A.J. Burnett Projection

[aburt19]

I was looking at the projection for A.J. Burnett and was a little surprised. I watched him in several games and was not
impressed with what I saw. I felt that his final numbers for 2009 were not a reflection of what I saw (meaning that his
numbers were to going worsen in 2010). But I figured somewhere in the 4.50 range for ERA, 1.33 WHIP and maybe wins going
up. But the projections show him with an ERA a little over 5.00 and his WHIP at 1.37. I looked at his splits and figured that
maybe after the ASB he really had bad numbers. But while they weren't as good as the first half, they weren't really bad.

What was the reasoning behind the numbers being this high? I figured that he was a good candidate to bring up at the
auction to get someone to spend money because a lot of owners will look at last year and bid based on that.

[Todd Zola]

It is still quite early in the process, as Gary alludes to in the "Papelbon Plan" thread in the strategy section, the manner we are doing things is going to give you a little more info (if you choose to use it) than we have before.

Details will be forthcoming mainly from Gary, but here is the Cliff Note's version.

There are a few metrics that usually regress to the mean, especially for pitchers. The most well known is BABIP. Others are HR/FB and LOB% (the number of runners allowed that come around to score.)

While some pitchers seem to establish their own level (good or bad) anti- to the league average, given a large enough of a time frame, most will eventually regress in these areas to the mean.

For the past few years, Burnett has enjoyed a better than league average LOB%. That is, a lower percentage of his runners score, which lowers his actual ERA.

The higher than expected ERA is a result of our regressing his LOB% to the mean, along with a trending of his peripherals (K/9, BB/9) in the wrong direction and playing half his games in a HR park.

There is no "correct" level to regress. On this run of pitchers, we regressed everyone 50%. This is most noticeable with the top closers as their ERA is higher than intuitively one might expect.

I do not have access to our engine presently, but when I do, I will reduce Burnett's level of regression and provide the new numbers for comparison.

It is then up to the user to decide their philosophy. Can someone like Burnett bear down and pitch better with runners on base? Or is the high LOB% more luck?

We are still in the process of hammering this out, but there is a good chance that we supply (for the advanced user) sets of projections that differ by the level of regression. You can then think of this as a range of performance. A 0% regression completely relies upon the baseline the pitchers has established with respect to BABIP, LOB%, etc. A 100% will use the league average. As suggested above, we elected to use the 50% level for pitchers, meaning there is equal weight for the pitcher's historical levels and the league average.

If you wish to make the effort, you can adjust a projection to taste using these regression results. Over the course of the winter into spring, we will likely do this ourselves with the set we provide as the primary projection set. Our philosophy is to change these few and far between as otherwise the model becomes too subjective so why use a model? When we do make a change, it will be noted and explained.

Please do not treat this as a stop sign answer, discussion over.

Anyone -- please feel free to comment, ask questions, etc.

[Guest]

Todd is going to laugh at this. I apologize for the inside joke but AJ Burnett is now the Luis Gonzalez of pitchers.

I screwed up a line of the code in the projection machine. He's first alphabetically and a bad formula only copied to the line below him. I apologize.

The projection would have been (and will be) 4.23/1.37

That said, Todd's analysis is perfect of what things will look like going forward from a projection stance.

[Todd Zola]

At least I didn't have to try to explain this to a national radio audience

[cwk1963]

At least I didn't have to try to explain this to a national radio audience

That was pretty funny.

Todd, you mentioned there is no "correct" level to regress. Does this mean it's usually all over the board from year to year or there's no historical data to really be able to compare the levels on a yearly basis?

[Todd Zola]

To let everyone in on the inside joke, a few years back, I messed up and had the 2 Luis Gonzales with the same stats -- the good one was still hitting well for Arizona and the other one was a utility infielder for Colorado. Old friend Jason was on Jeff Erickson's XM show and they were taking calls. Someone asked Jas why we were so high on the Luis Gonzales from COL. Instead of hanging my fat ass out to dry, Mr. Grey talked about the potential for more playing time in the Rockie infield, some developing power and the Coors effect.

With respect to regression, I will let the professional, Gary, correct me if I am wrong, but is has to do more with there isn't a single level everyone regresses with. Given a large enough of a sample, most pitcher's will have a BABIP of .290-.310. One year it might be .270 and .330 the next. Then .318 and then .282. The average is around .300.

Then there is LOB% and HR/FB as well.

I am going to talk about this in a piece at some point, but our friends at Baseball Info Solutions track some fascinating data, some of which explains why some pitcher's always seem to outpitch their xERA and/or their normal peripherals like K/9 and BB/9. They track things like fielding and the ability to control baserunning. Mark Buehrle for instance is incredible at both picking guys off and having them break towards 2nd to be thrown out by the 1B, which is not a pickoff but a CS. Take away those outs from his WHIP and his xERA comes down. There are other examples.

