Reds come back!

The Reds wrested victory from the jaws of defeat this afternoon by coming back in the bottom of the 10th inning for three runs against Eric Gagne and Salmon Torres. In doing so, they absolved Edwin Encarnacion and Jared Burton of rather sloppy performances in the top of the 10th that led to two Brewers runs. Yay!

Here's the win probability graph, courtesy of Fangraphs:
Eddie's throwing error cost the Reds 0.133 WPA. His offense (including his two homers) netted a total of 0.215 WPA. So a net positive day for Eddie, but not by a whole lot.

Paul Bako continues to have a fairytale-like April with the Reds, today leading position players with +0.420 win probability added with the stick. Someday, one has to think he'll return to Earth. But it's fun while it lasts!

Friday Night Fungoes: Dan Fox, Projections, and the Future, man!

Pirates sign Dan Fox

The Pirates continue to their overhaul of their front office by hiring Dan Fox, aka Dan Agonistes, who has written at both Baseball Prospectus and The Hardball Times. His official title is Director of Baseball Systems Development, so I am not sure if this is the same job that Tango advertised a while back--but it might be. The responsibilities sound similar--responsible for "integrating the array of quantitative and qualitative information in a way that makes both even more instructive."

Obviously I'm thrilled for Dan. This no doubt is incredibly exciting for him, and I'm sure the Pirates made a heck of a hire to get him. But this is kind of a bummer for me in a few ways:

1. The Pirates as an organization probably just got a bit better. As a Reds fan, that's not a good thing.
2. It means that we as a community lose access to an exceptionally good, broadly trained, and very resourceful analyst.
3. Dan Fox is one of the three main reasons that I decided to continue to subscribe to Baseball Prospectus this year. So now I'm down to PECOTA and Kevin Goldstein's work on prospects, though I'll give a nod to Nate Silver's occasional high-quality article as a secondary reason.

I have to say, I might start paying a little bit more attention to the Pirates in coming years. I like what I see and hear from their front office since they hired Neal Huntington, and I will be living minutes from their AA-franchise. I don't know if I could ever really leave the Reds and root for an NL Central foe, but you never know.


Projection System Showdown

I missed this when it happened, but (with a hat tip to studes) Tom Tango recently did a rather careful study of projection accuracy for hitters in the 2007 season.

Before I get into Tango's study, I do want to point out that his results are similar to those reported by Nate Silver last year (hitters & pitchers here). There are a few subtle methodological differences, though, and Silver focused more on rank order of the systems than the actual impact of any differences seen. The latter is probably the most important point, as you'll see...

Anyway, Tango followed two basic steps:

1. Adjust forecasted league-average OPS to match actual 2007 league-average OPS.
2. Determine the average deviation between expected OPS value and actual OPS value (i.e. the average residual).

He reported overall average error, as well as four groups based on total multi-year MLB plate appearance totals: high (grizzled veteran regulars), medium, low, and rookies (made debut in 2007). The results can be hard to sift through in that very long thread (starts on #29, ends on #103), so I opted to whip up a quick graph (hope Tango doesn't mind). Note: I inverted the y-axis to read more intuitively--the "higher" the points are (vertically), the better, because they show less error.You may want to open the graph in another tab to view it. I tried to stretch it to separate the lines as much as I could, within reason. Here are the primary conclusions of Tango's study:

• There isn't much difference between the different projection systems.
  
• Overall, the best system (PECOTA) provided OPS measurements that were, on average, 0.069 off from the actual values. That means that a player projected to have an 0.800 OPS will, on average, have an OPS between 0.731 and 0.869. Not particularly accurate, really.
  
• The worst system, aside from just projecting everyone to be league-average, had an average error of 0.075 OPS points. So, again, an 0.800 OPS player will, on average, have an OPS between 0.725 and 0.875. Not particularly worse than the "good" system.
• Free systems CHONE and ZiPS were a whopping 0.001 OPS units "worse" than the much-lauded PECOTA. Marcel was only 0.002 OPS units behind. And they beat for-cost systems like Bill James' and Shandler's.
  
• The best thing to do is to take an average across all nine projection systems. But even then, you only get an extra "point" of OPS accuracy over PECOTA and 3 points over Marcel. Again, BFD.
  
• Despite all the additional information that systems like CHONE and PECOTA have about minor league players, they're only marginally better than Marcel when projecting players with few MLB plate appearances.... and in those cases, Marcel can only projects league average.
• Forecasting playing time is hard.
  
• The Fans' community forecast didn't do better (or worse) than the objective systems in projecting OPS. However, they DID do substantially better when projecting playing time.
  
• Sal Baxamusa and Tango think that a system combining the Fans' projections of playing time and Marcel's simple projections of performance would probably outcompete anything out there in a head-to-head contest.

Overall, in terms of accuracy, we can do just as well--if not better--using CHONE, ZiPS, or Marcel (all free) as we can using the for-cost PECOTA, Bill James, or Shandler projections. I think I knew this was true in theory, but always figured I was still gaining something by using PECOTA over something like Marcel. Now, I'm feeling doubtful...

Two caveats to this study: 1) it only looks at hitter projections, and (most importantly) 2) it only looks at 2007 data.

Even so, I guess I'm not seeing a compelling reason to use PECOTA at this point. So, on my list of reasons to renew my BPro membership, I guess I'm down to Kevin Goldstein. ... I enjoy Kevin's column, but I'm not sure if that'd be enough if I were renewing today.


The Future of Sabermetrics, man!

