A type of distance metric which is a variation of the Manhattan metric. It includes a normalisation step which divides the Manhattan distance by the number of dimensions involved, giving $\text{MAE}=\frac{1}{n}\sum _{i=1}^n\left|a_i-b_i\right|$. MAE is used primarily in statistics to compare data sets, for example by testing the accuracy of predictions against a collection of measurements. It also makes it possible to compare distances measured in spaces with different number of dimensions. MAE employs symmetric penalisation, in that accuracy improvements and worsenings of identical magnitude cancel each other out, resulting in identical accuracy evaluations. If a specific context requires a worsening to matter more than an improvement, the Mean Squared Error or Root Mean Squared Error metrics are a better choice.