In the context of evaluating predictive models, how do mean squared error (MSE) and mean absolute error (MAE) differ in terms of sensitivity to outliers?

Mean squared error squares each error before averaging, so a large error from an outlier becomes much bigger and dominates the score; this makes MSE very sensitive to outliers. Mean absolute error simply averages the absolute size of errors, treating each deviation linearly, so a single extreme value only adds its magnitude to the total and is less influential. For example, if a model predicts 10 when the true value is 0, MSE would add 100 to the sum, while MAE would add only 10, showing the larger impact of outliers on MSE. Because of this, MSE is preferred when you want to penalize large mistakes heavily, whereas MAE is chosen when you want a more robust, outlier‑tolerant measure.