There were two Hardball Times articles this week that I thought did a tremendous job of setting the stage for a lot of the work we're likely to see in the coming years:

• Sal Baxamusa did a great job of putting forth an organizational scheme to help us understand baseball player valuation. I find those sorts of charts to be tremendously effective ways to organize my thoughts. The first chart he presented is basically what I worked through in my player value series last winter. What Sal does is take it the next step and show how we might better understand pitching, especially in light of the influx of pitchf/x data. We could do similar charts for hitting, baserunning, and fielding--all of which might be influenced by the "f/x" line of data (hitf/x, fieldf/x, etc).
• Mike Fast put forth a litany of questions that remain to be answered in baseball research. What makes his list useful is that all of these may be answered, at least in part, via the use of ball tracking data. Therefore, we may see tremendous progress on most of these within the coming year or two. Exciting times are ahead...

Signing Young Players Early

Skyking has a great post breaking down the contract extension of Evan Longoria with the Rays. He finds, as did Tango, that Longoria is almost certain sacrificing a huge amount of money by signing this deal today. On the other hand, he just guaranteed that he'll make $17.5 million, even if he goes and has a career-ending injury tomorrow.

As Sky points out, that first $17.5 million would probably worth much more to most of us then the $50 million that might come after it. He made a similar point after the Granderson signing. I'm sure that I'd have a hard time not taking the $17 million today, even if I knew I'd probably make three times that if I went year-to-year, simply because the risk could be so catastrophic.

I would not be surprised to see the Reds try to work out similar deals with Joey Votto and (especially) Jay Bruce this season or offseason. In fact, I expect them to do it: it makes too much sense NOT to do it. I'm not sure about Cueto, Volquez, and Bailey, though...pitchers are so darn injury prone that it's pretty dangerous to sign them long-term. Still, the kinds of discounts that teams are getting when signing players like Granderson, Tulowitzki, and now Longoria are so extreme that at some point it will make sense to lock up young pitchers as well.


More MLB customer service issues: Blackouts

Last week, I mentioned a series of recent issues related to MLB's tendency to not necessarily keep the best interests of their fans in mind. I forgot a big one: blackouts. Maury Brown posted a BPro article (I guess that's a reason to keep my subscription) about TV blackouts last weekend that is worth a read, though I frankly still just don't understand what MLB is thinking. Here's my favorite quote from it:

Only in baseball would there be a collective head nod to the idea that it's good business practice to restrict consumers' access to your product.

Awesome.

Also, I just became aware of a blog that has been started up in protest of MLB's blackout policy. This will be a good way to keep tabs on the issue. FWIW, they report that there may be some tangible effort to at least remove the most absurd blackout restrictions in the near future.


Reds Shut Down Scout's Blog

Finally, this week saw the birth and death of a blog by one of the Reds' scouts, Butch Baccala. He apparently was shut down due to concerns about giving out information that might put the Reds at a competitive disadvantage. To my eye Butch clearly knew where that line was and would not cross it. The Reds apparently disagreed.

I see this as yet another example of the Reds' penchant for absurd secrecy. And frankly, I find that both annoying and disappointing.

Dave from Louisville thinks I feel this way because I'm an academic. Thank goodness I'm never going to have to work in the corporate world, because it sounds awful.

Monday Night Reds Monitor - Through 13 April

I've put together some spreadsheets that I'm able to fairly quickly populate to get up-to-date stats on the 2008 Cincinnati Reds, including the sorts of statistics described in my player value series. The idea is to have a weekly look at the Reds stats throughout the season, with a minimal (for me) amount of commentary.

Tonight was the first run (though I'd already set up the spreadsheets) and required a fair number of tweaks, but even so I was able to get all the stats ready in 35 minutes. I expect that I'll be able to get it down to 20-some minutes in the future, which will let me update these season-to-date stats weekly. I will probably do this in lieu of the monthly Reds reviews I've done over the past few years, as those are just too time consuming and didn't really provide much genuine insight.

I've tried to include notes below each table about the meaning of some of the more obscure statistics, but if you have questions please do not hesitate to ask!

NL Central Update

Team	W	L	PCT	RS*	RS/G*	RA*	RA/G*	Pwins	W%for90	XtrapW
STL	9	4	0.692	61	4.71	47	3.61	8	0.544	112
MIL	8	4	0.667	63	5.28	50	4.21	7	0.547	108
CHN	7	5	0.583	57	4.77	58	4.85	6	0.553	95
PIT	6	6	0.500	64	5.30	66	5.47	6	0.560	81
CIN	6	7	0.462	51	3.96	50	3.88	7	0.564	75
HOU	5	8	0.385	53	4.08	55	4.23	6	0.570	62

Remarks: RS, RA, RS/G and RA/G are all park-adjusted. Pwins is the PythagoPat predicted wins for each team. W%for90 is the winning percentage a team will need from now on to reach 90 wins. XtrapW is the extrapolated number of wins the team will get, assuming they maintain the same winning percentage.

The Reds' miseries in Pittsburgh amplified it, but over the first two weeks the Reds' offense has been the worst in the NL Central. Fortunately, the pitching has been second-best, which has kept them out of the basement.....Pythagoras has them in third place, for whatever that's worth.

