Monday, February 28, 2005


According to dozens of wire reports, the Dodgers have come to terms with the following 40-man roster cats:
They've also come to terms with non-roster players Joe Thurston, Chin-Feng Chen, and Henri Stanley, whose names should be given more color; I propose Thirsty Joe, CFCin' is CFBelievin', and Henri Ennui. If any of those three can make the roster - which, with the Nakamura signing, would probably require beating out both Dr. No and J-Grab in addition to staving off 40-man masters like Myrowvia, Jason Rep Inc., and Kriss C-Ross - they would probably take the spot yielded by Darren Dreifort's eventual return to the 60-Day DL.

There's also another potential spot on the 40-man roster if Houlton is returned to the Astros. In other words, the chances of Thurston, Chen and Stanley to make the top 40 depend on the D.J.

To the best of my knowledge, that leaves the following 11 40-man roster cats yet to be issued contracts:

All-Time Dodgers GIDP

Jon's eminently quotidian post today had me checking out those Dodgers sortable stats, and I was 3/4 of the way done making a spreadsheet of the Dodgers all time leaders in GIDP per plate appearance. Well, funny thing happened on the way to the opera; my computer froze, and I lost my work. Heck, I thought, maybe I should just use the sabermetric baseball encyclopedia to generate this for me. Here's the all-time Dodgers leaders in GIDP per plate appearance relative to league average, minimum 1200 PA, according to Lee Sinins' Sabermetric Baseball Encyclopedia:

1 Mickey Owen 169 61 36
2 Lenny Harris 161 47 29
3 Eddie Murray 148 50 34
4 Babe Phelps 148 58 39
5 Paul Lo Duca 147 72 49
6 Joe Medwick 145 51 35
7 Roy Campanella 143 143 100
8 Carl Furillo 143 207 145
9 Don Zimmer 139 43 31
10 Bruce Edwards 139 42 30
11 Norm Larker 135 40 30
12 Tommy Davis 134 86 64
13 Frank Howard 130 60 46
14 Luis Olmo 130 36 28
15 Don Drysdale 130 34 26
16 Tony Cuccinello 127 41 32
17 Manny Mota 127 57 45
18 Mike Piazza 127 75 59
19 Eric Karros 126 165 131
20 Kal Daniels 125 26 21
21 Tim Wallach 123 37 30
22 Shawn Green 123 88 71
23 Wally Moon 121 64 53
24 Joe Stripp 119 55 46
25 Mark Grudzielanek 118 61 52
26 Steve Garvey 117 164 140

Nice to have Don Drysdale show up on this list. Lenny Harris, Eddie Murray, Mike Piazza, and Eric Karros are the names that bring me back, which certainly betrays my age.

Two thirds of the recently jettisoned offensive nucleus make the list in Green and Lo Duca. For the curious, Adrian Beltre comes in at #53, with a rate of 96 - just better than league average. Jackie Robinson checks in at #55. Brett Butler comes in with the lowest GIDP rate (29!), although part of that comes from having fewer PA with a runner on first.

Hmmm... this has got me thinking about Tom Ruane's finding that ground ball outs are more valuable than fly ball outs and strikeouts even when accounting for GIDP; is a ground ball out less valuable for a TTO-heavy team in a park that is unfriendly to groundballs? I'll try to take a look at this soon.

Sunday, February 27, 2005

Why Nakamura Is Suck

If you're like me, you've probably thought you read the above phrase at least five or six times over the past few days.

Comparing the Projections by Component

At Minor League Ball today, John Sickels started publishing some of his projections from a system he calls JSPS-2. He didn't mention any of the specifics, and I thought it would be fun to compare his numbers with ZiPS, Marcel, and PECOTA (I'm not the only one- Rob already has a comparison up at 6-4-2).

For this comparison, I wanted to look at the differences by component, rather than by counting totals, to see where the systems differ on each player's skillset.

Here's Dallas McPherson:
Dallas lwts/600 $BB $K $HR $H $XB $3B
JSPS-2 0.3 0.075 0.258 0.067 0.312 0.302 0.172
ZiPS 9.2 0.086 0.305 0.078 0.346 0.288 0.188
PECOTA 7.4 0.1 0.302 0.071 0.343 0.292 0.095
Marcel -3.9 0.087 0.216 0.051 0.304 0.244 0.091

Including Marcel isn't done to find the best projection; it's more of a baseline, since it uses very little data for players with as little major league playing time as McPherson and the reliability rating it assigns to its McPherson projection is among the lowest for any player.

