Policy and Complex Systems
accretion and link decay algorithms on the subnetwork to be updated . Accretion is run first and decay is run second . New links to be added and existing links to be cut are recorded using the population of network links contained in each organization . Once the links to be added and the links to be cut are determined and recorded , the model updates the network by adding and deleting the appropriate links to the population of active links . This newly updated network is referred as the forecast network . Since no links are actually added to or deleted from the network until after both the accretion and decay algorithms are run , changing the order to run decay before accretion would not influence the results ; all accretion and decay decisions see the same network when organizations make their accretion and decay decisions .
In the second process , the forecast network is compared to a new network that is empirically observed update to the existing
network . This empirically observed update is labeled as “ Network Growth ” in figure 1 and this second process is labeled as “ Model Performance Assessment .” A survey of the members of the Farm to Plate Network asked respondents to identify whether the links in the three subnetworks were “ new ,” meaning that the links had been formed in the last year and as part of the responding organization ’ s activities within the Farm to Plate Network . Those links that were not marked as “ new ” are treated as “ existing .” This survey was conducted twice , in 2012 and 2014 , and is discussed in detail , under Initialization . The existing network , which provides initial conditions , and the observed network are constructed in two different ways . Each way of constructing these networks uses a different combination of new and existing links : 1 ) using the networks of existing 2012 links to forecast the networks of new and existing 2012 links , and 2 ) using the networks of new and
10