|
Re: Table is still trash |
Sorry to butt in here: 1) Following a defined procedure such as that used for calculating magnitudes is not subjective. You can call it sub-optimal if you wish, but not subjective. 2) Because of the variation in magnitude scales, my preference is to use whichever value most favours the predictor. We are usually only talking small differences, and to rule out a prediction as valueless on a technicality looks poor. 3) As I argued in SRL a while back, the key issue is whether the predictor's record is better than chance. You can use whatever criteria for scoring near misses the predictor likes, but then apply the same criteria to a series of random predictions that match the scope of the real predictions. The significance of the predictor's results is a factor of how better they are than the random score. The easier the predictor makes the criteria for a hit, the easier it is for the random predictions to be hits as well, and that way the net significance value remains constant.
Follow Ups: ● Re: Table is still trash - Roger Hunter 09:13:35 - 3/26/2001 (6387) (0) |
|