THE BOOK--Playing The Percentages In Baseball

Great analysis!

I like this approach, but the Fielder market should also be factored into overpay discussions.  If no other offer was within $50M of the Tigers offer,* then the Tigers overpaid by a substantial amount, even if the $/WAR calculation works out favourably over the course of the contract.  In other words, they could have potentially had the same player, same production for less money.

*do we know what the other offer(s) was/were?

Nice analysis - Interesting perspective. 

Of the 10 comparables on the list, the worst total Rbaser in their age 25 to 27 seasons are Thomas and Thome at -6, and except for Bagwell (+12), all were in the range of -6 to 5.  Prince is -17.  Defensively, using Rtot over the same period, Prince is -16 and only Thomas (-36) and Thome (-20) are worse. Thomas’ last year with over 100 games at first was his age 28 season; Thome’s last was his age 33 season.

The Tigers are entering uncharted territory.  Fielder is so slow and has such little range, that I question that he will age as well as many on the list.  Only his dad was really similar in build, and aged the worst of that list.

I think Ilitch’s thinking is “I’m 82 and really don’t care about his production after I’m gone!”

There is absolutely no reason to remove Cecil from the list. I don’t think you should have even mentioned that. Why would you? An integral part of a player’s projection is the chance that he tanks or gets injured such that he has to retire early. That is the reason that the LAST think you want to do is eliminate a player who is like that from your list of comps, unless that list is short.

In any case, the list is short enough such that the certainty on that 31 WAR is quite low I would think.

And, I know you did not do this intentionally, but if we look at his prior seasons to age 25 and his average WAR per season is not nearly as high as 25-27. Even with an age adjustment, I think it is lower (I could be wrong). You cannot ignore information even if you didn’t do it intentionally. Here is an example: Let’s say that I look at 2 players’ last 3 years and they both have an average of 3 WAR per season. Then I find out that player 1 was 1 WAR per season in the 3 years before that and player 2 was 4 WAR in those prior 3 years. Who do you think will have the better projection?

So unless the other players on your list also had around the same average production in seasons prior to age 25, then you have to adjust your list or adjust Fielder’s projection as compared to the average of your list.

I want to add that using players with comparable numbers like that IS the best (and most transparent) way to estimate a player’s value going forward, in general. I say, in general, because it only answers the question, “What if I have a first baseman who averaged around 14 WAR in their age 25-27 seasons, what can I expect over the next 9 years?”

That does NOT necessarily answer the question, “What if I have a first baseman who is very slow, has also averaged 2.7 WAR per season at ages 22-24 (and whatever else you know about him that would be relevant to a short and long term projection)...?”

The second thing is that depending on the number of players in the comp list, there could be a much larger uncertainty in the projection than using a traditional projection technique or even expanding the list of comps and then tweaking the numbers. For example, if our list only has 8 players, surely we can use another 20 players who averaged 3.5 WAR per season rather than 4.5 to help us in our analysis…

Why would you remove Cecil when he is the most relevant comparison because he shares more in common with Prince than the other guys? It’s like trying to project what a kids height will be growing up by comparing him to other white adults, but excluding his mother from the comparison.

Cecil is a bit odd.  He missed one season (I think he was in Japan) at age 25.  It’s not clear therefore that he would have remained a comp had I looked at age 22-27.  So, he just didn’t fit into the rest of the group.

So, if someone wants to repeat for ages 22-27, I doubt Cecil will be in there, while I think at least 6 of the others will still be there., and the other 3 will probably be pretty close.  I think.

Basically, Cecil barely made it in.

Nice analysis, simple but appropriate.

I think some of the negativity is due to Princes weight and similarity to Mo Vaughn in appearance, call it a fat bias if you will.

Does that WAR include baserunning?

