Biasing by a constant amount might help. The thing about unbiased random noise is that it cancels itself out! And for the most part, we can compute exactly how

Biasing by a constant amount might help. The thing about unbiased random noise is that it cancels itself out! And for the most part, we can compute exactly how many data points are needed to achieve a particular level of precision in the result.