THE BOOK--Playing The Percentages In Baseball

The 2007 Washington Nationals are providing a case study in replacement level starting pitching.  And that was before all the injuries!

Tango (or Sean):  why do you think all pitchers in the rotation in April should be excluded?  It seems to me that someone like Jerome Williams is a replacement-level pitcher (per Scott’s observation), despite being in the rotation.  I prefer something like Nate Silver’s FAT definition, based on salary level.  But of course that’s a lot more work.

I like Sean’s definition and methodology although I too would simply include all starter performances and not exclude pitchers with more than half of their appearances (or IP) as relievers.

Guy, you can’t start including or excluding pitchers who were in the rotation at the beginning of the season.  What would be your criteria?  Sure, there are plenty of pitchers in the original rotations who were replacement-level because teams are not too good at evaluating pitching talent plus a certain percentage of back of the rotation pitchers may be replacement level because teams run out of non-replacement level pitchers before they fill all their starter spots.

But, by using Sean’s methodology, we guarantee that all of his pitchers are by definition replacement starters if, in the first place, we define replacement starters as pitchers not originally thought good enough for a starting role

Then again, we could also define replacement level starters as those pitchers who are not deemed good enough for original starting roles OR for sarting roles after injuries and other things shake out after the start of the season.

So, before you decide what methodology to use in choosing replacement pitchres, you have to define you what you mean by replacement pitchers - and that depends on what you are going to use “replacement level” for.  Since we only use it to determine $ value for players, there really iis only one good definition for replacement pplayers.  And that is those FA players who can be hhad for near min salary (actually a mil or lless since there is some value for “experience") and those pre-FA players who are deemed to have little or no value to the teams.  The first group is easy to identify and the second group is not.  But if it were me, I would look at all starters whether they be in the original starting rotation or not, who are “freely available” in terms of salary or in the case of pre-arb players, who are deemed to have very little value by teams.

Tango, I do not know what you mean by the Guthrie thing.

Also, I have said for years that a replacement starter is around .8 runs worse than an average starter.  And I don’t need to identify replacement starters to do that (although that would confirm my assertion). The reason that it is so “low” (good) is that the spread of talent among pitchers is small enough (smaller than among batters) that by the time we get to .8 runs worse than average, there are more than enough pitchers to go around, combined with the fact that there are plenty of pitchers in the “replacement pool” who are subsantially better than that but for some reason teams don’t like them.  And then teams will use pitchres who are worse than replacement level (again, because they misevaluate them), freeing up lots of better pitchers for the replacement pool.  Since, again, replacement level should have the same definition of “freely available” we have to include in the replacement pool, those pitchers (and batters if we are talking about position players) who technically are better than replaceent level but for whatever reasons are not recognized as such by teams.

Anyway, we really have to define very specifically what we mean by replacement level, which probably entails explaining what we want to use it for, before we start trying to figure out how to identify them and thus figure out their performance level.

Obviously, you have the one-year anomolies, since talent is not very liquid. 

However, I will also presume that Williams must have been injured last year.  After all, the guy had an ERA+ of over 100 for each of the three seasons from age 21 to 23, with a total of 63 starts.  That can’t possibly be a replacement-level pitcher.

The following year in 2006, he had 12 IP, with 5K and 10 NIBB.  I won’t bother looking it up, but I can guess he was injured.

In any case, a replacement-level player cannot be one of the 750 players on the active roster on Apr 1.  The illiquidity of players and the irrationality of some GMs doesn’t make that a perfect rule, but it’s one of the best ones.

MGL: If Sean has 100 (I don’t know what the number is) replacement pitchers, you should expect that each one makes up 1% of the replacement pool, when looking at their stats.  But, Guthrie is performing very well, and is racking up IP, enough that instead of the 1% that he should be weighted as, he’s weighted as 8% of the pool.  Clearly, what you need to do is account for each pitcher equally, since Guthrie is one pitcher, and not eight.  However, that will lead to other selective sampling issues.

***

As for your -0.80 runs level, it doesn’t seem that you apply the starter/relief adjustments from The Book.  A replacement-level pitcher is .380 as a starter, making the average MLB starter +.110 wins.  The exact same replacement-level pitcher is .470 as a reliever, making the average MLB reliever +.050 wins.

Taking out Guthrie:

5.03 ERA, 952 IP means 532 ER for all repl pitchers, including Guthrie.

Guthrie is 21 ER, 74 IP

That makes the rest of the squad as 511 ER, 878 IP, for an ERA of 5.24

This gives us a win % of .408.

