

Times caught stealing (CS) were not been recorded for the National League from 1876-1914, or from 1916-1919, and even more recently from 1926-1950.  The American League followed this course from 1901-1913, and from 1916-1919. 

DMB simulates stolen bases from the run and jump ratings.  The game does not key off the actual SB and CS stats; the event tables dont include these as factors.  Luke Kraemer, DMB tech support engineer said this about the matter on the DMB Forum, June 19, 2006:  SB and CS are mostly driven by the Jump and Steal ratings but the CM is aware of base stealers with extreme performances like Henderson, Coleman, Wills, etc. and makes some adjustments for these really aggressive runners.

Often, third-party pre-1927 database developers have determined their run and jump ratings from stolen base opportunities.

So, its not necessary to have actual CS stats for DMB to replay a season; however, CS stats will tune-up the stolen base game more realistically than rating players by stolen base opportunities, so I created estimates for the Wiki Seasons and for second editions of 1894 and 1895.  While rating players by stolen base opportunities works OK, my observations show it sets the steal ratings badly sometimes.  Besides, I wanted a better idea of how the players performed and a good estimate of times caught stealing provides this, while a 0 in the CS column would reflects nothing.

I didnt like the CS estimates from the June 26, 2006 edition of the 1902 Wiki Season, so I modified it greatly.  I worked on estimated times caught stealing from late December until late May, 2007.  In the end, I created estimated CS stats for 11,410 player profiles for seasons ranging from 1894 to 1919, not counting composites displayed in the player register.

I tried different methods and tested endless variations.  In the long run, I applied a modification of Bill Jamess speed score, SpS4, runs scored as a percentage of times on base: (R - HR)/(H + BB - HR).  The usual procedure is to apply this to both league totals and individual players.  I applied a different basis for the league total instead of simply applying the numbers from the singular season in question.  To explain why, I need to touch on another modifier:  the expected SB rate.

A typical CS estimate formula looks like this: 

[[1/(( player runs per times on base / league runs per times on base) X expected SB rate)] - 1] X SB = CS

Normally, the expected stolen base success rate applied is .55.  This is averaged from known CS stats from neighboring seasons:  1914-1915 and 1920-1925. 
David Laurila of Baseball Prospectus asked baseball historian Dan Levitt, the co-author (with Mark Armour) of Paths to Glory, and a contributing author to SABRS Deadball Stars of the American League and Deadball Stars of the National League the following question on April 29:
What is known about the success rate of steal attempts in the [Deadball] era?
Levitt:  At a time when fewer home runs made scoring from first less likely, taking a greater risk to get into scoring position was acceptable. Although I can't offer a definitive answer on how much less the break-even percentage might have been during the deadball era, the actual success rate was roughly 55 percent. Managers were much more willing to sacrifice an out to get the runner to second.
In creating the Wiki Season CS estimates for seasons spanning 1896 to 1916, I wasnt satisfied by assuming the expected SB rate as being .55.  Its a fast way to an estimate, but it was part of the problem in creating inferior CS estimates for last years 1902 Wiki Season edition.
As it turned out, I created not one but 23 different SB rates depending on player type and a players stolen base totals.  Cumulatively, my CS estimates would reflect a league SB rate ultimately, but a league rate would not be part of the formula.  (For players who stole no bases, but may have been caught stealing or being thrown out on a busted hit-and-run play, random numbers were generated for 4525 players.  These had been sorted into 17 sets determined by NFP.)
As I did with my new estimated strikeout figures for the Wiki Seasons, I sorted players by usage:  regular, bench, and pitchers, an idea from Michael Schells Baseball's All-Time Best Sluggers.  The Organizer note, Spawning Estimated Strikeouts for DMB, explains this idea and what NFPs are all about.
I did apply SB rates from 1914-1915 and 1920-1925, but I used them differently.  I found that the players from these eight years could be sub-grouped by stolen base totals.  For example, regular players with more stolen bases tend to have more success than players with less.  Its not a stretch to imagine that a hundred players each with sixty-plus stolen bases will tend to have much higher SB rates than a hundred players with only six each.  So I began grouping players into subsets of 30 or more samples, graduated generally by diminishing SB rates as the table shows:

