THE BOOK--Playing The Percentages In Baseball

VORP uses a different replacement level for catchers than other positions - I think it’s 85%. It doesn’t seem like RARP does. My gut feeling is that RARP is correct on this.

Based solely on your lists above, it appears that RARP may like walks more than VORP, with Giles, Cust & Wilkerson topping the delta list.

I notice that 3 A’s are on the second list and, in fact, every Oakland regular is rated higher by RARP than by VORP. The difference is an average of 1.1 runs/100 PA. That has to be something systemic like they’re using very different park factors.

I have no doubt that there are many nuances, some major, like the low walks for VORP, and some minor, like a park factor for a few parks.  Add in a differing replacement level by position, and, boom, it’s hard to tell why one is doing what it is doing, without alot of work.

That’s why I did a reimplementation of MLV and BRAA to do my comparison - that way I avoided the question of whether the differences I was seeing were due to different datasets, park factors, replacement levels, and just focused on the run estimation.

Also, what are you doing about players who played multiple positions? I believe VORP counts each player’s production by position seperately, while RARP rolls them all up into one value.

First off, I added up exactly the RARP and VORP numbers as BP has.

I assigned the primary position as Clay had to each player.

From a quick look at the 20 players with differences, I’d guess VORP to be the more realistic of the two metrics.

VORP also uses 75% of positional average for 1B (and DH), which is lower than the 80% of most other positions.  That seems dumb and overrates 1B, and means I disagree with Hugh.

Carlos Zambrano is an interesting test case, since he doesn’t walk at all.

Hugh, what makes you think VORP is “more realistic”? Because I don’t see it.

The gap between Zambrano and Lohse is 24 runs in VORP and 9 runs in RARP.  Those are the leader and trailer by both measures.

There should not be any kind of difference like that, especially since the overall pitchers as a group are both close to zero.

B-R.com, using Palmer’s LWTS, has a 17 run gap between the two.  If you adjust for the slight PA difference, you’ll get close to 20 runs, which is “close enough” to VORP and “far away” from RARP to give you pause on RARP for pitchers-as-hitters.

I do know that I have a big issue with Equivalent Runs at the low-end, right around where the pitchers hitting levels is, actually.  This may be a manifestation of that.

I presume this is what you’re talking about:

http://www.insidethebook.com/ee/index.php/site/comments/why_is_eqa_so_complicated/

If so, that’s a problem with EqA, not EqR. I don’t know if Clay uses EqA to figure out RARP or not. (You could do it either way, I think.)

But all linear run estimators should have some issues at the extremes, not just EqR.

It may be an EqR issue:

“4. EQR = (2*RAW/LgRAW - 1)*(LgR/LgPA)*PA “

If RAW is half the lgRAW, EqR = 0.  It might even get in the negative.

I’ve never looked into pitchers-as-hitters, but I would bet that there is an issue there, be it EqR or EqA.

At the very least, we can see that the sympton is there: having Zambrano just be 9 runs better than Lohse as a hitter is inconsistent with the Woolner and Palmer approach.  Seeing that I have had misgivings with EqA at the bottom-end, it’s almost certainly an EqR issue.

My comment at #8 is wrong.  VORP uses 85% of average for 1B/DH.  And #1 is wrong—it’s 75% for catchers.

Based on this, I’m not sure why DHs tend to appear overrated by VORP (see Aubrey Huff this year) or why the total catcher value always seems really low.

Got it.  DHs are overrated by VORP because they are compared to other DHs—which tends to be a rather uninspiring position—instead of being compared to a level of performance at least equal in quality to 1Bs.  This is a good launching point for explaining why positional adjustments based on fielding or player pool are better.

I feel like I’m in a two year time warp thinking this much about VORP.

Using the appearances table in the BDB and some linear weights, I managed to cobble together values that look a lot like VORP, but without some of the extra baggage:

http://www.editgrid.com/user/cwyers/Linear_Weights_VORP

I’m not exactly proud of the park factors - it’s just the Batting Park Factors from the teams table with a little regression thrown in. I am somewhat more pleased with how I figured replacement level - it’s close enough to how BP does it in aggregate, but on a position-by-position basis it’s based on Tango’s adjustments.

I just posted projections for all of the current minor league rfee agents, including the hightest level they played at this past season.

I hope this may be helpful in empirically deriving values for replacement level.

http://statspeak.net/2008/11/freely-available-talent.html

In the second sheet of your file, those numbers (like .121) are runs per PA?  If yes, your replacement level is almost exactly the same as mine, which is around -18 runs (below average) per 630 PA. How do you get these replacement numbers from your minor league FA?

Unless I’m missing something, I think MGL is reading what Brian wrote and looking at my spreadsheet. I didn’t use minor league free agents at all.

Those are R/PA. I set replacement level at -20 runs per 700 PA, and then adjusted by position from there, using the figures Tango came up with from your UZR data.

That said, a quick look at Brian’s spreadsheet gives me -19 BRAA for the players listed per 700 PA. (Just did a weighted average of wOBA, using AB as the weight - not perfect but workable. .306 wOBA as a group.) So pretty close.