The data is in the Bill James handbook -- which at this point contains so much more than just stats that it is a must-get for the advanced fantasy player, even though it is not a "fantasy" book.

The point is, with superior fielding, perhaps a pitcher's BABIP can indeed be lower than the league average, or at least regress to something lower -- which is the same as not regressing as strongly to the league average.

And perhaps someone like Buehrle can maintain a better than average LOB% as he erases many via his great move.

[Guest]

Todd, you mentioned there is no "correct" level to regress. Does this mean it's usually all over the board from year to year or there's no historical data to really be able to compare the levels on a yearly basis?

I'll do cliff notes here and then provide the full methodology in a site piece.

We use historical data to calculate the average regression of each skill set a pitcher posesses. We are currently doing more testing to see if different types of pitchers (age, role, etc) regress differently. Of course this is all "ON AVERAGE".

So say the average pitcher with a 3.00 ERA regresses to 3.50 in the next season (making it up)

If I have a pitchers with 3 straight years ERA of 3.00, 2.99, 3.01 - using the regression we'd project 3.50, but you have a 3 year history that says he doesn't regress. Therefore for this pitcher it might make sense not to regress him a lot. Alternatively a pitcher with less history or a different history you might regress more.

The reality is that we're going to be able to present a range of projections all of which are statistically valid, but at the end of the day, the user has to make a call obviously, which we have to provide our opinion on how to do so.

[aburt19]

It is still quite early in the process, as Gary alludes to in the "Papelbon Plan" thread in the strategy section, the manner we are doing things is going to give you a little more info (if you choose to use it) than we have before.

Details will be forthcoming mainly from Gary, but here is the Cliff Note's version.

There are a few metrics that usually regress to the mean, especially for pitchers. The most well known is BABIP. Others are HR/FB and LOB% (the number of runners allowed that come around to score.)

While some pitchers seem to establish their own level (good or bad) anti- to the league average, given a large enough of a time frame, most will eventually regress in these areas to the mean.

For the past few years, Burnett has enjoyed a better than league average LOB%. That is, a lower percentage of his runners score, which lowers his actual ERA.

The higher than expected ERA is a result of our regressing his LOB% to the mean, along with a trending of his peripherals (K/9, BB/9) in the wrong direction and playing half his games in a HR park.

There is no "correct" level to regress. On this run of pitchers, we regressed everyone 50%. This is most noticeable with the top closers as their ERA is higher than intuitively one might expect.

l

Please do not treat this as a stop sign answer, discussion over.

Anyone -- please feel free to comment, ask questions, etc.

After reviewing the pitching projections and comparing them to the historical data for the last three years, it looks like
the amount of the regression has taken the ERAs and maybe the WHIP a little too high overall. It's not a big problem
because when "Paul" (I know that's not the name for it, but I can't remember the name) comes out I can adjust the
pitchers that I think have been adjusted too much either up or down.

Thanks for getting the projections out this early. In these cold winter months when I'm cooped up, it helps to have
something that reminds me of spring and summer.

[Todd Zola]

After reviewing the pitching projections and comparing them to the historical data for the last three years, it looks like
the amount of the regression has taken the ERAs and maybe the WHIP a little too high overall. It's not a big problem
because when "Paul" (I know that's not the name for it, but I can't remember the name) comes out I can adjust the
pitchers that I think have been adjusted too much either up or down.

Thanks for getting the projections out this early. In these cold winter months when I'm cooped up, it helps to have
something that reminds me of spring and summer.

As alluded to in other places, there is a chance we can do you one better and you can do the adjustment by altering the level of regression. We have elected to do a 50% regression. If things proceed as planned, we may indeed offer a spreadsheet where you can input the level you desire. Less than 50 and the pitcher's historicals are favored. More than 50 and more regression to the mean is favored.

In addition, feel free to challenge us and perhaps convince us that we should manually adjust certain pitchers.

[aburt19]

As alluded to in other places, there is a chance we can do you one better and you can do the adjustment by altering the level of regression. We have elected to do a 50% regression. If things proceed as planned, we may indeed offer a spreadsheet where you can input the level you desire. Less than 50 and the pitcher's historicals are favored. More than 50 and more regression to the mean is favored.

In addition, feel free to challenge us and perhaps convince us that we should manually adjust certain pitchers.

One of the pitchers that I wondered about is J.P. Howell. The ERA projection is 4.114 and the WHIP is 1.343. In 2009 and
2008, his ERA was under three and his WHIP averaged in the 1.20 range. In 2007, his ERA was over 7.00 and his WHIP was
equally as bad. But that was as a starter and I would normally discount that year in forecasting 2010.