Regulars:

SB 		RATE
40+		.716
31-40 		.651
24-30 		.615
20-23 		.574
The four categories above include 1916 stats

17-19 		.575
14-16 		.542
11-13 		.539
9-10 		.521
8 		.509
7 		.489
5-6 		.458
4 		.409
3 		.375
2 		.326
1 		.193


Bench:

SB 		RATE
11-20 		.647
6-10 		.584
4-5 		.563
3 		.526
2 		.495
1 		.476


Pitchers:

SB 		RATE
2-7 		.707
1 		.804


Since I averaged SB rates this way from eight seasons between 1914 and 1925, the caught-stealing formulas league runs per times on base also reflects an average from all those years, and a not single season.

In February, while I was prorating stolen bases for 1894 to 1897--modern stolen base records began being kept in 1898, I realized having 1914, 1915, and 1920 to 1925 as the basis for league runs per times on base would not work as well as using the actual seasons being evaluated.  In the end, I rebuilt my estimated CS database with the league runs per times on base split into five player groups:

1894-1895 NL
1896-1897 NL
1898-1900 NL
1901-1919 NL
1901-1919 AL

The nineteenth-century seasons had to be sorted into three smaller sets instead of one 1890s group because of the volatility of this period; however, delineating the 1901-1919 Deadball era into two super league sets created a firmer foundation with more samples to average than my original eight-season set ranging from 1914 to 1925.  The resulting estimates produced less outliers and more consistent values.  Since I had decided to make all the Deadball years, plus seven 1890 seasons, as the basis for league runs, it was easy to consider creating CS for all 11,410 players from these many years with missing CS numbers. 

I finished the estimates, and I compiled the league totals for the first time.  I was surprised the SB rates were as high as they were considering the high CS totals I had seen for quite a few players during the many weeks I had doodled on this project.  Next, I evaluated the estimates by sorting the records by player careers.  Many results had been rounded down so that the estimate groups would more closely match the SB rate of the control groups. I had made an assumption about keeping the rates similar between the two groups, and now I was realized this was an unnecessary condition.  Around this time, I discovered a few hot-shot base stealers like Collins, Cobb, and Carey had known CS stats for 1912 and 191; however, their success rates werent going to rewrite the record books.  

Heres how the estimates looked after round one in mid May:


Year	Lg	SB	CS	rate
1894	NL	2489	1601	0.609
1895	NL	2359	1579	0.599
1896	NL	2334	1463	0.615
1897	NL	2331	1527	0.604
1898	NL	2069	1479	0.583
1899	NL	2677	1634	0.621
1900	NL	1686	1103	0.605
1901	NL	1402	760	0.648
1902	NL	1362	905	0.601
1903	NL	1561	803	0.660
1904	NL	1573	963	0.620
1905	NL	1598	1024	0.609
1906	NL	1463	1028	0.587
1907	NL	1322	1012	0.566
1908	NL	1385	1100	0.557
1909	NL	1506	1136	0.570
1910	NL	1594	1114	0.589
1911	NL	1692	1040	0.619
1912	NL	1572	974	0.617
1913	NL	1578	1071	0.596
1914	NL	1440	1065	0.575
1915	NL	1191	995	0.545
1916	NL	1328	1094	0.548
1917	NL	1145	985	0.538
1918	NL	1029	788	0.566
1919	NL	1167	955	0.550
1901	AL	1451	603	0.706
1902	AL	1315	695	0.654
1903	AL	1154	707	0.620
1904	AL	1205	887	0.576
1905	AL	1328	908	0.594
1906	AL	1533	1088	0.585
1907	AL	1455	1091	0.571
1908	AL	1350	998	0.575
1909	AL	1545	1100	0.584
1910	AL	1670	1118	0.599
1911	AL	1722	997	0.633
1912	AL	1807	1084	0.625
1913	AL	1672	1126	0.598
1914	AL	1666	1375	0.548
1915	AL	1444	1040	0.581
1916	AL	1425	1172	0.549
1917	AL	1268	1048	0.547
1918	AL	960	843	0.532
1919	AL	912	771	0.542

The overall SB rate for all seasons was .593.  For 1901 to 1919, the percentage was .589.