Hitting

Last	First	PA	%BB	%K	%LD	BABIP	AVG	OBP	SLG	ISO	OPS	PrOPS	lwts_RC	R/G	RAR
Patterson	Corey	46	7%	4%	15%	0.189	0.262	0.304	0.667	0.405	0.971	1.039	8.8	7.19	4.5
Phillips	Brandon	57	9%	19%	29%	0.375	0.308	0.368	0.481	0.173	0.849	0.786	8.9	6.46	4.1
Bako	Paul	35	14%	17%	25%	0.458	0.367	0.457	0.500	0.133	0.957	0.753	6.5	8.97	4.0
Keppinger	Jeff S	58	10%	3%	14%	0.298	0.320	0.386	0.500	0.180	0.886	0.885	8.6	6.28	3.9
Griffey Jr.	Ken	54	15%	13%	19%	0.297	0.273	0.389	0.409	0.136	0.798	0.821	6.4	5.05	2.0
Dunn	Adam	50	26%	20%	11%	0.192	0.167	0.380	0.250	0.083	0.630	0.851	4.9	4.14	0.8
Hopper	Norris S	20	5%	0%	15%	0.313	0.313	0.389	0.313	0.000	0.702	0.718	2.4	4.79	0.7
Encarnacion	Edwin	49	20%	16%	7%	0.172	0.179	0.347	0.333	0.154	0.680	0.849	4.7	3.88	0.5
Valentin	Javier	18	11%	6%	7%	0.267	0.250	0.333	0.313	0.063	0.646	0.616	1.8	3.93	0.2
Votto	Joey D	26	0%	15%	36%	0.364	0.308	0.308	0.308	0.000	0.616	0.705	1.9	2.75	-0.5
Hatteberg	Scott	27	19%	7%	25%	0.200	0.190	0.333	0.238	0.048	0.571	0.839	1.9	2.70	-0.5
Freel	Ryan	24	4%	8%	16%	0.238	0.227	0.250	0.273	0.046	0.523	0.634	1.0	1.39	-1.4
Castro	Juan	10	10%	0%	0%	0.000	0.000	0.100	0.000	0.000	0.100	0.581	-0.6	-1.63	-1.7

Remarks: PrOPS estimates OPS based on batted ball data, and deviations between the two are often due to "luck." LWTS_RC are estimated runs created based on linear weights. RAR is runs above replacement player, without a position adjustment (that is done with the fielding data). All runs estimates are park-adjusted.

The summed linear weights across all players is 57 runs, which is 6 more than they've scored--almost a half-run per game more. Situational hitting hasn't been good thus far to say the least....Patterson, Bako, Phillips, and Keppinger all had a fine first two weeks.....PrOPS thinks that Bako will turn back into a pumpkin soon, and isn't impressed with Phillips.....On the other hand, PrOPS thinks that Patterson and Keppinger have been legit, and that Dunn, Encarnacion, and Hatteberg have had awful luck thus far.....Dunn and Eddie's line drive rate has been dreadful, though.....Ryan Freel has picked up where he left off last season, and Juan Castro managed to do a tremendous amount of damage to the Reds despite having only 10 PA's.


Total Player Value (Hitting + Fielding)

Last	First	Pos	RAR	Fielding	PosAdj	TtlValue
Patterson	Corey	CF	4.5	2.0	0.2	6.8
Bako	Paul	C	4.0	1.2	0.6	5.8
Keppinger	Jeff S	SS	3.9	0.7	0.3	4.9
Phillips	Brandon	2B	4.1	-1.5	0.1	2.7
Votto	Joey D	1B	-0.5	2.7	-0.3	1.9
Dunn	Adam	LF	0.8	1.3	-0.4	1.7
Castro	Juan	SS	-1.7	3.0	0.1	1.3
Griffey Jr.	Ken	RF	2.0	-0.4	-0.4	1.2
Valentin	Javier	C	0.2	0.0	0.2	0.4
Hatteberg	Scott	1B	-0.5	0.9	-0.3	0.1
Freel	Ryan	CF	-1.4	0.2	0.0	-1.2
Hopper	Norris S	LF	0.7	-2.1	-0.1	-1.5
Encarnacion	Edwin	3B	0.5	-2.9	0.1	-2.3

Remarks: RAR is the same as above, and is park-adjusted. Fielding is the average runs saved estimate between ZR and RZR. Position adjustments are adjustments of the run value of a player's positions, pro-rated for playing time. Total value is just the sum of all of these numbers, and is an estimate of total run value above a replacement player.

These fielding estimates have the Reds at +5 runs overall.....Patterson's plus performance at the plate and field has him leading the Reds position players in value, along side the even more surprising Paul Bako.....Jeff Keppinger has kept on hitting, and has been adequate at shortstop.....Votto and Castro's fielding numbers look too high, and are an indication of small sample size issues.....The opposite is (unfortunately) probably not the case with Encarnacion, who comes in well below-replacement level over the first few weeks....unfortunately, the the Reds don't really have other options as long as Gonzalez remains injured.


Pitching

Last	First	IP	K/9	BB/9	HR/9	HR/F	%GB	BABIP	ERA	FIP	OPSa	BsR	BsR/G	RAR	FIPRAR
Harang	Aaron	21.0	6.4	2.1	0.9	10%	42%	0.200	2.14	3.37	0.579	6.4	2.73	7.5	4.6
Cueto	Johnny	19.3	11.2	0.5	1.9	19%	35%	0.171	3.72	3.09	0.550	5.7	2.66	7.1	4.9
Cordero	Francisco	5.0	9.0	1.8	0.0	0%	17%	0.167	0.00	1.55	0.285	0.3	0.54	4.3	2.7
Volquez	Edinson	10.3	7.9	4.4	0.0	0%	48%	0.286	0.87	2.66	0.560	2.7	2.35	4.1	3.1
Lincoln	Mike	6.7	5.4	0.0	0.0	0%	62%	0.227	1.35	1.75	0.442	1.0	1.39	2.7	2.0
Burton	Jared	6.7	14.8	1.3	2.7	47%	64%	0.000	4.05	3.67	0.494	2.1	2.88	1.6	0.6
Mercker	Kent	3.7	4.9	4.9	2.4	24%	44%	0.000	2.45	6.71	0.595	1.3	3.20	0.7	-0.9
Fogg	Josh	9.0	6.0	3.0	3.0	24%	41%	0.214	7.00	6.59	0.888	6.0	5.97	0.0	-1.2
Affeldt	Jeremy	4.0	6.8	6.8	2.3	47%	82%	0.182	2.25	6.68	0.833	2.5	5.61	-0.3	-1.0
Weathers	David	4.7	0.0	9.6	0.0	0%	33%	0.263	3.86	6.14	0.750	2.9	5.54	-0.3	-0.9
Coffey	Todd	8.3	2.2	1.1	2.2	24%	50%	0.281	7.56	5.70	0.883	5.5	5.99	-0.9	-1.2
Arroyo	Bronson	15.7	8.0	2.9	2.9	26%	39%	0.286	5.17	5.92	0.947	12.4	7.11	-2.0	-1.0