ZiPS and PECOTA both project very high $H's for McPherson. Although it's certainly a skill McPherson has excelled at coming up, I'm not sure if that's a reasonable expectation. This is the same concern I voiced about Delwyn Young a week ago. Sickels' system also expects McPherson to put the ball into play more than the others do, projecting lower K and BB rates than ZiPS or PECOTA. PECOTA really likes McPherson's ability to draw walks while ZiPS bets high on both home runs and triples.

To take a look at how hard each system projects McPherson to hit the ball, I calculated the average run value of McPherson's batted balls for each projection and then multiplied by 400 to approximate a season's worth of batted balls (we'll call this measure r400b for short, okay?). ZiPS is highest at +47 runs, PECOTA says +42, and JSPS-2 says +30.

Here's Jeremy Reed:
Reed lwts/600 $BB $K $HR $H $XB $3B
JSPS-2 -3.9 0.09 0.125 0.018 0.314 0.243 0.212
ZiPS -2.3 0.099 0.104 0.023 0.302 0.209 0.172
PECOTA -3.3 0.086 0.109 0.026 0.303 0.228 0.095
Marcel 11.3 0.096 0.155 0.034 0.339 0.228 0.077

Again, Marcel is only for color- it assigns a very low reliability to its Reed projection. ZiPS and PECOTA are very similar, with the only substantial difference coming from walk rate. JSPS-2 slots in between the two in walk rate but projects a good deal more strikeouts. JSPS-2 also projects high $H and $XB rates, somewhat compensated for by a lower HR rate. Using the same r400b measure from above, there's virtually no spread here: ZiPS -3, Pecota -1, JSPS-2 +0.

Reed's hype rollercoaster has been a lot of fun to watch, as his amount of hype is directly proportional to his batting average on balls in play:

NCAA, 2002: .342
Low-A, 2002: .339
High-A, 2003: .341
AA, 2003: .413
AAA, 2004 (Charlotte): .281
AAA, 2004 (Tacoma): .313
MLB, 2004: .426

My unstudied thought is that the .302-.314 $H range in these Reed projections is probably about right. I'd like to see what the spread in $H under JSPS-2 is, as the similarity between it's $H projections for McPherson and Reed is equal parts encouraging and disconcerting. I'm not really sold that the Mariners would be better off with Reed playing in the majors this season; I don't know that it's better for his development, and I don't see how it benefits Seattle enough to offset the impact of starting his service clock earlier.

In any event, I'm waiting with baited breath to learn more about how John Sickels is doing his projections and what the rationale behind them is.

Friday, February 25, 2005

My Least Favorite Stat

Projected VORP.

This stat simply should not exist.

If you read this site, there's a good chance you're familiar with VORP (Value Over Replacement Player), a stat that quantifies a player's offensive value relative to a replacement-level offensive contribution at their position. VORP is a highly-contested stat in itself; in my humble opinion, it's waaaaaay overused by those who like it but completely misunderstood and foolishly discarded by its mainstream detractors. The question of quantifying a replacement-baseline as well as the question of measuring value are both problematic. Baseball Prospectus, for example, uses raw stats (1B, 2B, 3B, HR, outs, BB, etc.) in determining value, when a true measure of value would also evaluate context and add win probabilities and so forth. The difference is largely negligible, but it's there.

The problems with VORP, however, are easily dealt with if you know what you're getting into and you know about the context in which a player's VORP was earned. It's a useful summary.

However, a player's projected VORP, a stat I've seen bandied about on sites I frequent quite a bit since the release of BP's PECOTA spreadsheet, tells me very little. Here's an exercies: rank these four players in terms of who you would want to have playing third base for your team in 2005:

Player A: 19.3
Player B: 17.2
Player C: 17.2
Player D: 13.9

Doesn't tell you much, does it? You'd probably gravitate toward listing them based strictly on VORP, but you wouldn't be very confident in your decision. Now let's say I give you EqMLVr, the offensive rate stat upon which VORP is based:

Player C: .096
Player A: .069
Player D: -.017
Player B: -.039

If we prorate that production to 150 games, the difference between Player B and Player C is a full 20 runs, yet they're the same in projected VORP. Now, in a sense, projected VORP is right in creating that difference, and I'll explain why in a second. But if all you have is VORP, you don't know why the difference should be there.

Player A is Andy Marte, player B is Russ Branyan, Player C is Jose Valentin, and Player D is Eric Hinske. Now that you know that, you should be able to intuit why Valentin's VORP outstrips his MLVr by so much: he's counted by the system as a shortstop. Given that the difference between Valentin and Branyan in defense can be reasonably projected at about 20 runs, VORP seems to get this right. However, it gets it right for the wrong reasons: it uses playing time projections and defense-neutral position adjustments. If Valentin fielded like Derek Jeter and Branyan fielded like Scott Rolen, VORP projections would be getting this one clearly wrong.

