About two weeks I posted a study of the degree to which playing mostly for one team gives a player an edge in Hall of Fame selection, as opposed to a player who moves around from team to team, from city to city, from uniform to uniform, from franchise to franchise, from dugout to dugout, from apartment to apartment, from woman to.  . . where was I? 

            Oh yes; the effects of rapid migration on a player’s Hall of Fame chances.  The study concluded that players who spend most of their career with one team have a very large advantage in Hall of Fame selections.   In response to this, some readers have suggested that this effect might be a side-effect of time-line discrepancies.  In other words.   .oh hell; I’ll find a source and quote it:

Guy123
Following up, it does seem likely that some of what Bill's study found is an era effect rather than a one-team effect. His appendix lists non-HOF players in the study but not the HOFers, so I can't precisely replicate his study. But using WAR, players with 55-69 career WAR have about a 50% chance of making the HOF (higher for those closer to 69, lower for those close to 55). Of the 65 eligible players in that range, 35 are in the HOF and 30 are not (54% in).

The two pools are similar in average career WAR (HOF 63, non 60), plate appearances (9248, 9021), and OPS+ (126, 125). But there is one enormous difference between them: the years in which they played. The average HOF debuted in 1956 and played his last season in 1974; the non-HOF players, on average, debuted in 1975 and finished in 1992. That's an astonishingly large difference: the typical non-HOF player began his career a year after the average HOFer retired!

            So I decided to study that; that is, I decided to study whether the effects earlier documented could, in fact, be caused by the fact that. . .well, the fact that more players are elected to the Hall of Fame over time.   In studying this issue, I came to a much better understanding of what is happening in the data, and that is the goal here, so I will thank Mr. 123 for the suggestion.  Other than that, I am not responding to him or to his research; rather, I am addressing the issue itself. 

            The timeline issue and the one-team issue ARE, in fact, related, but the relationship is not what has been suggested.  It is rather more the opposite.  It is more probably true that timeline bias caused the effect to be UNDERSTATED in the other study, rather than that it caused it to be overstated.  Which is what would normally happen.  If you have two biases operating within a field of data, what would normally happen is that the two would conflict with one another, causing both to measure as lower than they actually are unless they are specifically disentangled from one another.  Only if the two are aligned along multiple axis would the one bias increase the measurement of the other. 

            This is what really happens in the data, in three bullet points:

1)     Among the earliest groups of players—the players before Babe Ruth, essentially—there is little or no Hall of Fame selection advantage for the players who are more easily associated with one team. The advantage is pretty close to nothing. 

2)     The selection bias in favor of players associated with one team or with little career movement appears about 1920, and grows wider over time, until it becomes a truly enormous sorting criteria—much larger than measured before-- in the selections of the most recent candidates. 

3)     What almost certainly happens over time is that the Hall of Fame selections "catch up" to the players who were overlooked by Hall of Fame electors in the short run.  The better-qualified players who are not selected for the Hall of Fame in their first 30 years post-retirement tend to get picked up later, which tends to reduce the one-team advantage.   Over time, players like Bobby Abreu, Kenny Lofton, Fred McGriff, Scott Rolen, Dale Murphy and John Olerud are likely to be added to the Hall of Fame, and this will reduce the frankly enormous advantage that one-team or near- one-team players have.   Reduce, but never eliminate, or not for more than 100 years. 

Among the earliest groups of players—the players before Babe Ruth, essentially—there is little or no Hall of Fame selection advantage for the players who are more easily associated with one team. The advantage is pretty close to nothing. 

            Which makes sense when you think about it.  The Hall of Fame opened in the late 1930s, with regular selections beginning in the 1940s.  Let’s talk about the players of the 1880s.  You have to remember:  almost nobody voting for Hall of Fame selections in the 1940s actually saw anybody from the 1880s play.  There’s a 60-year gap, to begin with, but also:         

o   There was no television, either in the 1880s or the early 1940s, so no sportswriter ever saw one of those players play on television.

o   Attendance at games in the 1880s was miniscule, compared to modern numbers.  Games were often attended by dozens or hundreds of fans, very few games by 10,000 or more.

o   Most people in the US in 1885 lived on farms or in small towns.  Relatively few lived in cities, so relatively few had any access to major league games.

o   Players in the 1880s bounced around more than players of any other era, more than players today, because the situation was relatively chaotic.  Leagues formed, franchises folded and leagues folded and franchises jumped from town to town at a much higher rate.  It wasn’t a stable situation. 

The great majority of the players from the 1880s and 1890s who are in the Hall of Fame were actually selected after 1960.   It’s a different process.  The voters had no actual memory of seeing the players play or of the time when they did play, no "I loved him when I was a kid" element in it.   It’s more of an analytical review of old records, "analytical" meant here in a broader sense.

