In their 2005 study, The Longevity of Baseball Hall of Famers Compared to Other Players, Abel and Kruger (2005) used major league baseball (MLB) player data from Sean Lahman’s Baseball Database (2022) to evaluate the role of awarded achievement on longevity. Using a Cox proportional hazards model, Abel and Kruger (2005) found that median post-induction survival for Hall of Famers was 5 years shorter than for noninducted players (18 vs. 23 years, respectively).
This workbook reproduces the analysis with the 2021 release of the database (v2022.2) and updates the analysis with the newer data. Differences in players included in the data and their attributes resulted in slightly different coefficient estimates. The Authors’ findings were generally confirmed except that instead of an expected five additional years of life for 20-year-old MLB players, the Lahman-based data produced estimates of only three additional years. Fitting the models with data updated through the 2021 season resulted in lower measured mortality risks that are in closer alignment with the Authors’ findings. A 20-year-old MLB player can expect an additional 4.5 years of additional life over comparable U.S. males. A Cox proportional hazards model with time-dependent covariates confirmed the Authors’ findings that height, weight, and handedness are not associated with risk of death, and career length is inversely associated with risk of death.
Baseball players elected to the Hall of Fame are likely to have relatively high self-esteem due to repeated achievement.
Abel and Kruger (2005) analyzed data from the 2002 Lahman database, Version 5.0 (2002). The present analysis uses the same database, and also the R version of the 2021 edition of the Lahman database. The Lahman R package v10.0-1 (2022) is continuously updated and available for download on CRAN.
# 2002 Lahman database (v5.0)
MasterV5 <- read_csv("../data/lahman_50-2k/Master.csv")
HallOfFameV5 <- read_csv("../data/lahman_50-2k/HallOfFame.csv")
FieldingV5 <- read_csv("../data/lahman_50-2k/Fielding.csv")
# 2021 Lahman database (v10.0-1)
data("People", "HallOfFame", "Fielding", "Appearances", package = "Lahman")The Master data set from v5.0 is replaced with
People in v10.0-1. It includes both players and non-players
(executives, announcers, and umpires) inducted into the Hall of Fame.
Actual players are identified by presence in the
Appearances data set (v10.0-1) or a non-null
playerID (v5.0). Of the 20,370 rows in the data set, 20,167
are players.
Abel and Kruger (2005) filtered for players in the Hall of Fame who had been inducted while still alive as of 2002. They matched each Hall of Fame player to players of the same birth year who were also alive in the induction year.
hof_v5 <- HallOfFameV5 %>%
inner_join(MasterV5, by = "hofID") %>%
filter(
inducted == "Y",
# Inducted as a player _and_ has a playerID identifier (need both)
category == "Player", !is.na(playerID),
yearid <= 2002,
coalesce(deathYear, 9999) > yearid
) %>%
mutate(inductedAge = yearid - birthYear) %>%
select(playerID, birthYear, deathYear, inductedYear = yearid)
ctrl_v5 <- hof_v5 %>%
inner_join(MasterV5, by = "birthYear", suffix = c(".hof", "")) %>%
filter(
!is.na(playerID),
coalesce(deathYear, 9999) > inductedYear,
playerID != playerID.hof
) %>%
anti_join(hof_v5, by = "playerID") %>%
select(playerID, birthYear, deathYear, inductedYear, playerID.hof)
fielding_v5 <- FieldingV5 %>%
group_by(playerID, position = POS) %>%
summarise(.groups = "drop", minYear = min(yearID), maxYear = max(yearID), n = sum(G)) %>%
group_by(playerID) %>%
slice_max(order_by = n) %>%
mutate(seasons = maxYear - minYear + 1)
dat_v5 <- bind_rows(hof = hof_v5, ctrl = ctrl_v5, .id = "hof") %>%
inner_join(MasterV5 %>% select(playerID, height, weight), by = "playerID") %>%
inner_join(fielding_v5 %>% select(playerID, position), by = "playerID") %>%
mutate(
bmi = weight / if_else(height == 0, NA_real_, height)^2 * 703,
induction_year_age = inductedYear - birthYear,
# follow-up time is earlier of two dates: death or data set date.
fu_year = coalesce(deathYear, max(MasterV5$deathYear, na.rm = TRUE)),
fu_time = fu_year - inductedYear,
fu_status = if_else(is.na(deathYear), 0, 1)
)Abel and Kruger (2005) report,
One hundred and forty-three (143) players in the Hall of Fame had been inducted into the Baseball Hall of Fame while still alive as of 2002. These were age-matched with 3,430 players. The mean age of the Hall of Famers at time of induction was 57.5 years (SD \(\pm\) 12.4). A total of 1,695 of these players had died by the end of 2002.