I'm not saying that his projected ERA to be below 3.00. I think that his ERA the last two seasons was lower than what it should
have been. I could see a 3.40-3.60 range for the ERA and a 1.25-1.28 WHIP.

[Todd Zola]

I'll ask Gary to chime in if he has some time as this is more up his alley.

Here's the deal with Howell. I'm not so sure the issue is starter versus reliever.

Hi '07 - '09 peripherals....

K/9 B/9 HR/9 K/BB
07 8.6 3.7 1.4 2.3
08 9.3 3.9 .6 2.4
09 10.7 4.5 .9 2.4

His K-rate has improved, but at the expense of his BB-rate. His K/BB is basically identical. Gary has done some research showing all K/BB are not the same, but in the range we are talking, the difference in effect of a 8.6/3.7 and 10.7/4.5 is not enough to account for the vast ERA swing.

So...

The disparity in HR is pretty significant, but can it be chalked up to Howell starting versus Howell relieving? I'm not so sure.

The REAL difference in the 3 campaigns is the respective hit rates.

07 69 in 57 IP
08 62 in 89.3 IP
09 47 in 66.7 IP.

I don't have the BABIP data, but it is pretty obvious that if you consider 1H/IP about average, in 07 he was above average or perhaps "unlucky" while the past 2 seasons he was below average or perhaps "lucky".

Again, is this due to hit coming from the bullpen in 2008 and 9? Hmm.

What is the primary difference, other than perhaps stretch/windup between starting and relief, especially for southpaws? Maybe percentage of LHB one faces?

Let's take a look.

07 54L 163R 25%L
08 165L 203R 45%L
09 74L 117R 39%L

So yeah, he faced more RHB as a starter and he was hit a little more, but if you look at the numbers, not enough to account for the disparity in BABIP, though the HR% is affected.

So where does that leave us.

Two questions.

One is should 2007 be weighed similarly to how we weigh the 3rd year in all our projections or should we soften the weighting to account for the switch from SP to RP?

The other is how much do we regress his BABIP? Presently, we do a 50% pitching regression. If we contend he has the ability to control BIP and HR/FB, and use 0% regression while not altering the weight of 07, his numbers are 3.93/1.33.

Still maybe a little higher than one would intuitively imagine.

Personally, I am not comfortable discounting 2007 as his K and BB rates suggest he is the same pitcher. I can live with the 0% regression, and going with 3.93/1.33.

I think the take home message, more than anything with Howell is expect some level of regression. Can we put an exact number on it? Sorry, no. But with the signing of Soriano, Howell is likely moving back to the 7th/8th inning role where he will go more than an inning on occasion and face more righties.

[aburt19]

I'm not that conversant with "regression analysis", so I'll ask a dumb question or two.

What is it that you are regressing? BABIP, K/9, BB/9, HR/9, ERA, WHIP or what?

Does it really make sense to use the same regression for the entire population?

Is the same regression done on the hitters? If so is it based on BABIP, HR/FB, GB/FB rates, K rates, BB rates, or what?

Is this regression analysis the same that's been done in the past? It appears that it's had a greater emphasis this year.
The same factors would have been in effect for J.P. Howell (just an example) last year and in that year he would have
had only one season of good stats with a lousy 2007 being in the more recent past. Yet his ERA for 2009 was projected
to be a 3.84.

I agree that it's not possible to completely ignore a year when looking at the last three years data to get a baseline for the
current projection. But there is no doubt that some pitchers for some reason just shine in the bullpen after being a mediocre
starter (Dennis Eckersley comes to mind).

Again J.P. Howell is just an example. The same thing applies to every pitcher.

I'm not trying to be a problem here. I just want to make sure that I understand the methodology of the projections.

[Todd Zola]

We will be posting a series of essays regarding your questions with the next update. Is it okay if we wait a week, let you digest them and we will welcome any follow up questions?

But to address a couple of your points. It is easy to cherry pick players and make anecdotal arguments to make a point. We prefer to look at things in a statistical view.

Now again with Howell, looking JUST AT HIS PERIPHERALS, K/9, BB/9 and HR/9, he was basically THE SAME PITCHER in 2007 as he was in 2008 and 2009, he just got different RESULTS.

RESULTS are not always a direct reflection of the skills that went into them. As pointed out, the primary differences with Howell '07 and Howell '08/'09 were HR/9 and hits allowed -- both of which have elements of luck. It should be noted at this point that the Rays defense has been amongst the league's best, so a low hit rate is not all luck, it is partially based on outstanding fielders backing the Tampa staff.

