Sensitivity Analysis Methods

Since population density of some hosts was coarsely estimated, we tested the model for sensitivity to inaccuracies in density estimation by randomly choosing a density for each species in each fragment from a range equal to our estimate ± 10% for mice and chipmunks and our estimate ± 50% for all other species. A range of predicted NIP was then generated for each fragment by calculating NIP using the actual species detected in that fragment and the randomly generated densities, iterated 100 times. Since inaccuracies in density estimation are likely to apply to some species in some fragments and not necessarily to all, randomized density estimates varied independently for each species in a given fragment. The actual species richness and the identity of the species present remained unchanged for each fragment. The resulting range should assess how robust the model is to errors in density estimation.

Sensitivity Analysis Results

The results of the sensitivity analysis indicated that the model is robust to considerable inaccuracy in density estimation. When the density of each host species was allowed to vary independently by ±10% (for mice and chipmunks) or ±50% (for all other species) of our original estimates, the maximum predicted NIP of the 100 iterations for each fragment was comparable to the original model prediction (deviating by 7% of the prediction on average), and the correlative relationship between the maximum predicted NIP for each fragment and observed NIP was preserved (P = 0.008; R = 0.51). The minimum NIP of the 100 iterations per fragment more substantially underestimated NIP when compared to the original model estimates, deviating on average by 24% of the original predictions, indicating that the model is not entirely insensitive to multiple errors in density estimates. Instances where the sensitivity analysis estimates differed greatly from those predicted by our measured densities occur when the randomly generated densities of influential dilution and reservoir hosts were changed simultaneously in opposite directions (for example, when mouse density was reduced and squirrel density was increased). Nevertheless, like the maximums, the minimums of the 100 iterations were significantly correlated with observed NIP (P = 0.02; R = 0.45) and the randomly generated densities produced NIP estimates that were significantly correlated with observed NIP in 99 of the 100 iterations (range of P values: 0.0015 – 0.053; range of correlation coefficients: 0.59 – 0.38).