17 May 2012

Orioles 2005 and 2012 seasons: Apple and Oranges? (Pt I)

Baseball is a difficult sport to predict.  The season is long, guys get injured, and players all of a sudden develop skills or lose them.  This can make it difficult for a long suffering follower of the Baltimore Orioles to feel comfortable with their current success.  If you believe in fWAR and project it out, you wind up with 44.3 expected fWAR for the whole team.  Add that to a replacement level expectation of 48 wins and you get a 92 win team.  If you are more of a pythagorean runs type of person, then you have an expectation of 89 wins if they keep this season rolling.  Or, perhaps, you are a Bill Parcels guy and believe your record is an accurate description of yourself . . . then you think this is a 102 win team.

So with 89, 92, and 102 wins, we have quite a bit of difference from my early season projection of 68 wins.  To be clear...there is no way I can see this team as anything much less than 75 wins.  I think the two big keys this year the rest of the way are that the peripherals look decent for many of the players and the upper minors are not completely devoid of talent.  Mind you, the upper minors do not have great talent, but it is talent that you can use in a pinch.  That really has not been the case for the Orioles for a long while.  It is good to have a guy like Chris Tillman as a 6th, 7th, or 8th option in your rotation as opposed to being your third or fourth guy.  It is also good to have someone like Brad Bergesen as DFA fodder as opposed to being in your starting rotation.  We can certainly disagree with some of these moves, but I think we have to recognize that there is a higher and broader base level of talent sitting at Norfolk buffering real star talent lower in Bowie and Delmarva.

That said, I am also a great believer in regression to the mean.  I have yet to feel comfortable with a reassessment of my season projection, so my current projections are simply the current record with a projected .420 winning percentage on the rest of the season.  That brings my projection to 77 wins.  Honestly, that is a pretty exciting season in and of itself.  Sometime Camden Depot writer Daniel Moroz tried to address where this season will be heading on his own site, Camden Crazies.  Looking at how performance has related to underlying statistics, he came up with a rough estimate that the Orioles are about 6 games over their heads at the moment.  That would lead to an expectation of something like 86, 89 or 99 final wins roughly when you prorate that.

Again, the assumption here is that there is a set talent level associated with runs or wins and that this talent level remains continuous (e.g., injuries do not drastically affect performance, strength of schedule).

This makes me wonder about 2005.  It was a season where the Orioles geared up by trading for Sammy Sosa to shore up right field and banking on further production and growth from their existing roster.  Javy Lopez had a great first season with the Orioles in 2004.  Rafael Palmeiro produced slightly below league average.  Tejada and Mora could be argued as the best SS and 3B combo in baseball.  David Newhan broke out big in 95 games as a super utility guy slashing 311/361/453.  Rodrigo Lopez was very good and the trio of Ponson, Cabrera, and Bedard flashed plus pitching from time to time.  Jorge Julio, BJ Ryan, and John Parrish also made for a formidable bullpen.  Additionally, the Orioles had John Maine and Hayden Penn mulling around in the minors, waiting for their chance.  Now, the expectations were not incredibly high, but there was a good deal of hope for this team.

How did the team look at the end of May?  In terms of pythagorean, fWAR, and win marks...the expectation could have been for final season win tally projections of 95, 106, and 98 wins, respectively.  At the end of May, the Orioles sat with an offensive fWAR of 10.7 and a pitching fWAR of 7.5.  Those would be on pace for finals of 34 and 23.8, respectively.

fWAR Pace Actual
Roberts 0.362 0.449 0.642 3.4
4 12.7 6.7
Tejada 0.313 0.361 0.588 -1.3
2.1 6.7 5.1
Mora 0.284 0.344 0.501 1.2
1.6 5.1 4.1
Lopez 0.273 0.316 0.483 -0.8
0.9 2.9 1.9
Gibbons 0.269 0.32 0.546 1.8
0.9 2.9 2.5
Matos 0.274 0.377 0.393 -1.1
0.7 2.2 1.2
Raffy 0.271 0.361 0.449 -3.6
0.1 0.3 -0.1
Sosa 0.239 0.309 0.394 -1.6
-0.2 -0.6 -1.2
Bigbie 0.235 0.27 0.311 -0.2
-0.4 -1.3 0.1

