Policy and Complex Systems
the differences between the forecast and observed networks , as seen in Figure 1 . Finally , we review the performance of our algorithm to see how its forecasts could be improved . The model is described in this section , according to the Overview , Design Concepts , and Details ( ODD ) protocol ( Grimm et al ., 2006 ). We developed our ABM using AnyLogic 7 ( AnyLogic University Edition 7.0.3 ).
Overview Purpose
The purpose of this model is to test an algorithm for building a network based on agent decisions , with the agents in this model being organizations . If validated , then an agent-based model that uses our algorithm can provide accurate forecasts of the network ’ s future growth and development through modeling agents ’ decisions to add or remove links . State Variables and Scales
The model comprises three hierarchical levels : network links , organizations , and a “ main object ” or modeling environment . Network links operationalize our networks within the model and allow for managing the network when the model is adding and removing links . Network links are formulated in two different manners . First , AnyLogic ’ s Java libraries offer a state variable for organizations , by which the organizations can be directly connected via a network link and the presence of this connection directly tested . The model uses this form of link object to maintain active network links in memory . A population of networks links also exists within each organization and records both active and non-active links . These network links include five state variables : fromID , toID , toAdd , toCut ,
and startTime . The variables fromID and toID record the organizations that a given network link connects . The variable startTime records the model time for when a link is created . The variables toAdd and toCut record whether this link should be added or removed from the list of links maintained in memory , respectively .
Organizations contain seven state variables : orgID , orgName , orgAcronym , capacity , sector , jurisdiction , and jurisLevel . These state variables allow us to link the organizations to the ERGM and then use the ERGM to drive agents ’ decisions to add or remove links . The first three of these , are identifiers that link to the empirical Farm to Plate network data . orgID records an integer number identifier , which is also used in the fromID and toID state variables for network links . orgName and orgAcronym contain the full name and unique text identifier for each organization in the Farm to Plate Network . The other four variables record the attributes used in the ERGM , matching to capacity , sector , jurisdiction , and jurisdictional level , respectively . The attributes which we use and their application to governance networks , like the Farm to Plate Network , are defined and developed in existing literature ( Koliba et al ., 2010 ; Scheinert & Comfort , 2014 ; Koliba et al , 2015 ). Capacity is defined as an organization ’ s ability to influence its environment and operationalized through organizational staff sizes and budgets ( Scheinert , 2012 ). Sector records an organization ’ s sector , including public , private , and non-profit organizations , but , in this case , also separates out federal- , state- , and municipal-level public agencies as well as educational / research organizations ( Koliba et al , 2015 ). Jurisdiction records an organization ’ s basic geographic area of operation , based on political boundaries . Values used to distinguish levels of
8