THE BOOK--Playing The Percentages In Baseball

Yes, I was thrilled that he did it the “right” way, before the fact and not after the fact.

I have always said that replacement is about .75 runs over average.

This study showed a run or so, as Tango says. But that is with inefficient player evaluation by teams.  I am quite certain that I (or a smart team) can find replacements that are closer to 3/4 of a run worse than average.  Quite certain.

I dunno what the “.75 over average” means, unless you talk in terms of the replacement’s role as a starter and reliever.

If you mean a replacement’s ERA based on pitching two-thirds of his innings as a starter and one-third as a reliever, then I can accept the .75 as reasonable.

***

For example, the average starter in 2007 had an RA of 4.99, and the average reliever was 4.55 (overall league average is 4.83).  Component RA would be similar.  Anyway, this works out to a winning percentage of .485 for starters and .528 for relievers.

If we presume a .090 win difference between the roles of starter and reliever for the same pitcher, then a .528 average reliever would be .438 as a starter.  Obviously, the top relievers are not emergency starters.  If we take out the ace relievers, the rest of the relievers are probably .500 as relievers.  So, a .500 reliever is a .410 starter. The magpie .413 number seems reasonable to me.

In other words, an “average pitcher” would have an RA of 5.13 as a starter (.470 win%) and 4.23 as a reliever (.560 win%).

If MGL is saying that a “replacement pitcher” would have an RA of 0.75 above that (5.88, 4.98 respectively), I’m ok with that (implies win% of .405, .485).

I prefer using .380 and .470.

At one point I took ZIPS projections for the last couple years and tried to establish “replacement” levels for pitching (and actually for each position as well), under the assumption ZIPS would give an accurate picture of talent supply for baseball.  I used ZIPS (I could use PECOTA if I’d subscribe to BP) because it contains minor league players, and alots playing time as if the player actually played a full year, whether they do or not.  I’m not sure this is the “correct” way to figure replacement levels, but it made sense and I came up with very similar numbers to these.  I believe (I’ll need to look when I’m at my computer) I used 6 starters per team (as in 5 starters, 1 top prospect/extra pitcher/injury contingent) and averaged the remainder as well as those top 180 starters to get the numbers.  The number’s were almost exactly the same as in this article.  I believe the “average” (top 6 per team) starter had a projected RA of 4.96 and the replacements 5.75.

Whoops! that was ERA, not RA, and I’m not 100% on my numbers, I don’t have my spreadsheets

Also, I don’t the method I mentioned was entirely correct, it was more an exercise in toying with a projection system, that included minor league players, to determine what the actual supply is, or if it viewed the “replacements” as being close to what would be expected.  I’m sure there’s a much better way to do it than grabbing the top X (I used 6 starting pitchers per team, maybe top 5, or maybe all pitchers, I dunno), but it was easy to do with Excel and a few minutes.

My go to guy for a .500 pitcher is Oil Can Boyd:
http://www.baseballprospectus.com/dt/boydoi01.php

BP has his pitching runs above AVERAGE as +21 over 1390 IP (or 0.14 per 9IP better than average).  His runs above replacement is +335, or +2.17 runs per 9IP above replacement!

Carl Pavano:
http://www.baseballprospectus.com/dt/pavanca01.php

His runs above average is -3 over 1039, or essentially zero.  That is, average.  His RAR is +267 for an average of +2.31 runs per 9IP average replacement.

These numbers imply a win% of around .300 for a replacement-level pitcher.

***

Compare that to VORP:
http://www.baseballprospectus.com/statistics/sortable/index.php?cid=280461

Jon Garland, Miguel Batista, Jarrod Washburn all were league average pitchers according to VORP, and they were about +25 to +26 runs above replacement.  With about 22 full starts each (198 IP / 9), that makes their runs above replacement per 9IP at around +1.15.  This implies a replacement win% of .400.

***

How about it, BP?  Isn’t it confusing to have two wildly different definitions of replacement level?  Ditch one (the wrong one).

I was looking at BP’s RAR for a couple of pitchers the other day and I thought it seemed extremely high. They say the replacement level RA for pitchers is 6.11 (compared to 4.50 average) ...

http://www.baseballprospectus.com/glossary/index.php?mode=viewstat&stat=135

Off the cuff guess here ... is this just making up for the low replacement level for position players because of the fielding component .. i.e., putting pitchers on the same level as position players?

I suspect that’s the reason it is done.

On the other hand, giving the defense a .300 win percentage and the offense a .300 win percentage gives you a team win percentage of .155, which is consistent with Clay’s public position (25 wins per 162 GP).

Even if you WANT to have a .155 team winning percentage, you can’t forget about the fielders!  To get a .300 defense win percentage, you could have a .335 pitching win percentage and .460 fielding win percentage.

You (Tango) had the replacement starters as 1 run higher than league average in ERA (or RA I guess).

I claim that the ones who were replacements this year were NOT the best starters available, because all teams collectively are not efficient at evaluating pitcher talent.

The smarter teams would be able to choose better talent as replacements.  All I am saying is that given a smart team, a replacement starter (as a starter only) SHOULD be only .75 to .8 runs above average and not 1.0.  That’s all.  Obviously 1 run above average is what repalcements actually were, but that is not much of an interesting number.

If you are advising a team about salaries and what have you, you would advise them based on what THEY and YOU could put on the field as a replacement player, not what all teams collectively would.  And if you are smart and the team is smart or at least listened to you, you would use YOUR replacement level and not the actual replacement level.

