Global patterns in coronavirus diversity

Anthony, Simon J.; Johnson, Christine K.; Greig, Denise J.; Kramer, Sarah; Che, Xiaoyu; Wells, Heather; Hicks, Allison L.; Joly, Damien O.; Wolfe, Nathan D.; Daszak, Peter; Karesh, William; Lipkin, W. I.; Morse, Stephen S.; Mazet, Jonna A. K.; Goldstein, Tracey

A binary regression model with logistic link function was used to investigate whether the degree of host switching varied by region. The dependent variable was host switching, and either cospeciation or sharing was the reference group. Region was the independent variable. We used the generalized estimating equations (GEE) (Liang and Zeger 1986) method in order to account for multiple event types within one family and derive a robust estimate of the standard errors. Event data for all four reconstructions were aggregated for the analysis. Where host-virus associations were duplicated, one was removed. If the duplicated event types did not agree [which can occur given the sensitivity of these methods to variation in topology and number of taxa in each tree (Conow et al. 2010 )], we randomly selected one of the two possibilities and repeated the randomization 100 times. The GEE estimates from the 100 random iterations were then combined using rules established by Rubin (1987) . The analysis was repeated to explore variation based on bat family (independent variable). Some families only possessed one type of event, which would introduce quasicomplete separation into the GEE model, so they were excluded from the corresponding models. However, the numbers of bat species within these excluded families were very small (<5) compared to the overall number of bat species for the whole study, so we felt this would not impact any overall trends.