Relating Choice of Agent Rationality to Agent Model Uncertainty - an experimental study

The importance of model uncertainties arising from different assumptions about human behavior as opposed to parameter uncertainty is often neglected in integrated models for policy development. In this study, so-called agent model uncertainty is estimated in relation to the choice of agent rationality. A classification scheme is proposed which allows us organize decision models according to their deviation from full rationality. Five decision models covering the whole range from full rationality to maximum deviation from rationality (random decisions) are classified. They are then used in an existing integrated model simulating crop fertilizer usage and related threshold policies for groundwater protection. Using this model and the different decision models, two hypotheses are tested: 1) that agent model uncertainty increases with increasing deviation from rationality and 2) that agent model uncertainty increases for all decision models similarly and uniformly in response to an increase in noise in the model. Results are analyzed with respect to changes in policy and with respect to the level of weather influence on crop yield. Results show that agent model uncertainty varies with deviation from the purely rational in a non-linear way. Hypothesis two also does not hold. The degree of sensitivity of results with respect to uncertain parameters that the agent needs to consider is very much dependent on the decision model. Therefore it is suggested to test agent-based models for robustness and validity with respect to agent model uncertainty by using different categories of decision models that sample the range of possible rationalities.