Now, playing time projections are important for many reasons. For one, since PECOTA uses comparable players, a playing time projection can tell you about how often similar players stayed in the lineup. For another, the amount of playing time in the projection can demonstrate the context from which the projection was derived; Olmedo Saenz won't maintain his offensive rates as an everyday player and so forth.

That being said, the lack of opportunity Branyan received with the two Ohio-based organizations doesn't have much bearing on how good he will be if given a shot at a full-time job. As such, projected VORP is telling us about noise and not talent.

Now, we shouldn't expect VORP to tell us about talent; it's a value measure. The problem is that there's no reason to project value; when looking to the future, we want talent. If you're choosing a player for the season, you want to know their talent and their ability to manifest that talent (i.e. their ability to stay in the lineup and so forth). VORP combines them for us in an ad hoc manner. We don't need that.

For instance, if we know J.D. Drew is an injury risk and have a good idea of the relative probabilities of how many games he'll play, we'll already account for that in building the team by getting a quality fourth outfielder like Ricky Ledee and having players like Cody Ross and Chin-Feng Chen in the wings as right-handed hitters who will be average or better offensively if used as the starters against LHP. Thus, the generic replacement level isn't really germane in considering the actual amount of value that Drew adds to the team nor the value lost if he's injured. VORP essentially averages out Drew's injury risks to get playing time and then multiplies that by his offensive talent level to get VORP. What we really want to do is model Drew's overall contribution to the team; if he plays a full season, the team will score this many runs; if he misses 80 games, they'll score this many runs. Then we weight each scenario by its probability. This is a much better model for projected value.

On top of that, VORP acts as if it knows about a player's defense and baserunning by adding a positional adjustment. It doesn't. Wouldn't you rather know a player's position, defensive ability, and offensive ability separately then to merge them all together?

While VORP is a useful shorthand to describe a player's contribution in a season, it's far from complete. When looking at projections, our goal should be to stay as far away from shorthands as possible; they're distracting. If we care enough about a player's performance to want to project it, how is it possible that we don't care enough about the player's performance to contextualize it in determining value?

Of course, projected VORP doesn't make any claim to tell us the whole story. But I can't think of any reason to use it, and it is taken by many to actually tell most of the story of a player's future. Since all the data to calculate a player's projected VORP can be derived from PECOTA's numbers, projected VORP is only useful insofar as it saves you a little bit of spreadsheet programming. It is counterproductive in that it discourages attempts to combine that data in more useful ways.

Man, I'm cranky this week.

Thursday, February 24, 2005

The Whining You Don't Want to Hear

Too much school. Too much other stuff. Baseball to come soon.

If you haven't checked out The Baseball Analysts, the new home of Bryan Smith and Rich Lederer, you probably don't read my site anyway so this will fall on closed ears. But if you haven't, give it a whirl.

To tide you over, here's a bit from Ken Gurnick that interested me:

Kazuhisa Ishii, sounding confident despite fighting for a spot in the rotation, said he has resumed throwing the splitter he abandoned after arriving in the United States four years ago.

"It was a good pitch for me in Japan, but it didn't feel right with the American baseballs when I came here so I stopped throwing it," said Ishii. "I practiced a lot in Japan with American baseballs over the winter, and I think it can be good for me again."

Anyone got anything on this one?

Friday, February 18, 2005

Delwyn Not-So-Young

Don't have much time today, but I wanted to pose a discussion question for anyone interested. Dayn Perry just ranked Delwyn Young, who spent last season as the second baseman in high-A Vero Beach, as the 36th best prospect in baseball. Young also was mentioned in a thread on the most underrated prospects in baseball at John Sickels' great new blog, with Sickels himself agreeing.

This came as a surprise to me, as I'd actually thought of Young as perhaps overrated. Young's put up great raw numbers, drawing walks at each level he's played at, putting up .200+ ISO's at each level, and putting up batting averages in excess of .280 at each level.

However, I'm not too sure I believe what Delwyn has done is sustainable. He strikes out a lot, and his $H (a.k.a. batting average on balls in play) has been between .414 and .424 at each level, which is close to off the charts. Even if that's largely due to his talent and not his environment and luck, it's pretty inconceivable that he'll be able to field a $H consistently over .350 in the major leagues. Young's excellent power certainly suggests he's smacking the ball pretty hard, but I think his raw numbers are setting expectations too high.

