Table 2. Performance of the models during empirical cross-validation.
Unlinked model | Linked model | ||
Proportion of isolates correctly assigned | Actual | 0.52 | 0.64 |
Predicted | 0.82 | 0.64 | |
Bias | Chicken | −0.10 | −0.03 |
Cattle | −0.13 | 0.00 | |
Sheep | 0.20 | 0.03 | |
Pig | 0.01 | 0.00 | |
Bird | 0.00 | −0.01 | |
Rabbit | 0.00 | 0.01 | |
Sand | −0.01 | 0.00 | |
Water | 0.03 | 0.00 | |
RMSE | Chicken | 0.11 | 0.04 |
Cattle | 0.14 | 0.07 | |
Sheep | 0.21 | 0.07 | |
Pig | 0.01 | 0.01 | |
Bird | 0.03 | 0.02 | |
Rabbit | 0.02 | 0.02 | |
Sand | 0.01 | 0.01 | |
Water | 0.05 | 0.01 | |
Coverage | Chicken | 12 | 93 |
Cattle | 19 | 97 | |
Sheep | 5 | 97 | |
Pig | 76 | 99 | |
Bird | 86 | 99 | |
Rabbit | 73 | 100 | |
Sand | 93 | 99 | |
Water | 84 | 99 |
The unlinked and linked models are defined in the Methods. The predicted proportion of isolates correctly assigned assumes that isolates are assigned to their most probable source a posteriori. Bias, RMSE (root mean squared error) and coverage are reported for the proportion of isolates estimated to originate from each source. Coverage was defined as the number of simulations, out of 100, in which the true proportion fell inside the 95% credible interval.