Spawning Estimated Strikeouts for DMB

Highlights--estimates include modifications by:
1) Activity of comparable players with recorded Ks from 1894 to 1919.
2) Player age
3) Park Factor


Seven of the Wiki Seasons would have missing batter strikeouts unless I had estimated them. These particular Deadball Wiki years begin with 1897 and end with 1907. 

Strikeouts for batters were not recorded from 1897-1909 in the National League, and from 1901-1912 in the American League.

Why estimate strikeouts? Without strikeouts entered in a batters event table, the DMB game engine will produce about half the league whiffs expected, as I recall. This is true even though strikeouts are known for pitchers, and therefore the league SO totals are known. Replaying a season with missing strikeouts is unacceptable to me, as it must have been for our DMB masters who estimated strikeouts for early teams, like the 1902 Pirates, the 1904 Giants, and the 1906 Cubs, featured on the All-time Greatest Teams. Also, K estimates were involved in the formation of the All-time Greatest Players disk. (The boys at Baseball Prospectus dont try to estimate missing stats.)

So how does one estimate strikeouts?

Tom Tippett wrote Estimating missing statistics for the AGP 2003 and 2006 disk in the Organizer note: Rating the Players:

In the early years of baseball history, certain statistics that we now take for granted today were not kept. As a result, we were forced to make estimates where the actual data was missing.

We hoped to develop good estimates of the missing stats by looking at other stats that have been recorded for all of baseball history. For example, perhaps we could estimate sacrifice flies by looking at things like how often a batter 
put the ball in play, homered, and grounded into double plays.

So we compiled a database consisting of thousands of modern player-seasons with a minimum 400 plate appearances or batters faced. Using scatter plots, correlation tests, and multiple regression, we looked for relationships between (a) the stats that are available for all seasons and (b) the stats that are now available but were missing for some of the early years.

When strong relationships were found, we used those relationships to estimate the missing stats. When we were unable to find any relationship at all, or where the relationship was very weak, we estimated the missing stats based on 
league averages.

Unfortunately, and to our surprise, we found only a few strong relationships despite all of our effort. As a result, many of the missing stats from the late 1800s and early 1900s are estimated based on league averages.

Thats what I had to do, estimate strikeouts based on averages. I dabbled with regression, and I made pretty scatter charts, and while there is a relationship between higher batting averages and lower strikeouts and another link exists between playing time and strikeouts, the correlation is too low to be meaningful.

Heres what I didnt do. I didnt simply deliver a fricassee of the league percentage of strikeouts/outs and multiply that by batter outs, though this stuff gets mixed in. While it does reproduce strikeouts well enough on the league level, it creates little variation from player to player. 

A league average is composed of the averages from many different players who exhibit a wide range of talent including their ability for whacking pitches into play instead of whiffing. I divided pitchers from non-pitchers. Why? Pitchers strikeout more. Most cant bat. Then, I separated regular players from the non-regulars. Regular players strikeout less. 

There are some exceptions. Three starting pitchers were dual players who also played other positions and hit reasonably well. One was Win Mercer who stuck his head in the oven and turned on the gas as his punishment for playing with whores and cards. These three qualified as regular players. A few others dual players count as non-regulars and not pitchers.

I used ideas from Michael Schells Baseball's All-Time Best Sluggers to determine who the regular players were for a season by applying the minimum number of NFPs per game from Table 2.2 on page 16. For example, reading from this chart, a NL player needs to average 2.2 NFPs per game for seasons between 1896 and 1938, while an AL player needs 2.4 NFPs for seasons between 1901 and 1907. 

NFP means net-facing pitchers. Its the total of a players AB, BB, and HBP experience, a skimmed version of plate appearances. NFPs work better than PAs when one is comparing players from early seasons because sacrifice hits, sacrifice flies, and catchers interference are mostly not part of their historical record.

I created a database to analyze the strikeout experience of 5594 players as available from 1894 to 1919. By 1920, the Lively Ball era begins and strikeout dynamics change, so I stopped there with my samples.