I think I’m not explaining myself well enough with “average”.

The average starter had a RA of 4.99 in 2007.  This is a fact.

The average pitcher *as* a starter would have likely had a RA of 5.18.  That is an estimate.  That is, the average starter did have an RA of 4.99.  The average reliever *as a starter* would have likely had an RA of 5.55 (up from the 4.55 they *did* have as a reliever).  So, two thirds of 4.99 and one third of 5.55 gives us 5.18.

So, when you MGL say +0.75 in RA above average, is that above the 4.99 average or above the 5.18 average?

Would your average replacement pitcher *as a starter* have an RA of 5.74 or 5.93?

If the average starter has an ERA of 4.99 then I am saying that I can find a pitcher from the scrap heap that would have an ERA of 5.74 as a starter.

The caveat is that I don’t know that I can find one for EVERY team (IOW, there may not be enough to go around), but I think I can find one for MY team.  And that is only because there are not enough smart teams around.  Many of them are using “worse-than-replacement” starters in various starting roles.

I looked at my spread sheet last night, it was 2006-2007 ZIPS.  The average starting pitcher.  What I had was the top 300 (2 years, 5 starters per team) OVER age 27 (ie, likely still not a prospect) + a 30% contingency for injuries/attrition), so top 390, they averaged RA of 4.98 (so, the average starter was expected to have an RA for 4.97), then I looked at the next 60 + 30% attrition contingecy, and came out at 5.96 RA as the average.  If I did not use the 30% attrition (I’m not even sure the correct number, it was a WAG), got an RA of 4.84 for the first group and 5.60 for the second group.

I don’t think ZIPS was the best tool here, for a few reasons, non of which are knocks against ZIPS, but it jives with you numbers. 

Maybe the reason the 30% made the numbers so clase was the inefficiency MGL is talking about.

One thing to keep in mind about these “replacements”:  it includes guys named Clemens, Martinez, Morris, etc.  That is, these are not all replacement in the traditional sense of “freely available talent.” Plus, plenty of them are pre-arb players who were “free” (virtually) but not “available.” Whether it changes the results, I don’t know.  Mulder hardly came “free” to StL, but on the other hand pitched much worse than replacement level.

Tango:  is your RA average weighted by IP?  And if so, how much of a problem is selective sampling?  (those who succeed keep pitching, those who fail don’t)

I only looked at pitchers who had GS/G projected >= 90%, to avoid RA inflations.  A better way would probably to have taken any pitcher w/ GS/G projected >=67% (or so) and adjusted the RA numbers based on GS/G.  Also, ZIPS factors in defense and park, so using unadjusted RA is not correct, but since my goal was not a study, but rather a rough guage of replacement level, I think it’d work.  I actually started the spread sheet to look at replacement levels for hitters by position, but I see problems with that unless you had several years of projections.

#14 Guy, those reasons are why I looked at a pre-season projection for all players rather than results. 

I figured if I looked at players 27 and older I’d avoid a lot of the “prospect” status, ie, if you’re 27 and older, and projected to pitch at replacement level, chances are a team would be more willing to give you up for peanuts, and avoids guys like Clemens getting into the mix.  It wouldn’t account for older pitchers who are overpaid or overvalued (thus hardly free), but not very good any more though.  Also by looking at the projected RA rather than actual RA you avoid looking after the fact, and ZIPS assumes the pitcher gets to pitch for the said team whether it happens or not, which avoids the selective sampling based on innings pitched. 

What I did was toy with a spread sheet (oddly a couple days before this post, which always seems to happen), and does not count as a study.  My goal was more to toy with replacement levels based on projections rather than actually determining replacement levels.  My idea was a projection eliminated several selective sampling/bias issues with results based analysis.  Originalyl I was trying the guage the differences between average position players and replacements, but realized defense was a bigger issue than I realized, and without accurate defensive projecitons, the idea was mostly futile, which is why I shifted to looking at starting pitchers.

5.74 RA in a league of 4.83 RA implies a win% of .417.

Magpie’s study that is the thrust of this thread shows the replacements at 5.79 (win% of .413).

VORP says around 5.94 (win% of .400).
My model says 6.20 (win% of .380).
Clay Davenport implies 7.35 (win% of .300).

You definitely have some support with actual data.

I’ll argue that the magpie data does have some selection bias (the better replacements have more IP, and therefore more weight).

I don’t think we have much disagreement here.  Your “0.75 above starter average” implies 19% runs above the overall league average for all pitchers.  My number is at 28%.

The number of pitchers on the scrap heap is a legitimate question.  We’d have to assume there’s at least 30 of them there.

I did a similar study midway through the year (Tango linked to it so its here somewhere) but I made a point to remove any pitcher who would have been an opening day starter like Pedro but for injury, Roger Clemens as he’s anything but free talent, and also anyone from Baseball America’s top 100 preseason prospect list.

Do that and you’ll get a bit higher ERA for your replacements.

Guy, you are right that I shouldn’t be using the term replacement with respect to magpie’s study, since that’s not what he studied.  The better term for his pitchers are the “emergency starters”, and some of them find a home as a rotation starter.

What you could do with the magpie data is determine the salary and years of service data, and compare that to the average at those levels.  That’ll tell you how much you are paying for emergency, and how much above replacement those guys are costing you.

The numbers I cite (4.99, 4.55) are simply the overall league averages that you can find at b-r.com.