So after all this, I decided to rework the estimates by rounding most of them straight up.  Then, the career numbers were creamier and more consistent.  If the CS estimates were to be any lower, I would have had to fudge the numbers to match public perception, but I was happy with the figures.

Here are the league SB rates from the final round:


Year	Lg	SB	CS	rate
1894	NL	2489	1643	0.602
1895	NL	2359	1644	0.589
1896	NL	2334	1516	0.606
1897	NL	2331	1609	0.592
1898	NL	2069	1522	0.576
1899	NL	2677	1690	0.613
1900	NL	1686	1139	0.597
1901	NL	1402	809	0.634
1902	NL	1362	975	0.583
1903	NL	1561	817	0.656
1904	NL	1573	1011	0.609
1905	NL	1598	1078	0.597
1906	NL	1463	1099	0.571
1907	NL	1322	1072	0.552
1908	NL	1385	1160	0.544
1909	NL	1506	1199	0.557
1910	NL	1594	1155	0.580
1911	NL	1692	1064	0.614
1912	NL	1572	987	0.614
1913	NL	1578	1109	0.587
1914	NL	1440	1124	0.562
1915	NL	1191	995	0.545
1916	NL	1328	1163	0.533
1917	NL	1145	1002	0.533
1918	NL	1029	837	0.551
1919	NL	1167	986	0.542
1901	AL	1451	621	0.700
1902	AL	1315	697	0.654
1903	AL	1154	734	0.611
1904	AL	1205	936	0.563
1905	AL	1328	948	0.583
1906	AL	1533	1131	0.575
1907	AL	1455	1136	0.562
1908	AL	1350	1053	0.562
1909	AL	1545	1174	0.568
1910	AL	1670	1178	0.586
1911	AL	1722	1019	0.628
1912	AL	1807	1114	0.619
1913	AL	1672	1201	0.582
1914	AL	1666	1375	0.548
1915	AL	1444	1040	0.581
1916	AL	1425	1153	0.553
1917	AL	1268	1097	0.536
1918	AL	960	911	0.513
1919	AL	912	796	0.534
1920	AL	752	707	0.515
1920	NL	969	862	0.529
1921	AL	685	545	0.557
1921	NL	803	771	0.510
1922	AL	681	515	0.569
1922	NL	755	634	0.544
1923	AL	741	606	0.550
1923	NL	824	657	0.556
1924	AL	749	581	0.563
1924	NL	754	659	0.534
1925	AL	711	582	0.550
1925	NL	672	517	0.565
1926	AL	664	509	0.566
TOT		69735	49719	0.584

The total above relates only to seasons from 1894 to 1919.  The Deadball era estimated CS rate was .580.  The NL was .578, and the AL rate was .582.


Here are a few observations about the estimated league rates in light of those early baseball years.  Improvement in catcher protection and defense dont appear to be significantly improving after 1894.  I expected the nineteenth-century SB rates to be a little higher.  Perhaps with fewer samples for compiling league runs per times on base for 19th-century seasons, the estimates arent as strong.  The estimates do indicate that the late 1890s was not an era for hog-wild base stealing.

Two items stand out about base-stealing rates in the early 1900s:  One, the foul strike rule gave players less opportunities to steal.  Two, the disruption to National League rosters by the nascent American League made both circuits uneven in talent, and good base stealers took advantage of opposing weaker catchers. 

The new cork-center ball made batting averages jump in 1911 and 1912.  Runs scored increased.  My SB rates jump for both leagues and for both years, too.  Whether the improvement in base stealing is an aberration of a formula that applies runs scored to CS estimates, or it is a natural consequence of higher on-base averages, has not been studied.

One thing for sure, by 1913, pitchers had the edge again with the introduction of the emery ball.  For the rest of the decade, Deadball resumed and stolen base rates fell.

The late Deadball seasons with estimated CS values appear right in line with the known factors from 1914, 1915, and 1920-1925.

See the organizer note, About the Wiki Seasons Project, to learn more about the wiki concept for DMB database development.

Robert Bofors
July 1, 2007