***

So, when Guthrie is 8% of the weight, the repl pitchers are .428.  When Guthrie is 0% of the weight, the repl pitchers are .408.  Presuming we have 100 starters in the repl pool, Guthrie should be 1% of the weight, and therefore, if we take 1/8th oof .428 and 7/8th of .408, and we get: .410.

And that’s pretty much what we expected.

If I was trying to do a perfect study, I would just look at games started instead of removing the guys who were primarily relievers.  Then I could include Colby Lewis’s one bad start - he’s definitely a replacement pitcher.  But that would take 10X more time than I had, I just got season stats from dougstats.com, put it in a spreadsheet, and picked my replacements.

I can look at grouping them by innings as Tango suggested, another idea is to look at the median instead of the mean, so we don’t overweight Guthrie.

using MGL’s 0.80 runs in a 4.25 ERA context, a replacement pitcher would be a .420 W%.

A .380 pitcher allows about 30% more ER than average, or about a 5.60 ERA in 2007.

"In any case, a replacement-level player cannot be one of the 750 players on the active roster on Apr 1.”

If all teams had equal resources, this would be a valid operating assumption.  But some teams clearly do put “freely available talent” in their rotations.  (I believe Williams was signed by WAS as a FA, but maybe I’m misremembering that.) Anyway, I’d be closer to MGL’s definition: salary under $600K or so if FA.  And throw in pre-FAs who have no value to their team, if you can identify plausible criteria for that (not sure you can).  That said, I’m not sure including these players will change the results much. 

* *

A -0.8 runs starter is clearly not a replacement player.  That means he’s +.2 runs as a reliever, which isn’t terrific but definitely gets you a job in MLB and a salary well above the minimum.

All pitchers:
- since year 2000
- at least 28
- makes 500,000$ or less

That gives me between 55 and 71 pitchers each year.

Anyway, that’s 438 pitchers (about 63 per year), the IP total is about 9% of a team’s total.  The % of starts per game pitched is 14%.

Their ERA is 4.42.

All pitchers in that time period had an ERA of 4.35.

If I make it at least 30 years old, and their ERA is 4.44.

At least 32, and making 500,000 or less?  118 pitchers, with an ERA of 4.43.

35 and older?  48 pitchers, ERA of 4.40.

How about between the ages of 28 and 30 (so as to get around some of the age bias)?  4.42 ERA for 283 pitchers.

There are plenty of pitchers available on the scrap heap.

Guy, a replacement-level starter is NOT a replacement-level PITCHER.

As I said, we have to talk about replacement-level pitchers AS starter and AS reliever.

A repl pitcher is .380 as a starter and .470 as a reliever.

I also posted this on my site:

Grouping them by innings:

Top 3rd (5 pitchers) had a 3.59 ERA

Middle 3rd (10 pitchers) 4.91

Bottom third (30 pitchers - 2/3 of the group) 6.49 ERA

This last group is the pitchers who get one or 2 starts, get hammered, and send back to the minors. The first group, well, some of them got lucky, some may be showing they were better pitchers than teams thought they were this past offseason.

Median ERA is 5.74

The simple average of those three groups is 5.00 ERA.

If we weight them by the number of pitchers, we get 5.82.

Rally noted the median is 5.74.

One must question why are not more relievers starting.  An average reliever will certainly post a better than 5.74 ERA as a starter.

Keep in mind selective sampling.  Just because we get X results looking at pitchers who start and relieve doesn’t mean all relievers will only lose that much effectiveness.

But the Cardinals sure are giving their former bullpen a lot of starts.  So maybe more of them should start.

Looper 4.66
Wainwright 4.70
Wellemeyer 4.88 (5.71 starting)
Thompson 5.28 (5.80 starting)
Keisler 5.65

Tango/10:  If we assume these pitchers threw about 2/3 of their IP in relief, then their 4.42 ERA becomes more like 5.10 as starters.  And I assume you have the “Guthrie” problem in your sample as well—those who succeed got to pitch more innings (unless you weighted all pitchers equally).  So it seems to me that your “freely available pitchers,” pitching as starters, are probably in the 5.30-5.50 range at best.  So yes, there are plenty of pitchers available on the scrap heap, but you’re not happy when your favorite team has to use them. 

* *

I agree with Rally/14.  A lot of relievers would see their ERA rise by far more than 1.00 if forced to start (obviously, Mariano couldn’t have posted ERAs under 3.00 the last 4 years as a starter).  Presumably there’s a range, with the one run gap being typical of the subset of pitchers allowed to perform in both roles.  I think it’s a fair assumption that relievers are generally pitchers with above-average gaps—otherwise, they’d be starters—so many of them would face a very stiff penalty if converted.  Conversely, to the extent there are pitchers with very small starter/reliever gaps, these guys are either starters (it would be a waste to use them as a reliever) or not in the majors.