Remarks: BsR are base runs for a given pitcher, based on hitting events (not earned runs). RAR is base runs above replacement player, using a different standard for starters and relievers. Relievers with saves get a leverage-index boost in their RAR value. FIPRAR is a DIPS-based estimate of runs above replacement, using Tom Tango's Fielding Independent Runs as the runs estimator.

Base runs predicts the Reds should have given up 49 runs, 1 shy of their actual total.....There's a fairly bimodal distribution here, with great performances and lousy ones, and little in between....Aaron Harang and Johnny Cueto bring top honors, but Cordero and Volquez have been outstanding in less work.....Kudos to Mike Lincoln for a tremendous first two weeks....There are huge mismatches, in the bad way, between ERA and FIP for Mercker, Weathers, and Affeldt, which makes me worry about the bullpen.....Base Runs agrees on Weathers and Affedlt...has Weathers really still not had a strikeout this season?!?

Thanks to the Hardball Times, who supplied most of the statistics used above. ZR data came from ESPN.com.

Markov: Dusty Baker's lineups aren't half bad

Update: Due to several errors with how I was using John Beamer's Markov model, this study is frankly a load of hooey. I am leaving it here for archival purposes, but please disregard pretty much everything here. Sorry about that--I had done several checks to be sure I was "doing it right," but it turns out that the specific ways I was thinking about the input were absolutely incorrect. Embarassing to say the least, but that's what happens with research now and then...

I'll post an updated version of this study when I get a chance. The results are far less surprising, and much more in line with other work on lineups than these were.

There was a chorus of complaints from Reds' faithful over the Reds' opening-day lineup, and this has continued in reaction to subsequent lineups over the Reds' first two weeks of play. I asked folks to submit to me what they felt the opening day lineup should have been in a "Can you make a better lineup than Dusty Baker" contest of sorts.

I've now taken those lineups and, with the help of PECOTA '08 projections, I've plugged them all into John Beamer's Markov Chain spreadsheet that came with the Hardball Times 2008 Annual. The results surprised me, and I expect that they'll surprise a good number of you as well.

Background - Evaluating Lineups (you can skip if you want)

The past several years have seen increasingly sophisticated work done in the area of lineup construction. Among the most widely-publicized tools that stems from this effort is David Pinto's lineup tool, which I first discussed roughly two years ago. It is based on a set of studies that used linear regressions to relate player OBP and SLG at each lineup position to runs scored. Those regressions indicated that we should make some changes to how lineups have traditionally been designed. Around the same time, The Book was published; it used a different approach, but arrived at many of the same conclusions. Check out my old post for details.

The problem with approach upon which Pinto's tool, in particular, is built is that a lineup is an incredibly dynamic thing. Regressions just report typical relationships between OBP and SLG across different lineup spots based on how MLB managers have filled out their lineup cards in the past. That's a different thing from having a tool that you can use to try out radically different lineups, as Pinto's tool permits one to do. For example, if you hit the pitcher 1st, that will have large effect on the run-producing opportunities of the #2 and #3 hitters compared to hitting an on-base machine like Scott Hatteberg 1st. In other words, the player placed into each of the lineup slots will have direct effects on the opportunities of all other players in the lineup, and the consequences of a seemingly minor substitution might not be immediately apparent. Regression simply will not capture these interactions.

Enter Markov Chains. Markov chain models are models that organize complicated processes into steps. At each step, you input a certain probability that a variety of specific events will occur. The result is a series of branching event chains, which are then summarized according to their probabilities. John Beamer described them here in his introduction to his model.

It turns out that this is a great way to model a lineup's performance--if you can input the chances of different offensive events happening for each player (using, for example, PECOTA projections), you can have Markov step through a lineup throughout a game, keeping track of outs and innings, all the while keeping track of the entire range of possibilities in terms of offensive production. In other words, if you can design a Markov model that fits the game of baseball, and give it accurate data, it will provide you with more precise estimates of offensive production by a lineup than with the regression coefficients can ever hope to do because it actually attempts to simulate baseball.

In the 2008 THT Annual, John Beamer released an excel-based Markov chain model that is apparently capable of accurately modeling baseball just as I described above. It is probably the best publically-available tool for lineup analysis.

2008 Opening Day

First, let's look at the Opening Day lineup, isolating our choices to only those players that Dusty chose to play (i.e. no Joey Votto, no Jay Bruce, etc). That way, we can focus more on the effect of moving players around the batting order, and less on player substitutions.