I’ve always said that using comps is the ultimate (and only “real” way) to do a projection. However, unless you can get a very large sample of players who share almost all the relevant characteristics with the player in question, it is going to yield a result which has a much lower certainty than that produced by a good, comprehensive projection methodology that uses lots of players of different stripes to develop that methodology.

Some combination, like presumably Pecota uses, seems like the perfect (at least as perfect as we can get) answer to me, but judging from Pecota’s results and problems over the years, that doesn’t seem to work too well either.  Then again, it could just be Pecota and not the idea itself (combining a classic projection methodology with a “comp” one)…

@7:  Yes, Cecil was in Japan (Hanshin Tigers)

@9:  In the numbers I looked at, baserunning was included in WAR

I agree with MGL/10.

After all, imagine you are lucky enough to have Andruw Jones as your comp if you are 26 years old, but unluckly enough to have him as your comp if you are 30 years old.  Or you have Edgar as your comp if you are 38 years old.  Do we want to forecast Strasburg, Felix and Kershaw based on their comps to Gooden?  And what lucky strike-zone challenged 25 yr old pitcher is going to get RJ as his comp?

Yes, if you get 10 or 20 or 100 comps, that mitigates some of that.  But, does the comp-process really buy you anything over-and-above Marcel?  Maybe there’s some particular combination of speed-youth, or position-age which a standard forecast doesn’t consider, but the comp-process (implicitly) models.  Sure.  But, where you win there, you may lose elsewhere.

The PECOTA results shows that there’s little to no gain there.  And I think Dan said ZiPS also uses comps?  (I seem to remember seeing him reference comps I think.)

There is one place where comps helps, and that’s in visualization.  The article I wrote about Jeter for ESPN two years is a perfect example.  I can still remember the particulars of that.  And this one of Prince Fielder will last a while in my memory.  It gives the forecast a shape and flavor.  It’s alot harder to argue against a Prince Fielder forecast if I rely on specific example of history, rather than me saying that the pattern of history can be reduced to a Marcel-model.  No one likes to debate a black box.  We all want to debate the specific circumstances that perhaps makes Prince and Jeter unique, something a black box simply models out of existence.

But a chocolate bar is still a chocolate bar.  It’s tasty, but that’s still not a meal.

I note that the r-squared of the comparables ("old" vs. “young") is 0.27.  Maybe a combination forecast should be 70-75% Marcel and 25-30% comparables.  (Maybe.)

The variability of WAR in the future is much greater than the variation in the past.  (Take the standard deviation as a percentage of the mean—it’s 23% for the young hitters and 48% for their older selves).

Also, lowering the standards for good predictions might be in order.  The standard deviation for the old sluggers is 48% of its mean, so if you assume the comparables are a true sample and predict Fielder to have 31.2 WAR going forward, you are really predicting that 2/3 of the time he will have between 16.4 and 46.1 WAR.

Yes, definitely, there will be wide variation in what we will observe.  Six of the ten are within 10 WAR or so of the mean, and that’s what I expect would happen.

Median is 35.5, if that obviates the Cecil factor. Also makes the $200MM mark more palatable.

Since we’re only talking about 10 data points, obviously, we’re subject to the one extreme data point.  If what happened to Cecil normally happens 10% of the time to any kind of superstar in their 20s, then fine, keep Cecil in. 

A better thing would be to look at ages 23-27 for 1B with at least 16 WAR AND those ages 25-27 with at least 10 WAR, AND at least 600 PA at age 27, and look at the last 30 players to match, then you’ll get a better set of results, and more “smooth”.  We can see if Cecil-like players are 10% of the sample or not.

Does someone want to do this?

And you have to look at the average, not the median.

From what I gather at BR.com, there are only 13 people that meet that criteria, two of which are Pujols and Miguel Cabrera. The other 11 are Palmeiro, Todd Helton, Frank Thomas, Eddie Murray, Norm Cash, Will Clark, Fred McGriff, Steve Garvey, Earl Torgeson and Mattingly.

Average WAR from 28-36: 28.0