The Kemp Speed Theory: Do bigger players slow down earlier?

Derek -

This is neat. But what happens if the player gets much taller while he’s playing?

More seriously, do the curves look the same if we look at, say, only those players who had at least 5% SB/SBO rate in their first year?

Jonathan,
Height can certainly change, but much less than weight, and very little over two years, I’d think.

As for adding a speed parameter, I remember trying this at one point but got a smaller sample size than I was comfortable with (after all, there aren’t that many 6’3 burners).  I don’t remember exactly what I was requiring, though, so I’ll take a look again and play with how lax I’m being with the requirement.

Will,
I am using most players available in a given year, but not all.  I am excluding players who play the final season of their career in Year 2 to eliminate some bias, although players whose first season is in Year 1 were included.

I’m not sure if I’m understanding you correctly, but wouldn’t only including players who play in the majors from age 24 to 36 be *creating* bias?  Players who enter the league at 24 are probably a little different than players who enter at 27 or 28, and a player will necessarily be good (at least to some degree) to have a 12 year long career.  Not to mention that this would probably lead to some small samples.

Nice analysis, as usual!

One would think that most teams have kept some sort of records of players weights over the years.  Now that Bill James is an insider (and since THT has some connection with him), maybe he can use that access to get all the teams to donate their players’ weights to Retrosheet (or SABR, whatever), as far back as they got them. 

Obviously, some teams would be anal and have them back to Aught-nine of the 20th Century and others will only have them for the past decade, but at least that could be made available, though perhaps HIPAA might interfere with that today, don’t know if weight is considered a personal bit of health data.

Lastly, I was wondering how hard it would be to split your data into eras and compare.  Clearly, the fitness of players is dramatically different from today to 100 years ago. 

Maybe you can compare pre and post WWII:  I know that vets brought back ampethamine from the war, so perhaps some of them brought back fitness too (I’m thinking Ted Williams here as the prime example).  Or even split post-WW II into two or more eras, choose your splits.  I would also wonder how things have changed since the game changed in the 1993 timeframe, when, as Eric Walker calls it, the Silly-ball era started.

Derek:

Say you have 2 guys. One has 20 steals each year between ages 20 and 30 and 10 steals each year between ages 30 and 40. The other guy has 20 steals each year between ages 20 and 30, but was forced out of the league because his speed declined dramatically and he would have had 0 steals between 30 and 40.

The actual age curve would should be 20 from ages 20 to 30 and 5 from ages 30 to 40, but if the curve is only calculated with players in the league, the curve is biased upwards in the 30-40 sample to 10 instead of where it should be at 5.

Put more simply, by including players with short careers, you are probably flattening the age curve. Your age curve doesn’t represent the decline in speed as a player ages, you’re showing the decline in speed CONDITIONAL on still playing. Those are two very different things.

Matt Kemp may have a very long career, and the age curve one year ahead probably has almost no bias given Kemp’s current age. This means that your analysis on him is probably right on. However, you can’t use your methods to project 10 years out because you’re not doing it quite right.

That being said, I like where you’re going with the analysis

Derek:

Yeah, I guess the age curve you need has everything to do with how you’re going to apply it. If the question is how do people age, then you should only use people who are in the league for every year of the wage curve. If we ask the same question, but feel certain that the guy we care about will be in baseball 10 years from now, then your implicit conditioning assumption is satisfied and it’s fine to use the conditional age curves you’ve created here.

Again, nice work!