Originally Posted by
JellySlayer
Again, some of them are pretty far off. A troll barbarian gets ~10-12 points higher To than the predicted values, and a similar amount of St more. In less extreme cases, you still see fairly consistent differences of several points in all stats. An average human fighter starting with an average 17 To instead of a predicted 14 is a fairly significant effect. Also, it's worth pointing out that background and questions should only affect the scatter, not the averages, since some of these will lower your stats, not raise them. The Guidebook values don't predict the averages correctly though--they systematically underestimate the PC's stats.
More to the point, as I've described in the indicated thread, I think that the algorithm used to generate the table is incorrect, and have demonstrated a method that seems to correctly predict the starting stats of a given race/class combination with significantly higher accuracy. Given that we can predict the starting stats with much higher accuracy than the Guidebook table, I don't see any reason not to take advantage of that.