Comments on: Ramon Hernandez and a catcher leading the way in WPA This and that about baseball stats. Tue, 16 Jul 2013 17:01:55 +0000 hourly 1 By: Johnny Twisto Sun, 12 Jun 2011 01:47:58 +0000 Sorry if it's been asked before...does the WPA account for fielding events of significance

It does not. All credit is given to the pitcher. It will be a great step when someone can devise an objective, consistent way to divide credit with the fielders too. We brainstormed a bit a couple months ago on how that might be done.

Rudi's catch was worth about 0.07 WPA, improving Oakland's chance of winning from 85 to 91% (rounding errors, apparently). OAK was up 2-0 with 1 out and none on in the bottom of the 9th. I guess the real value of the catch is what the win expectancy would be if he doesn't make it. CIN has a runner on 2nd(?), 1 out, tying run up. No idea what OAK's WE would be there -- 75%? So the swing is between ~75 and 91.

By: nesnhab Sat, 11 Jun 2011 12:16:11 +0000 The batter who hit Rudi's catch was Denis Menke...IIRC

By: nesnhab Sat, 11 Jun 2011 12:14:01 +0000 Sorry if it's been asked before...does the WPA account for fielding events of significance...for example, Joe Rudi's leaping catch in 9th, G2, 1972 WS? And, what was that catch (out) considered of any significance by the WPA compared to other events that day?

By: Johnny Twisto Fri, 10 Jun 2011 04:55:46 +0000 I think a Markov is used, that is referred to in The Book.

By: kds Fri, 10 Jun 2011 03:49:54 +0000 JT and BSK, I think it is likely that a Markov chain simulator is used to generate the WPA figures.
Home team down by a run loads the bases in the bottom of the 9th with none out, in a 5 runs/gm environment they have a 74.92% chance of winning. So a triple play would get the pitcher .7492 WPA. I think this is the highest possible for one play by the pitcher/defense. (Given the run environment.) Thanks to Topper009 for that link, lots of fun.

By: John Autin Thu, 09 Jun 2011 14:26:51 +0000 @40, Doug -- That situation happened in reverse to Matt Holliday in 2009. With Oakland, he played 93 of 94 games through July 23, then was traded to St. Louis. He played with the Cards the next day, their 99th game -- so he "lost" 4 potential games. Holliday missed just 1 game with Oakland and 1 with St. Louis, but played just 156 games. Had the trade gone the other way, he might have played 164 games.

These days, with free-agents-to-be often dealt at the deadline, I don't think it's uncommon for an everyday player to be traded during the season.

By: Doug Thu, 09 Jun 2011 04:29:02 +0000 @6, @34.

Another way a player could end up playing more than 162 is if he is traded in mid-season from a team that had played more games than the team he is traded to.

My hunch, though, is this is pretty unlikely to happen since you don't usually trade a player who's been in your lineup every game. Probably, he'd lose some playing time first before you traded him. Could be possible, perhaps, if it was a salary dump or an unsignable free agent. But, even then, if a deal was close to being consummated, that player would probably be sat down just prior to the deal going down, to avoid risk of injury.

By: BSK Thu, 09 Jun 2011 04:00:35 +0000 JT-

I actually picked up "The Book" from the library recently and have been working my way through it (as "math brained" as I am, stats was never my strong suit). I should plug B-R for inspiring me to do so. It was my understanding the the Win Expectancy and Run Expectancy were entirely empirical, though I'd probably still consider them such even allowing for minor tweaks. It actually makes a lot of sense that Win Expectancy would not be based solely on W-L records after certain base-out-run-inning situations have occurred. But it is still pulled from historical data, namely Run Expectancy.

I got into it a bit with Andy once regarding this same issue with CoolStandings (I think it was over the infamous Phillies post). It is something that has generally always bothered me about such approaches to predictions. Saying something has "never" happened before doesn't mean it can't happen or won't happen; only that it is probably something very unlikely to happen. Now, when a data sample gets large enough, we probably can make some generalizations, but there will always be noise in the data.

As I noted before, we can look at a coin toss and *know* (assuming a fair coin) the likelihood of a given flip coming up heads or tails. From there, we can determine the likelihood of any combination of flips. We can compare this to the historical data, but the theoretical odds will *never* change. This is not true of baseball. As you note, run environments play a huge factor. Overcoming a 7-run deficit in the mid-90's is very different than doing so in the ead Ball era. How different? We can probably approximate, but never know for sure.

In the end, we're arguing (if we even are arguing!) about numbers a place or two too far to the right of the decimal point to be truly meaningful. What matters most is that we understand what the stats are attempting to tell us and whether or not they are successful in those endeavors (or, really, the people behind the stats... the stats are simply what they are).

I'll keep going through "The Book" and hopefully can add more to the conversation.

By: Johnny Twisto Thu, 09 Jun 2011 03:41:41 +0000 BSK, I suppose it's more accurate to say it's theoretical based on empirical. (Well, maybe that's true of everything theoretical....)

Win expectancy is based on run expectancy. Run expectancy is based the observed runs scored from each of the 24 possible base-out states in an inning. But it is not based solely on those actual observations, it is tweaked a bit. This is done both to account for insufficient sample size and to account for different run environments.

I suggest reading the first chapter of The Book. (I see you can read a piece of it at Amazon. Anyone know if there's a way to read more? I feel like Tango has often referred people to some free site to read the book, but I don't remember if that was Amazon or something else that has even more of it available.)

And I know you're not being snarky. If you have more questions, I'll do my best to answer them, though we may be butting up against the limits of my abilities here....

By: Neil L. Thu, 09 Jun 2011 03:13:38 +0000 @36
Thom-13, the count is not available in BRef for the Roger Freed game.

In a gamelog the count is shown in the fifth column from the left. You can see that for the Freed game that column is blank.