As the "fandom" element of how players are viewed fades and the analytical review of old records takes over, the issue of whether a player stayed with one team or bounced around becomes irrelevant.   But in the short term, it is one of the dominant sorting keys as to who gets elected and who doesn’t. 

The First Study

I did two studies of this issue, and the second study is the better one and the one that leads us to more solid conclusions, so I wish I could report on that one first, but. . .doesn’t work.  I’ll have to start with this one.

In the first study, I divided the players into six groups along the time line, with each of the six being sub-divided into three parts.  For each player, I identified his "center season", the center being the first season after which his career Win Shares so far are greater than the rest of his career.   In other words, Henry Aaron through 1962 had 281 career Win Shares, but had 362 remaining after 1962, so 1962 is not the center season of his career.  In 1963 he added another 41 Win Shares (his career best), which brought his total 322, and then he had 321 remaining, so 1963 is the center season of his career.  It is common for the center season of a player’s career to also be his best season, or his highest Win Shares.   Of course, you could do the same research sorting by year of birth or year of retirement, wouldn’t matter much, but it seemed to me a little bit better to identify the center of the career, so that’s the way I did it. 

Then I sorted players by their center seasons, as follows:

Up to 1907               Group 1

1908 to 1932           Group 2

1933 to 1958           Group 3

1959 to 1978           Group 4

1979 to 1994           Group 5

1995 or later            Group 6

The lines were drawn so that there would be about 170 players in each group.   Equalizing the size of the groups was a dominant concern here, whereas in the second study, it wasn’t.

Then I sorted each of those into three sub-groups by their "one team identification percentage" relative to the era.   In each group, there were exactly 50 players who had "high" one-team identification, exactly 50 players who had low one-team identification, and somewhere between 64 and 75 players who had "medium" or mid-range one-team identification. 

Then I studied those players by the same method explained in the previous article.  Lou Whitaker was in Group 5A.  That is: a player from the 1979-1994 time period who had a one-team identification of 100%.  Gary Sheffield was in group 6C:  A post-1994 player with a very, very low one-team identification percentage, 14.4%.   Whitaker thus is entered into the Group 5A totals as 0 over .704, meaning that he had a 70.4% chance of going into the Hall of Fame, but has not so far, and Gary Sheffield was entered into Group 6C as 0 over .922.   Tony Oliva was entered as 4A, 1.000/.211—

"4" meaning that he played in the era 1959 to 1978,

"A" meaning that he had a high one-team identification percentage (100% with the Twins),

"1.000" meaning that he has been selected to the Hall of Fame, and

".211" meaning that his chance of being selected to the Hall of Fame, just based on his career Win Shares, was 21.1%.

In all six time-line groups, the selection percentage in Group A was higher than the selection percentage in Group C.   Controlling or eliminating the time-line bias, the selection advantage of the players with higher one-team identification persisted, and in fact INCREASED in most of the study.  

Here is a summary of the data, reminding you that "A" means players with a high one-team identification, and "C" means players with a low one-team identification.  

In other words, among players whose careers are centered in 1907 or earlier, the 50 players who had the highest one-team identification had an expectation of 10.4 Hall of Fame selections based on their Win Shares, but actually had only 9 players selected, or 87% of expectation.   The percentage is higher in Group A than in Group C, but not really higher; it’s just higher by a tiny, meaningless margin.   Here’s the rest of the chart:

            You may note that the selection rate for Group 6C, which is Gary Sheffield’s Group, is 12%, against a norm of 100%.  The second-lowest selection rate for any group of players is 57%.   All groups from the Babe Ruth era are over-represented in the Hall of Fame, and all groups since 1979 are under-represented. 

            This data is not consistent with the theory that the time-line is the driver and the one-team effect is a passenger—more or that later—but also, the data does not suggest that this is somehow related to steroids.  I am always leery of discussing who my be a steroider for fear I might accidentally accuse some innocent by-passer, but if Sheffield is a steroid suspect, then I believe he would actually be the ONLY steroid suspect in Group 6C who might otherwise be in the Hall of Fame.    Juan Gonzalez and Sammy Sosa are in Group 6A, and almost all of the others (Manny Ramirez, Carlos Delgado, Jason Giambi, Raul Mondesi, Barry Bonds, Mark McGwire, Ken Caminiti and Rafael Palmeiro)—are in Group 6B.   Some of the steroid guys are in Group 5.  It doesn’t seem like steroids could be the hidden cause of the 6A/6C selection discrepancy. 

Why I did another study

            There were several things that bothered me about this study, so I decided to try to fix them and run another, essentially the same but with a few flaws corrected or reduced a little. 

