and rec’d

There are two things “he” doesn’t like about the Astros so far. One, the run differential doesn’t support the record. We’ve all been through all the Pythag issues. I know it’s not perfect, but it shouldn’t be ignored. Two, based on UZR/RZR and tRA, the Astros “should” have given up 23 more runs than they actually have. (And scored 8 more, to be fair.) One thing Justin doesn’t consider is strength of schedule.

http://www.beyondtheboxscore.com/2009/7/22/958219/btb-power-rankings-through-tuesday

Avoids the issue of the quality of RJ’s article, sure, but addresses the same topic.

Personally, while I don’t particularly like the Astros, I’ll be rooting for them, as close races are highly entertaining.

but, as it turns out, I read DQ’s crtique at the same time that I was ruminating over the interesting discussion in this thread at BBTF, which was started by MGL’s borderline-arrogant contention that much of the sabermetric community doesn’t really understand “regression to mean.” Since RJ’s article is really another version of the “regression to mean” argument, it was interesting thinking of the article in the context of some of the arguments back and forth in the BBTF thread. Regression to mean is based on the concept that an overall population tends to move toward the population mean over time. And that implies that most members of the population will move toward the population mean over time. But that means most, but not all, members should regress in the direction of the mean. The question is, how do you go about identifying players and teams which are playing at their true talent level? RJ Anderson’s article implies that this can be proven, in the case of the Astros—something which I think overstates the state of our knowledge.

The concept of “luck” is probably overused (and sometimes misused). In the context of RJ’s article, luck is all performance which doesn’t reflect the player’s true talent level. The problem is that we don’t know what the player’s true talent level is, and we can only make inferences from whatever statistical data we can observe. But true talent level varies over time, can be affected by injuries and repairs of injuries, players’ ability to make changes in their games, etc. In the case of the Astros, I see plenty of instances where our ability to judge player’s current talent level is difficult. Forecasting systems like ZIPS, PECOTA, etc., are weak at forecasting young players with little ML experience, and Arias and Fulchino are Astros players who probably were subject to this projection weakness. To some extent, Pence and Bourn fall in the same category. I have written previously about Tejada and the difficulty in forecasting the rate of decline for players like him, who sometimes even experience near-peak or peak performances in late years. I think it is very possible that Tejada, like Biggio, is adjusting his style of hitting as he gets older, and again, this is difficult to reflect in projection systems. None of the player projection systems handle injuiries and post surgery performance well. And the Astros are depending on two starting pitchers who are post-injury and post-surgery cases. The impact of TJ surgery, like Ortiz’s, in particular can make a ZIPS projection irrelevant. I think it is worth noting that the community WAR projection, which incorporates more detailed knowledge of the Astros’ players, projected a W/L record a couple of games above .500, which is considerably better than the “mainstream” forecasts.

If RJ’s article had simply concluded “the Astros might be this good, but I wouldn’t bet on this performance continuing,” I doubt that many of us at TCB would have reacted so negatively. I think the smug tone, though, that “we know this team is bad,” makes us think that the article is more preconception than analysis.