Essentially, you are calculating z-scores for each category and summing everything up to get a total ranking. The problem is that the z-score method assumes that all categories are normally distributed. If categories are normally distributed, you could then actually convert z-scores to percentiles. For example, a z-score of +2 is at the 98th percentile. However, some categories (e.g., stolen bases and saves) are clearly not normally distributed. Thus, players with very high totals in these categories (i.e., outliers) are going to have total scores that are inflated.

Since projections tend to predict more clustering around the mean than actually occurs (read: smaller standard deviations), it will tend to inflate the value of a marginal stat.

If in the last three years, a marginal SB is worth about 1.5 HR, but your projections have a marginal SB worth 1.2 HR, then you’re going to underestimate SBs in your valuation.

I still see some problems with this, but would it be better in assessing individual values of players? I guess in fantasy maybe we don’t care that the playing time for the entire player population totals correctly, since we are only concerned with the top 300 or so players?

As I understand it, most advanced projections use conservative playing time estimates, so that overall the league AB & IP approach what would be a normal total. Since the projection systems aren’t projecting injuries, this even things out on a league-wide scale.

Of course, I’m still new at understanding all this stuff. Maybe I’m missing something.

This results from using the standard deviations of the projections when doing the standard score calculations rather than using the standard deviations of actual season data.