from Amazon
Tuesday, February 26, 2008
I duplicated Rally’s study on First Basemen and Saving Errors
From 00 to 07, I charted all successful throws from all infielders, including pitchers and catchers, to the first baseman, including the back end of DP’s. I did not include throws where the first baseman made an error or the runner was safe on a hit (late throw, but no error). I only recorded a throw as successful if it led to an out.
I also looked at ROE’s where the fielder made a throwing error. I used STATS data.
So I had lots of “pairs of players” where one player was the first baseman and the other was another infielder. For each pair, I had number of total throws and number of those total throws that were a throwing error.
I then used a similar “with and without you” method that Tango uses and Rally used. First of all, I only included those pairs where a fielder made at least 20 throws to any one first baseman.
To figure the “with and without you” for each first baseman, I went through every fielder who made at least 20 throws to him and subtracted the error rate (throwing errors divided by total throws) for that combo from the corresponding error rate for that fielder (at the same position) and every other first baseman. I weighted this number (the error rate difference) by the minimum of the two “number of throws.” I think Tango always weights by the actual number of throws made to that first baseman, and I am not sure what Rally did.
So then I got a “weighted rate difference” for each first baseman and simply multiplied that by the total number of throws to that first baseman by all of the fielders combined. That gives me a total number of errors above or below average for each first baseman. Each of those is worth around .75 runs I think (the run value of a throwing error plus the absolute value of an out).
Here are the best and worst since 00, in “errors saved per 1000 throws,” with at least 2000 throws to him. 1000 throws is about the equivalent of one full year. To get runs saved per year, just multiply by .75.
Best
Name No. throws Errors saved per 1000 throws (per season)
Mientky 4189 6.0
Sexson 6004 5.2
Berkman 2005 4.8
Conine 3100 4.8
Olerud 4481 4.2
T. Clark 2813 3.1
T. Lee 3996 3.1
Teixera 4335 3.0
Worst
Name No. throws Errors cost per 1000 throws (per season)
Karros 2986 -5.8
Jeff Bagwell 4931 -4.2
Julio Franco 2010 -4.1
Casey 6225 -3.3
Thome 4476 -3.2
Hatteberg 3967 -2.3
Keep in mind that these are sample numbers. If I had to guess a regression rate to estimate true talent, I would say 50% at 2000 (wild guess). Interestingly, if I sum up all the first basemen who had no more than 300 throws in those 8 years, which are mostly fill-ins, they are a combined -2.8 per 1000. So, saving bad throws is definitely a skill that comes with experience, although players with more than 300 but less than 1000 throws did just as well as, if not better than, more experienced players. That might mean that there are quite a few full-time first basemen who are bad at this skill, but are still first basemen because they hit well and cannot play anywhere else, if that makes any sense.
I did not adjust for the pool of “other first basemen” for each first baseman. I agree with Tango that it is likely around zero for most of the players, especially those with large samples of throws themselves. Of course, as with all of the “with and without you’s” you are technically comparing a player to the rest of the players and not to the league as a whole (including himself). Kind of like computing a park factor without including the “other park correction factor” which allows you to include a portion of the home park data in the pool of road park data (so that you can compare a park to ALL the parks in a league and not just the “other parks"). In this case, it shouldn’t really matter much. I’ll put the entire list on Google docs later today. Maybe someone can do an ICC (intra-class correlation) in order to figure the correct regression rate to go from these sample numbers to an estimate of true talent based on the number of throws, although I am not sure if you would regress based on the actual number of throws to each first baseman, the total of the “minimum of the two”, or the total of the harmonic means of all the pairs. Probably the last one.