Below is a table of the Reds' opening day players, along with their 2008 PECOTA projections broken down into my favorite set of diagnostic stats:

Last	First	PA	%K	%BB	BABIP	AVG	OBP	SLG	ISO	OPS	lwts_RC	R/G	wOBA
Patterson	Corey	458	16%	5%	0.306	0.268	0.307	0.402	0.134	0.709	56.3	4.76	0.300
Keppinger	Jeff	514	6%	8%	0.317	0.305	0.364	0.418	0.113	0.781	70.6	5.82	0.336
Griffey	Ken	435	16%	11%	0.283	0.268	0.350	0.480	0.213	0.830	65.7	6.20	0.354
Phillips	Brandon	629	16%	6%	0.302	0.274	0.325	0.444	0.170	0.769	86.4	5.43	0.324
Dunn	Adam	579	25%	16%	0.298	0.261	0.388	0.549	0.288	0.937	103.0	7.73	0.392
Encarnacion	Edwin	561	16%	8%	0.308	0.285	0.356	0.493	0.208	0.850	88.5	6.51	0.354
Hatteberg	Scott	278	9%	11%	0.296	0.285	0.368	0.440	0.155	0.808	40.3	6.12	0.352
Valentin	Javier	208	12%	9%	0.287	0.269	0.333	0.424	0.155	0.757	27.1	5.22	0.327
Pitcher	Pitcher	363	35%	3%	0.202	0.123	0.156	0.153	0.030	0.309	---	---	---

Note: Markov pays no attention to plate appearances, except to generate frequencies of each even happening. Therefore, it doesn't matter if a player has 200 or 500 PA's, what matters are the number of singles, doubles, strikeouts, walks, etc, per plate appearance. Also, Markov doesn't know about lefty/right splits--it just assumes an average pitcher, who is mostly (but not entirely) right-handed. :)

I received 14 different lineups from users both here and at RedsZone, and I threw in a few of mine own as well. There are actually 9! = 362,880 possible lineup combinations of these nine players (counting the pitcher slot) that we could theoretically try, but I think these represent a good part of the diversity of recommendations that most folks might like to try. Below, I list all of those lineups, as well as the Markov-based estimate of runs per game that those lineups could be expected to provide. The +/- Baker column lists the season-level differences in performance of each lineup compared to Dusty Baker's true opening day lineup (in italics).

Name

1st

2nd

3rd

4th

5th

6th

7th

8th

9th

R/G

R/162G

vBaker

Bluzer-OD

Keppinger

Hatteberg

Phillips

Dunn

Griffey

Encarnacion

Patterson

Valentin

Pitcher

4.90

793.6

4.5

Chris-OD2

Keppinger

Encarnacion

Dunn

Phillips

Griffey

Hatteberg

Patterson

Valentin

Pitcher

4.89

792.9

3.8

Pickoff-OD

Keppinger

Encarnacion

Dunn

Griffey

Phillips

Hatteberg

Patterson

Valentin

Pitcher

4.88

790.5

1.4

Baker-OD

Patterson

Keppinger

Griffey

Phillips

Dunn

Encarnacion

Hatteberg

Valentin

Pitcher

4.87

789.1

0.0

AVG-rank

Keppinger

Encarnacion

Hatteberg

Phillips

Patterson

Griffey

Valentin

Dunn

Pitcher

4.86

788.1

-1.0

brad-OD1

Keppinger

Hatteberg

Griffey

Dunn

Encarnacion

Phillips

Patterson

Valentin

Pitcher

4.86

787.6

-1.5

SLG-rank

Dunn

Encarnacion

Griffey

Phillips

Hatteberg

Patterson

Valentin

Keppinger

Pitcher

4.86

786.8

-2.3

OPS-rank

Dunn

Encarnacion

Griffey

Hatteberg

Keppinger

Phillips

Valentin

Patterson

Pitcher

4.84

783.4

-5.7

OBP-rank

Dunn

Keppinger

Hatteberg

Encarnacion

Griffey

Valentin

Phillips

Patterson

Pitcher

4.83

782.7

-6.4

texasdave-OD

Keppinger

Hatteberg

Dunn

Encarnacion

Griffey

Phillips

Valentin

Patterson

Pitcher

4.80

777.7

-11.4

Chris-OD

Keppinger

Hatteberg

Encarnacion

Dunn

Griffey

Valentin

Phillips

Patterson

Pitcher

4.79

775.7

-13.4

jinaz-OD

Encarnacion

Dunn

Keppinger

Griffey

Hatteberg

Phillips

Valentin

Patterson

Pitcher

4.77

772.2

-16.9

jinaz-OD-exploit

Encarnacion

Dunn

Hatteberg

Griffey

Keppinger

Phillips

Valentin

Patterson

Pitcher

4.76

771.5

-17.6

Trace's Daddy-OD

Hatteberg

Dunn

Griffey

Phillips

Encarnacion

Keppinger

Valentin

Patterson

Pitcher

4.76

770.7

-18.4

justincredible-OD

Hatteberg

Keppinger

Griffey

Dunn

Phillips

Encarnacion

Patterson

Valentin

Pitcher

4.75

770.0

-19.1

OesterPoster-OD

Hatteberg

Keppinger

Griffey

Dunn

Phillips

Patterson

Encarnacion

Valentin

Pitcher

4.74

768.0

-21.1

mlbfan30-OD

Hatteberg

Keppinger

Dunn

Encarnacion

Griffey

Phillips

Patterson

Valentin

Pitcher

4.72

764.8

-24.3

Degenerate-OD

Hatteberg

Keppinger

Dunn

Encarnacion

Griffey

Phillips

Patterson

Valentin

Pitcher

4.72

764.8

-24.3

fareast-OD

Hatteberg

Keppinger

Phillips

Griffey

Dunn

Encarnacion

Valentin

Patterson

Pitcher

4.70

761.4

-27.7

brad-OD2

Hatteberg

Dunn

Encarnacion

Griffey

Phillips

Keppinger

Patterson

Valentin

Pitcher

4.69

760.6

-28.5

joel-OD1

Hatteberg

Dunn

Encarnacion

Griffey

Phillips

Patterson

Valentin

Pitcher

Keppinger

4.68

758.9

-30.2

redsmanrick-OD

Hatteberg

Dunn

Encarnacion

Griffey

Keppinger

Phillips

Patterson

Valentin

Pitcher

4.66

755.5

-33.6

All I can say is "wow." Dusty's lineup didn't come out as #1, but it was darn hard to beat, and only an estimated 5 runs per season behind Bluzer's top-rated lineup.

