professionals generally find the two samples demonstrate similar responses in both agricultural ( Cummings , Holt , & Laury , 2004 ; Fooks et al . 2016 ; Messer , Kaiser , & Schulze , 2008 ) and non-agricultural ( Vossler , Mount , Thomas , & Zimmerman , 2009 ) contexts , it may still be a valid exten-
185