Updated ZiPS Projections

Jacobs isn’t a great example because his sample size will be sufficiently large that a month of data shouldn’t change much even if it’s pretty extreme, but we’ll go with it anyway. Let’s say you project Jacobs to hit 24 home runs in 2009 and through 25% of the season, he’s still sitting at 0. Your projection for the year will no longer be 24 HRs because you KNOW what his production is for 25% of the season. Assuming your assessment of his ability to hit home runs hasn’t changed, you’ll apply the same HR rate and project him to hit 18 HRs for the year (75% of 24). But you don’t just average your same projections against the known 25%—that 25% is useful data, so you can use it to update your projections and get a better idea for what you expect going forward. Maybe Jacobs’s 40 games without a HR has convinced your model that it shouldn’t expect a 24 HR/season rate going forward—it should only expect a 20 HR/season rate. Using the new projected rate for the rest of the season, you’d now project Jacobs’s season HR total at 15 (75% of 20).

A month of data doesn’t tell you much, but it’s not like it’s wrong to update your model more than once a year, nor would updating your model be anything like “admitting mistakes” in your previous projections. Projections are just estimated probability densities. It’s really a silly criticism to suggest that the fact that projection may be updated midseason to incorporate new data indicates a flaw in the system.