data(Fielding)
# Basic fielding data
require("dplyr")
# Roberto Clemente's fielding profile
# pitching and catching related data removed
# subset(Fielding, playerID == "clemero01")[, 1:13]
Fielding %>%
filter(playerID == "clemero01") %>%
select(1:13)
# Yadier Molina's fielding profile
# PB, WP, SP and CS apply to catchers
Fielding %>%
subset(playerID == "molinya01") %>%
select(-WP, -ZR)
# Pedro Martinez's fielding profile
Fielding %>% subset(playerID == "martipe02")
# Table of games played by Pete Rose at different positions
with(subset(Fielding, playerID == "rosepe01"), xtabs(G ~ POS))
# Career total G/PO/A/E/DP for Luis Aparicio
Fielding %>%
filter(playerID == "aparilu01") %>%
select(G, PO, A, E, DP) %>%
summarise_each(funs(sum))
# Top ten 2B/SS in turning DPs
Fielding %>%
subset(POS %in% c("2B", "SS")) %>%
group_by(playerID) %>%
summarise(TDP = sum(DP, na.rm = TRUE)) %>%
arrange(desc(TDP)) %>%
head(., 10)
# League average fielding statistics, 1961-present
Fielding %>%
filter(yearID >= 1961 & POS != "DH") %>%
select(yearID, lgID, POS, InnOuts, PO, A, E) %>%
group_by(yearID, lgID) %>%
summarise_at(vars(InnOuts, PO, A, E), funs(sum), na.rm = TRUE) %>%
mutate(fpct = round( (PO + A)/(PO + A + E), 3),
OPE = round(InnOuts/E, 3))
Run the code above in your browser using DataLab