Pecota 2013

[Pickles]

Nate cheated.

You can't cheat baseball predictions.

Not to derail this thread, but I haven't heard any kind of "cheating" allegation against Silver. Care to explain.

[flashjordnk]

This stuff is filler after all. Speculative and abstract. Baseball is played on the field,projections mean little but fodder for web sites and water coolers. Im not ANTI STATS just so you know, but

I think the likelihood of Nate Silver(one of the founders of PECOTA) calling an election and comparing that to calling a pennant race is just not wise. Too many things can happen. And BTW,

politicians dont usually get injured like baseball players.

I mean not to derail too, but (a) I want to know how Silver cheated.

And two, I don't think predictions post primaries are as difficult to call. But I think trying to predict both parties candidates pre-primary completion and a winner would be similar.

On the other hand, I don't think Nate thinks of the predictions as the same. I think he'd tell you his predictions would be more like taking the results of the year, meshing them against his models to predict to start the year and then adjusting them (daily) would be more similar to what he does in politics.

[El Gordo]

Not to derail this thread, but I haven't heard any kind of "cheating" allegation against Silver. Care to explain.

I think he means that if Nate Silver is your partner and you find strange underwear in your drawer, there's a 29% chance he cheated on you.

[LookitsPuck]

We only scored 712 runs last year. So, the runs scored is probably in line.

We allowed 705 runs last year. So a near 80 run regression (roughly worth 8 wins) is an interesting take.

We really need bigger bats, our line-up just doesn't cut it, really.

[El Gordo]

We only scored 712 runs last year. So, the runs scored is probably in line.

We allowed 705 runs last year. So a near 80 run regression (roughly worth 8 wins) is an interesting take.

We really need bigger bats, our line-up just doesn't cut it, really.

We need more men of base. The bats are big enough.

[Frobby]

We only scored 712 runs last year. So, the runs scored is probably in line.

We allowed 705 runs last year. So a near 80 run regression (roughly worth 8 wins) is an interesting take.

We really need bigger bats, our line-up just doesn't cut it, really.

I don't see how the last sentence follows from what precedes it. If we allow 785 runs, that will be the problem, not the hitting.

[weams]

This stuff is filler after all. Speculative and abstract. Baseball is played on the field,projections mean little but fodder for web sites and water coolers. Im not ANTI STATS just so you know, but

I think the likelihood of Nate Silver(one of the founders of PECOTA) calling an election and comparing that to calling a pennant race is just not wise. Too many things can happen. And BTW,

politicians dont usually get injured like baseball players.

Nate Silver is a great analyst of trends. He seems pretty spot on with numbers. I am really surprised that both Orioles and As fare so poorly in one of his models. There MUST be some things that are not being taken into consideration. I wonder if in season HGH testing will skew any on the historical trending. Are their unaccounted for variables?

I hope that Buck and Dan can be that.

[Can_of_corn]

Nate Silver is a great analyst of trends. He seems pretty spot on with numbers. I am really surprised that both Orioles and As fare so poorly in one of his models. There MUST be some things that are not being taken into consideration. I wonder if in season HGH testing will skew any on the historical trending. Are their unaccounted for variables?

I hope that Buck and Dan can be that.

They haven't been "his" models for a few years now.

[flashjordnk]

The thing to remember is that when a team like the Orioles surges ahead in performance like this, any statistical system is going to assume regression because the career numbers would represent a bleaker picture. But in this case they do so without any analysis as to why, other than the fact that their previous performance was bad. The model can't/won't ask the question, WHY did they improve.

The model seems pretty heavy on regression.

Hammel is a really interesting guy in the mix this year. He's always been a popular sabermetric guy. In 2009 and 2010 he had solid seasons with the Rockies, including sub-4.00 FIP years and according to fangraphs was an 8 WAR player during that time, BBref gave him 2.2 WAR for the same period.

It's not at all surprising that Hammel's ERA is a bout .2 ERA lower than his career averages. With the way models and regressions work, that shouldn't be a surprise. PECOTA doesn't necessarily account for the development of a new pitch or a new pitching approach and is willing to shirk his good performance last year based on his career. Bill James for example is also predicting a reduction in the K/9. To me that's the portion that Hammel is going to have to prove. His contact rates last year were about 7% lower than his career. Was that an anomaly or was that sustainable.

Assuming a bigger workload than last year and regression, that's 40 additional runs that these metrics are attributing Hammel to allow. Is it wrong/unfair of them to do so? Absolutely not. But I think what we saw from Hammel last year wasn't just a flash in the pan. I worry more about his knee and his health. But think we saw what a lot of people thought he could be for a long time.

Chris Tillman. Same analysis. Here's a guy with a 4.73 career ERA. Granted over many fewer innings. Never putting more than 65 innings alone in one season on the books in the MLB prior to this he's a guy I would have thought they'd projected a bit more positive for this upcoming year. He's a guy who was as high as the #22 prospect by BA. With a composite sub 4.00 ERA in the minors, you'd think they could project something closer to 4.00 for him in 2013. I think he's obviously due for some regression, but again there is some pretty substantial grey area here. 60% of his innings have been logged at a 5+ ERA so you can see why a projection model might have doubts for the guy. BUT there I'm guessing Tillman has much higher tails for his outlying performances available to him.