Think about that. Baker's lineup violates one of the biggest "rules" for lineup construction that us stat people harp on--his leadoff hitter is projected to have a miserable 0.307 OBP this season. And yet, the interactions between players in his lineup are such that his lineup results in more wins per season than most other variants...at least, according to Markov. My own lineups, which I designed based largely on the lineup chapter in The Book, rated as a fairly middle-of-the-pack lineup, and came out a good 17 runs (~1.5 wins) behind Baker's model. And some of the user-submitted lineups, which look very reasonable to my eye, came out more than 30 runs per season behind Baker's. Again, "wow."

A few other observations and interpretations:

• The range of performances of different lineups was about 38 runs per season, despite all lineups featuring the exact same players. That's almost 4 wins worth of variation! More than I expected to see.
  
• Lineups with Keppinger leading off did better than lineups with Hatteberg leading off, despite them being rather similar hitters according to PECOTA--both are high OBP guys, though Hatteberg projects to have more power.
• Most of the "best" lineups have Phillips batting in the 5th spot or higher in the lineup. Many of the "worst" lineups (like mine) have Phillips batting in the 5th spot or lower.
• None of the lineups that bat Dunn in the #2 spot do very well, despite his crazy-high OBP.
  
• My "idiot" lineups, in which I just ranked players by a rate stat like OPS, did pretty well for themselves (better than my "smart" ones). Ranks by AVG did particularly well, despite batting Dunn 8th!
  

Markov also can report how often each lineup slot will lead off an inning for any lineup configuration. Here is Baker's lineup, as well as the top-5 and bottom-5 lineups, broken down by how many innings each slot led off (on average):

	Lineup Slot - # Innings led off
Lineup Name	1	2	3	4	5	6	7	8	9	StDev
Baker-OD	1.76	0.77	0.81	0.95	1.04	0.86	0.89	0.85	1.06	0.30
Top 5 Lineups
Bluzer-OD	1.80	0.78	0.79	1.08	0.95	0.85	1.15	0.88	0.72	0.33
Chris-OD2	1.81	0.77	0.79	1.03	1.02	0.88	1.13	0.83	0.73	0.33
Pickoff-OD	1.86	0.77	0.78	1.03	0.97	0.89	1.14	0.83	0.73	0.35
AVG-rank	1.78	0.77	0.68	1.01	1.05	0.89	1.13	0.92	0.78	0.33
brad-OD1	1.8	0.8	0.8	1.1	1	0.9	1.1	0.9	0.7	0.33
Bottom 5 Lineups
Degenerate-OD	1.80	0.77	0.79	1.38	0.93	0.86	0.90	0.83	0.73	0.36
fareast-OD	1.82	0.76	0.68	1.37	0.95	0.82	0.90	0.85	0.85	0.36
brad-OD2	1.84	0.77	0.79	1.38	0.93	0.80	0.86	0.90	0.72	0.37
joel-OD1	1.82	0.82	0.67	1.37	0.96	0.80	0.87	0.90	0.78	0.37
redsmanrick-OD	1.84	0.77	0.79	1.38	0.93	0.81	0.87	0.89	0.72	0.37

Observations:

• Baker's lineup is very different from the others:
• It has the lowest frequency with which the leadoff hitter would lead off innings, though only by a small amount. Still, it might be enough to help diminish the problem of Patterson leading off.
  
• It has the highest frequency with which the #3 hitter would lead off innings. The #3 hole is the spot in the lineup that most frequently bats with two outs and runners on in real baseball, but Baker's lineup might reduce this effect compared to the others.
  
• Baker's lineup also has the pitcher leading off innings more often than any other lineup. That can't possibly be a good thing, can it?
  
• Top-5 lineups vs. Bottom-5 lineups
• The #4 slot leads off innings an awful lot for the bottom-5 set of lineups. This is typically a power hitter spot (either Griffey or Encarnacion in these cases), so you'd normally want runners on base when they're hitting. This problem is minimized in Baker's lineup.
• One really neat finding: the standard deviation among lineup slots in the frequency with which they led off was consistently lower for "good" lineups then for "bad" ones. This might mean that the "bad" lineups have more bottlenecks that tend to kill innings, resulting in certain lineup spots being more likely to lead off the next inning. So, perhaps good lineups distribute the best hitters around the lineup more than poor ones?

...I'd like to do a bit more with these data, breaking down the different lineups based on their rate stats, but this has taken long enough to get written as it is. So, let's move on to the Best of the Organization lineups.

Best of the Organization Lineups

The other set of lineups I requested involved folks' choice of any players from within the Reds' organization. I received 26 of these lineups. Here they are, again along with Dusty Baker's opening day lineup, as well as a few of his others from the first week.

