Undervalues walks by as much as OPS does? Didn't see that coming.

Good read. Also was happy to see the definition of BaseRuns in the comments. I'm going to be trying that out.

Dear Mr. Schlub,

Thank you for your letter to Baseball Prospectus. As you know, we value the input of all our readers.

However, we are sorry but we are unable to fulfill your request for an autographed photo of Will Carroll in a bikini.

We do find this concept of linear eights to be of great interest and we would like to subscribe to your newsletter.

Thank you again for taking the time to write.

Mama Prospectus

PS Everyone knows that BP is Baseball Primer although we appreciate the compliment.

PS Everyone knows that BP is Baseball Primer although we appreciate the compliment.

BTF should send BPro a cease and desist letter. ;-)

Bet maybe 20% of people around here remember that.

good
old
days

I asked someone who would know, and he said VORP is built to use any run estimator, so while the VORP at BPro may do what Tango says, and I believe it does, I think that this difference may be similar to the difference between RC and RC-Tech.

Anyone know why nobody's tried geometric weighting? [IIRC that would involve each run element being dependent on the magnitude/frequency of other run elements] I'm just surprised that most analysts are still using variations on a run-estimating system which is about a third of a century old.

agree with Chris (post 5). Basic RC weighs walks a little light (just under half of single). RC-Tech, which gives walks about 1/4 value of a single in the TB total, gives walks the value of about 60% of a single.

John/6: are you aware of BaseRuns?

Tom/7, Chris/5: yes, (some of) the RC-tech version is better than the basic RC version. Had Keith used that, I really wouldn't have much of an issue. Note that the current RC-tech version that James employs way overvalues the single, as Patriot has demonstrated on his blog a few weeks ago.

I asked someone who would know, and he said VORP is built to use any run estimator, so while the VORP at BPro may do what Tango says, and I believe it does, I think that this difference may be similar to the difference between RC and RC-Tech.

Right. No reason you can't sub in Base Runs as the run estimator in VORP. I might take a stab at that today.

Does the Runs: (1.73*OBP + SLG) * 0.27 * PA BaseRuns relationship hold over time? Obviously not to the high error environment of 100 years ago, but say in the 1930 NL or the 1919 AL?

jwb/10: that relationship to BaseRuns (or Markov) holds fairly strong, except for the HR. The run value of the HR should be pretty static around 1.4 runs regardless of (reasonable) run environment (Pedro to Bonds). However, that quick equation you are citing has no mechanism to stick to that, and therefore, the run value of the HR will differ wildly from what it should be, the more extreme the run environment.

Right. No reason you can't sub in Base Runs as the run estimator in VORP. I might take a stab at that today.

This is important, because I love the construct of VORP. Of course, Keith was one of the most open USENETters a decade ago when I tried my hand at stats.

Chris, it shouldn't be too hard. Patriot gives us the formula for theoretical team BaseRuns, which is essentially what Marginal Lineup Value is. VORP is a player's MLV minus a replacement player's MLV over the same number of PAs. I'd say you're looking at half an hour to back-calculate the component stats you need for your replacement players, another half-hour to implement TT BsR in Excel with everybody's stats. Really the hard part is the data gathering; I know the equation for TT BsR looks a bit rough, but that's why you let the computer do the heavy lifting.

Really the hard part is the data gathering; I know the equation for TT BsR looks a bit rough, but that's why you let the computer do the heavy lifting.

Harold is helping me out. He's a good chap.

Colin: after you do that, compare that a straight Linear Weights, and list all the players with more than a 2 run difference. That'll either show you where you may have a bias (in either LWTS or BsR), or, if the list is null, how BsR is really the same as LWTS.

Colin: after you do that, compare that a straight Linear Weights, and list all the players with more than a 2 run difference. That'll either show you where you may have a bias (in either LWTS or BsR), or, if the list is null, how BsR is really the same as LWTS.
Page 1 of 1 pages

I already have a spreadhseet with XR, so I can do that when Harold scores me teh BsR spreadsheet.

Chris, you would need to compare BsR to the intrinsic Linear Weights values it would generate using the Plus1 method (like here). One question at hand is if the "Theoretical Team" (TT) construct (team runs with and without player) buys you anything more than 1 or 2 runs even for extreme players.

If you compare TT construct using BsR to some linear method other than LWTS-generated BsR, you won't know if the difference is because of the different weights of the linear method, or the difference between TT and the linear method.

Tom - I'm pretty sure that Patriot did just that at the link provided, at least for A-Rod.

Right. Now a matter of doing it for everyone. If you find that every single player is within 2 runs using the TT construct, does it make any sense, effort-wise, to implement TT? If you do find at least 1 player, say 3 runs or more of error, then I'd say, yes, it'd be worth it.

This is the same as FIP and DIPS. FIP is the linear shortcut for the more complicated DIPS process. Since DIPS is so far more complicated to implement than FIP, I'm happy with what FIP provides as a shortcut.

While the TT construct of run creation is definitely the correct thing to do, if LWTS as a shortcut gets you 100% of the way there (+/- 2 runs), then I'd say the same thing applies.

if LWTS as a shortcut gets you 100% of the way there (+/- 2 runs), then I'd say the same thing applies.

That's provided that the deltas are not systematic. If the deltas all go one way, then you have to look at why that is.

-- MWE

Okay, but if you have a difference between TT BsR and empirically-generated linear weights like the one Ruane has, which one would you give preference to? My gut feeling is the empiric linear weights, at least for hitters.

Right, in terms of the bias. And, it would almost certainly be a systematic bias, not random.

Though, preferably it's a bias that would affect all great players the same, since, typically, you are comparing players of the same quality. If Dunn and Bay and Miguel Cabrera are all biased at +1.5 runs because the bias is at the quality level, it'd basically be a wash. If Miguel is given +2 too much and Melky is at -2 too much the other way, that's a 4-run gap, but you probably are not comparing those two guys to begin with. Dunn being nicked on walks and HR, while an equally good batter getting undeserved bonus on singles and doubles would be more problematic.

In any case, 100% of players within +/- 2 runs, that's probably within a range that most people would say "same thing". Just a matter of deciding what your threshhold is for "same thing".

Colin: you can tweak the values of the TT BsR to make it inline with Ruane's numbers. Once you do that, the TT BsR is the preferred model. But, like I said, it probably won't matter since you'll get virtually the same results as with Ruane's numbers.

Trying to follow this conversation is making my head hurt. I really wish I wasn't so horrible at mathematics...

Put a normal stat line in VORP. See what you get. Now, add a single into VORP. See what you get. The difference between the two "gets" is the impact of the single. Do this for each event. That'll give you how much "weight" each event has on VORP.

We know what the relative weights of each event should be. As it turns out, the weighting of the walk is undervalued by a fair amount.

That help?

It does. Thanks.

I really wish I wasn't so horrible at mathematics...

Don't be so hard on yourself. For a chimp, your math is pretty good.