Growing Collaborations
generally become less likely to decay while homophilic pairings are slower to decay than heterophilic pairings ( Burt , 2000 ). Network bridges are more likely to decay than nonbridge links for the first three years that the link exists and are less likely to decay than non-bridge links after three years ( Burt , 2002 ). Additional research would be needed to combine decay based on attribute pairings with decay based on network bridges ; Burt does not offer any conclusions about how these formulas could be used together . where :
Exponential Random Graph Models
ERGM is designed to elicit the varying influence of , among other potential model terms , attribute values , homophily , transitivity , preferential attachment in a network ’ s formation but is sensitive to specification ( Handcock et al ., 2014 ), meaning that its results are dependent on which of these possible model terms are included when the ERGM is estimated ( Hunter & Handcock , 2006 ; Hunter , 2007 ; Hunter , Handcock , Butts , Goodreau , & Morris , 2008 ; Hunter , Goodreau , & Handcock , 2013 ). Heterophily is treated as a reference case to homophilic pairings . ERGM estimation calculates a logistic regression that measures the influence that attribute value , attribute pairings , transitivity , and preferential attachment have on the set of observed ties , using the following functional form ( Goodreau et al ., 2009 ):
Estimation platforms provide terms for nonpaired attribute values , homophilic pairings , and transitivity ( Hunter et al , 2008 ). 2 The influences of attribute values , homophily , and transitivity , measured as logit coefficients , may then be used in a logistic formula to produce estimates ( Kennedy , 2003 ) that can be interpreted as the probability that each potential network link is observed .
Modeling Network Development in the Vermont Farm to Plate Network
We test the effectiveness of an
algorithm parameterized by the results of an ERGM for forecasting network growth . We build an agent-based model ( ABM ) that executes this algorithm to forecast change in the Farm to Plate Network and then compare our forecast growth to empirical changes that occurred at the same time . We calculate an error rate for our algorithm by measuring
2
Every pairing between two nodes can be described as either heterophily or homophily . Since ERGM estimation uses terms for homophily and each node ’ s own attribute values , the influence of heterophily is included in the ERGM ’ s intercept term .
7