Actually, negative scores isn't the right way to do it. What we want is a number representing the "value" of a predictor's ability. It should range from 0 to 1.0, no value to perfectly accurate.

But I suppose a negative score could mean someone who was always wrong.....

Roger