Why isn't the response, "Scrap it"?

What I can tell you though is that when a predictor is made more complex, it must add more predictive value, by a significant amount. If it does not, then my only response can be just two words:

Occam’s Razor.

Or maybe it's an indication that the underlying behavior is more complex that your simple model of the relationship.

-- MWE

What I can tell you though is that when a predictor is made more complex, it must add more predictive value, by a significant amount. If it does not, then my only response can be just two words:

Occam’s Razor.

I don't necessarily agree with this. All these FIP/xFIP etc are already way past the point that the avg. fans (even stat inclined) can calculate them himself. With modern computing power, who cares how complex it is?

If a full projection system like ZiPs/Oliver/Cairo is necessary to get the best estimate, why do I care if it's complex?

If a full projection system like ZiPs/Oliver/Cairo is necessary to get the best estimate, why do I care if it's complex?

Yeah, it's not like I make you guys run ZiPS! Whether you like or don't like the projections, for the end-user, the only complexity is looking at a website.

Yeah, it's not like I make you guys run ZiPS! Whether you like or don't like the projections, for the end-user, the only complexity is looking at a website.

Question, Dan. Is ZiPs a full blown regression/forecasting model?

What I can tell you though is that when a predictor is made more complex, it must add more predictive value, by a significant amount. If it does not, then my only response can be just two words:

Is Fip better than component era by a significant amount? I thought it was barely noticeable difference.

Tom's postulate:
differences in pitching projections are not primarily driven by differences in ERA estimator systems
differences in pitching projections are primarily driven by differences in change-over-time, park-and-defense effect adjustments, and possibly minor-league weighting of estimator systems

(edited)

I like having all of the estimators available. I think the mistake is to fall in love with one particular estimator as the be all, end all method. My philosophy: look at all of the results and use your judgement in how to weight them. If one method produces an outlier result for a pitcher, think about why that may be the case, and either reject the outlier or use the outlier to temper your view of the other methods (depending on the circumstances).

Generally, FIP and x-FIP are pretty good (and produce similar results to SIERA or tERA), but I think there are occasions when factors like groundball tendency are important, and the more complex estimator may provide additional useful information.

Question, Dan. Is ZiPs a full blown regression/forecasting model?

I've heard Dan say that ZiPs relies on Monto Carlo simulations, which I believe doesn't rely upon any linear or non-linear regression models.

I've heard Dan say that ZiPs relies on Monto Carlo simulations, which I believe doesn't rely upon any linear or non-linear regression models.

How does he get the means and distributions for the simulations? i.e. how do you translate past stats into future probabilities w/o regression?