These nine players were the primary starting nine.  If you believe in the performance to date for these guys, it would seem quite natural that this 2005 team would make the playoffs.  You have Brian Roberts on pace to produce one of the greatest seasons ever.  Tejada and Mora provide two additional star quality players.  Lopez, Gibbons, and Matos contribute as solid average performers.  Bigbie is a guy you can rotate out or trade for a better piece.  Raffy and Sosa are detriments to the team whose experience and contracts prevent replacement.  Roberts' performance will easily cover them.  If the pace would have held out, with all of these guys players, and replacement players contributing 0 fWAR, this would be a 30.9 fWAR offense.  That would have put them second in the AL behind the Indians and in front of the Red Sox.

Instead, the team wound up with 20.3 fWAR from this group.  Roberts could not sustain hitting one out of every four fly balls out of the yard and suffered an injury toward the end of the season.  That cut 6 wins off the pace.  Simply looking at BABIP, Roberts (.392), Tejada (.358), and Mora (.347) ranging about .046 to .081 over their career average BABIP.  That should have looked ripe for a regression.  Everyone else was within .030 of their career rates.  The offense should have simply looked like it was subsisting on three guys who may be a little lucky.  Introduce, regression, injuries, and positive drug tests . . . the team collapses to an average offense over the course of the season and a poor one after May.  After putting up 10.7 fWAR in the first two months of the season, the Orioles accrued 9.3 over the final four months.  The meme is that what killed the Orioles that year was the pitching falling apart, but the change in fortune (i.e., lost wins) was about 60/40 on the shoulders of the offense.

In part 2, I will take a look at the 2012 Orioles offense, which will probably be rather redundant with Daniel's analysis...who knows?


Daniel said...

Just wanted to note that " Looking at how performance has related to underlying statistics, he came up with a rough estimate that the Orioles are about 6 games over their heads at the moment." is not completely accurate. The 6 games was only for the position players - post going up on the pitchers today - http://www.camdencrazies.com/2012-articles/may/how-for-real-are-the-orioles-pitchers.html - where it's probably another ~5-8 wins. Also, the 6 shouldn't be pro-rated - the difference between their pace and where they end up, so all 6 should come out of the rest-of-season.

Great post in general though.

Anonymous said...

There are a couple of things that need to be mentioned with the 2005 team that weren't.

First of all, this team had a historic collapse. If I remember correctly, no other team in history went from as far above .500 to as far below .500 as that team did. To think that this squad will fall apart at that level is assuming another historically awful half of a season.

Secondly, the Palmeiro steroid suspension and subsequent drama around Tejada and b-12 injections definitely contributed to the decline. Remember that Gibbons, Matos, Roberts, Bigbie, Palmeiro, and Tejada were all mentioned in the Mitchell report, and the fact that one of their teammates just got busted for steroid use had to have had an impact on them. They were all probably worried about their own futures.

I realize the team was already fading at that point, but this definitely could have played a role in turning a month long slump into a historically awful half a season.

In other words, there were a number of non statistical contributors to that team's collapse. I too am a believer in regression to the mean, but this comparison is apples to oranges in my opinion.

Matt P said...

Take a look at 2010. The team went 34-23 from August to October. They put up an 11.4 fWAR in that time frame or averaged over a whole season 32.4 fWAR. According to that, they should have won 80 games. If you look at the rest of the season, they had 19.9 fWAR total(68 total wins).

Basically, I'm using this anecdote to propose the following.

Players performance in baseball isn't linear. Instead of doing roughly the same each month, we should expect major swings in performance. Accordingly, the way teams perform in baseball also isn't linear. They may have most of their production and therefore most of their wins in only two months. You probably noticed that this happened in 2005 also. Now, in some seasons, the situation isn't as extreme as it was in 2005 and 2010. Some years the distribution of WAR is more linear than in others. For example in 2011, there was very little variance. In 2009, there was some but not a major amount(we had a 35-42 run followed by a 29-56 run). But there's usually some variance.

Therefore, I propose that saying the games that the Os have already won are in the bank is probably incorrect. I mean yes, those games have been won. But given past experience, we shouldn't be surprised to see the Os to go on a 25-14 run at some point in the season. It's happened before without the team overachieving. The fact that it's happened now (like 2005) and not later in the year (like 2010) should be irrelevant. When you made a projection, it should have accounted for this run automatically. However, something like extreme injuries is something that is difficult if not impossible to project.

