Skip to contents

Corsi, Fenwick and PDO in R: Hockey Shot-Share Metrics with fastRhockey

Saiem Gilani, Ben Howell

Source: vignettes/articles/hockey-advanced-metrics-corsi-fenwick.Rmd
hockey-advanced-metrics-corsi-fenwick.Rmd

Shot-share metrics measure which team had the puck going toward the right net. Goals are rare, about three per team per game, so a season of goal differential is noisy. Shot attempts happen ten to twenty times as often, and a team that out-attempts its opponents tends to out-score them over time. This article builds the standard shot-share metrics from fastRhockey NHL play-by-play, for teams and for individual players, at 5-on-5.

What you’ll learn

  • How Corsi, Fenwick, shots, goals and expected goals relate to each other.
  • Which fastRhockey columns identify a shot attempt, the shooting team and the strength state, including one trap with blocked shots.
  • How to compute CF%, FF%, SF%, GF%, xGF%, Sh%, Sv% and PDO for every team at 5v5.
  • How to compute on-ice Corsi for individual players from the on-ice skater columns.

If you are new to the package, start with Getting started with fastRhockey.

What are Corsi, Fenwick and PDO?

Metric Counts Share
Corsi (CF / CA) All shot attempts: shots on goal, goals, missed shots and blocked shots CF% = CF / (CF + CA)
Fenwick (FF / FA) Unblocked attempts: Corsi minus blocked shots FF%
Shots (SF / SA) Shots on goal, including goals SF%
Goals (GF / GA) Goals GF%
Expected goals (xGF / xGA) The sum of the xg of unblocked attempts xGF%

Shooting percentage (Sh%) is GF / SF. Save percentage (Sv%) is 1 - GA / SA. PDO is Sh% + Sv%. League-wide every goal is one team’s shooting and the other’s goaltending, so league PDO is exactly 1. A team far above 1 is usually riding hot shooting or goaltending that tends to fade.

Why 5v5? Power plays inflate one team’s attempts for reasons unrelated to how well it plays at even strength, and empty-net time distorts both sides. Analysts compare teams at 5-on-5, where both sides have five skaters and a goalie.

Which columns do I need? 

library(fastRhockey)
library(dplyr)
library(ggplot2)

pbp <- load_nhl_pbp(seasons = 2026)

pbp %>%
  filter(event_type %in% c("SHOT", "MISSED_SHOT", "BLOCKED_SHOT", "GOAL")) %>%
  count(event_type)
#> # A tibble: 4 × 2
#>   event_type       n
#>   <chr>        <int>
#> 1 BLOCKED_SHOT 43664
#> 2 GOAL          8836
#> 3 MISSED_SHOT  41811
#> 4 SHOT         69507
  • event_type: SHOT, MISSED_SHOT, BLOCKED_SHOT and GOAL are the four kinds of shot attempt.
  • event_team_abbr: the team credited with the event. For BLOCKED_SHOT it is the blocking team. The shooter is event_player_1_name and the blocker event_player_2_name. You must flip the team to get the shooting side.
  • strength_state: skaters on the ice as "<event team>v<opponent>". For 5v5 the order doesn’t matter.
  • season_type: "R" for the regular season, "P" for the playoffs.
  • period_type: drop "SHOOTOUT" rows; shootout attempts are not part of the game’s run of play.
  • xg: expected goals for unblocked attempts (see the NHL xG model article).
pbp %>%
  filter(event_type == "BLOCKED_SHOT") %>%
  select(event_team_abbr, event_player_1_type, event_player_1_name,
         event_player_2_type, event_player_2_name) %>%
  head(3)
#> # A tibble: 3 × 5
#>   event_team_abbr event_player_1_type event_player_1_name event_player_2_type
#>   <chr>           <chr>               <chr>               <chr>              
#> 1 CHI             Shooter             Brad Marchand       Blocker            
#> 2 CHI             Shooter             Mackie Samoskevich  Blocker            
#> 3 FLA             Shooter             Connor Murphy       Blocker            
#> # ℹ 1 more variable: event_player_2_name <chr>

How do I compute team Corsi and Fenwick at 5v5?

First, label each regular-season attempt with its shooting and defending team.

attempts <- pbp %>%
  filter(
    season_type == "R",
    period_type != "SHOOTOUT",
    event_type %in% c("SHOT", "MISSED_SHOT", "BLOCKED_SHOT", "GOAL")
  ) %>%
  mutate(
    shooting_team = if_else(
      event_type == "BLOCKED_SHOT",
      if_else(event_team_abbr == home_abbr, away_abbr, home_abbr),
      event_team_abbr
    ),
    defending_team = if_else(shooting_team == home_abbr, away_abbr, home_abbr)
  )

attempts_5v5 <- attempts %>% filter(strength_state == "5v5")

nrow(attempts_5v5) / nrow(attempts)
#> [1] 0.7918631

About 79% of regular-season attempts happen at 5v5.

