By scoring proportions, we're ignoring the inherent difference in variability between days with differing numbers of counts (and perhaps ignoring the fact that a proportion's sample variance is a function of the proportion?). For a first pass this is fine with me. But in the long run, I'd like us to be able to score predictive counts too.
By scoring proportions, we're ignoring the inherent difference in variability between days with differing numbers of counts (and perhaps ignoring the fact that a proportion's sample variance is a function of the proportion?). For a first pass this is fine with me. But in the long run, I'd like us to be able to score predictive counts too.