In other words, if you're not sold that players on this team are genuinely better than expected, then there's really no reason to expect that this team win more than 68 games. It's just that they've won them earlier than we would expect. The assumption that there is a set talent level associated with runs or wins and that this talent level remains continuous seems to be incorrect based on the years we've studied.

The Pittsburgh Pirates last year went from 53-47 to 72-90, a 19-43 run. The Indians were 30-15 before falling to 80-82. Then of course, there's the Orioles in 2003 went from 50-53 to 71-91(a 21-38 run). Or the Orioles in 2002 who went from 63-63 to 67-95(4-32). Sure, these cases aren't the same as the 2005 Orioles with their 42-28 record to start the season, but they're not so different. All of these are cases of teams who looked better at one point then they ended up performing.

Jon Shepherd said...

Matt - That is not correct. A 68 win projection team is expected to win 42% of games that have yet to be played. Their ability to win is not affected by what they did in the past.

Think of it this way, a coin is expected to be heads 50% of the time. If you flipped it 50 times and it wound up being heads, what would you expect over the next 50 flips? It would not be 50 tails. It would remain 50/50.

Jon Shepherd said...

Anon - It was a historic collapse, but I think we need to recognize a few things: (1) AAA had little talent in 2005 and (2) so much of the beginning of the season was based on Roberts overperforming.

Also, the Mitchell report was not released until after the season and no one had been told about it during the season. With Raffy being caught...it would have affected all teams equally. PEDs were (and are) everywhere.

Jon Shepherd said...

Daniel - Thanks.

Point taken on the batters. The pro-rated issue...the game was a long one last night. I need to get back to my spreadsheet and remember what I was doing. Work has been crazy and I have taken up training for a half marathon.

Matt P said...

I don't think your coin example is the best.

A coin being flipped is random. Each outcome is unrelated to the others. They're all independent of each other. Naturally, in this case, it doesn't matter if you have had fifty heads already, you'd still expect 25 heads and 25 tails.

On the other hand, a baseball team goes hot and cold. Just look at the Orioles these past twelve years. At times, for a long amount of time, this team has done relatively well. Yet inevitably, they've hit a cold streak where they lost a high percentage of games.

I would argue that this indicates each outcome is slightly dependant on others. I understand this sounds like it must be incorrect, but I'm pretty sure we can show that this happened for every Oriole team from 2000-2010.

While this sounds incorrect, it does make sense. The Os historically have had little depth. When our starters were fresh, they were able to do well because they were MLB quality. When they got tired and injured, we had little in the minors who could help. Consequently, our team struggled.

There was an article on Grantland talking about how the Orioles have performed each month for the last twelve years. Rany noted that we've been .500 for April and then for every month after that this team has gotten worse(with the exception of August and September). There's no reason to assume that won't happen again next year.

Jon Shepherd said...

Matt...the way these projections work are that they predict future performance, not past performance. The projection simply is not how you put it. It does not understand injuries or thin depth. It suggests that there is a talent level. Just because you win a lottery today does not mean you get hit by a train tomorrow.

I think you are writing about something besides the projection because your statements simply are not correct if we are simply talking about performance projections.

Matt P said...

I agree that projections don't take injuries into account. That's why they need to be tweaked given accurate injury info.

I think you're assuming that win data should be normal. It seems you're arguing that we should expect a similar win percentage in April, May, June, etc. I'm arguing that we shouldn't expect to see a similar win percentage in each of these months. This is because players aren't coins and they have good months and bad months.

We saw this happen for the Orioles in 2005. That's why the Os earned most of their WAR in the first two months.

I propose a test for my theory. Using Fangraphs, I was able for each season and team from 2000-2011(360 season teams total) to download a dataset with each teams record and pitching fWAR. I was also able to download more datasets with each teams record and pitching fWAR for April 2000, April 2001 all the way to April 2011.

I did the following. I turned the records into win-percentage and merged the columns so that I had a dataset with the season, team, season win percent, april win percent, season pitching fwar, and april pitching fwar.

