As baseball writers, analysts, and fans get set to lay out their predictions for the next six months, I find it useful to look back at last year and see what I can learn from the picks I made a season ago. I’ve been doing this exercise for a few years, and sometimes there are valuable lessons to be had. Sometimes, to be honest, there aren’t. Baseball’s funny that way.

The first thing to do is check the run environment. In my predictions, I had 23,358 runs being scored last season, which was high by about 3.5 percent; the actual total was 22,584. That’s the global error, so to make comparisons, I’ve gone ahead and adjusted all of my RS and RA predictions upward by that amount. Call it park effects for prognosticators.

Let’s make something clear: I’m predicting runs scored and allowed, and the records are simply a function of that. I make some minor adjustments to correct for rounding errors and the possible impact of a particularly strong or weak bullpen, but for the most part, I’m concerned with runs. We simply don’t have much evidence that outside of the effect of a bullpen, teams can distribute their runs in a way that gives them a leg up on the Pythagorean formula. So my predictions, and my evaluations of them, focus on runs. Nailing a team’s record but being off by 70 runs of differential is a bug, not a feature.

My Net Error Score (or NES), which is the total number of runs I missed by on all 60 predictions, was 3,517. That’s a tick worse than last year’s figure, and middle of the pack for the years in which I’ve done this. Here are the best and worst projections by that metric. “Pred” is the adjusted figure:

             Actual          Pred*
Team        RS    RA       RS    RA    NES
Royals     691   781      698   762     26
Red Sox    845   694      819   678     42
A's        646   690      661   718     44
Angels     765   697      781   727     46
Indians    805   761      805   710     52

Rangers    901   967      812   821    235
Pirates    735   884      638   755    226
Twins      829   744      650   713    211
Phillies   799   680      821   841    183
Cardinals  779   725      704   816    166

*Figures may not all match due to rounding.

The presence of the A’s and Indians up top isn’t much consolation. The A’s were actually a bit better than that for three months before trading away two of their top starting pitchers; by rights, that probably should have been a bigger miss. The Indians, of course, were so bad in close games that they traded away CC Sabathia while still having a positive run differential, and all things considered they might have won the division had they not done so.

The misses are something of a type. I wasn’t that far off on the differentials for the Rangers and Pirates, but completely blew their run environments, even after adjustments. Both teams’ staffs were horrible, bad enough to waste what ended up being better offenses than expected. The Twins famously had an amazing year hitting with runners in scoring position, which is why I missed their runs scored by more than one per game. The Phillies’ bullpen, which I expected to be below average, was excellent, and their defense was much improved as well. That’s how you miss runs allowed by one per game.

I think the above approach is the best way to evaluate predictions, but you could also give someone a pass for gaps that are of a kind, like the Pirates and Rangers above, and argue that nailing the run differential is what you’re really trying to do. In that case, the top and bottom look like this:

             Actual          Pred*
Team        RS    RA       RS    RA    NES
Brewers    750   689      823   768      6
Dodgers    700   648      758   700      6
Red Sox    845   794      819   678     10
A's        646   690      661   718     13
Angels     765   697      781   727     14

Cardinals  779   725      704   816    166
Tigers     821   857      891   770    157
Orioles    782   869      648   885    150
Twins      829   744      650   713    148
Rays       774   671      785   822    140

*Figures may not all match due to rounding.

Even here, the accurate predictions don’t really say much about any skills I might possess. To hit those numbers, the Brewers had to trade for a guy at midseason who pitched at a Cy Young-caliber level for three months. The Dodgers dealt for someone who hit .396 with power for two months. Without Sabathia and Manny Ramirez getting involved-and no, I didn’t see those deals coming in March-those picks would have been much further off. See also the A’s, who had to dump at midseason to get that close to my prediction.

On the flip side, you have the well-covered aging of the Detroit Tigers and the magical RISP talent of the Twins. There’s also the Rays, who had one of the greatest single-season turnarounds in run prevention in baseball history.

I’m seeing a lot of mistakes in evaluation, but what I’m not seeing is an actionable pattern. When I golf, I sometimes go an entire round pushing shots left, or leaving putts short, or fluffing chips. (Actually, I sometimes do all of these things at once.) This isn’t so bad, because if you’re making one mistake over and over, you can identify the error, correct it, and move on. But when you’re making different mistakes each time-alternating slices and hooks, hitting irons fat and thin, spraying putts all over the green-then you’re in trouble, because you can’t even figure out what to fix.

That’s what I’m seeing here. I missed a lot of stuff last year, none of which necessarily shows a blind spot, and some of which is just stuff that, no excuse implied, everyone missed. What the Rays did, what the Twins did, isn’t the kind of thing that you can see coming, which is why-say it with me now-they play the games.

One last thing I want to look at, however, is the very basic task of getting the arrows pointed in the right direction. As analysts, we should be able to classify teams as “going to be outscored” and “going to outscore their opponents” as a very basic measure of competence. Two years ago, I got just 18 teams right. Last year, I jumped all the way up to 19. This bugs me, because I have to think that a dart-throwing fourth-grader could get at least half of the teams right. It’s a little like picking the NCAA field; you’re not really predicting 65 or even 34 slots. You’re picking the last four or so, because most of the field is set for you. If you get two or three wrong, you’ve just wasted weeks of your life.

In baseball, everyone agrees that the Yankees, Red Sox, and Cubs are going to outscore their opponents, and that the Padres, Pirates, and Nationals will be outscored. You can probably even extend those categories by two or three each. The questions come in the middle. You come to BP to find out about the middle.

I’ll take my stab at that middle this week. Hopefully I’ll do better this time, but as you read my prediction sets and those of others this week, keep in mind that despite my breakdown of the numbers above, you want to worry less about the numbers and more about the analysis. It’s the words, the thought process, that matters, and I’ll say that even after I run my Net Error Score into double-digits this year.

You need to be logged in to comment. Login or Subscribe
You must be a lefty? Thanks for the work.
No predictatron this year?
No, but HACKING MASS is coming soon.
Oh..that sucks!
I'm not sure adjusting the runs environment globally is a fair adjustment. I understand why you do it, and that you want to concentrate on relative differences and on runs scored and runs allowed, but I don't know that you can just assume that runs scored and allowed globally is not something you are also predicting, or more importantly that this missed prediction is going to be spread equally amongst all teams. Was it really the case that globally each team scored and allowed 3.5% less than you predicted or is it maybe the case that certain teams or certain divisions had more or less runs scored.
what happened to Predicatron this year?
If in your predictions, total runs scored was high by 3.5 percent, the proper adjustment should be adjusting predicted RS and RA downward instead upward, isn't it?
That was my first thought too. Can we get a clarification?
I went back and checked. I wrote it wrong: I adjusted the predictions downward by 3.5 percent correctly in the spreadsheet. As to mbodell's point...I actually was thinking about this yesterday myself. I'm not making a global error, in all likelihood, although I am estimating the run environment to some extent. However, the error in that estimation may not be distributed evenly. It's something I may change in future years.
I think that predictions made by PECOTA, while informative, have a real problem. For some reason they just don't add up. You look above, and see that your prediction for the Rangers offense was off by over a hundred runs and the defense by the same amount in the same direction. This was not because of the run environment. Its because everybody the Rangers sent to the mound sucked last year. And most everyone they sent to the plate (after April) hit. Could anyone have predicted Milton Bradley to hit .345 with power and walks? for Hamilton to hit like an MVP most of the year? for Millwood AND Padilla to get hurt?