Group vs Knockout Scoring at a Live World Cup, with groupby

Every World Cup produces the same claim around this stage: "the knockouts tighten up." It's a perfect first analytics question because it's checkable with one groupby — and because the answer, on real data, is messier and more interesting than the cliché. We'll test it on a genuine snapshot of the tournament happening right now: the first 96 completed matches of the 2026 World Cup, through the round of 16, bundled as wc2026_results.csv so everything runs offline.

This builds on Grouping, Pivoting and Reshaping — the general tool — and applies it to a single sharp question. The rhythm here (derive columns, group, aggregate, chart with sample sizes showing) is the same one behind every stage-split, season-split or home/away-split table you've ever read.

Go deeper with the free textbook: Introduction to Soccer Metrics at DataField.dev.

1. Load the matches and derive the columns you wish existed

   The CSV has one row per completed match: date, stage, the two teams, and the two scores. The three columns the question needs — total goals, did both teams score, was it goalless — don't exist yet, so make them. This derive-then-aggregate rhythm is most of real analysis.

   import pandas as pd
   
   df = pd.read_csv("wc2026_results.csv")
   df["total_goals"] = df["home_goals"] + df["away_goals"]
   df["both_scored"] = (df["home_goals"] > 0) & (df["away_goals"] > 0)
   df["goalless"]    = df["total_goals"] == 0

   One row per real match

         date stage          home               away  total_goals
   2026-06-11 group        Mexico       South Africa            2
   2026-06-12 group   South Korea            Czechia            3
   2026-06-12 group        Canada Bosnia-Herzegovina            2
   2026-06-13 group United States           Paraguay            5
   ...
   96 matches, stages: group, round of 32, round of 16

2. One groupby, five answers

   Group by stage and aggregate five ways at once with named aggregations: how many games, how many goals, goals per game, the share where both teams scored, and the goalless share. reindex puts the stages in tournament order instead of alphabetical — a tiny step that saves every reader of your table a double-take.

   stage_order = ["group", "round of 32", "round of 16"]
   by_stage = (
       df.groupby("stage")
         .agg(games=("total_goals", "size"),
              goals=("total_goals", "sum"),
              goals_per_game=("total_goals", "mean"),
              both_scored=("both_scored", "mean"),
              goalless=("goalless", "mean"))
         .reindex(stage_order)
   )

   The cliché, confronted

                games  goals  goals_per_game  both_scored  goalless
   stage                                                           
   group           72    215            2.99        0.542     0.097
   round of 32     16     42            2.62        0.562     0.000
   round of 16      8     23            2.88        0.500     0.125
   
   Knockouts pooled: 24 games, 2.71 goals/game (groups: 2.99)

   Read the middle column honestly. Scoring did dip when the knockouts began — 2.99 in the groups to 2.62 in the round of 32 — then bounced most of the way back (2.88) in the round of 16. Pooled, the knockouts sit about a quarter of a goal below the groups. That's a tightening, but a mild one: nothing like the strangled, goalless football the cliché promises.

3. Chart it, with the sample sizes on the bars

   Three numbers deserve a chart with a warning label: the stages have wildly different sample sizes (72, 16, and 8 games), so print n on every bar. A rate from 8 games is a reading, not a truth — the honest chart says so on its face.

   fig, ax = plt.subplots(figsize=(8.6, 5.4))
   bars = ax.bar(by_stage.index, by_stage["goals_per_game"])
   for b, (_, row) in zip(bars, by_stage.iterrows()):
       ax.annotate("n=%d games" % row["games"],
                   (b.get_x() + b.get_width()/2, 0.12), ha="center")
   ax.axhline(df["total_goals"].mean(), ls="--")   # tournament average

   

4. Interrogate the cliché's strongest prediction: 0-0s

   If knockouts really strangle football, goalless games should pile up there — teams playing for penalties. Count them by stage.

   zeros = df[df["goalless"]]
   print(zeros.groupby("stage").size())

   Where the 0-0s actually live

   Goalless games: 8 of 96 (8%)
   stage
   group          7
   round of 32    0
   round of 16    1
   
   The cliche says knockouts breed 0-0s. This snapshot: 7 in 72 group
   games, 1 in 24 knockout games - the knockouts have been DECISIVE, not cagey.

   Seven of the tournament's eight goalless games happened in the group stage. The knockouts went 23 straight games without one before the round of 16's final tie broke the streak. On this snapshot, elimination football has been decisive, not cagey — the opposite of the cliché's strongest claim.

Troubleshooting