On top of that, Young has been somewhat old for each of his leagues. A 22-year-old in A-ball is certainly behind the curve for premium prospects. Given that reports of his defense aren't particularly good, I have trouble thinking Young projects as an above-average player at second base - or, for that matter, anywhere else.

Contrast Young to Willy Aybar. Aybar's batting averages have been in the .270 range outside of his first trip through the FSL, and his ISO's have consistently been about .150. His walk rate is pretty much the same as Young's. However, Aybar is a year younger than Young and has been playing a league or more ahead of Young in each season. Their translated performances in 2004 were of roughly equal value, and Aybar's still a year younger. Given also that, if I'm not mistaken, Aybar's defensive outlook looks to be substantially better in the long-term, I'm not sure that Young is that much more valuable. Perry ranked Aybar #98.

True, Young clearly has the better power potential of the two. But given Young's strikeout problems and given that they're not coupled with an outstanding walk rate, he might have a lot of trouble getting on-base enough. I have a hard time thinking it's reasonable to expect Young to be a better than average major league position player (not that there's anything wrong with that).

What do y'all think?

Wednesday, February 16, 2005

Dodger Stadium

New article about Dodger Stadium at the Hardball Times by this guy. Check it out.

Some Like It Hot: The St. Valentin's Day Massacre

From Jon's writeup of Jose Valentin today:

[F]rom July 1, 2004 through the end of the season ... Valentin was awful every which way - particularly so against right-handed pitching, which the Dodgers are counting on him to face at least at the outset of the 2005 campaign. 170 plate appearances is a tiny sample size.

As with Choi, you must always take sample size into account while judging a player. On the other hand, Valentin’s second-half features more at-bats than Choi had, and it comes attached to a player who has been declining for most of the century.

I absolutely cannot agree with this assessment. 175 plate appearances against RHP in one half of a season is a pretty tiny sample size. If you took Derek Jeter's first 175 PA of the year and attempted to conclude that he'd lost it, you'd be dead wrong. The fact that a failure in a player's performance is more recent is of extremely limited predictive value - see Mitchel Lichtman's study on hot and cold streaks. It is much more useful to look at Valentin's season as a whole; in-season trends are of very limited significance.

To say that Valentin had more PA than Choi is a little crazy, since Choi's sample size was microscopic. The lesser of two wrongs is still wrong.

Furthermore, though I think this effect was unintentional, Jon seems to be punishing Valentin for hitting poorly against his core constituency, right-handed pitchers. But his relative success against left-handed pitchers can't be thrown out the window because he won't be asked to face many LHP this season; those plate appearances still reflect his overall hitting skill. It's much sounder to actually use his full .162/.241/.350 line and adjust it upward based on the number of PA he had against LHP. His true talent against RHP relative to his true talent against LHP almost certainly did not change over this period of time.

In addition, there are a number of factors outside of Valentin's actual talent that could be intervening here. It could easily be that Valentin was facing tougher pitchers over that sample. It could also be that Valentin faced more fly ball pitchers over that sample, and the effect of that is almost as significant as the handedness of the pitcher, as Thomas Hanrahan has shown. He also could have had many more games in ballparks that reduced his effectiveness. I don't know if any of this is true, but over such a small sample size there's a good shot that some combination of the above is factoring somewhat heavily.

Furthermore, if we talk about a player's decline it's not very worthwhile for projecting a player's future to talk about a player's decline in total. Let's break down Valentin's performance by component:

$BB $K $HR $H $XB

1st H 0.085 0.261 0.134 0.301 0.277
2nd H 0.089 0.288 0.064 0.167 0.125

1st H 0.103 0.426 0.029 0.353 0.308
2nd H 0.074 0.36 0.125 0.167 0.231

1st H 0.09 0.306 0.11 0.314 0.283
2nd H
0.084 0.311 0.083 0.167 0.162

The walks and strikeouts stayed virtually the same. Home runs declined somewhat, and particularly against right handed pitchers. However, we're talking of a very marginal difference here - had Valentin had 11 HR against RHP in the second half and 12 in the first half, there would be no difference. Instead he had 16 in the first half and 7 in the second half. I'll leave to you to decide whether a difference of that magnitude constitutes true decline, keeping in mind both the size of the sample and the lack of controls (park factors, strength and type of opposition).

