Estimating host influence on cross-species transmission
Our goal was to see if some host species were more likely to be involved
in cross-species transmission of various parasite species. For each
parasite species and site, we constructed undirected networks of hosts
with edge connections weighted by the epidemic overlap value for each
host combination, while removing host species with no connections from
the network. Because some parasites produce long-lived spores capable of
delaying transmission (and for others we lack the necessary natural
history information), we summed the epidemic overlap values across years
for each site. Since we were purely interested in potential
cross-species transmission, we did not include loops as we did for the
networks described above (i.e., no connections of a host species to
itself). Additionally, we dropped very small networks (two or fewer host
species), because our measures of host influence were not applicable.
As a proxy for the influence of each host species on cross-species
transmission of different parasites, we calculated the eigenvector
centrality for the host nodes in each network. Eigenvector centrality
was useful because it considers the entire network and emphasizes second
order connections; a given node is more central if it is connected to
other nodes that themselves are more central (Bonacich 2007). If one
imagines a random walk across a network, the stationary distribution of
time spent at each node is proportional to the eigenvector centrality.
Eigenvector centrality scores range from 0 to 1, with 1 being most
central, and were calculated on weighted networks. We chose eigenvector
centrality because it emphasizes connections throughout the network,
which could be important in a community disease context where there
might be chains of transmission across different host species (Gómez et
al. 2013).