Name

1st

2nd

3rd

4th

5th

6th

7th

8th

9th

R/G

R/162G

vBaker

mikegrayson-Best

Bruce

Keppinger

Griffey

Dunn

Encarnacion

Phillips

Votto

Ross

Pitcher

5.01

811.3

22.2

brad-Best

Keppinger

Votto

Bruce

Dunn

Griffey

Encarnacion

Phillips

Valentin

Pitcher

5.00

810.3

21.2

Chris-Best1

Keppinger

Dunn

Griffey

Bruce

Votto

Encarnacion

Phillips

Valentin

Pitcher

4.99

808.8

19.7

jinaz-Best-exploit

Votto

Dunn

Bruce

Encarnacion

Griffey

Keppinger

Phillips

Valentin

Pitcher

4.95

802.0

12.9

ED44-Best2

Keppinger

Votto

Dunn

Phillips

Griffey

Encarnacion

Patterson

Valentin

Pitcher

4.95

801.9

12.8

Alex-Best

Keppinger

Votto

Dunn

Phillips

Griffey

Encarnacion

Patterson

Valentin

Pitcher

4.95

801.9

12.8

Alex-best3

Freel

Keppinger

Dunn

Phillips

Griffey

Encarnacion

Votto

Valentin

Pitcher

4.94

799.7

10.6

Baker-Game4

Patterson

Keppinger

Griffey

Phillips

Dunn

Encarnacion

Votto

Valentin

Pitcher

4.93

798.0

8.9

ED44-Best

Keppinger

Bruce

Dunn

Phillips

Griffey

Votto

Encarnacion

Valentin

Pitcher

4.92

797.7

8.6

Chris-Best2

Bruce

Keppinger

Encarnacion

Dunn

A_Phillips

Griffey

Phillips

Hanigan

Pitcher

4.90

794.0

4.9

justincredible-Best

Votto

Keppinger

Griffey

Dunn

Phillips

Bruce

Encarnacion

Ross

Pitcher

4.88

789.8

0.7

mlbfan30-B

Votto

Keppinger

Dunn

Encarnacion

Griffey

Phillips

Bruce

Valentin

Pitcher

4.88

789.8

0.7

Baker-OD

Patterson

Keppinger

Griffey

Phillips

Dunn

Encarnacion

Hatteberg

Valentin

Pitcher

4.87

789.1

0.0

Alex-best2

Hopper

Keppinger

Dunn

Phillips

Griffey

Encarnacion

Votto

Valentin

Pitcher

4.85

785.7

-3.4

redsmanrick-best

Votto

Dunn

Encarnacion

Griffey

Keppinger

Bruce

Phillips

Ross

Pitcher

4.85

785.6

-3.5

Enquirer-vsR

Votto

Dunn

Encarnacion

Bruce

Griffey

Phillips

Valentin

Keppinger

Pitcher

4.84

783.9

-5.2

jinaz-Best

Votto

Encarnacion

Dunn

Phillips

Griffey

Bruce

Keppinger

Valentin

Pitcher

4.83

782.0

-7.1

joel-Best1

Votto

Dunn

Encarnacion

Griffey

Bruce

Phillips

Valentin

Pitcher

Keppinger

4.81

779.8

-9.3

brad-Best

Votto

Dunn

Encarnacion

Griffey

Bruce

Phillips

Keppinger

Valentin

Pitcher

4.80

778.4

-10.7

joel-Best2

Keppinger

Dunn

Phillips

Bruce

Encarnacion

Griffey

Ross

Pitcher

Gonzalez

4.80

778.2

-10.9

joel-Best3

Keppinger

Dunn

Phillips

Encarnacion

Griffey

Patterson

Ross

Pitcher

Hatteberg

4.79

776.5

-12.6

DannyB-B

Keppinger

Dunn

Griffey

Encarnacion

Votto

Phillips

Hopper

Valentin

Pitcher

4.77

773.2

-15.9

texasdave-B

Keppinger

Votto

Dunn

Encarnacion

Griffey

Phillips

Ross

Hopper

Pitcher

4.76

771.3

-17.8

fareast-Best

Votto

Phillips

Griffey

Dunn

Bruce

Encarnacion

Gonzalez

Ross

Pitcher

4.76

770.5

-18.6

PastAndPending-B2

Freel

Keppinger

Bruce

Griffey

Phillips

Dunn

Votto

Ross

Pitcher

4.74

767.9

-21.2

Bluzer-B

Hopper

Keppinger

Dunn

Phillips

Griffey

Encarnacion

Valentin

Pitcher

Freel

4.74

767.4

-21.7

Baker-Game2

Patterson

Keppinger

Griffey

Phillips

Dunn

Encarnacion

Votto

Bako

Pitcher

4.73

766.3

-22.8

PastAndPending-B

Hopper

Keppinger

Griffey

Phillips

Dunn

Votto

Encarnacion

Valentin

Pitcher

4.72

764.9

-24.2

JamesB-B

Hopper

Keppinger

Griffey

Dunn

Phillips

Votto

Encarnacion

Valentin

Pitcher

4.72

764.4

-24.7

redsmanrick-best2

Keppinger

Phillips

Dunn

Encarnacion

Griffey

Ross

Votto

Pitcher

Hopper

4.71

763.2

-25.9

DannyB-B2

Hopper

Dunn

Griffey

Encarnacion

Votto

Phillips

Gonzalez

Ross

Pitcher

4.69

759.3

-29.8

Enquirer-vsL

Keppinger

Dunn

Phillips

Bruce

Encarnacion

Griffey

Ross

Gonzalez

Pitcher

4.68

758.7

-30.4

Baker-Game3

Freel

Keppinger

Griffey

Phillips

Dunn

Encarnacion

Hatteberg

Bako

Pitcher

4.63

750.1

-39.0

Thoughts:

• Player choice matters: here we see that Baker's opening day lineup can be beat regularly by employing Joey Votto and Jay Bruce in favor of Scott Hatteberg and Corey Patterson.
  