The other key difference was the number of hits on balls in play and the number of hits on balls in play that went for extra bases. This can be significant, yes; it can indicate a loss of bat speed or running speed or a change in type of swing. It can just as easily reflect changes in competition strenght and type, ballpark, opponent's defense, and luck, with luck being the most likely culprit. While the data may reflect an actual decline, the extent to which this data actually suggests an actual substantial decline in talent is very low. If anything, I think it suggests a serious decline in luck, so improvement may be more likely going forward than decline. Keep in mind, too, that his $H in the first half was much higher than it has typically been in the past, and that he has historically trended toward lower second half $H's:

2000: .294 1st half/.302 2nd half
2001: .321/.244
2002: .279/.248
2003: .268/.227

Now, it is significant that Valentin's $H's are declining from year to year; that does indicate decline. But it does not indicate overall decline, as the other components of his performance have not shown any substantial decline over the course of the century. If his $H had stayed constant over the past five seasons, Valentin's GPA's would look like this:

2000: .261
2001: .269
2002: .261
2003: .263
2004: .259

Given the relative variability of batting average on balls in play, it's actually fairly likely that Valentin's $H is set for a rebound.


Jon also states that "perhaps he is the hitter’s Derek Lowe - someone that general manager Paul DePodesta has figured will take unique advantage of Dodger Stadium." Jon hit this one on the head. I don't want to get into it here since I have a forthcoming article which will explain it in detail, but Valentin is the kind of hitter that benefits most from Dodger Stadium. The short version is that Chicago and LA both turn a lot of fly balls into home runs and stop them from being doubles and triples. Chicago does this moreso than LA, but LA does it much more than you think. However, Chicago is otherwise pretty much neutral, while almost all other aspects of Dodger Stadium hinder hitters. Thus, Valentin's relative value in Dodger Stadium is maximized.


In the comments to Jon's post, somebody quoted a previous post I'd made at length, and one of the things I'd pointed out in the post was that Valentin had been a much better performer with runners in scoring position. Jon took issue with that, arguing that batting with runners in scoring position is likely not a tangible skill. Here I'll first reference Andy Dolphin's study on clutch hitting, which is a must-read. Clutch-hitting is a skill with little spread in true talent, and a good deal of the observed spread is sample noise. However, for some players clutch hitting (and "choking") are real and significant factors. This is not necessarily an issue of performing under pressure; rather, it reflects how a batter's approach fits the situation. If a batter goes to the plate in a tie game with one out and a runner on third in the ninth and swings for a home run, the batter is not providing optimal value. A good batter can change their approach based on the situation. So there is certainly a skill involved.

Does Valentin have this skill? I would argue that he does. With runners on, he strikes out substantially less than he does with the bases empty, and his $H is much higher with runners in scoring position. Obviously, there are sample size issues here, but having a lower $K and higher $H in concert is pretty indicative of an alteration in approach.

Jon rightly points out that one should measure a batter's skill with runners in scoring position "by comparing players to each other, not by comparing a player's RISP stats to his own non-RISP stats. What kind of skill is it when a player is worse when the pressure is off?" That's an excellent point, but it's not relevant to the point I was making when I wrote about Valentin's RISP record. The point I was making is that Valentin's raw numbers undervalue him because his successes are concentrated in places where they are of more value. If you have two .250/.320/.410 hitters and one hits .245/.310/.400 with RISP and the other hits .290/.375/.480 with RISP, the second hitter is more valuable but the raw totals don't reveal it. This is a result not of clutch ability and high moral fiber but of the hitter adapting their skills to the game in better ways.

I'll use Valentin's 2002-2004 situational hitting to demonstrate this point. The base/out data is not freely available, but the baserunner data (i.e., none on, runner on first, etc.) is. Using Tango's base state linear weights, we can compare what Valentin's raw totals indicate to his actual totals. For the period, Valentin's raw lwts are -12, or -4.6 per 600 PA. Using base state data, however, he was +1 or +.3 per 600 PA. That's a 5-run swing per season or roughly half a win.

Now, sample size is still a major concern here, but Valentin had established a substantial trend here and the 02-04 data includes the only season from the 2000 and on data (i.e., what's publicly available) in which he did not demonstrate a large split, and the split for every season but 2003 was of roughly the same size. We should still regress the split, but my very cursory research suggests we should add a +3 RISP adjustment to Valentin's projection.

Once again, this is not an argument that Valentin is particularly clutch or of high moral fiber or whatever; this is an argument that his raw numbers slightly underestimate his value.

On top of all this, of course, is that Valentin is an excellent defender and very good baserunner; those components figure to add about 20 runs to his value (I have +13 for 3B range, +1 for double plays, +2 for baserunning, and +2 for runner advancement and avoiding double plays). Factor in position adjustment, and Valentin has to be a -25 or so batter or worse to be a below average third baseman. If you'll recall my comparison of Marcel, ZiPS, and PECOTA, he was -7, -9, and -25 in those projections. Given that none of those projections, with the possible exception of PECOTA, factors in his likely reduction in PA against southpaws or the +3 RISP adjustment, Valentin's a pretty solid bet to be above average.