Then count every attempt twice, once for the shooting team and once against the defending team, and summarise.

team_5v5 <- bind_rows(
  attempts_5v5 %>% mutate(team = shooting_team, side = "for"),
  attempts_5v5 %>% mutate(team = defending_team, side = "against")
) %>%
  group_by(team) %>%
  summarise(
    CF  = sum(side == "for"),
    CA  = sum(side == "against"),
    FF  = sum(side == "for" & event_type != "BLOCKED_SHOT"),
    FA  = sum(side == "against" & event_type != "BLOCKED_SHOT"),
    SF  = sum(side == "for" & event_type %in% c("SHOT", "GOAL")),
    SA  = sum(side == "against" & event_type %in% c("SHOT", "GOAL")),
    GF  = sum(side == "for" & event_type == "GOAL"),
    GA  = sum(side == "against" & event_type == "GOAL"),
    xGF = sum(xg[side == "for"], na.rm = TRUE),
    xGA = sum(xg[side == "against"], na.rm = TRUE),
    .groups = "drop"
  ) %>%
  mutate(
    CF_pct  = CF / (CF + CA),
    FF_pct  = FF / (FF + FA),
    SF_pct  = SF / (SF + SA),
    GF_pct  = GF / (GF + GA),
    xGF_pct = xGF / (xGF + xGA),
    Sh_pct  = GF / SF,
    Sv_pct  = 1 - GA / SA,
    PDO     = Sh_pct + Sv_pct
  ) %>%
  arrange(desc(CF_pct))

team_5v5 %>%
  select(team, CF, CA, CF_pct, FF_pct, xGF_pct, GF_pct, PDO) %>%
  mutate(across(ends_with("pct") | matches("PDO"), ~ round(.x, 3))) %>%
  head(10)
#> # A tibble: 10 × 8
#>    team     CF    CA CF_pct FF_pct xGF_pct GF_pct   PDO
#>    <chr> <int> <int>  <dbl>  <dbl>   <dbl>  <dbl> <dbl>
#>  1 CAR    4620  3112  0.598  0.589   0.564  0.536 0.984
#>  2 COL    4613  3523  0.567  0.559   0.576  0.629 1.02 
#>  3 TBL    3847  3377  0.533  0.524   0.539  0.563 1.02 
#>  4 UTA    4005  3532  0.531  0.529   0.528  0.532 1.00 
#>  5 OTT    3798  3387  0.529  0.537   0.549  0.522 0.992
#>  6 ANA    4184  3806  0.524  0.525   0.503  0.475 0.982
#>  7 VGK    3956  3604  0.523  0.528   0.546  0.503 0.983
#>  8 LAK    4117  3769  0.522  0.522   0.52   0.52  1.00 
#>  9 FLA    3758  3505  0.517  0.509   0.503  0.472 0.989
#> 10 NJD    4013  3800  0.514  0.513   0.49   0.426 0.971

A quick check: league-wide, goals for equal goals against, so PDO averages out to 1.

team_5v5 %>%
  summarise(league_sh = sum(GF) / sum(SF), league_sv = 1 - sum(GA) / sum(SA)) %>%
  mutate(league_pdo = league_sh + league_sv)
#> # A tibble: 1 × 3
#>   league_sh league_sv league_pdo
#>       <dbl>     <dbl>      <dbl>
#> 1    0.0952     0.905          1

Does shot share predict goal share? 

ggplot(team_5v5, aes(x = CF_pct, y = GF_pct)) +
  geom_hline(yintercept = 0.5, colour = "grey80") +
  geom_vline(xintercept = 0.5, colour = "grey80") +
  geom_smooth(method = "lm", se = FALSE, colour = "grey60", linewidth = 0.6) +
  geom_point(colour = "#531896", size = 2.5) +
  geom_text(aes(label = team), vjust = -0.8, size = 3) +
  scale_x_continuous(labels = scales::percent) +
  scale_y_continuous(labels = scales::percent) +
  labs(
    title = "5v5 Corsi share vs goal share, NHL 2025-26 regular season",
    x = "CF%", y = "GF%",
    caption = "Data: fastRhockey::load_nhl_pbp()"
  ) +
  theme_minimal()

Scatter plot of NHL team 5v5 Corsi for percentage against goals for percentage, 2025-26 regular season, with team labels.

team_5v5 %>%
  summarise(
    cor_CF_GF  = cor(CF_pct, GF_pct),
    cor_FF_GF  = cor(FF_pct, GF_pct),
    cor_xGF_GF = cor(xGF_pct, GF_pct)
  ) %>%
  round(3)
#> # A tibble: 1 × 3
#>   cor_CF_GF cor_FF_GF cor_xGF_GF
#>       <dbl>     <dbl>      <dbl>
#> 1     0.497      0.55      0.755

Here xGF% tracks goal share more closely than raw Corsi or Fenwick, because it weights attempts by danger. Teams far off the trend line are where PDO comes in.

