Fig. 4.

The support and contribution of PEDV diffusion predictors among 12 countries. Among the predictors being considered, four were estimated to have a very strong impact on the global spread of PEDV: live swine trade between provinces, sample size at origin, sample size at destination, and swine population size at destination. Support for each predictor is represented by an inclusion probability that is estimated as the posterior expectation for the indicator variable associated with each predictor (E[δ]). Indicator expectations corresponding to Bayes factor support values of 3, 20, and 150 are represented by a dotted vertical in this bar plot. The contribution of each predictor is represented by the mean and credible intervals of the GLM coefficients (β) on a log scale conditional on the predictor being included in the model (β|δ = 1). We note that there is a high degree of correlation between some of these variables including sample size (see supplementary fig. S11, Supplementary Material online). As sampling sizes are expected to have an impact on the number of location transitions, we considered origin and destination sample sizes as separate predictors in our GLM. However, even when taking into account sampling biases, we still found support for other factors in addition to sampling size predictors, which may suggest that they are robust (Lemey et al. 2014). Furthermore, including or excluding the sample sizes in the GLM analysis had minimal to no effect on the geographic reconstruction, further indicating robustness of the results. Only predictors whose credible interval excluded zero are shown.