Policy and Complex Systems
Towards an Agent Based Model of Network Growth
ABMs provide a means for a researcher to direct agent behavior while still allowing for emergent properties , like networks . This makes an agent based model a natural match for bottom-up forecasting of network growth . ABMs give space to each agent to make independent decisions . The researcher is left to determine how the agents makes those decisions , including how dependent , independent , or interdependent the agents ’ decision rules are . The strength of ABM approaches is that they allow for examining situations with a complex set of interdependent agents and rules ( Ostrom , 2005 ). Their weakness is that ABMs require extensive amounts of high-quality data to build , validate , and calibrate . Data that meet the needs of an ABM are often difficult to acquire in any social scientific context , including organizational partnership networks ( Janssen & Ostrom , 2006 ).
Consistent standards and protocols for using empirical data in ABMs are in short supply , particularly among social scientific applications of ABMs ( Janssen et al ., 2008 ), though some , such as the ODD protocol , are beginning to gain wide acceptance ( Grimm et al , 2006 ; Grimm et al , 2010 ; Janssen et al , 2008 ). While the ODD protocol helps in communicating model design , it does not speak of the challenges of linking the model ’ s structures , processes , and outputs to empirical data ( Grimm et al ., 2010 ). Janssen and Ostrom ( 2006 ) identify two sets of trade-offs for how to use data that a modeler must consider when designing an ABM : 1 ) context-specificity versus generalizability , and 2 ) few subjects versus many subjects ( see table 6 ). Under the first trade-off , the modeler either designs the model to apply to many cases broadly or designs the model to recreate a specific case with high fidelity to that case ’ s nuance . Under the second trade-off , the model chooses either to gain detailed data on decision-making processes from a few subjects or to acquire broad data on many subjects , but without the depth that is possible when the subjects are few in number . For Janssen and Ostrom ( 2006 ), these trade-offs create a foursquare typology that identifies four means of data acquisition and usage , shown in table 6 . ERGM analysis provides a version of stylized facts that can be transformed into an algorithm for forecasting network growth ; ERGM provides network analysts with a means to measure the relative influence of each behavior for selecting partners , thereby providing the rules for how a network developed ( Goodreau et al ., 2009 ). Examining the influence that each decision rule has on link selection provides insights to how specific networks formed .
Modeling Networked Governance with ABMs
The Vermont Farm to Plate Network plays a governing role with a list of 25 goals and a strategic plan for harnessing the network and its membership to achieve those goals ( VSJF , 2015a , 2015b ). Increasingly ABMs are being employed to study multiscale , multi-sector governance arrangements ( Pahl-Wostl , 2005 ; Janssen & Anderies , 2007 ; Janssen , et al ., 2008 ; Ostrom , 2010a , 2010b ; Veldkamp et al , 2011 ; Grimm et al ., 2006 ; Zia & Koliba , 2015 ). These ABMs rely upon ethnographically derived institutional rule structures extracted through extensive comparative case studies ( Pahl-Wostl &
18