Tuesday, June 14, 2005

ERV Boxscore vs. WPA Scoring

If you haven't seen it already, Studes over at Hardball Times did a review of Saturday night's Nats-Mariners tilt at RFK using the "Win Probability Added" method of scoring. Studes' work provides a nice comparison to DM's own ERV scoring of the same game.

For the most part, a non-statistical observer, and ERV fan, and a WPA scorer would see this game the same. Everyone and their brother knows that Guillen's single to score the go-ahead run in the 7th was the biggest play of the game. In a 2-1 game, it's easy to see.

There are subtle difference, however, and it's worth comparing the two. Byrd was a net negative under DM's scoring, while he came out as the fourth most valuable player in Studes' system. Ryan Church was a hugely valuable player according to DM's ERV/WV system, while he was only about as valuable as Byrd in Studes' scoring. Fascinating that similar approaches could yield different results.

The problem is, neither approach can tell you whether there was enough Dippin' Dots on hand Saturday night. I mean, the net value of the Ice Cream of the Future on a muggy DC evening was HUGE.


At 10:02 PM, Blogger DM said...

"Fascinating that similar approaches could yield different results."

Not that fascinating when you realize that my spreadsheet has a minor bug in it that short-changed Byrd's WV. Thanks for the post, which helped me identify it. I'm working on fixing it now.

Thanks for the post, generally, too. I've sent an email to Studes for more info.

At 10:04 AM, Blogger Studes said...

Thanks a lot for the link. I was going to comment that actually Guillen's hit wasn't the biggest play of the game, because the bases were already loaded with one out. A run in that situation was nearly certain.

The biggest play of the game, as DM has inferred (I think) was Byrd's triple with two out.

At 10:15 AM, Blogger DM said...

You're welcome, Studes. I took a look last night at Byrd's WV for the triple. He should have gotten 1.97, which makes it the second-most valuable play in the game behind Guillen's single, which was 2.91. My system (based in the increase each run provides on the probability that the team will win, on average)generally rates go-ahead runs more valuably than tying runs, and those that come later in the game, because the visiting team has less chance of coming back then.

At 4:07 PM, Blogger Studes said...

Sorry for my slow response, but I certainly agree that go-ahead runs and runs that are scored later in games are more important for winning.

But the difference between the two was the Guillen's hit came with the bases loaded and one out, while Byrd's came with two out and just a runner on third. IOW, a lot of the win value was already accounted for and on base when Guillen clubbed his single, based on the odds of someone knocking at least one of those runners home.

I'm wondering if this isn't the difference between your system and WPA. Is yours adequately assigning WV to those who got on base before Guillen's hit?

Just kicking it around a bit...

At 5:51 PM, Blogger DM said...

OK, here's the theory behind how Win Value assigns values to runs. (I had to rebuild my spreadsheet to accurately implement this, it was missing a few situations like Byrd's).

As explained, each run has a different value (e.g. the first run the top of the first is 0.61). I start with the RV for the play, then modify it to WV. For example, if the leadoff guy gets on, the RV for that play is 0.39, or 39% of a run. But we know that that run is only worth 0.61, so the WV for that play is 0.39*0.61, or about 0.24. It gets more complicated as the RV's go over and under 1.0 & 2.0, because the run after 1.0 is worth a different amount than the run up to that one. This helps deal with the "sac fly problem" is discuss in the main win value approach.

So, yes, Studes, buy starting with the RV I account for the runs earned getting the runner to 3rd. My current spreadsheet does now reflect that Byrd's hit was more valuable,but that was due to the bug in my sheet, not the theory.

Also, I figured out a way to display a running WV chart for the Nats over the course of a game. I'll try to post some, and we can compare them to WPA. I think they will be almost identical in many respects.

I like to picture the men on base has having the WV for the run they represent floating over their head, so that if Wilkerson walks to lead off the game for the visiting Nats, he would have a 0.61 floating over his head. (It's not precisely like that -- the runs are not discretely assigned to individual runners, but that illustration helps me a bit).

At 6:35 PM, Blogger Studes said...

Sounds good, DM. Thanks.


Post a Comment

<< Home