Monday, February 14, 2005

The ERA of Innocence

Noah Lowry=William Van Landingham

-Icaros in comments today at Dodger Thoughts

I don’t mean to pick on Icaros, a valued member of the Dodger blog community. However, I think it’s significant that someone like Icaros – who I know puts a lot of thought into what he writes when he leaves blog comments – would write something like that.

Lowry and VanLandingham do have several things in common. They both pitched in college and they both debuted for the Giants at age 23. VanLandingham pitched 84 innings with a 3.54 ERA in his rookie season while Lowry pitched 92 with a 3.82 ERA.

However, Lowry is left-handed and VanLandingham was right-handed. VanLandingham had a 6.0 K/9 and 4.6 BB/9 while Lowry had a much better 7.0 K/9 and 2.7 BB/9 in his debut. VanLandingham made up the difference by only allowing 4 home runs in his 84 innings, an anomalously low number even if he was, as I seem to recall, a heavy groundball pitcher. Even so, VanLandingham’s ERA benefited much more from defense and fortuity than did Lowry’s.

On top of that, the way Icaros evoked Van Landingham seemed to suggest that Lowry’s success was unsustainable. That’s an odd comparison since Van Landingham had his best season in his second year. Maybe Icaros meant to compare VanLandingham’s second season to Lowry’s first season. His K, HR and BB rate in his second season actually closely resemble Lowry’s 2004 performance. However, the raw numbers are clearly misleading; the run scoring environment Van Landingham played in was much more favorable to pitchers, and relative to league average VanLandingham’s performance didn’t match Lowry’s. On top of that, comparing VanLandingham’s sophomore effort at age 24 to Lowry’s debut at age 23 isn’t exactly apples to apples.

So the two are, through the very early part of their major league career, both above average pitchers pitching three years after being drafted out of college. There are hundreds of pitchers that meet that criteria, and there’s not much to suggest that within that population these two are very similar. In fact, their different peripherals indicate that they’re pretty different pitchers.

So why compare them? Because they’re Giants! Plain and simple. Even though what team two pitchers play for ten years apart has less predictive value than their breakfast cereal, it’s easy to compare the two. We get suckered into a comparison because of our emotions toward the players. I remember the last time the Giants had some hot stud young college pitcher – what a disaster that was! If Lowry flames out, we’ll point to the VanWarningsigns and congratulate our collective memory. If he succeeds, we’ll probably forget that we once made the comparison.

This speaks to what I consider to be perhaps Bill James’ greatest insight. Baseball is full of numbers, he argued, but the typical application of numbers in baseball was based not on the contextual meaning of those numbers but on the emotional and associational impact of the numbers. .304 average? Sign me up! 31 home runs? Wow, a big bopper! James argued that his goal was to strip away that veneer and determine which numbers were important and why.

Treating two pitchers as similar because they played for the same franchise is clearly folly, and I’ll bet that Icaros realized that soon after his initial posting when other Dodger Thoughts commentators described differences between VanLandingham and Lowry.

But still, they seem so alike with those mid-3 ERA’s and IP totals. The thing is, ERA’s and IP are firmly entrenched in our way of thinking about pitchers despite being of limited utility. While most readers of this site are probably quick to mock anybody who trots out Run and RBI totals, ERA and IP aren’t too different from those stats.

In terms of predictive value, the number of innings a pitcher throws in one season doesn’t heavily influence the number he’ll throw in future seasons. There will be correlation, of course, because pitchers are used in fashions that reflect their skill set and because some pitchers have recurring injury issues. However, those are reflected in other statistics as well, and IP doesn’t tell you anything substantial about a player’s true talent level that couldn’t be unearthed with other metrics.

Similarly, ERA doesn’t have great predictive value. A player’s ERA’s will often end up being similar in many years, but it’s the components of ERA that matter in determining how effective a pitcher is. To project pitching performance, it’s necessary to look at how often a pitcher strikes batters out, walks batters, and so forth; it’s not necessary to look at the player’s ERA. The difference between a player’s ERA and what their component figures would predict can be useful, but not because of anything inherent in ERA; ERA is just a way of checking how the other parts of the system add up so that we can see how a pitcher deviates from the standard model in terms of concentration of baserunners and runs and so forth. Those are things which are at the margins of what we look for in a pitcher anyway, though, so ERA’s value in such an endeavor is limited. Moreover, the actual data on how a pitcher’s baserunners and runs are concentrated is superior to extrapolations from ERA, so ERA is at best an imperfect intermediary.

