2.7 Sensitivity analysis
To more fully explore the relative importance of each model parameter in
affecting ASF persistence, we also performed a regression-based global
sensitivity analysis (Saltelli et al., 2008; but see Nsoesie et al.,
2012 for an application on individual-based epidemiological models).
Once more, we ran 1,000 iterations of the model, this time randomly
selecting each parameter from a uniform distribution ranging from the
optimized parameter value to its 150% values. For each iteration, we
recorded the number of days of virus persistence in the population.
Then, we visually checked the linearity of the relationship between the
input parameter and the resulting virus persistence. Following, we
standardized all input parameter values using the z-score method
(Kreyszig, 1979), and performed a generalized linear regression using
virus persistence as the response variable and the standardized model
parameters as predictors. To account for the overdispersion in the data
we used a quasi-Poisson distribution for the response variable. The
regression coefficients of each predictor provided an estimate of the
sensitivity of ASF persistence to changes in that parameter (Saltelli et
al., 2008). The z-score standardization made all regression coefficients
comparable, although the initial model parameters were measured on
different scales.
RESULTS