A certain skill level USUALLY translates to a certain result. We are carrying over a skill level from 2007 that is perfectly in line with the skills displayed in 2008 and 2009.

The RESULTS from 2007 and 2008/2009 were clearly different.

The regression is primarily for things "out of the player's control", most notably BABIP and HR/FB for pitchers. These are the 2 areas Howell was most fortunate, so they are also the areas he will be most hurt when we regress.

Our philosophy and model is to look at the big picture, use our global research and apply it consistently to all pitchers.

If we go through each pitcher and subjectively alter them, why develop and test a model? If we always make the same alteration, based on the same stat, that gets built into the model.

Does this mean of there are 10 pitchers with Howell's stats and skills, all 10 will regress as we suggest?

No, we will miss a few.

But we don't know FOR SURE which ones, or that would be part of the model.

Keep in mind the projections are only PART of the story. It is still December. Sometime in January, we will release accompanying profiles with a sabermetric lean to them, which will help keep the projection in perspective, especially with cases like Howell.

Now, all this said, will we ever go against the model and make subjective changes?

Yes.

But it is my opinion that the evidence is circumstantial with respect to Howell that 2007 should be discounted because he was a starter. His K/9 and BB/9 were in line with what he displayed the past 2 seasons. The results he incurred need to be reflected in the possibility of results he can incur again.

Now, with all THAT said...

What aburt or anyone else for that matter can do is completely different than those producing a global set of projections.

You guys can be as subjective as you want!!!

It is simply our philosophy to produce an OBJECTIVE set of projections as possible, employing intensive research based on multiple seasons of data, giving the most probably statistical result based on the available input.

MOST PROBABLE!!!

Not 100% definite.

You can downgrade a guy because you heard he parties too much at night and has a large percentage of day games. You can upgrade a guy because he had a kid and think he will be more mature. That's the fun of this game. We feel it is a DISSERVICE to include anecdotal or subjective analysis of that nature in the FORMAL PROJECTION SET.

Truth be told, others disagree with that last statement. Some of these others are customers that wish we would "go out on a limb." more often. Some of these others are fellow members of the industry that pride themselves on "going out on a limb."

To wrap this up (for now, again, we will be providing some in-depth descriptions of the model next weekend), in professional wrestling, there is a term "Kayfabe". This is used to describe when the wrestler is out of character. For years, Gary and I have talked about doing a "kayfabe" piece, where we each take off the sabermetric hat and "shoot" (another wrestling term). Basically, we would go over some of our personal gut feelings and subjective analysis. For a few reasons, we have not yet done this. If I were a betting man, I would put a few sheckles on reading a piece or two this spring where we shoot from the gut on a few players.

[Guest]

Back from Christmas break.

As Todd says I will be completing a thorough site piece soon on the regression process. However, we can discuss Howell here for now.

Our methodology for dealing with pitching projections has changed somewhat from 09 to 10 so for now lets ignore the 09 projection in the discussion (though thumbs up to you for holding us accountable to it).

Howell is a BABIP and LOB% wonder, simple as that. He relies on a .260ish BABIP and 77-80% LOB% in order to put up those numbers.

I don't like the rise in his walk rate, even though the K rate went with it and he's a good G/F guy.

You could argue we should pull him downwards, I could argue that with that K, BB, and HR rates he is a 4.10 guy with league average LOB and BABIP rate, ignoring the regression of the K, BB, and HR rates we are adding to the projection.

Basically you're paying for him to continue to defy gravity somewhat. Not that it's a terrible bet given the last 2 years but the projection isn't terrible even if it's maybe a tweak higher than you'd expect.

[aburt19]

[quote="GaryJ"]Back from Christmas break.

As Todd says I will be completing a thorough site piece soon on the regression process. However, we can discuss Howell here for now.

Our methodology for dealing with pitching projections has changed somewhat from 09 to 10 so for now lets ignore the 09 projection in the discussion (though thumbs up to you for holding us accountable to it).



/quote]

Thanks. for the response. I really wasn't trying to hold you accountable for 2009 because everyone doing projections is
off on some players. It was more to highlight what you have confirmed is a change in methodology in handling the pitching
projections.

This wasn't really to comment on the projection for Howell. He was just one that stuck out as an example. It was more
about the projection process and not a specific player. In fact it was really about the regression process. I've always been
very satisfied with the projections in the past. It just seems like the change in methods for handling pitchers has certainly
affected a fairly large number of pitchers. I'm not saying that is either good or bad, just that it has affected them.

[Guest]

You are right, though it's an evolving process which is why we're very open book with everything we do.

In kayfabe terms I wouldn't pay for more than 3.75 from Howell but there's upside there from a practical sense.