Last spring, I developed a new offensive metric I called PrOPS, which is short for predicted OPS. Obviously, it’s a method for predicting OPS, which is a good shortcut method for measuring a hitter’s productivity. But I didn’t develop the statistic to be just another option to choose from when judging a player’s offensive value. After all, I believe Bill James was largely correct in saying that “the world needs another offensive rating system like Custer needs more Indians (or, for that matter, like the Indians need another Custer).”
Instead, I intended to find some information that is lost in the noise of common baseball statistics. My goal was to use some of the new batted-ball data published by The Hardball Times to find information that outcome-based metrics (i.e., batting average, on-base percentage and slugging percentage) miss. In particular, I was curious about the problem of luck. When a cylindrical bat hits a round ball lots of funny things can happen to influence the outcome.
Just because a hitter does some things that normally lead to good/bad outcomes doesn’t mean those outcomes will always happen. Good process can result in bad outcomes, and bad process can result in good outcomes. If these outcomes don’t cancel out, but pile up in one direction, outcome-based metrics can distort a player’s true performance. PrOPS is a tool for extracting the noise from the statistics that we normally find to be quite reliable predictors of run production.
The method for generating PrOPS involves using batted-ball data, along with a few other important factors, to predict the offensive output of players using linear regression. The projections are thus based on the way players hit the ball and not actual outcomes. PrOPS credits players who hit the ball similarly with outcomes typical of all players of the comparable hitting profiles. This method strips out some of the luck that pollutes outcome-based hitting statistics, because no player receives direct credit for the actual outcomes at the plate.
One of the uses of PrOPS is that it can spot fluke seasons. For instance, it’s often unclear if players with abnormally good and bad years are experiencing improvement, decline or just runs in random bounces. When compared to outcome metrics, deviations from predictions indicate that a player has been lucky or unlucky. In The Hardball Times Baseball Annual 2006 I test PrOPS’s usefulness in identifying fluke performances using four seasons of data (2002-2005).
It turns out that PrOPS does a decent job of spotting lucky and unlucky performances. One advantage of working with the data from many seasons—the first version was based on 2004 data only—was that I was able to refine my approach to generate more accurate estimates of the impact of batted-ball types on hit outcomes, which greatly improved the PrOPS metrics. It turns out that some of the corrections that I needed to make in the previous version—speed, for example—were no longer necessary.
Over the summer, I posted weekly updates of PrOPS until I just got out of the habit of doing it. But when the free agent and trade markets began to heat up, I found it useful to look at the newer 2005 PrOPS numbers for guidance. While hanging out over on Mac Thomason’s Braves Journal—the unofficial online home for Braves fans—I kept posting PrOPS lines of guys in whom the Braves might have some interest, because many of these players were coming off odd years.
For example, PrOPS said Julio Lugo didn’t play as well as his 2005 numbers, and Edgar Renteria’s down year in Boston was right on target with his 2004 in St. Louis. (Not that it makes losing Andy Marte any easier to take.) A few of my online friends found the numbers to be helpful as well, so I decided it was time to post the final 2005 PrOPS numbers for all players on THT for others to see. The statistics are listed separately for the AL and NL, and I have sorted them by team. If you want to see my analysis of these numbers, pick up a copy of the Annual.
But I didn’t stop at just updating the old numbers. In addition, I decided to take the metric a little further. After all, if PrOPS has the power to distinguish between fluke and true performances, then I should be able to use it to predict future performances. Thus, I developed a PrOPS-based projection system. It’s not fancy, and it doesn’t attempt to predict anything more than the OPS for a player in the following season. I used two years of data to project OPS for all players with a minimum of 130 plate appearances in the previous season based on PrOPS, player ages and the home parks’ effects on hitters.
I also calculated 95% confidence intervals for the projections. For rookies or other players with fewer than 130 plate appearances in 2004, I generated projections using a single year of data. I have less confidence in these numbers, but they are better than no numbers at all. I did not generate projections for players with fewer than 130 plate appearances in 2005. (Sorry, no Bonds projection.) The projections assume the player plays for the same team as he did in 2005. The 2006 PrOPS Projections are listed alphabetically.
How accurate were the projections for past seasons? Using a method of comparison reported in the 2004 Baseball Prospectus for comparing OPS projections of competing projection systems, I tested the PrOPS projections on a similar sample of players—I could not replicate it exactly. The PrOPS projections performed about as well as the top systems. I wouldn’t put much stock in this test in terms of stating any one system is superior to any other; however, I think it shows that the PrOPS projections are worth using along with other systems.
Please enjoy the 2005 PrOPS and 2006 PrOPS projections. If you are interested in further discussion of PrOPS, check out my article “Giving Players Their PrOPS: A Platonic Measure of Hitting” in The Hardball Times Baseball Annual 2006. I welcome comments and suggestions.