The estimated strikeouts reflect player performance from the tested seasons by NFP. If I can make sense of one of my endless spreadsheets, it appears that regulars have basic, estimated strikeout averages ranging from 10 to 13 percent. Non-regular strikeouts range from 12 to 30 percent, while pitchers have SO averages from 14 to 36 percent. To smooth the data, I used moving averages composed of 100 player values. 

As hitters with high batting averages tend to strikeout less, regular players who hit above one deviation, for this project thats about .317, were modified to reflect K averages of eight percent. For the test group, 180 regulars batted .317 or higher out of 5594 players.

But thats not all the modifications introduced to increase variability and perhaps give the numbers more pseudo-realism. Three other ideas from Schells book were examined and two were added as modifiers to the estimated strikeouts created from the NFP study.

The first one is adopted from the Age Profile for Strikeouts, page 54, of Baseballs All-Time Great Sluggers. As Schell says, Players improve their strikeout rates as they age. For the old-time players, K rates drop about 2.2% per year. An age factor was added to each player. A 35 year-old hitter strikes out 15% percent less often than does a 28 year-old player. Actuarially speaking, a 22 year-old player is likely to strikeout 14% more than someone six years older.

The second modification is a park factor for strikeouts taken from Appendix H, Regular Player Event Averages. These were adjusted for estimated Ks by a teams rate of home/away games. 

So you dont like this idea? Unlike DMB park factors which are rooted in seasonal player performance, Schells factors reflect physical changes in the park or rule changes that affected park play. These factors dont change so much from year to year. In the AL, Chicagos South Side Park had a 91 SO park factor for 1902-1909. Detroits Bennett Park had a 91 PF for 1901-1908. During the same period, the As Columbia Park had a whopping factor of 121tied with 1876 Louisville for the highest strikeout PF in Schells appendix. 

In the National League, Brooklyns Washington Park had a 106 strikeout PF for 1902-1912, while Cincinnatis Palace of the Fans was rated only an 88 for 1902-1911.

Like age factors, SO park factors were easy to match up with the 3360 Wiki player stats in my WikiStrikeouts database.

One experiment that didnt work was trying to apply a home run park factor taken from Schells Regular Player Event Averages. Along with that, I tested Schells rare event adjustment for homeruns. The idea of applying a HR park factor is that expected strikeouts rise with homers. In a twist, hitters who play in homerun-robbing parks would have been modified upwards while guys whacking lots of taters in band boxes would have been docked. 

Schell had a method for estimating strikeouts, too. Well, sort of. For the 1893-1919 era, Schells formula for regular players with 550 NFPs is 64 SO + 0.49 x adjusted homeruns. It works for him when relating the best sluggers across eras, but for evaluating seasonal strikeouts for all players, its a misapplication producing hot numbers. The same conclusion applies to a rare events adjustment.

The league strikeout averages were not applied until it was time to adjust the player averages. In some seasons, team total strikeouts are known, and these were applied to the adjustments. In the end, the strikeout estimates were rounded off and fine-tuned a K here and there so the final numbers matched the historical record. 

The 1902 database that was released last summer has strikeout estimates. These have been revamped for a second edition.

I compared my stuff with the estimated strikeouts from the Deadball teams of DMBs AGT-1. Those teams have good estimates. I like mine more since the Bofors values capture pitchers, limited-use players, and rarefied, Hall of Famers better. Even so, these are still man-made estimates and cant capture the extremes of everyday baseball.

One final note: For 1897, I looked at players from 1894 to 1896 to see if there were guys who consistently hit well and produced low strikeout averages every year. I figured I would find a dozen or more, and I would be crafty and modify the players strikeout estimate for 1897 to reflect his good hitting from preceeding years. Too bad, I didnt find a single instance where I could justify this.

Robert Bofors
Edited from the DMB Forum
Feb. 7, 2007

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