I propose the following. If you're correct, then we should notice a strong correlation between a teams record at the end of April and a teams record at the end of the season. However, if I'm correct, we should notice a weak or even no correlation between a teams record at the end of April and at the end of a the season.

When I did the stats, I had a correlation coefficient of .55. If you take teams that win fewer than 70 games (the type which we're discussing), the correlation coefficient drops to .38.

Considering a weak correlation is .5, this would seem to indicate that most wins in April shouldn't be considered "in the bank".

The way to determine how many would be by downloading the win percentage for each month and doing a regression. I suspect that the amount would be negligible.

Jon Shepherd said...

I just do not see how your views are related to the simple projection system. I think to make the projection system more complex than it is would not be sensible.

Matt P said...

If you want to keep things simple, then it simplifies down to this:

We can't forecast anything using actual records until around the end of June. Until then, it makes sense to just consider any actual results before then as aberrations that will be corrected as the season goes on.

If I flipped a coin two hundred times, I would expect roughly one hundred heads and one hundred tails, but I wouldn't expect it to go head-tail-head-tail...

Jon Shepherd said...

I fundamentally believe that there should be no expectations that positive or negative aberrations will be equaled out by an opposite aberration. Such a concept is a belief that guys are 'due.' I think that concept has been repeatedly shown to be inaccurate.

A hundred heads in a row does not mean that the next flip is a tail. The coin is a coin. The projection system is a projection system.

Matt P said...

Ok, my coin example was a poor one.

"I fundamentally believe that there should be no expectations that positive or negative aberrations will be equaled out by an opposite aberration."

Let me ask you a question then.

Suppose the Orioles went 5-1 (like they did in 2011). Clearly, this was a positive aberration. Do you believe that this wasn't or wouldn't be equaled out by an opposite aberration? Certainly you expect even poor teams to have a winning streak or a positive aberration, right?

I think we'd have to be arguing over sample size, no?

"I think that concept has been repeatedly shown to be inaccurate."

Interesting. Looking at the Orioles, I presumed the opposite.

Look at the 2011 Orioles. They were 30-31 at one point and then ended the season 39-62. Why wouldn't you say this was a positive aberration (.491 winning average) followed by a negative aberration (.386 average).

Or the 2010 Orioles. Went 32-73 to start the season and ended it 34-23. Why isn't this a positive aberration followed by a negative?

Or the 2009 Orioles. 34-40(.460) followed by 30-58(.341).

Or the 2008 Orioles: 60-62(.492) followed by 8-31(.205).

Or the 2007 Orioles: 50-55(.492) followed by 19-38(.333).

Or the 2006 Orioles: 38-44(.463) followed by 32-48(.400).

Or the 2005 Orioles: 42-28(.600) followed by 32-50(.390).

Or the 2004 Orioles: They had a 37-51 stretch(.421) followed by a 21-13 (.618).

Or 2003: We were 27-27(.500) followed by a 44-64 stretch(.407)

Or 2002: 63-63 followed by 4-32.

That's arguably ten years straight and almost certainly nine out of ten years straight with either a positive aberration followed by a negative one or vice versa.

It's been rare that we've put up such a good stretch for such a long period... it's probably only happened thrice over these past ten years. However there have been many cases where we've played above our abilities and ultimately crashed.

I haven't looked at other teams... but it looks the opposite to me.

Jon Shepherd said...

With an unlimited sample size, the true value of a team will be realized. I do not think the assumptions required for that perspective are valid in this scenario. I do not believe that the average value over a whole season must equal the expected value.

European Prof said...

If you flip a coin and get 50 straight heads, the most probable outcome of the next 50 flips is not 50 tails, or even 25/25 heads/tails, but 50 more straight heads because there is sufficient sample size in the first 50 flips to show that assumptions about the integrity (quality) of the coin were erroneous.

Jon Shepherd said...

Of course, the example given is one of hyperbole to make a simple point.

That said, if one chooses to be painfully literal then a coin can be perfectly weighted and you could wind up with 50 heads. That chance is about one in a quadrillion or so.

ptcello said...

Where is part 2?
very interesting article, and it is also interesting to note that as of today, your prediction is still 77 wins.
I hope the orioles don't make any dumb trades, but I think they'll still finish at or above .500.
30-37 over the remaining seasons means 81 wins, and barring a complete collapse or the Second Coming, I don't think that can happen.