Policy and Complex Systems
network context to which ERGM can apply . Future research should repeat the modeling process in other contexts , such as other policy problems , like energy and water , as well as outside of policy implementation . We demonstrate that an ABM that grows a network from the bottom up can forecast that network ’ s growth in one context . Future research should consider if our conclusions hold for ABMs built outside of a food systems or policy context .
The primary building block of an ABM is its agents and their decision mechanisms ( Axelrod , 1997 ; Pahl-Wostl & Ebenhoeh , 2004 ; Janssen & Anderies , 2007 ; Grimm et al . 2006 ; Zia & Koliba , 2015 ). Our algorithm ’ s weakness is that the algorithm cannot recreate agents ’ true decision processes , only their outcomes . ABMs that apply even simple decision processes can produce complex results , often with unexpected emergent patterns ( Campbell et al ., 2014 ; Epstein , 2006 ; Janssen & Ostrom , 2006 ). Researchers may then encode the decision processes of relatively simple agents to create effective forecasting models in those contexts . Models using more complex agents , including human beings and organizations are usually limited in their scope since the full decision processes , including the capacities for learning and adaptation ( Axelrod & Cohen , 2000 ; Choi & Robertson , 2014 ; Comfort , Boin , & Demchak , 2010 ), are often difficult , if not impossible , to anticipate in the scope of a pre-programmed ABM . This limitation is what has led to the use of ABMs in theory building but limited their application in theory testing and forecasting ( Epstein , 2006 ). Advanced modeling methods , including artificial neural networks ( ANNs ) are now emerging ( Zhang , Patuwo , & Hu , 1998 ; Schmidhuber , 2015 ), but have to see wide application in social scientific contexts . Future research should consider integrating
ANNs and ABMs . Conclusions
Understanding how networks change can provide insights into how those same networks will behave . Goal-directed interorganizational partnership networks , by definition , seek to obtain some kind of goal , whether setting a policy agenda , raising public awareness , providing information and education , implementing policy , or any combination of these goals , as well as many other potential goals . The Vermont Farm to Plate Network pursues 25 different stated goals . Its ability to pursue its goals is dependent on the voluntary efforts of its members and the resources that the members devote to their efforts within the Vermont Farm to Plate Network . If we wish to understand how network structures influence performance , we will need to understand not just how to measure those structures at a single period , but how to measure their change .
Network theorists have been studying this change ( Barabasi et al , 2002 ; Perc , 2010 ; Tomassini & Luthi , 2007 ), but , so far , their efforts have largely been limited to documenting how network measures have changed within a single network over time , using dynamic network analysis . This approach has paved the way for understanding network growth and provided a solid basis for documenting this change . However , this approach has not yet been able to offer insights into the mechanisms that drive network growth and development . Our research shows that ERGM provides insights into these mechanisms by measuring the influences of different agent strategies in selecting collaborators and that these insights can be applied to forecast growth . Our algorithm , when paired with an algorithm for network decay ( Burt ,
20