A Simple Request Regarding Niese

From what I gather, the buzz is … ellipses … mean business.

PITCHER wins are.

In those games, the Mets have averaged 6.5 runs scored, well done Livan.

Holds are pretty stupid too

Alan Embree came into a game this year, picked off a runner to end the inning before throwing a pitched, got pinch hit for next inning, and got the win. He got a win without throwing a pitch, brilliant stat.

K% and GB% tell you exactly what they claim to, the percentage of strike outs and ground balls a pitcher gets. That result is completely dependent on the pitchers ability when given a large enough sample size. It is definite, there is no way to say, if pitcher A was on another team in another ballpark the results would be different, that is what a pitcher is capable of. In the rules of the win loss record statistic, a pitcher can start 33 games in a year, go 9 innings, give up one run and only one run every single game and go 0-33. Pedro Martinez once pitched a perfect game through 9 innings, his team didn’t score until the tenth. Sure he won the game, but it was entirely possible he could have lost that game in the tenth or received a no decision. The number of wins a pitcher has is not indicative of a pitchers performance. Look at Steve Trachsel in 05, he was terrible but won 15 games because the run support the Mets gave him. By looking at K%, BB%, GB%, etc, you find out the pitchers abilities independent of teams performance.

The point is you have a set of data, and you analyze it. You can always get more precise, and you try to, but you’re dealing with so many numbers that if you can eliminate a level of precision or imprecision as gratuitous or unnecessary, you take advantage.

The problem with situational dependence is that some people talk about it like its the holy grail of statistical analysis, which couldn’t be further from the truth. On the micro scale of game theory, the situational element is important, very important. It effects various levels and elements of realtime decision making and reaction. The goal of statistical analysis is that you want to understand how those realtime situations correlate to various particular outcomes. What a stat like GB% tells you that a particular outcome is likely from all situations that fall into the set: [Pitcher X faces a batter]. Could we be more precise and ask questions about if that pitcher is any more likely to generate a groundball, say, with a runner at first and one out? Sure, but the amount of extra info you get by asking that question will tell us relative to just asking “what is a pitchers GB%” is much smaller than the difference between asking “what is a pitchers GB%?” and “What is a pitchers W-L record?” And generally the questions “What is a pitchers GB%?” and “What is a pitchers GB% with a runner on first and one out?” will have virtually the same answer if you let the sample size get big enough. This is because most pitchers would usually like to get groundballs anyway. Unless Ichiro circa 2001 is at the plate, a groundball is usually a favorable outcome. So perhaps the more variant and interesting question to ask is: “What is a Pitcher X’s GB% in situations where a groundball is not a favorable outcome?”

In the end, the point is you can get all the same information that W-L provides you from other statistical sources and so much more, so it becomes a fairly useless stat, at least in terms of analysis.