My data set matches the Hall of Fame players closely, but has about twice as many age-matched controls.
# Players in Hall of Fame. Compare to 143.
dat_v5 %>% filter(hof == "hof") %>% nrow()
## [1] 144
# Age-matched players. Compare to 3,430.
dat_v5 %>% filter(hof == "ctrl") %>% pull(playerID) %>% n_distinct()
## [1] 6430
# Mean age of Hall of Famers. Compare to 57.5, SD 12.4.
dat_v5 %>% filter(hof == "hof") %>%
summarize(
mean = mean(induction_year_age),
sd = sd(induction_year_age)
)
## # A tibble: 1 × 2
## mean sd
## <dbl> <dbl>
## 1 57.4 12.5
# Deaths. Compare to 1,695.
dat_v5 %>% filter(deathYear <= 2002) %>% pull(playerID) %>% n_distinct()
## [1] 3434mdl <- coxph(surv())
dat2002 %>%n_distinct(ctrl_v5$playerID)
ctrl_v5 %>% ggplot() + geom_histogram(aes(x = birthYear), binwidth = 1)
ctrl_v5 %>% count(birthYear)
MasterV5 %>% filter(deathYear == 0)
summary(hof_v5$deathYear)
hof_v10 <- HallOfFame %>%
inner_join(People, by = "playerID") %>%
filter(
inducted == "Y",
category == "Player",
coalesce(deathYear, 9999) > yearID
) %>%
mutate(inductedAge = yearID - birthYear) %>%
select(playerID, inductedYear = yearID, birthYear, inductedAge, deathYear)
hof2002 <- hof %>% filter(inductedYear <= 2002)
cohort_0 <- hof %>%
inner_join(People, by = "birthYear", suffix = c(".hof", "")) %>%
filter(!is.na(debut)) %>%
# semi_join(Appearances, by = "playerID") %>%
filter(
# !is.na(debut),
# deathYear != 0, # must be a placeholder val.
coalesce(deathYear, 9999) > inductedYear
) %>%
anti_join(hof, by = "playerID")
cohort <- cohort_0 %>%
select(playerID, birthYear, deathYear, nameFirst, nameLast) %>%
unique()
cohort2002 <- cohort_0 %>%
filter(inductedYear <= 2002) %>%
select(playerID, birthYear, deathYear, nameFirst, nameLast) %>%
unique()
dat <- bind_rows(
HOF = hof %>% select(playerID),
CONTROL = cohort %>% select(playerID),
.id = "hof"
) %>%
inner_join(People, by = "playerID") %>%
inner_join(career, by = "playerID") %>%
mutate(bmi = weight / if_else(height == 0, NA_integer_, height)^2 * 703) %>%
select(playerID, hof, bmi, position, seasons, everything())
dat2002 <- bind_rows(
HOF = hof2002 %>% select(playerID),
CONTROL = cohort2002 %>% select(playerID),
.id = "hof"
) %>%
inner_join(People, by = "playerID") %>%
inner_join(career, by = "playerID") %>%
mutate(bmi = weight / if_else(height == 0, NA_integer_, height)^2 * 703) %>%
select(playerID, hof, bmi, position, seasons, everything())Hall-of-Famers died significantly earlier than their controls. The median length of post-induction survival for the Hall of Famers was 18 years (95% CI = 15.0–21.0) versus 23 years (CI = 22.1–23.9) for matched controls (Odds ratio [OR] = 1.37; 95% CI = 1.08–1.73).
The Methods and Materials section of the article elaborates the age-matching criteria (page 960).
[age-matched players] were alive at the time of induction when induction occurred for their case-matched cohort. Only players were included in our analysis (executives, announcers, umpires were not included; managers were included who had also been players).
The code chunk below identifies every player who shares the same age as a Hall of Famer and was alive at that Hall of Famer’s induction year. Note that a player may have been alive for one Hall of Famer’s induction, but not for the later induction of another Hall of Famer with the same birth year.
