The Kemp Speed Theory: Do bigger players slow down earlier? by Derek Carty October 21, 2009 Will Matt Kemp’s size cause him to lose his effectiveness on the basepaths in 2010? (Icon/SMI) A few weeks ago, Eriq Gardner noted how many fantasy analysts are already penciling Matt Kemp in as a top-five pick for 2010. Eriq played devil’s advocate a bit and discussed some of the reasons why Kemp might not be such a slam dunk. One such reason was his speed. To quote: Speed: As mentioned above, Kemp is on a path toward surpassing 35 SB this season, an extraordinary achievement for a player who is 6-foot-3 and approximately 225 pounds. Players measuring those dimensions aren’t typically speed demons and when they do surpass 30 SB, as Alex Rodriguez did in 1998, it tends to be followed by a few years of more moderate steals production. In 2006, Baseball Prospectus writer Kevin Goldstein wrote this about the then-prospect outfielder: “At 230 pounds, Kemp’s plus speed could dissipate quickly.” Reportedly, Kemp showed up to spring training this year in excellent condition, and his success rate on the base-paths this year (81%) shows no cause for concern, yet we’ve likely seen the best from Kemp in the steals department. This theory intrigued me, and I wanted to take a deeper look into it. Is this actually the case? And if it is, what’s the extent of it? Age curves To start, let’s look at an age curve for three groups of players: league average (all players), players 6-3 or taller, and players 5-10 or shorter. These groups will be known as “average,” “tall,” and “short,” respectively, from this point forward. The stat we’ll examine will be SB/SBO (steals divided by opportunities to steal), or the rate at which a player both attempts a steal and succeeds given that he reaches first base. We’ll use data from 1919 to 2008. To form the graph, we’ll look at year-to-year changes and display them as a percentage of Year 1 so that all three groups of players will start at the same place and will be easier to compare. The main takeaway here is that “tall” and “average” players maintain the speed they had at age 21 longer than “short” players, who start trending downward at age 23. Tall players start that downward trend at age 24, but it’s much less pronounced as they’re able to keep at least 93 percent of their Year 1 speed all the way until age 28. Once those tall players start their decline, however, they face a steeper drop than the short players. To illustrate this a bit better, here’s a chart showing raw year-to-year changes as opposed to the gradual aging approach we just took. We’ll also condense our age range to 24-37 to use ages with a little bit larger sample and to hone in a little bit more on what we’re looking at. In this light, we see that short and average players behave very similarly. The short players show some wider swings, but that’s simply a sample size issue. The pattern is essentially the same. Tall players, however, follow a much different pattern, as we started to see in the initial age curve. Hopefully this graph makes it a little clearer. Each year from age 27 through 32, tall players unfailingly see a drop in their speed. Then there’s a bit of a resurgence at age 34 (almost certainly a sample-size issue—in all likelihood, there is probably a plateau for ages 33 to 35) and then some more decline. To circle back on the short players for a moment, there is one noticeable difference between them and average players. At age 33, notice that their line begins to slope upward. This doesn’t mean that they gain speed, but rather they lose it at an increasingly smaller rate. In fact, from age 33 to 37, short players lose a total of just 6 percent of their speed. After that, of course, they decline. Summing it all up Essentially, short and average players see their skills decline at a pretty steady rate, short players easing up a bit from 33 to 37. They seem to lose roughly 5 percent of their speed per year until they reach 33. Tall players behave differently, seeing little overall change from 21 to 25, dropping a bit and leveling off until 27, then taking a nosedive until 33. They level off again from 33 to 35, then plummet until the end of their careers. Application to Matt Kemp It looks like Kemp’s big body won’t hinder his speed much, at least in 2010. (Icon/SMI) Now let’s turn our attention back to Kemp. The poster boy for Eriq’s theory has fit our age curve pretty well up to this point in his career. He’s posted SB/SBO rates (including MLEs) of 18%, 14%, 22% and 20% at ages 21, 22, 23 and 24. Aside from that outlier at age 22 (which was accumulated during a somewhat small sample of 455 at-bats), he’s been pretty consistent, just as the age curve tells us (especially if we were to regress each of his rates to the mean). So what can we deduce about Kemp (who turned 25 at the end of last month) going forward? Well, I think it’s relatively safe to say that his speed will stay in tact, for the most part, next year. Unless he puts on some weight, he should remain in that “initial plateau” area for tall players (lasting from age 21 through 25). After 2010, these age curves tell us to expect a small dip until age 27, then a precipitous fall off. Overall, the Kemp Speed Theory seems to hold some real credence, it’s just that Kemp himself hasn’t reached the point where he’s likely to be affected. Side-note on caveats and bias You probably noticed that I didn’t use weight as a parameter, as Eriq’s theory suggested. While I think this would be an important variable, unfortunately the data we have available to us doesn’t allow it. You see, a database doesn’t seem to exist (at least publicly) that assigns a weight to a player for each individual season. Instead, we only get something like career-to-date or end-of-career weight data. This will create problems if we try to use it for age curves. For example, when Barry Bonds was 25 years old and stealing 40 or 50 bases per year, he probably weighed around 150 pounds. At the end of his career, he weighed around 240 pounds. If we were to create a weight parameter in our age curve, Bonds would not be lumped in with the 6-2, 150-pound guys at the age when he actually was 6-2, 150 pounds. Instead, he would fall into the 6-2, 240-pound bucket at every age—even though that’s not who he was at age 24. This creates lots of problems and bias. Using only height does introduce some problems, but not nearly as many, and it’s mostly just an offshoot of not having weight. For example, we have no idea which players are gaining weight and slowing as a result. If we’re predicting the future for a modern-day player, we’ll know that he’s maintained his weight, so ideally we’d want to eliminate guys who added weight from our study, but we simply aren’t able to do that. Instead, we’ll have “tall players who gain weight” and “tall players who maintain weight” all lumped together, despite the fact that “tall players who gain weight” will likely be skewing our results a bit. Overall, though, using just height is much sounder than including weight. At some point I may run these age curves again, including a weight parameter, using data from just the past four years or so to eliminate some of the issues with weight, although that might just lead to a small-sample-size issue. There’s also some selection bias inherent with age curves in general, and I’ve taken some precautions to avoid them, but some just can’t be completely eliminated, so I wanted to make note of it. Finally, because we’re using stolen base opportunities as our denominator, our sample is much smaller than if we were using something like at-bats or plate appearances. I included 90 years worth of data to compensate, but the sample sizes are still less than ideal, especially for ages on the extremes. The general points should probably hold, though. Concluding thoughts I’m not yet ready to say that I’m drafting Kemp in the top five, but I’m not nearly as worried about his speed as I might have been a few weeks ago. If you guys have any questions, feel free to ask away.