Growing Collaborations
To ensure computational comparability between the networks in both periods , separate versions of the networks in each period are constructed using only the organizations that appear in both periods , identified as the intersection networks . Since the intersection networks are limited to only the organizations that appear in both periods , they have the same number of nodes , as seen in table 1 . The 2012 and 2014 editions of the survey collected data on three kinds of interaction within the Farm to Plate Network : information sharing , project collaboration , and resource sharing . Each of these forms a separate subnetwork , as defined by Scheinert , Koliba , Hurley , Coleman , and Zia ( 2015 ), where a network with multiple types of interaction can be disaggregated into the separate types of interaction . The separate types of interaction then form networks of their own referred to as subnetworks . A union network for each year was formed by including the links found in at least one subnetwork for that year . Each subnetwork was analyzed and modeled independently to provide additional testing space for our model . Table 1 provides a summary of the intersection subnetworks that we use in this study .
Input
The model uses two data sources for inputs . First , as discussed under Initialization , the model uses the network data collected in the Farm to Plate Network survey for initialization . This same data is used at the beginning of each repetition of the accretion and decay algorithms , such that the model is re-initialized for each repetition of the algorithms .
Second , the model uses estimated ERGM results to parameterize the accretion algorithm . Doing so ensures that our estimated ERGM drives agents ’ decisions about forming new links . The ERGM is estimated using the ergm ( Hunter et al , 2008 ; Handcock et al , 2014 ) and statnet ( Handcock , Hunter , Butts , Goodreau , & Morris , 2003 ) packages in R ( R Core Team , 2014 ). The model uses one parameter for each of the terms in the ERGM , meaning one parameter for each value of each attribute and for each homophilic pairing of each value of each attribute , along with a term for the influence of the network ’ s density and transitivity . The results of estimation are discussed below and the chosen results are included in Appendices B and C . When existing 2012 links are used to forecast new and existing 2012 links , the ERGM results in appendix B are used . When new and existing 2012 links are used to forecast new and existing 2014 links , the ERGM results in appendix C are used .
Submodels
Link Accretion : The accretion algorithm cycles through all 30,800 organizational pairs seeking structural holes , where links can be added . When the model
13