DMB Forum
Feb. 12, 2007

Modern stolen base records began being kept in 1898. For the Wiki Seasons of 1896 and 1897 [and earlier seasons like the second editions of 1894 and 1895], I'm adjusting or prorating the stolen bases downward by comparing stats from the next three seasons, before the foul ball rule changed in the National League for 1901. Is this blasphemy? 

RobertE13, DMB Forum, Post 153, (2/3/07 2:03 am): Notice the dropoff in SB success estimates from 1897-1898. That's because prior to 1898 players were credited with a SB when they went from first to third on a single or otherwise took an extra base. So it's likely that success rates really weren't that different in the 1890s. All those extra steals really gum up trying to figure these things out.

RobertE13, DMB Forum, Post 155, (2/3/07 10:59 pm): Prorating is probably best. It won't be perfect, because there are some different skills, and taking the extra base is necessarily dependent on the hitting farther down in the batting order. But it's probably better than giving Hugh Nicol 138 steals.

Prorating 1896 reduces stolen bases to 69% of the total listed in Major Leagues records. For 1897, prorating adjusts the SB numbers to 80%. In both cases, the adjusted rate is 1.33 steals per game. 

Houston, we have a problem...

Looking at the records from 1894 to 1900, I suspect prorating circumcises too much SB tissue. Baseball appears to be evolving quickly during the 1890s so prorating may not be the best answer.

Next, I was more enamored by trying to find a solution by comparing runs scored as a percentage of times on base: (((R - HR)/(H + BB - HR)). I liked exploring this method because runs, hits, and walks were way up for 1896 and 1897, giving base stealers more opportunities. However, the rates produced may be too high if all Im doing is comparing their ratios and stopping. For 1896, this method reduces stolen bases to 93%, while for 1897, steals drop to 94%. The results for 1897 dont look bad to me, steals per game change from 1.67 per game to 1.56. Comparing that rate to the figures from 1899 and 1900, it seems acceptable considering the game was changing faster than computer technology changes today. If I had to choose between these two methods, Id pick the second.

But, I dont know, I'm the not math man here--just show me the answer and Ill stick the adjusted steals into DMB. 

For my third conjurement, I prorated the stolen bases by times on base, it that's the description for it, which gave me numbers up the middle: Combining the game-adjusted sums of 1898-1900 stolen bases divided by times on base (H+BB-HR) resulted in a .1126 SB rate. Then, I applied that rate to the 1896 and 1897 time on base totals. This SB/(H+HB-HR prorating reduced 1896 stolen bases to .759 of the current record. For 1897, the rate was .871. Now the SB/game rate is 1.46 and 1.45. Speciously appealing, maybe thats it, but Im dubious.


DMB Forum
April 21, 2007

This part of the Wiki project ran into early March. The story tells you something about why the Wikis drag on.

Play-by-play data from 1911 was too distant in time to be helpful in the long run. Working next on a linear weights angle impressed me that a play-by-play parsing would have been of limited value.

Earlier, I checked in with stat guru Bradon Heipp about creating an Estimated Runs Produced skeleton for determining the values of stolen bases by linear weights.

danagonistes.blogspot.com/2004/10/brief-history-of-run-estimation_27.html

gosu02.tripod.com/id16.html

About his ERP skeleton, Brandon said: 

The ERP formula does not work nearly as well in the 19th century as it does today. One thing that should be accounted for definitely in olden times is reaching base on errors, which in theory should be a little more valuable then a single. However, I did not adapt ERP for that.

Brandon suggested I prorate the stolen bases steals per times on base.

I couldn't resist exploring the matter further by mixing in errors. So I took a crash course in Linear Weights 101. I read more Web information and picked up Pete Palmer's and Jim Thorn's book, The Hidden Game of Baseball. And I found that linear weights is fairly easy stuff.

Unfortunately, fiddling with errors as part of a skeleton didn't work, and I don't think it can work without blowing up linear weights with the resulting explosion spawning a new life form.

Essentially, linear weights won't work for distilling missing values like Deadballs caught-stealing values.

Shane Holmes engaged Dave Studeman at hardballtimes.com about the matter. Shane said, Studeman advised you to prorate based on time on base. Surprise, surprise, right?

Afterwards, Shane wrote Tom Tango and received a thoughtful reply about treating a SB opportunity as 80% of a single and 60% of a walk or hit batter.

So with the stat gurus going three for three, advising a prorating path, thats what I did with the provision of selecting Tangos SB/Opp delineation for its mystical cache. For prorating these seasons, I compared them with 1898, 1899, and 1900.

Tangos method produced estimates similar to my third variation which are not as severe as a straight-up SB prorating.

Furthermore, I adjusted the rates to reflect player groups as defined by Michael Schell in Baseballs All-Time Best Sluggers. Regulars for Schell are position players with a minimum of 2.6 NFPs (net-facing pitcher, similar to plate appearances) per game for 1894 and 1895 and 2.2 NFPs for players from the 1896 and 1897 season. I adjusted NFPs per team, not merely by league. As you see, I plan to updgrade 1894 and 1895 fairly soon.

Here are the outcomes:

 
Year       Group          Prorate
1894       regular        0.76
1894       bench          0.86
1894       pitcher        1.15

1895       regular        0.79
1895       bench          0.84
1895       pitcher        1.20

1896       regular        0.72
1896       bench          0.85
1896       pitcher        0.99

1897       regular        0.85
1897       bench          0.94
1897       pitcher        0.98


With this method, it appears that the regular players were the ones who had the wheels and experience to grab most of the extra bases that still count as SBs in todays records. Pitchers showed little change. (For 1894 and 1895, I didnt need to adjust pitcher stolen bases.)

The result of tuning prorated stolen bases by subgroup is not quite as simplistic as displayed here, but it clearly provides a better estimate of what happened in those late 19th-century seasons.


Robert Bofors
Edited June 23, 2007