            First, I noticed that I had accidentally included in the data three or four players who should have been excluded, because they played a little bit in 2016, thus were not yet Hall of Fame eligible.  They were minor players who had no real Hall of Fame chance, and they don’t damage the conclusions in any meaningful way, but. . . an error is an error, so there is that.

            Also, I noticed that I had somehow excluded Jim Palmer from the study; I don’t know how that happened.   I added him back in.

            The thing that bothered me most, honestly, was the "era" groupings.   1933 to 1958 was an era, and 1959 to 1978 was an era.   Mickey Mantle’s center season was 1958 and Willie Mays’s was 1962, so Mantle wound up in Group 3, and Mays in Group 4.  They were born in the same year, 1931, as were Ernie Banks and Eddie Mathews. There are no two Hall of Famers in history who are more closely associated with one another than Mantle and Mays, and it seems frankly wrong to put Mantle in one era and Banks, Mathews and Mays in a different era.   That kind of thing is an inevitable consequence of drawing lines across history, so I had talked myself into believing that that was OK, but then I noticed that Harvey Kuenn and Joe Vosmik were listed in the same "era".  Joe Vosmik and Harvey Kuenn are actually quite similar players, but I am sorry; Joe Vosmik and Harvey Kuenn are NOT in the same era.  They’re just not.   Mantle and Mays are in different eras but Joe Vosmik and Harvey Kuenn are in the same era?  No. 

            So I decided to re-group the players by era.  And then, since I was committed to doing the work over again, I decided to address something else that was bothering me. 

            We can easily improve the predictive value of Win Shares as a Hall of Fame indicator by including some consideration of Most Valuable Player Awards and Cy Young Awards.   Roy Campanella had only 207 Career Win Shares, the same number as Bret Boone, Bob Friend or Tommy Davis, but we know that Campanella is ahead of those other men, as a Hall of Fame candidate, because he won three MVP Awards.  Randy Johnson earned only 326 Win Shares in his career, whereas Bobby Grich had 329, but we know that Johnson was more likely to be selected than Grich was because Johnson won 4 Cy Young Awards.  

            I decided to include in the Hall of Fame estimates, counted as Win Shares:

(1)  25 points for each MVP Award, and

(2)  15 points for each Cy Young Award.

There are any number of reasons why this might be said to be not proper analytical method.  Who says 25 points is the right amount for an MVP Award?  Who says it isn’t 15, or 40, or 47?  If MVP awards are credited as Hall of Fame enticements, what about batting titles?  Didn’t Tony Oliva’s batting titles push him over the line? 

All valid points, but the thing is, I am 100% certain that including these "points" in the system as ersatz Win Shares makes the Hall of Fame probability estimates more accurate.  I don’t have a shred of doubt about that.   Roger Maris had 223 career Win Shares, so we had him listed with a 12% probability of Hall of Fame selection, the same number as Todd Zeile and Harvey Kuenn.  When we add 50 points for his two MVP selections, however, Maris’ Hall of Fame probability estimate increases from 12% to 30%.  That seems a lot more realistic.   By the inclusion of these "points", Sandy Koufax increases from 7% to 27%, Ernie Banks from 59% to 85%, Bob Gibson from 56% to 82%, Carl Hubbell from 49% to 73%, Dennis Eckersley from 45% to 65%, Rollie Fingers from 6% to 11%, etc., etc.  I haven’t any doubt that these are helpful adjustments.

Of course, when I increase the percentages for some players, this requires that I DECREASE the percentages for those players who DIDN’T win an MVP award or a Cy Young Award and re-calibrate the system, which was a lot of work, but don’t mind me, my kids are grown and there’s a pandemic; what else do I have to do?   As Roger Maris goes from 12% to 30%, Harvey Kuenn and Todd Zeile, Maris’ partners at 223, drop from 12% to 9.5%.  

With the re-adjustment, the top group of players goes from 425 (Win Shares) to 450 (Augmented Win Shares), and a couple of those not selected are no longer in the top group, so the top group goes from 92.2% to 94.5%.   I used the same process as before, but just re-ran the process with augmented data, thus getting slightly different Hall of Fame expectations for almost everybody.   Lou Whitaker, for some reason, goes from 70.4% to 70.5%.

The Second Study

Then I ran the second study, using essentially the same process as before, but this time with SEVEN groups, rather than six, but with no "middle group", so that the number of sub-groups in the study decreased from 18 to 14.   Any player whose one-team percentage was higher than the period norm was considered in the "High" group, and any player whose one-team percentage was lower than the period norm was considered in the "Low" Group.  (Fortunately, since there is a complex calculation underlying the one-team percentage, there are essentially no ties, so everybody is either above the line or below it.   Nobody has to be left out.)

