ATP & WTA Match Data: Descriptive Overview

Tennis
R
Data Analysis
Surface trends, serve statistics, age distributions, and all-time leaders across 55+ years of professional tennis
Published

March 10, 2026

Code 
library(tidyverse)
library(scales)
library(gt)

theme_set(
  theme_minimal(base_size = 13) +
    theme(
      plot.title = element_text(face = "bold", size = 16),
      plot.subtitle = element_text(colour = "grey40", size = 12),
      panel.grid.minor = element_blank(),
      legend.position = "bottom"
    )
)

surface_cols <- c(
  "Hard"   = "#2171B5",
  "Clay"   = "#CB4335",
  "Grass"  = "#27AE60",
  "Carpet" = "#8E44AD"
)

tour_cols <- c("ATP" = "#1A5276", "WTA" = "#C0392B")

Data

The data comes from Jeff Sackmann’s GitHub repositories, which contain match-by-match results for every ATP and WTA tour-level match since 1968.

Code 
data_dir <- here::here("data-projects", "data")
dir.create(data_dir, showWarnings = FALSE, recursive = TRUE)

download_tour <- function(tour_prefix, base_url, years = 1968:2025) {
  cache_file <- file.path(data_dir, paste0(tour_prefix, "_matches.csv"))

  if (file.exists(cache_file)) {
    return(read_csv(cache_file, show_col_types = FALSE))
  }

  df <- map_dfr(paste0(tour_prefix, "_matches_", years, ".csv"), \(f) {
    tryCatch(
      read_csv(paste0(base_url, f), show_col_types = FALSE,
               col_types = cols(.default = "c")),
      error = \(e) tibble()
    )
  })

  write_csv(df, cache_file)
  df
}

atp_raw <- download_tour(
  "atp", "https://raw.githubusercontent.com/JeffSackmann/tennis_atp/master/"
)
wta_raw <- download_tour(
  "wta", "https://raw.githubusercontent.com/JeffSackmann/tennis_wta/master/"
)
Code 
clean_matches <- function(df, tour) {
  df |>
    mutate(
      tour = tour,
      tourney_date = as.Date(as.character(tourney_date), format = "%Y%m%d"),
      year = year(tourney_date),
      surface = str_to_title(surface),
      surface = if_else(surface %in% c("Hard", "Clay", "Grass", "Carpet"),
                        surface, NA_character_),
      winner_age   = as.numeric(winner_age),
      loser_age    = as.numeric(loser_age),
      w_ace        = as.numeric(w_ace),
      l_ace        = as.numeric(l_ace),
      w_df         = as.numeric(w_df),
      l_df         = as.numeric(l_df),
      w_svpt       = as.numeric(w_svpt),
      l_svpt       = as.numeric(l_svpt),
      w_1stIn      = as.numeric(w_1stIn),
      l_1stIn      = as.numeric(l_1stIn),
      w_1stWon     = as.numeric(w_1stWon),
      l_1stWon     = as.numeric(l_1stWon),
      w_2ndWon     = as.numeric(w_2ndWon),
      l_2ndWon     = as.numeric(l_2ndWon),
      w_bpSaved    = as.numeric(w_bpSaved),
      l_bpSaved    = as.numeric(l_bpSaved),
      w_bpFaced    = as.numeric(w_bpFaced),
      l_bpFaced    = as.numeric(l_bpFaced),
      minutes      = as.numeric(minutes),
      tourney_level = case_when(
        tourney_level == "G" ~ "Grand Slam",
        tourney_level == "M" ~ "Masters",
        tourney_level == "A" ~ "Tour 250/500",
        tourney_level == "F" ~ "Tour Finals",
        tourney_level == "D" ~ "Davis/Fed Cup",
        TRUE ~ tourney_level
      )
    )
}

atp     <- clean_matches(atp_raw, "ATP")
wta     <- clean_matches(wta_raw, "WTA")
matches <- bind_rows(atp, wta)

