# Also in the Article

Bias
This protocol is extracted from research article:
Using digital surveillance tools for near real-time mapping of the risk of infectious disease spread
NPJ Digit Med, Apr 16, 2021;

Procedure

The bias of forecasts is a measure of the tendency of a model to systematically under- or over-predict71. The bias of a set of predictions $I^jt$ at time t at location j is defined as

where the mean is taken across the N draws. H(x) is the Heaviside step function defined as

The above formulation can better be understood by considering the following extreme scenarios. If every projected value $I^jt$ is greater than the observed value $Ijt$, then the Heaviside function is 1 for all i = 1, 2, …N, and $meanHI^jt−Ijt$ is 1. The bias for a model that always over-predicts is therefore 1. On the other hand, if the model systematically under-predicts, then $meanHI^jt−Ijt$ is 0 and the bias is -1. For a model for which all predictions match the observed values exactly, the bias is 0.

Note: The content above has been extracted from a research article, so it may not display correctly.

Q&A