Here are the time-line Groups:

Group 1         Up to 1899   (Cap Anson era)

Group 2         1900 to 1917  (Honus Wagner era)

Group 3         1918 to 1935  (Babe Ruth era)

Group 4         1936 to 1952  (Joe DiMaggio era)

Group 5         1953 to 1970  (Willie Mays era)

Group 6         1971 to 1987 (Mike Schmidt era)

Group 7         1988 or later (Barry Bonds era)

            A less problematic grouping.  Here is the data from the study:

            The data shows that, of players who had HIGH one-time identification percentages, 140 made the Hall of Fame, whereas for those who bounced from team to team, only 96 did so—with essentially the same aggregate expectation. 

As in the earlier study, there is no significant advantage for the one-team or few-teams players in the early history.   The Babe Ruth era is 1918 to 1935, and Hall of Fame selections began shortly after that.   Once the voters are voting on players that they actually saw play and actually may have rooted for once, the bias immediately becomes apparent in the data, and grows dramatically larger over time, increasing to about a 160% bias effect in the most recent era, or 2.6 to 1, although the 160% figure is based on a relatively small sample (31 Hall of Fame selections.) 

There is no way to interpret this data to say that the time-line bias is creating the advantage for players who played for only one team.   It is not creating it; it is not contributing to it.   If anything, it is hiding it.  Comparing players of generally equivalent accomplishment to their contemporaries, there is a very significant Hall of Fame selection advantage for the players who were easily identified with one team.  

Three Notes in Closing

First, the "Center Year" chart is just really interesting.  It’s fun; I don’t know why.  The list of players with 150 or more Win Shares whose center year was 1958 is Mickey Mantle, Harvey Kuenn, Pete Runnels, Al Smith, Del Crandall, Lew Burdette. Smoky Burgess, Johnny Temple and Vic Power.   The list for 1968 is Willie McCovey, Billy Williams, Bob Gibson, Ron Fairly, Tony Oliva, Jim Perry, Tim McCarver, Don Buford, Dennis Menke and Sudden Sam McDowell.  You can almost make a team out of those guys.  I don’t know why; it’s just fun.

Second, I have not given you the period norms for one-team percentages.   I saved them for a reason:  they’re relevant to the conclusion.  This chart combines the average one-team identification percentages from the two studies:

As you can see, the average one-team identification started out at about 45% in the 19^th century, increased steadily until about 1945, when it was near 70%.  It then declined slowly even before free agency arrived, and went down sharply at the start of free agency, mid-1970s.  

But what this means is, the data will not support an interpretation that the time line is responsible for the separation in data noted in the original study.   You have to think about what I am saying, so. . .think about it.   The implication is that the fact that more players with high one-team percentages have been selected to the Hall of Fame is because, over time, one-team percentages have gone steadily downward.   If one-team percentage have NOT gone steadily downward, then the passage of time would not cause a split between the Hall of Fame selection percentages of high and low one-team percentages. 

And one-team percentages have NOT, in fact, gone steadily downward over time.  The two lowest figures in the list of thirteen are actually for the OLDEST groups.   When you are looking for a downhill slope in the data, the fact that the data starts at the bottom of the hill is a terrible obstacle.  In essence, the argument is that the momentum of the split is created by the data skiing downhill.  But since the data has to CLIMB the hill before it skis down the hill. . . .different problem. 

If we split the data at 1950, then the average one-team identification percentage is 59.1% in the first half, and 57.2% in the second half.  If we split the data into the first half of the players and the second half, they divide between 1958 and 1959, and the percentages are 60% (up to 1958) and 56.1% (since 1959.)  AND A VERY SIGNIFICANT MAJORITY OF THE HALL OF FAMERS ARE IN THE FIRST HALF.  145 of the Hall of Famers are in the first half, 91 in the second half. 

Even if you interpret the time-line bias in the light most favorable to the theory, by ignoring the data before 1920, the time-line bias is still nowhere near as large as the split between the players with high one-team percentages and low one-team percentages. 

So what does that mean?   My third point here, my conclusions:

            If we split the data in half, the bias in the timeline is in the range of 5 to 10%.   But the difference between the homebodies and the vagabonds is 45%, more or less. 

            A bias in the data of 5 to 10% cannot possibly cause a SECONDARY bias, in a vaguely related measurement, of 45%.   It just doesn’t make sense. 

            So there are two reasons why this theory cannot be correct:

1)     That if we compare players only to others who played at the same time, the one-team type players still do far better in Hall of Fame selections,

2)     That it implies that the time-line bias is creating a bias in the one-team vs many-teams selection which is much larger than the time-line bias itself.