In addition, ERA is useless without reference to the scoring environment in which an ERA is recorded. A 3.00 ERA in a 6 runs per game environment is very different from a 3.20 ERA in a 4.5 runs per game environment, but ERA itself doesn’t elucidate the difference.

While IP and ERA don’t offer much utility in deciphering a pitcher’s true talent level and predicting future performance they do, on face, represent a decent measure of actual observed value. The number of innings a pitcher eats is significant and the average of runs allowed is useful. However, ERA is a poor measure of value; run average is much better, because pitchers have a substantial impact on the amount of errors committed behind them and since runners scoring after reaching on error still have to be moved over. Run average is the useful measure of value, and while RA should always be adjusted for defense in some manner to determine a pitcher’s actual value, ERA makes an adjustment for defense which is entirely ad hoc and doesn’t tell us anything about the pitcher’s value relative to other pitchers.

IP is a pretty good measure of value, although the number of outs the pitcher is responsible for is ultimately what we are looking for and IP takes some short cuts that detract from our ultimately attaining that knowledge. But in another sense, IP doesn’t give a very good sense of value since it treats every inning pitched as the same. While the majority of innings happen in contexts that are close enough to make the distinction meaningless, innings pitched at different times have very different values. Michael Wolverton’s Support-Neutral pitching stats are a way to acknowledge this, and the work of Tangotiger and others on leverage index and win probability added does the same thing. Even then, the overall value of a pitcher to the team depends on how good and how rested the rest of the pitching staff is. Thus, innings pitched, while meaningful for basic calculations of value when more detailed data is not available, is not the best way to assign value.

I don’t mean to say that we should discard IP and ERA, though I wouldn’t mind too much if we did. It’s just that they both represent small parts of the story but measure things that are close enough to what we want to know that we settle for them. That act of settling is, in most cases, not necessary, and it always brings with it a strong risk of distorting our perception.

The point is, the numbers that we look at in attempting to compare players should be tested on two fronts. The first is the data's ability to determine the player's true talent level and thus future performance. The second is the data's ability to determine the player's actual value over a period of time. ERA and Innings Pitched do neither very well, and the franchise a player is employed by probably doesn't do either one at all.

Thursday, February 10, 2005

re: Around the Horn: Bullpen

Quick reactions to Ken Gurnick's Around the Horn: Bullpen, though this feels redundant to me:

Of all the "Moneyball" assumptions about Dodgers general manager Paul DePodesta, the one he really laid to waste was the minimization of the closer.

That's the one that had Eric Gagne fans dreading that DePodesta would include the franchise's greatest reliever in one of the many winter housecleaning moves to redistribute payroll, the way the Oakland A's did with Jason Isringhausen and Keith Foulke when DePodesta was an assistant there.

One of the great "Moneyball" misconceptions is that the sabermetrically-inclined devalue the importance of a dominant closer. That's inaccurate. The argument is that certain pitchers are overvalued because they have been used in the ninth inning. That's without question true; Jose Mesa and Dan Kolb come to mind from just this past offseason.

Eric Gagne is simply a great relief pitcher. He pitches most in the 9th inning, and pitching in the 9th inning is typically very valuable. So part of his recent value comes from him having pitched in the 9th inning in the recent past, but his ability - and thus his future value - doesn't derive from him having pitched in the ninth inning.

Jose Mesa, on the other hand, is a solid reliever but no great shakes. By virtue of recording a lot of saves last season, the Pirates awarded him a $2.5 million contract this season, pretty out of line with his actual value to Pittsburgh.

Pitching successfully in the 9th inning is a skill unto itself, but the distribution of talent among major league relievers in this skill is such that it's not a skill worth considering in most cases. A group of pitchers whose talent means a 3.50 ERA will probably all fare the same as "closers" because the added requirements of pitching in the ninth inning aren't substantially different from what pitching successfully in the rest of the game requires. There may be a few pitchers who excel or flop with a small lead in the ninth, but determining who they are is tricky and most people's attempts to do so are based on small samples. Looking at who has been a closer in the past will tell you next to nothing, and it's the last question of any relevance I would ask about a reliever.

Thus, the notion that closers are overvalued is not an argument that teams don't need good relievers or that good relievers shouldn't be used to protect leads. It's an argument that relievers become overvalued once they are deemed closers.

On top of that, the examples Gurnick cites as closers the A's included in "winter housecleaning moves to redistribute payroll," Isringhausen and Foulke, were both free agents, and the A's made a very strong effort to retain Foulke. Gagne, on the other hand, still had two arbitration-eligible seasons. Apples and pomelos.

