Chat: 9/16/15 6:42am

Chat: 9/16/15 6:42am

Interests: Baseball Independent
9/16/15

Shouts

1
okdan's picture
9/16/15 10:06pm

<p>Hey Doc, the site is getting cluttered up with lots of content from authors who have never participated in the community. It's sad to see. I'm gonna stay away for a while and unsubscribe unfortunately from the RSS feed. Hoping to see more Doc and others, and less these SEO-driven WWE and algorithmically written Hawks posts.</p>

2
9/16/15 5:35pm

<p>Hey guys, I just started an account to write about some boxing and thought I'd introduce myself in here. If you have any writing tips or fights you want to see written up, you should let me know!</p>

3
anonymous (not verified)
9/16/15 2:47pm

<p>Glad to hear it Matt! I look forward to seeing what you shake out of the numbers tree. ~G</p>

4
SABR Matt's picture
9/16/15 12:46pm

<p>Small note: for those of you who contributed to my purchase of a MATLAB license to do advanced sabermetrics research in ways that MySQL struggles to handle (especially on data that is sequenced and the order matters), THANK YOU. Now that I have finally been able to afford a computer that can do the job, I am finding MATLAB far better suited to analyzing #Baseball data than either R or MySQL ever were. Some interesting things will be happening soon here.</p>
<div class="indented">Reply - Fungineer - 9/16/15 4:08pm<br>Great to hear Matt. I was one of the contributors and have been eagerly awaiting your findings. However, I have no issue with you publishing them in the (presumably) offseason since it will give us something to talk about besides the identity and methods of the incoming GM and the usual daily free agent/trade rumours we have to triangulate</div><div class="indented">Reply - Browns8625 - 9/17/15 4:12pm<br>Your sabr stuff gas been amazing I must say. It us tge new trend in baseball, and for good reason</div>

5
SABR Matt's picture
9/16/15 7:09am

<p>For those unfamiliar with how Elo ratings are calculated, the formula is pretty simple for chess, because each game is a win, loss or draw. In base, there are degrees of winning once you go past the game-level outcomes. A home run is a bigger win than a walk...a GIDP is a much bigger loss than a K. This means that when we calculate the change in Elo rating for a given plate appearance, we have to rate how much of a win it was for each player (I propose to do this with a pythagorean approach comparing the run expectancy for the average play in the given starting state to the runs created by the actual play, with each linearly adjusted so that the worst possible outcome has a run value of zero).</p>
<div class="indented">Reply - Browns8625 - 9/17/15 4:13pm<br>Any way you coukd draw out a real life problem.. I get it, but not on your level. Amazing stuff</div>

6
SABR Matt's picture
9/16/15 6:50am

<p>Also note - I am of the opinion that park factors should be additive, not multiplicative...that a hitter's park adds X number of runs per game per side - it does not multiply the scoring by some percentage larger than 100...so the alternative to context-adjusting the probabilities of various events is to context-adjust the Elo rating itself by adding a +/- Z to the event winning percentages (the step right before you find the Elo change score...the W% is leverage and match-up #Independent...if I did this to the Elo ratings or the change scores for each play, I'd be calculating averages that are biased by strength of match-ups and leverage index) based on the average event W% for the given context.</p>
<div class="indented">Reply - Fungineer - 9/16/15 4:29pm<br>Hi Matt - I'm not entirely sure I understand this and I'm fairly sabermetric so maybe it's worth including your methods (in laymans terms) when you publish your findings?</div><div class="indented">Reply - SABR Matt - 9/16/15 8:09pm<br>Hey fungineer...yeah...my apologies...I won't publish findings without a white paper describing how I got them...I was noodling here to see if I could work out some of my issues by trying to explain them to the room...LOL Probably shouldn't do that...but it did help me think things over.</div>

7
SABR Matt's picture
9/16/15 6:42am

<p>philosophical / methodological problem - does one want to context-adjust match-up based Elo ratings for #Baseball players?</p>
<p>Point #1 - A hitter's park will make the average batter look more successful than the average pitcher - the hitters will get more PA "wins".<br />
Point #2 - The relative frequency of events in a league is dependent on the scoring context of the league, to some degree...but also on the talent in the league.<br />
Point #3 - A home run in a hitter's park is still a "win" for the batter and some players transcend the parks (Nellie Cruz, e.g.)<br />
Point #4 - Context-adjusting the probability distribution of events that could transpire in a given situation will be exceedingly difficult, but not impossible.<br />
Point #5 - These Elo ratings could prove highly informative regarding clutch performance and even regarding minor league prospects and their likelihood for major league success...or they could wind up being just "fun". And it is difficult to know until I've attempted them.</p>
<div class="indented">Reply - Browns8625 - 9/17/15 4:18pm<br>This is great stuff and good additional factors.. reminds me of when nfl began to look more at qbr than regular passer ratings. Just added and accounted for much more</div>