Which teams are running hot or cold? 

team_5v5 %>%
  mutate(team = reorder(team, PDO)) %>%
  ggplot(aes(x = PDO - 1, y = team, fill = PDO > 1)) +
  geom_col(show.legend = FALSE) +
  scale_fill_manual(values = c(`TRUE` = "#531896", `FALSE` = "#A6E5F3")) +
  scale_x_continuous(labels = function(x) sprintf("%.3f", x + 1)) +
  labs(
    title = "5v5 PDO by team, NHL 2025-26 regular season",
    subtitle = "Shooting % + save %; league average is 1.000",
    x = "PDO", y = NULL
  ) +
  theme_minimal()

Bar chart of NHL team 5v5 PDO for 2025-26, centred on 1.

How do I compute on-ice Corsi for players?

A player’s on-ice Corsi counts every attempt by either team while that player was on the ice. fastRhockey NHL play-by-play carries the skaters on the ice for every event in home_on_1 … home_on_7 and away_on_1 … away_on_7, with player ids in the matching *_id columns. Goalies are in home_goalie and away_goalie, not in the skater slots.

slots <- 1:6

on_ice_long <- function(df, side) {
  name_cols <- paste0(side, "_on_", slots)
  id_cols <- paste0(name_cols, "_id")
  base <- df %>%
    mutate(row = row_number(), team = .data[[paste0(side, "_abbr")]]) %>%
    select(row, event_type, shooting_team, xg, team, all_of(name_cols), all_of(id_cols))
  names_long <- tidyr::pivot_longer(base, all_of(name_cols), values_to = "player") %>%
    select(row, event_type, shooting_team, xg, team, player)
  ids_long <- tidyr::pivot_longer(base, all_of(id_cols), values_to = "player_id") %>%
    select(player_id)
  bind_cols(names_long, ids_long) %>% filter(!is.na(player_id))
}

on_ice <- bind_rows(
  on_ice_long(attempts_5v5, "home"),
  on_ice_long(attempts_5v5, "away")
)

player_5v5 <- on_ice %>%
  group_by(player_id, player, team) %>%
  summarise(
    CF = sum(shooting_team == team),
    CA = sum(shooting_team != team),
    xGF = sum(xg[shooting_team == team], na.rm = TRUE),
    xGA = sum(xg[shooting_team != team], na.rm = TRUE),
    .groups = "drop"
  ) %>%
  mutate(CF_pct = CF / (CF + CA), xGF_pct = xGF / (xGF + xGA))

player_5v5 %>%
  filter(CF + CA >= 1500) %>%
  arrange(desc(CF_pct)) %>%
  mutate(across(c(CF_pct, xGF_pct), ~ round(.x, 3)), across(c(xGF, xGA), ~ round(.x, 1))) %>%
  head(10)
#> # A tibble: 10 × 9
#>    player_id player            team     CF    CA   xGF   xGA CF_pct xGF_pct
#>        <int> <chr>             <chr> <int> <int> <dbl> <dbl>  <dbl>   <dbl>
#>  1   8480817 K'Andre Miller    CAR    1597   988  73.8  55.6  0.618   0.57 
#>  2   8482702 Logan Stankoven   CAR    1300   816  67.7  46.7  0.614   0.591
#>  3   8478427 Sebastian Aho     CAR    1308   822  66.4  49.8  0.614   0.571
#>  4   8480336 Sean Walker       CAR    1754  1106  79.4  59.3  0.613   0.572
#>  5   8480830 Andrei Svechnikov CAR    1286   822  64.7  50.2  0.61    0.563
#>  6   8477940 Nikolaj Ehlers    CAR    1221   796  58.8  43.7  0.605   0.573
#>  7   8482809 Jackson Blake     CAR    1283   840  67.8  47.9  0.604   0.586
#>  8   8482093 Seth Jarvis       CAR    1099   723  53.6  42.2  0.603   0.559
#>  9   8481606 Jordan Spence     OTT    1377   907  67.9  42.8  0.603   0.613
#> 10   8475791 Taylor Hall       CAR    1174   787  59    44.3  0.599   0.571

Grouping by team as well as player_id keeps a traded player’s stints separate. The CF + CA >= 1500 filter keeps regulars; on-ice samples for depth players are small and noisy.

Rates per 60 minutes need time on ice. Play-by-play tells you who was on the ice for each event, not for how long. For CF/60 and similar rates, add 5v5 time on ice from load_nhl_shifts().

What about the PWHL?

The PWHL has a ready-made live function, pwhl_game_corsi(), that returns on-ice Corsi and Fenwick for one game from the HockeyTech feed, alongside pwhl_game_shifts() and pwhl_player_toi(). These call the live API, so they are not run here.

pwhl_game_corsi(game_id = 27)
pwhl_player_toi(game_id = 27)

For full PWHL seasons, load_pwhl_pbp() and load_pwhl_shifts() give you the inputs to build the same metrics yourself. The PWHL and college hockey guide has the details.

Caveats

  • Score effects. Teams that are trailing push for attempts and teams that are leading sit back, so raw Corsi flatters trailing teams. Analysts often adjust for score state and home ice; the home_score and away_score columns give you what you need.
  • Scorer bias. Missed and blocked shots are recorded by arena scorers, and some rinks record more than others.
  • xG coordinates. The xg column inherits a known NHL coordinate issue, described in the NHL xG model article.

Where to go next