The world knows well Gagne's achievements, but it's easy to overlook the spear carriers that make it possible. For the better part of two seasons, Guillermo Mota set up Gagne, but DePodesta dealt Mota at the trade deadline last July because that's what it took (along with Paul Lo Duca and Juan Encarnacion) to get Brad Penny.

Well, he also got Hee Seop Choi and Billy Murphy, although Murphy was squandered on Steve Finley. Mota certainly pitched well for the better part of two seasons, but the only time he was the "set-up man," so to speak, was in 2004; that 8th inning role was Paul Quantrill's in 2003.

The reason DePodesta was willing to include Mota and tamper with the chemistry of the best bullpen in the game was his confidence in relatively unknown Mota-clone Yhency Brazoban, who quickly hop-scotched from Double-A to Triple-A to Gagne's new setup man in a matter of months.

Must be a nature vs. nurture thing, since Mota's clone is three inches shorter and thirty pounds lighter. Brazoban is much more of a power pitcher than Mota, although they were both converted position players who were traded to the Dodgers. Maybe Brazoban is a Felix Rodriguez clone, too.

In any event, while Brazoban is good he's also easily the current Dodger most overrated by Dodger fans, non-Izturis category. Anyone check out his walk rate? When Brazoban is posting a 3.90 ERA in August and voices murmur about how the Dodgers ruined another can't-miss prospect, I'm going to be very angry.

Carrara is fearless and durable with the kind of mental makeup to rebound from a bad outing.

He would have to be after his 2003 campaign.

Duaner Sanchez might be the most intriguing arm behind Gagne's. He was picked up off waivers last winter, and his first full Major League season was a huge success with a 3.38 ERA over 80 innings.

Sample size + defense + park factor. Duaner is backing up Yhency on the overrated index. He's not bad at all, but he hasn't had a K:BB over 2 in two and a half seasons above AA. He has a ways to go.

Wilson Alvarez has the resume to be the fifth starter, but Tracy believes he is better suited as this stage of his career as a left-handed reliever, an area where the Dodgers lack depth. Non-roster candidates to back up Alvarez include Mike Venafro and Kelly Wunsch.

Alvarez is far too good of a pitcher to be used as just a lefty-killer. He also has shown a reverse platoon split of late, though that's probably more small sample hijinks than anything else. While the notion of there being an 8th-inning guy is highly problematic, if anyone should be it it's him.

Also, the quality of the lefties that DePodesta acquired for free this offseason is solid. Frank Brooks, whom Gurnick doesn't even note has blessed paw status, would be a pretty good LOOGY if given the shot. His AAA numbers last season don't look too good because the Pirates tried converting him back to a starter for a little while. Also, Kelly Wunsch has been a pretty good major league LOOGY for several years, though he's too liberal with the BB and HBP to be a top-tier lefty.

Tuesday, February 08, 2005

That Meager H

The "meager" puns - phonetically inaccurate as they are - continued to follow me with my debut at The Hardball Times. Check it out. There should also be a Dodger-relevant article from me up at THT later this week.

Saturday, February 05, 2005

Multiplicity: The Fourth Outfielder Challenge

Taking a page out of a Michael Keaton movie I've never seen, I'm looking for folk who want to write for the Fourth Outfielder. The reasons why should be clearer in a few days.

This job description is subject to change, but here it goes:

Write one or more entries most weeks. The topic of the site is baseball in general and the Los Angeles Dodgers in particular. The site's principle methodology is applying sabermetric insight to the operations of the Los Angeles Dodgers. An interest in sabermetrics and some familiarity is required; expertise is not.

The ideal candidate will be skilled in applying baseball knowledge creatively and focusing on the interrelation of multiple concepts.

Compensation: none.

If you'd like to write here, e-mail me telling me why and what you expect to contribute. Writing samples (baseball or otherwise) are a plus, but not mandatory.

If you want to show off your skills, here are a few questions you can attempt to address to get my attention or get on my good side. Choosing one and going in-depth is probably more purposeful than attempting to briefly answer each.

1. How does Jayson Werth's disproportionate number of plate appearances against LHP last season affect his projections?

2. How much money should the Dodgers have been willing to pay Norihiro Nakamura?

3. What kind of value did the Dodgers receive on Eric Gagne's re-signing?

4. The 2005 Dodgers are in Houston and Cesar Izturis is at bat with Milton Bradley on first, none out, and Brad Lidge pitching. It's the ninth inning and the Dodgers trail 5-4. J.D. Drew is on deck with Kent, Choi, Werth, and Valentin to follow. Should Izturis bunt?

5. Rank the top eight players in the Dodgers' farm system by value.

Start your engines.

This page is powered by Blogger. Isn't yours?