Overview

Code 
matches |>
  group_by(tour) |>
  summarise(
    `Total matches`    = comma(n()),
    `Year range`       = paste(min(year, na.rm = TRUE), "\u2013",
                               max(year, na.rm = TRUE)),
    `Unique winners`   = comma(n_distinct(winner_name)),
    `Unique tournaments` = comma(n_distinct(tourney_name)),
    `Median duration (min)` = round(median(minutes, na.rm = TRUE), 0),
    .groups = "drop"
  ) |>
  gt() |>
  tab_header(title = "Dataset overview") |>
  cols_label(tour = "Tour")
Dataset overview
Tour Total matches Year range Unique winners Unique tournaments Median duration (min)
ATP 194,996 1967 – 2024 4,384 4,188 97
WTA 158,092 1967 – 2024 4,535 5,843 93

Matches per year

The ATP has consistently staged more tour-level matches than the WTA, though both tours expanded rapidly through the 1970s and early 1980s before stabilising.

Code 
matches |>
  filter(!is.na(year), year >= 1970, year <= 2025) |>
  count(year, tour) |>
  ggplot(aes(year, n, colour = tour)) +
  geom_line(linewidth = 0.8) +
  geom_point(size = 0.6, alpha = 0.5) +
  scale_colour_manual(values = tour_cols) +
  scale_y_continuous(labels = comma) +
  labs(
    title = "Matches per year",
    subtitle = "ATP and WTA tour-level matches, 1970\u20132025",
    x = NULL, y = "Matches", colour = NULL
  )

Total matches played per year by tour

Surface distribution over time

The decline of carpet courts and the ascent of hard courts is one of the most visible structural changes in professional tennis. Clay has held remarkably steady as a share of total matches.

Code 
matches |>
  filter(!is.na(surface), !is.na(year), year >= 1970, year <= 2025) |>
  count(year, surface) |>
  group_by(year) |>
  mutate(pct = n / sum(n)) |>
  ungroup() |>
  ggplot(aes(year, pct, fill = surface)) +
  geom_area(alpha = 0.85) +
  scale_fill_manual(values = surface_cols) +
  scale_y_continuous(labels = percent) +
  labs(
    title = "Surface share over time",
    subtitle = "The decline of carpet and rise of hard courts",
    x = NULL, y = "Share of matches", fill = "Surface"
  )

Share of matches by surface type over time

Surface breakdown by tour

Code 
matches |>
  filter(!is.na(surface)) |>
  count(tour, surface) |>
  group_by(tour) |>
  mutate(pct = n / sum(n)) |>
  ungroup() |>
  ggplot(aes(tour, pct, fill = surface)) +
  geom_col(position = "dodge", width = 0.7) +
  scale_fill_manual(values = surface_cols) +
  scale_y_continuous(labels = percent) +
  labs(title = "Surface distribution by tour", x = NULL, y = "Share", fill = "Surface")

Surface distribution split by tour

Tournament levels

Code 
matches |>
  filter(!is.na(tourney_level),
         tourney_level %in% c("Grand Slam", "Masters", "Tour 250/500",
                              "Tour Finals", "Davis/Fed Cup")) |>
  count(tour, tourney_level) |>
  group_by(tour) |>
  mutate(pct = n / sum(n)) |>
  ungroup() |>
  ggplot(aes(reorder(tourney_level, pct), pct, fill = tour)) +
  geom_col(position = "dodge", width = 0.7) +
  scale_fill_manual(values = tour_cols) +
  scale_y_continuous(labels = percent) +
  coord_flip() +
  labs(
    title = "Match distribution by tournament level",
    x = NULL, y = "Share of matches", fill = NULL
  )

Match distribution by tournament level

Age distribution of winners

The peak winning age has shifted upward over the decades, particularly on the men’s side, reflecting improved fitness, sports science, and longer career spans at the top.