• The difference between the best and worst lineups in this case was about 5 wins. That's lower than I expected given the 4-win range in the opening day dataset, but then again the personnel differences among these lineups aren't that dramatic.
  
• I was gratified to see that my "exploitative" lineup, which takes advantage of the lack of information in the model about L/R splits and strings together Votto, Dunn, and Bruce in the top-3 slots, did quite well. So I'm not completely hopeless...
  
• Mike Grayson's top-rated lineup has Jay Bruce and his projected 0.336 OBP in the leadoff slot, again indicating that the issue of OBP in the 1-hole is less of a big deal than it's often made out to be.
  

Let's look at the distribution of how often hitters lead off innings from this dataset and see if we see a similar trend to the prior dataset:

	Lineup Slot - Innings led off
Lineup Name	1	2	3	4	5	6	7	8	9	StDev
Baker-OD	1.76	0.77	0.81	0.95	1.04	0.86	0.89	0.85	1.06	0.30
Top 5 lineups
mikegrayson-Best	1.77	0.77	0.78	0.94	0.98	0.88	0.91	0.90	1.06	0.30
brad-Best	1.80	0.78	0.77	1.07	0.89	0.88	1.17	0.88	0.77	0.33
Chris-Best1	1.79	0.77	0.77	1.03	0.89	0.88	1.17	0.87	0.83	0.32
jinaz-Best-exploit	1.78	0.77	0.77	1.38	0.87	0.89	0.87	0.88	0.79	0.35
ED44-Best2	1.80	0.78	0.79	1.08	1.02	0.83	1.14	0.84	0.72	0.33
Bottom 5 lineups
JamesB-B	1.77	0.78	0.79	1.03	0.97	0.79	0.91	0.92	1.05	0.31
redsmanrick-best2	1.79	0.80	0.73	0.95	1.06	0.91	1.12	0.84	0.80	0.32
DannyB-B2	1.75	0.76	0.78	1.03	1.02	0.88	0.77	0.93	1.07	0.30
Enquirer-vsL	1.83	0.76	0.67	1.36	0.95	0.81	0.89	0.91	0.82	0.37
Baker-Game3	1.76	0.77	0.78	0.95	1.05	0.86	0.91	0.87	1.07	0.30

Not the same trend. Mostly. The top lineups continue to have a fairly low standard deviation (at least compared to the bottom 5 from the opening day lineups), but we see low variation here among the bottom lineups as well. What's different?

Well, the bottom lineups tend to have inferior players to the better lineups this time around. Norris Hopper appears frequently instead of Corey Patterson in these lineups, which is a poor trade according to PECOTA. And Baker's game 3 lineup includes Freel and Bako instead of Patterson and Valentin.

...

Well, there's obviously a lot more to do on this front. I feel like we've really just started to scratch the surface on this issue. But I wanted to end with a summary of some of the tentative conclusions that I'm taking from this work.

1. There isn't one best way to make a successful lineup. Lineup interactions are such that there may be several very different styles of lineups that each result in similar overall performance.

2. In general, playing the best players is more important than lineup order. That's not to say that lineup order doesn't matter, but a poorly structured lineup with great players can beat a perfectly structured lineup with weak players.

3. Spreading one's best (and worst) hitters out to prevent bottlenecks may be an overlooked yet highly influential means of improving the performance of a lineup. The first dataset indicates to me that it might be more important than some of the other things we tend to worry about (like OBP in the leadoff slot). At the very least, it's worth further study.

4. Seemingly small differences between players can cause substantial differences in lineup performance when they are swapped between lineup spots. It may be that different sets of players may require substantially different lineup designs for optimal production.

5. We fans may not know as much about lineup construction as we think we do. In my case, at least, I'm feeling pretty clueless after working through this project. This post by J.C. Bradbury seems pretty apt. So, I'm inclined to give Dusty Baker the benefit of the doubt with his lineup order at this point...though I reserve the right to complain about personnel choices!


That's about as far as I'm willing and able to go for now. If you have suggestions, specific tests (or specific lineups) you'd like to see done, etc, let me know. I'm happy to continue working on this stuff as my (limited) time permits.

A few caveats and notes:

• All of the above results are obviously dependent on the quality and reliability of the Markov model upon which they're based. It seems really solid, but we should always keep this in mind. Again, I'll refer to the Bradbury post.
• As implied above, a different set of players--say, #1-#9 lineup splits for NL teams...not to mention AL teams--might result in very different findings. Such is the nature of this sort of thing.
• For pitchers and their pinch hitters, I just used 2007 Reds' pitcher totals. Maybe I should have used #9 hitter totals instead, as they would recognize the contributions of pinch hitters. But I didn't think about that until about a minute before hitting "publish post."
• With some assistance from John Beamer (thanks John!), I modified the spreadsheet to automatically input the actual % times each hitter led off an inning, based on the output of the model. You'll need to do this as well if you want to replicate my results. Drop me a line and I can help ya out on that.
• I used the standard baserunning tables in the spreadsheet, and I did click the Update SBA button after constructing each lineup. I have no idea how much modifying those baserunning tables might affect the results.
• I also just wanted to again thank John Beamer for publishing his Markov model with the Hardball Times Annual. If nothing else, it's provided a lot of food for thought!
  

Photo by Getty Images/Jonathan Daniel

Additional discussion about this project can be found at:

• RedsZone - SunDeck
• RedsZone - Old Red Guard
  
• Baseball Think Factory