I agree wholeheartedly with the last par. in #15 above.  I am pretty sure that there are many relievers whose ERA would be more than a run worse if they started, and plenty whose ERA would be less than a run worse, plus I am pretty sure that there are all kinds of reasons who certain pitchres are geared more for starting or for relieving, not the least of which is number of pitches in their arsenal, at least according to CW.  Plus there are stamina issues.  Plus when we say “one run worse” as a starter, that depends on whether the reliever is a one inning pitcher like a closer, or a middle reliever who may average 2 IP per apperance.  Plus, as Tango showed in The Book, there appears to be a difference between pitchers who are converted or are swingmen or are emergency starters, etc.

IOW, I think that an analysis of replacement level is much better being done separately for relievers and starters.

And again, the more than I think of it, the more you have to use salary ($ value to a team) to identify your replacement level players since that is what we are using replacement level for - to determine the value of a player as compared to a player who can be had for a low salary.  If your definiton of replacement level is different, then you can use a different criteria for identifying replacement level players.

That is one reason why “my” replacement level is usually better than other analysts’ - because in my pool, I include many players who are actually pretty good players but are not considered as such by the teams, for whatever reason.  And the fact, as I mentioned already, that teams will use pitchers like Lima (and many others), who are a lot worse then plenty of players who do not have jobs, which also increases the talent level of the pool of players who are not being used or are being used but are not given much value by the teams.

Just to reiterate above, if you want a practical definition of replacement level, and you want to identify all the players in that talent level pool, you simply have to put yourself in the position of any or the average team and ask yourself, O.K., if I wanted to acquire a player at X position (1B, SS, bullpen, short or long, starter, etc.), for near minimm salary, either a FA who is deemed to not have much value, or a young player who is also deemed to not have much value (as I said, harder to identify because most young players have low salaries, regardless of talent), what pool do I pick from?  There is one qualification to that:

The answer (the overall level of talent in that pool) depends on whether you are a typical GM/owner/manager, or whether you are a good sabermetrician with a good projection system.  If you are the former, you are going to come up with a worse level of talent.  I lean towards answering the question from the perspective of the sabermetrician with good projection skills or perhaps the combination of a good sabermetrician and a good set of scouts and traditional baseball minds in general.  IOW, I want to know the best level of talent that is freely available using the best talent evaluation skills available (probably a combination of the sabermetrician and the best scouts/traditional baseball minds).

Further to my #10 above, where I showed that the ERA of pitchers 28 and older, with a salary of 0.50 million$ and less is 4.42.

This group had 14% of their games as starts.

Let me break this group of cheap old guys into exclusive starters and not exclusive.

There were 45 pitcher seasons where they pitched only as starters.  They averaged 136 IP each.  Their ERA was 4.52.

There were 393 pitcher seasons where these cheap old guys pitched at least 1 game in relief (7.5% of their games were as starters).  Their ERA was 4.40.

How does this compare to MLB from 2000-2006? If I look at the entire population of MLB pitchers who pitched 10% to 100% of their games in relief (average of 7.3% of their games as starters), these 1761 pitchers had an ERA of 4.42.

MLB pitchers with less than 10% of their games in relief (883 pitcher seasons), these 883 pitchers had an ERA of 4.30.

In both cases, the cheap old guys (aged 28 and older, making 500,000$ or less) had an ERA of 0.10 worse than their peers.

Tango, what is your point?  I think for one thing we’d really have to look at some of these guys to see why they were so cheap but pitched so well.  why do you think that the starters making .5 mil or less (presumably replacement pitchers) pitched so well?  I mean if a team can pick up a starter for .5 mil who will pitch to a 4.52 ERA it would be stupid for most teams to spend more money on any starter.  Something does not seem right here.

Plus if all non-closer and set-up relievers are paid around .5 mil or less (which may be close to the case), then there really is no such thing as a replacement middle or long reliever.

Guy/15,

As we are speaking about non-closer relievers, the assumption that the gap would be more than a run does not seem right to me.  We’ve pulled out from the sample the pitchers who most often throw an inning or less with commonly 2 pitches, and left the long men.  Admittedly, the LOOGies are left (pun not intended), and one suspects that as a class, they’d suffer more than a run hit if moved to the rotation.

In any event, the real question concerns replacement level pitcher in a starting role.  Alternatively, one has to account for the chaining effect on the pen.  Tango’s figure of .380 seems plausible.