Code 
matches |>
  filter(!is.na(winner_age), !is.na(year), year >= 1970) |>
  mutate(decade = paste0(10 * (year %/% 10), "s")) |>
  ggplot(aes(winner_age, fill = tour)) +
  geom_density(alpha = 0.5, colour = NA) +
  scale_fill_manual(values = tour_cols) +
  facet_wrap(~decade, scales = "free_y") +
  labs(
    title = "Winner age distribution by decade",
    x = "Age", y = "Density", fill = NULL
  )

Winner age density across decades

Match duration

Duration data becomes reliable from around 1990. The IQR band shows match lengths have been remarkably stable on the ATP tour, with a slight upward drift in median duration through the 2010s.

Code 
atp |>
  filter(!is.na(minutes), !is.na(year), minutes > 0, minutes < 600,
         year >= 1990) |>
  group_by(year) |>
  summarise(
    median_min = median(minutes),
    q25 = quantile(minutes, 0.25),
    q75 = quantile(minutes, 0.75),
    .groups = "drop"
  ) |>
  ggplot(aes(year, median_min)) +
  geom_ribbon(aes(ymin = q25, ymax = q75), fill = "#2171B5", alpha = 0.2) +
  geom_line(colour = "#2171B5", linewidth = 0.9) +
  labs(
    title = "ATP match duration over time",
    subtitle = "Median with IQR band (since 1990)",
    x = NULL, y = "Minutes"
  )

ATP match duration: median with IQR band

Aces and double faults

Ace counts have trended upward over time, consistent with the increasing emphasis on serve speed and the shift toward hard courts.

Code 
atp |>
  filter(!is.na(w_ace), !is.na(w_df), !is.na(year), year >= 1990) |>
  mutate(total_aces = w_ace + l_ace, total_df = w_df + l_df) |>
  group_by(year) |>
  summarise(
    avg_aces = mean(total_aces, na.rm = TRUE),
    avg_df   = mean(total_df, na.rm = TRUE),
    .groups  = "drop"
  ) |>
  pivot_longer(c(avg_aces, avg_df), names_to = "stat", values_to = "value") |>
  mutate(stat = if_else(stat == "avg_aces", "Aces", "Double faults")) |>
  ggplot(aes(year, value, colour = stat)) +
  geom_line(linewidth = 0.9) +
  scale_colour_manual(values = c("Aces" = "#2171B5", "Double faults" = "#CB4335")) +
  labs(
    title = "Aces and double faults per match (ATP)",
    subtitle = "Combined winner + loser totals",
    x = NULL, y = "Average per match", colour = NULL
  )

Average aces and double faults per match (ATP)

All-time match wins leaders

Code 
top_n_wins <- function(df, tour_name, n = 15) {
  df |>
    count(winner_name, name = "wins") |>
    slice_max(wins, n = n) |>
    mutate(tour = tour_name)
}

bind_rows(
  top_n_wins(atp, "ATP"),
  top_n_wins(wta, "WTA")
) |>
  mutate(winner_name = fct_reorder(winner_name, wins)) |>
  ggplot(aes(wins, winner_name, fill = tour)) +
  geom_col() +
  scale_fill_manual(values = tour_cols) +
  facet_wrap(~tour, scales = "free_y") +
  labs(title = "All-time match wins leaders", x = "Tour-level wins", y = NULL) +
  theme(legend.position = "none")

Top 15 match wins leaders (ATP and WTA)

Grand Slam titles

Code 
slam_titles <- function(df, tour_name, n = 15) {
  df |>
    filter(tourney_level == "Grand Slam", round == "F") |>
    count(winner_name, name = "titles") |>
    slice_max(titles, n = n) |>
    mutate(tour = tour_name)
}

bind_rows(
  slam_titles(atp, "ATP"),
  slam_titles(wta, "WTA")
) |>
  mutate(winner_name = fct_reorder(winner_name, titles)) |>
  ggplot(aes(titles, winner_name, fill = tour)) +
  geom_col() +
  scale_fill_manual(values = tour_cols) +
  facet_wrap(~tour, scales = "free_y") +
  labs(title = "Grand Slam titles", x = "Titles", y = NULL) +
  theme(legend.position = "none")

Grand Slam titles (top 15 per tour)

First serve win % by surface

Grass courts show the highest first serve win percentage, reflecting the low bounce and faster pace. Clay is the lowest, where rallies are longer and returns more effective.

