Underperformance Metric: Who’s At Risk For Missing Expectations

A few weeks ago, I began the process of determining an underperformance metric. In the article, I laid out the groundwork determining the drop off in plate appearances (PA) and production (wOBA). With these thresholds, I created several metrics, each with its own advantages and disadvantages. I’m not setting the values into stone yet but I’m getting closer to a solution. I’ve found a few value I like better than others.

In the original article, I found fantasy owners considered a drop in 220 PA from 600 PA (37% drop) and of 0.035 wOBA from .350 wOBA (10%) to be the thresholds. I didn’t mess with these two values. Besides the pair, I wanted to know when both occurred. Additionally, from a discussion in the comments, I found when either PA or wOBA thresholds where met and when both dropped close to, but not over, the thresholds. This value (called Minor Drop) I found to provide the most overall value.

With the four outputs, I considered several methods to predict the underperformance chances and ended up using logistic regression. Logistic regression takes the historical inputs (only PA, wOBA, and age for now) and if the hitter underperformed. With these inputs, it outputs the player’s percentage chance of missing his projection. I ran the logistic regression for both the actual (e.g. 220 PA) and percentage drop (e.g. 37%).

I’ll just start with the final outputs and go over some of the observations on the different outputs.

Note: I used the final Steamer projections which takes into account that some players are hurt to start the season. The calculated under performance chances are likely a bit on the low side, especially with PA. The hitters who will get hurt in spring training aren’t known yet.

Actual drop

Plate appearances: The top values for this metric are crappy players with a ton of projected playing time. These players could lose their job from injury or suckitude. On the other end are talented players who will likely see a drop only if they are injured.

wOBA: This list is the exact opposite of previous one. The key here is that the higher a player’s wOBA, the more it can fall. Alcides Escobar is projected for a .272 wOBA. It’s tough for it to fall another 35 points.

Mike Trout (38%)and Bryce Harper (37%) lead this list with their underperformance rate near 37%. This seems high. I took the hitters projected for a .400 wOBA and 600 PA and then the actual results. If was 24 hitters in all and nine dropped for a percentage of 37.5%. 4

I didn’t like these results because most hitters were still productive. For this reason, I decided to run percentage drop.

Minor Drop: This value is just a bit lower than the previous two values. After finding the all the values, I found this one (or the Percentage Minor Drop) to be the one to concentrate on.

Major drop: The players with the highest rates are young, bad hitters with little-projected playing time.

Percentage drop

Plate appearances: With the percentage drop, hitters with few projected plate appearances are the most likely to drop. A hitter projected for 200 PA only needs to drop 70 PA to 130 to make the list. Also, among hitters with the same projected PA, those with projected lower wOBA can lose time as managers and owners would like to win and the crappy player is removed.

wOBA: While Trout and Harper are still high on the list, they aren’t at the top anymore. Instead, young bad players lead the list. These players don’t have a steady job and if they struggle early, teams will move on.

The hitters with the lowest underperformance rates are bad hitters with fulltime jobs. Their wOBA just can’t drop much further.

Now Rhys Hoskins leads this list. He shouldn’t. He’s only projected for 317 PA. If he’s healthy, it will be closer to 600. I’ve informed those in charge to raise the number.

Minor Drop: This list is a little weird in that the highest breakdown rate belongs to bad players with no fulltime job. Then it moves to talented players with jobs (both wOBA and PA can drop). It ends with bad players with fulltime jobs (wOBA can’t drop).

Major drop: These rates almost mirror the Minor Drop numbers so both don’t need to be utilized.

Conclusion

I’m generally happy with the results and it’s time to trim down the options. I’m not going to deal with eight different outputs for underperformance. It’s overkill. I’m going to cut it down to the Percentage Drop values for wOBA, PA, and Minor Drop. These values minimize some of the biases I noticed using the Actual Drop values.

I’m really tempted to just use the Minor Drop value for simplicity. For a few mock drafts, I will see how the three work and maybe then move down to one variable later. While I find the values helpful, I would love any suggestions. Were the initial underperformance values wrong? Are the Actual drop values a better option? Nothing is set in stone so it is a valuable time for suggestions. Thanks.



Print This Post

Jeff writes for RotoGraphs, The Hardball Times, Rotowire, Baseball America, and BaseballHQ. He has been nominated for two SABR Analytics Research Award for Contemporary Analysis and won it in 2013 in tandem with Bill Petti. He has won three FSWA Awards including on for his MASH series. In his first two seasons in Tout Wars, he's won the H2H league and mixed auction league. Follow him on Twitter @jeffwzimmerman.

newest oldest most voted
Bobby Ayala
Member

Can someone remind us how Steamer determines projected dropoffs in playing time? That’s a very important part of this metric, and would seem to be the hardest thing for a model to project. Thanks!

sabrtooth
Member
Member
sabrtooth

Seek and ye shall find!

http://steamerprojections.com/blog/projecting-playing-time/

Would not be surprised if this has changed a lot. One of the reasons that I like using the FG Depth Chart projections is that you get the cold, hard, rational rate data mixed with the subjective playing time judgements, which I think is a nice blend due to the kinds of information that might inform a playing time projection.