5. Troubleshooting Models

5.1. Is the model wrong, or am I?

If a model doesn’t meet expectations then it is possible that either expectations or the model are out of line with reality. The difficulty is, of course, how to tell which. In cases where predictions can be compared to actual measurements, these can be used to determine what is wrong with the model. Are all outliers in a cycle model on a certain type of road, for example? The model is probably over or underestimating the effect of traffic. All in a certain part of town? The model could be misestimating the weight of a particular origin or destination. Isolated errors? These can be caused by the spatial network model misrepresenting what is actually there; for example, roads disconnected in the model when they are connected in reality.

In other cases, we have no predictions with which to be sure, but the model doesn’t seem to match expectations. Two techniques can be used to drill into why the model predicts what it does; and hence help decide whether the results are likely to be correct.