Code 
atp |>
  filter(!is.na(w_1stWon), !is.na(w_1stIn), !is.na(surface),
         w_1stIn > 0, year >= 2000) |>
  mutate(first_serve_pct = w_1stWon / w_1stIn) |>
  ggplot(aes(surface, first_serve_pct, fill = surface)) +
  geom_boxplot(alpha = 0.7, outlier.alpha = 0.05) +
  scale_fill_manual(values = surface_cols) +
  scale_y_continuous(labels = percent) +
  labs(
    title = "First serve win % by surface (ATP, 2000+)",
    subtitle = "Winners only",
    x = NULL, y = "1st serve win %"
  ) +
  theme(legend.position = "none")

First serve win % by surface (ATP, 2000+)

Surface edge over time

Has professional tennis become more or less surface-neutral over the past 30 years? The “surface edge” is measured as the server’s share of all serve points played — a high value means big serves dominate (fast surface); a low value means returns and rallies rule (slow surface). The gap between Grass and Clay is the best single indicator of how differentiated the three main surfaces are.

Code 
edge_df |>
  ggplot(aes(year, srv_pct, colour = surface)) +
  geom_line(linewidth = 0.7, alpha = 0.45) +
  geom_smooth(method = "loess", span = 0.4, se = FALSE, linewidth = 1.2) +
  scale_colour_manual(values = surface_cols) +
  scale_y_continuous(labels = percent) +
  facet_wrap(~tour) +
  labs(
    title = "Server's share of serve points by surface",
    subtitle = "Higher = surface favours the serve (faster court). Loess smoother.",
    x = NULL, y = "Serve win %", colour = "Surface"
  )

Server’s share of serve points by surface (1995–2025)
Code 
gap_cols <- c(
  "Grass - Clay" = "#27AE60",
  "Hard - Clay"  = "#2171B5",
  "Grass - Hard" = "#8E44AD"
)

gap_df |>
  ggplot(aes(year, pct_pts, colour = gap)) +
  geom_hline(yintercept = 0, linetype = "dashed", colour = "grey60") +
  geom_line(linewidth = 0.7, alpha = 0.45) +
  geom_smooth(method = "loess", span = 0.4, se = FALSE, linewidth = 1.2) +
  scale_colour_manual(values = gap_cols) +
  scale_y_continuous(labels = \(x) paste0(round(x * 100, 1), " pp")) +
  facet_wrap(~tour) +
  labs(
    title = "Surface edge over time",
    subtitle = "Narrowing gap = surfaces converging; wider gap = more specialised play",
    x = NULL, y = "Percentage point gap", colour = "Comparison"
  )

Surface edge: gap in serve dominance between surfaces (1995–2025)

Data completeness

Serve statistics are only consistently recorded from the early 1990s onward. This is worth keeping in mind when doing any time-series analysis involving ace counts, break points, or serve percentages.

Code 
matches |>
  filter(!is.na(year), year >= 1970, year <= 2025) |>
  mutate(has_serve_stats = !is.na(w_ace)) |>
  group_by(year, tour) |>
  summarise(pct_complete = mean(has_serve_stats), .groups = "drop") |>
  ggplot(aes(year, pct_complete, colour = tour)) +
  geom_line(linewidth = 0.8) +
  scale_colour_manual(values = tour_cols) +
  scale_y_continuous(labels = percent) +
  labs(
    title = "Serve statistics availability",
    subtitle = "Percentage of matches with ace/DF/serve data",
    x = NULL, y = "Share with stats", colour = NULL
  )

Share of matches with serve statistics recorded