# "Hall-of-Famer’s had a mean (+/-SD) BMI of 25.2 +/- 1.8 versus 24.7 +/- 1.5 for controls"
dat %>% group_by(hof) %>% summarize(.groups = "drop", n = n(), meanBMI = mean(bmi, na.rm = TRUE), sdBMI = sd(bmi, na.rm = TRUE))
t.test(bmi ~ hof, data = dat)
dat %>% gtsummary::tbl_summary(by = hof, include = bmi, statistic = list(bmi ~ "{mean}, {sd}")) %>% gtsummary::add_p()
# "A higher percentage (64%, n = 92) of the Hall of Famers were also dead..."
# I have n = 90.
hall_of_famers %>% mutate(is_alive = if_else(coalesce(deathYear, 9999) > 2002, TRUE, FALSE)) %>% janitor::tabyl(is_alive)
# "...compared with controls (47%, n = 1,603) who were the same age at the time of the Hall of Famer’s induction..."
# I have n = 52%, n = 3,401.
cohort %>% mutate(is_alive = if_else(coalesce(deathYear, 9999) > 2002, TRUE, FALSE)) %>% janitor::tabyl(is_alive)
cohort %>% count(playerID, sort = TRUE)Does this match the article? Not even close. cohort
should have 3,430 players. Instead, it has
scales::comma(nrow(cohort)) players. I tried defining
age-match as sharing both birth year and birth month, but that yielded a
result set that was way too small.
nrow(cohort)What went wrong? Let’s audit 20 random players of
cohort. Since cohort has so many more players
than it should have, at least one of these players should be
unjustified.
set.seed(1234)
random_rownums <- runif(20, 1, nrow(cohort)) %>% as.integer()
matches <- cohort[random_rownums, ] %>%
inner_join(hall_of_famers, by = "birthYear", suffix = c(".cohort", ".hof")) %>%
filter(coalesce(deathYear.cohort, 9999) > inductedYear,
playerID.cohort != playerID.hof) %>%
select(playerID.cohort, nameFirst, nameLast, birthYear, deathYear = deathYear.cohort,
playerID.hof, inductedYear)A player is age-matched if 1) there was a Hall of Famer inducted while alive who shares the same birth year, and 2) they were alive during the induction year too. Are these players correctly age-matched correctly? Yes, all 20 are age-matched.
matches %>% select(playerID.cohort) %>% unique() %>% nrow()
People %>% filter(hofID == "adamsbo03h")Let’s look at some of the matches. Ethan Allen was still alive when Red Ruffing was inducted in 1967. Both were born in 1904. Art Bues was alive when Tris Speaker was inducted in 1937, and both were born in 1888. Douglas Whammy shared the same birth year as three Hall of Famers inducted while still alive. Douglas Whammy was still alive as of 2002.
matches %>% rename(inducted = inductedYear) %>% arrange(playerID.cohort)Why is my data set so much longer than Abel and Kruger’s - what am I missing here?
smith (2011) commented on the Abel and Kruger (2005) study with two criticisms. One, the result seems implausible on its face because it contradicts a wide body of research, and attributes a large effect to a single event, especially considering that the control group are also elite athletes. The second criticism is that the data set is flawed. smith (2011) explains that the Lahman database has incomplete data, including missing death dates. Abel and Kruger (2005) were assuming everyone with a missing death date was still alive, and since most missing death dates are for obscure players, their data biased upward the non-Hall of Fame players’ lifetimes.
Unfortunately, I cannot demonstrate this flaw in the data, and Smith (2011) only points out that players born in the 1800s must certainly be deceased.
Master %>% filter(birthYear == 0) %>% nrow()
Master %>% filter(deathYear == 0) %>% nrow()
Master %>% filter(birthYear < 1900, is.na(deathYear)) %>% nrow()
Master %>% filter(birthYear < 1900, is.na(deathYear)) %>% nrow()
People %>% filter(playerID == "adamsji01")
People %>% filter(birthYear < 1900, is.na(deathYear)) %>% nrow()
Master %>%
filter(!is.na(birthYear), birthYear != 0, deathYear != 0) %>%
mutate(
alive = if_else(is.na(deathYear), "Alive", "Deceased"),
age2002 = coalesce(deathYear, 2002) - birthYear
) %>%
ggplot() + geom_histogram(aes(x = age2002))Ernest L. Abel & Michael L. Kruger (2005) The Longevity of Baseball Hall of Famers Compared to Other Players, Death Studies, 29:10, 959-963, DOI: 10.1080/07481180500299493. PDF.
Lahman, S. (2002). The Baseball Archive v.5.0. Available at https://www.seanlahman.com/baseball-archive/statistics/.
Friendly M, Dalzell C, Monkman M, Murphy D (2022). Lahman: Sean ‘Lahman’ Baseball Database. R package version 10.0-1, https://CRAN.R-project.org/package=Lahman.
Smith G. The Baseball Hall of Fame is not the kiss of death. Death Stud. 2011 Nov-Dec;35(10):949-55. doi: 10.1080/07481187.2011.553337. PMID: 24501860. PDF