Furthermore, a model can achieve an excellent fit to data for reasons that have nothing to do with the model’s fidelity to the underlying biological process. For an extreme example, in [12] we show that nearly any positive function is the I(t) output of an SIR model with variable transmission rate (t). However, if the output of a model does not mimic the data (e.g., Figure 6), then the model is likely to be missing at least one important component. In this sense, a deficient model may be more helpful than a model that fits the data.