I think this is one of the circumstances where it'd be really nice to see a confidence projection of these guys. PECOTA runs thousands of models and composites them. But let's say the models reflect his current career path, a 65% chance of 5+ ERA and a 35% chance of a sub 3.00 ERA, sure the "average" will result in a mid 4.00's projection. In reality I imagine if PECOTA ran 10,000 iterations of the season, Tillman would have had fewer 4.60 ERA seasons than 5.00 ERA seasons. Because they have to aggregate the data, there is definitely some information lost in there.

Adam Jones hasn't OPSed less than 785 in the last two seasons and yet going into his year 27 prime age year, PECOTA projects him at 772.

I'm not saying these models are wrong, but taken on a micro level, there are some clear areas that leave their "predictions" in pretty strong doubt if you ask me.

[waroriole]

Not to derail this thread, but I haven't heard any kind of "cheating" allegation against Silver. Care to explain.

I don't even know how he could cheat that, unless he rigged every voting booth.

[Frobby]

. In reality Adam Jones hasn't OPSed less than 785 in the last two seasons and yet going into his year 27 prime age year, PECOTA projects him at 772.

I'm not saying these models are wrong, but taken on a micro level, there are some clear areas that leave their "predictions" in pretty strong doubt if you ask me.

I remember in February 2009, PECOTA predicted a huge dip in Nick Markakis' performance, which seemed to make no sense as he was entering his age 25 season and had improved each year he'd been in the majors. There were howls of protest here, including from me. But PECOTA was right.

It's important to remember that the "weighted mean projections" we post and react to on here are just what they say they are. They're just the weighted midpoint of all the possible projected outcomes. Some guys are going to do much better, and some will do much worse. PECOTA actually gives projections at different levels of probability for each player, but those details aren't usually posted here, nor are they published in BP's annual, they are only available online for paying subscribers to BP.

[DrungoHazewood]

Chris Tillman. Same analysis. Here's a guy with a 4.73 career ERA. Granted over many fewer innings. Never putting more than 65 innings alone in one season on the books in the MLB prior to this he's a guy I would have thought they'd projected a bit more positive for this upcoming year. He's a guy who was as high as the #22 prospect by BA. With a composite sub 4.00 ERA in the minors, you'd think they could project something closer to 4.00 for him in 2013. I think he's obviously due for some regression, but again there is some pretty substantial grey area here. 60% of his innings have been logged at a 5+ ERA so you can see why a projection model might have doubts for the guy. BUT there I'm guessing Tillman has much higher tails for his outlying performances available to him.

I think this is one of the circumstances where it'd be really nice to see a confidence projection of these guys. PECOTA runs thousands of models and composites them. But let's say the models reflect his current career path, a 65% chance of 5+ ERA and a 35% chance of a sub 3.00 ERA, sure the "average" will result in a mid 4.00's projection. In reality I imagine if PECOTA ran 10,000 iterations of the season, Tillman would have had fewer 4.60 ERA seasons than 5.00 ERA seasons. Because they have to aggregate the data, there is definitely some information lost in there.

I think Tillman's projection is pretty straightforward. Last year he had xFIP, FIP, and SIERAs over 4.00, the delta between his ERA and those estimators driven by BABIP of .220-something. His career numbers in each of those are near 5.00. So PECOTA sees a guy who had a huge step change in ERA, a smaller gain in component ERA estimators, and spits out a weighted mean projection in between his 2012 FIP and his career FIP.

We can hope that he's a guy who regularly outpaces his component projections, but I don't think we can reasonably assume that to be true. Prior to '12 his components and real ERA tracked pretty closely.

[flashjordnk]

I think Tillman's projection is pretty straightforward. Last year he had xFIP, FIP, and SIERAs over 4.00, the delta between his ERA and those estimators driven by BABIP of .220-something. His career numbers in each of those are near 5.00. So PECOTA sees a guy who had a huge step change in ERA, a smaller gain in component ERA estimators, and spits out a weighted mean projection in between his 2012 FIP and his career FIP.

We can hope that he's a guy who regularly outpaces his component projections, but I don't think we can reasonably assume that to be true. Prior to '12 his components and real ERA tracked pretty closely.

I'd be willing to say it's slightly more complex than that. I'm not saying he's going to sustain his growth, obviously there were plenty of trends to indicate it's not sustainable. 221 BABIP is literally not sustainable, espsecially with the 21% LD's. The K rate wasn't great either. I obviously think of the bunch he's the biggest one due for some regression.

But he is a guy that's gaining some velocity back, pitched better conceptually last year and is a guy that was pretty highly touted at some point. His swing strike % jumped to 8.1 last year up from a 6.9% career average.

As Frobby points out the weighting is something I'm not privy to as a non-subscriber but I think this is an instance where some continued natural improvements might stave out the jump to a mid 4s ERA type guy.

I think it's pretty tough to project anyone that's samples are all limited to under 90 innings in the MLBs, but I do completely understand where PECOTA's results are coming from. Just pointing out there are things inside the numbers that might show some reasoning for complete regression to career averages that might get missed.