Advanced Statistics: A Few Discussion Points

There is a statistical advantage to hitters in certain defensive alignments, such as when the infield is in double play depth or playing against the bunt. Essentially, in certain situations, the adage about runners on base causing havoc for the pitcher/defense can hold true. Is this statistical advantage used in calculations, such as wOBA? If not, how can defensive alignment be incorporated into an analysis of a hitter?

I don’t think so.

There is a demonstrated advantage hitters have with men on base, but I don’t think it’s accounted for in wOBA. As for the infield positioning thing… no
I bet some teams have a house stat or something that basically takes away all a guys cheap hits, but it isn’t in wOBA

RT @TBrownYahoo Rick Ankiel agrees to a one-year contract with Royals worth $3.25 million guaranteed. Second-year mutual option worth $6 million.

the original formula doesn’t have them, and I don’ t think Stat Corner (the first site to implement woBA) uses them. tango later noted that you could add them in to the “original formula” by using +.25 for SBs and -.5 for Cses

however, the most commonly cited implementation is af FanGraphs, and that DOES include SBs (although they don’t include ROEs). It should also be noted that the FanGraphs version doesn’t use the standard weights of the “original,” but uses Tango’s script (that anyone who works with Access or MySQL can get if you search) for generating custom weights for each season’s run environment.

see this post by Eric Seidman (now of BP) based on the work of Russell Carleton (now also of BP) done at the now sadlly defunct MVN site StatSpeak, which kicked f—-ing ass.

specifically on your point 3….

first of all, using something like SAS isn’t going to change the results. Whether you do an OLS regression in excel or SAS, you will get the same result (and excel is actually easier; I use both regularly at work). That said, some more advanced techniques would become available as you ween yourself off of Excel, but programs like R, Stata, etc. are just as capable, while much cheaper (base SAS is like $6k per user, plus extra for all the add-ons and stuff). A program like R is free and, as far as statistical capabilities needed for sabrmetrics, equivalent.

As for using more regression analysis, the short answer is of course it would be a great step forward. It has always bothered me the lack of statistical rigor most sabrmetrics is done with. Things like error analysis, distributions, and simple linear regressions are often either ignored, or very poorly done. Not to say there isn’t some good people out there doing it, but a vast majority of work you see on blogs doesn’t get much more advanced than a high school level stats class. Again, don’t get me wrong, that doesn’t mean current analysis out there is wrong, but it does mean “Truth” and “Fact” about baseball is a lot less concrete than some would have you believe (on either side of the stats debate). Part of the problem with it is the understanding gap between the baseball fans who demand this analysis, and the researchers who supply it. I could (and have tried to) talk all day about the economic fallacies of linearly priced WAR and salary caps, but unless you have a excellent understanding of economics, it gets discounted as an unknown and the simple explanation and rule is what gets propagated around the net.

That said, to be more specific with your desire for better projection systems, one thing in particular that I have always thought about is properly estimating weights for past years. You often hear about waiting past years by 5/3/1 or some variation of that to predict next year. The thing about that is those numbers are completely arbitrary. it would not be terribly hard for someone with access to all the data to actually run a regression and estimate the coefficients/weights of past years. Even with respect to age curves, position, etc. there is a lot of room to improve.

Going a step further however, a lot of projection systems actually already do use some regression analysis. Pecota comes to mind first, Silver used an extensive set of regressions (the way I understand it) to map players onto their comps. But I don’t know too much more detail.

Certainly a good post jsolo, important questions to think about.

Here is a great article on predicting attendance with regression:

MLB Attendance

Done with extra variables and t values compared. I think it is good for finding the values that matter with a bunch of unknown inputs.

Also I used it to find effects on parks for runs scored:

Park Factors

Number 2

A problem I had was that the higher the outfield wall the more runs were scored, which doesn’t make sense.

There is a place for regression, but it also it results need to be examined for common sense.

I think it is more important to ask a question and then find the solution how ever you can. Publish results. Get torn apart. Go back and fix.

The Unified Field Theory is nigh