Journal on Policy & Complex Systems Volume 1, Number 2, Fall 2014 | Page 100

�����������������������������������������������������������������
Furthermore , while it is possible to interrogate the model over long periods of time , it is unlikely that the parameter estimates remain fixed over many decades . With this in mind the simulation at this stage should be seen as more than a proof of concept , but not sufficient at this point to drive policy choices .
How Might We Embrace Complexity in Studying SDOH ?

Because they involve tangled patterns

of causality with multi-level , multiscale , and dynamic processes , and feedback between and within levels , SDOH are a prime target for the use of complex systems simulation techniques . Importantly , as we have argued here , in several instances , ��� adopting a complex systems approach to SDOH scholarship may , in some instances , be misleading and provide incorrect answers . The example above illustrates the potential for these techniques to simulate the effects of policies that are difficult to implement , combinations of policies that would be even more difficult to implement , the potential for following the effect of policies over time , over the life course , and over generations , and the ability to examine the importance of various etiologic pathways . Of course , there are many issues that can be raised by the use of such techniques and numerous barriers to using them ( Galea et al ., 2009 ; Galea et al ., 2010 ). In what follows we will focus on both general issues and issues that particularly arise with regard to the use of such techniques to study SDOH .
The simulation is best thought of as an abstract but not generic representation of the phenomena being studied . As such , decisions , often based on imperfect information , must be made about what to include in the model and what to leave out , and the answers arrived at will be based on the specific questions that are of concern . The fidelity of the model will depend on the extent to which sagacious choices are made — in the illustration above , the absence of gender could be seen as a critical flaw — but , resource issues aside , the explicit nature of the ABM means that new properties of agents , of the environment in which they live , or new knowledge of pathways thought to be critical can always be added . Thus , the simulation effort can be thought of as an incremental effort , although stopping rules may be difficult to enforce as new findings emerge from the simulation . We note , however , that the alternative to these approaches is often to grapple with complex phenomena with��� an explicit , or at best a theoretical , representation . This of course poses its own challenges ; perhaps surpassing those imposed by those incurred through the use of complex systems approaches .
The fidelity of the simulation effort is also highly dependent on the presence and quality of the data that are used to inform parameter estimates in the model . In the BMI simulation illustration above there are over a hundred parameters that drive the model ’ s behavior , and in such cases it is often difficult to find extant studies to inform the value of those parameters . Some of the data might come from cross-sectional studies and some come from longitudinal studies with differing lengths of follow-up , the covariate adjustments may differ for study to study and there may or may not be information to inform model estimates of parameter variability . In addition , parameter estimates may come from different study populations and have been collected at different periods of time . Because of the broad disciplinary terrain traversed by SDOH studies , the problem is compounded even more by the different analytic and
98