Reading Your Match History: Center and Spread
Module 3 · Descriptive Statistics

Reading Your Match History: Center and Spread

Summarize a season of games with mean, median, and mode, measure consistency with variance, standard deviation, IQR, and coefficient of variation, and see how one stomp game can distort an average.

A Season Is a Dataset

Your match history is a dataset: each game is an observation, and each stat (gold per minute, KDA, damage, vision score, game length) is a variable. Descriptive statistics compresses dozens of games into a handful of numbers that answer two questions:

  1. Center: what’s a typical game for you?
  2. Spread: how much do your games vary?

Types of Variables

TypeExamplesSummaries that make sense
ContinuousGold per minute, damage dealt, game lengthMean, median, SD, histogram
Discrete countKills, deaths, towers destroyedMean, median, mode, bar chart
CategoricalRole, champion, win/lossCounts, proportions, mode

You can average GPM. You can’t average “Mid lane” and “Jungle.” Picking the right summary starts with knowing the variable type.


Measures of Center

Here are 20 games of gold per minute (GPM) for one mid laner:

412 455 388 501 437 466 420 398 445 520
431 455 470 405 455 489 362 441 478 430
  • Mean (arithmetic average): add them up and divide by n. 8,858 / 20 = 442.9
  • Median: sort the values and take the middle. With 20 values, average the 10th and 11th: (441 + 445) / 2 = 443
  • Mode: the most frequent value. 455 appears three times.

Here the mean and median agree closely, which suggests the data is roughly symmetric.

When the Mean Lies

Now add one game where you stomped lane and finished at 780 GPM:

  • New mean: 9,638 / 21 ≈ 459.0 (up 16 points)
  • New median: 445 (up 2 points)

One unusual game moved the mean eight times as much as the median. The mean uses every value’s size; the median only uses their order. That makes the median robust to outliers. When a distribution is skewed (a long tail on one side), report the median, or report both.


Measures of Spread

Two players can have the same average and very different games. Consider KDA over 15 games:

Mean KDAMedian KDAStandard deviation
Player A (steady)3.073.10.29
Player B (coin-flip)3.072.42.27

Same mean, but Player B swings between 0.5 and 7.3. Spread is the statistic that tells them apart.

Variance and Standard Deviation

Variance measures the average squared distance from the mean:

s² = Σ (xᵢ − x̄)² / (n − 1)

Standard deviation is its square root, s = √s², which puts it back in the original units (GPM, KDA). For the 20-game GPM sample, s ≈ 39.2 GPM.

Why divide by n − 1 instead of n? Because we measured distances from the sample mean, which sits closer to the data than the true long-run mean does. Dividing by n − 1 (Bessel’s correction) corrects that bias when you treat your games as a sample of how you play in general.

Rules of Thumb for SD

For roughly bell-shaped data, about 68% of values fall within 1 SD of the mean and about 95% within 2 SD. For our GPM data, that predicts most games between 443 ± 39, about 404 to 482. Counting directly, 14 of 20 games (70%) land in that range.

Interquartile Range (IQR)

The first quartile (Q1) is the value 25% of the way through the sorted data; Q3 is 75% of the way. IQR = Q3 − Q1 covers the middle half of your games. For the GPM data, Q1 = 418 and Q3 = 467, so IQR = 49. Like the median, the IQR ignores extreme values, which makes it the robust partner to the SD.

A common outlier rule flags values below Q1 − 1.5 · IQR or above Q3 + 1.5 · IQR. Here that range is roughly 345 to 541, so the 780 GPM stomp is flagged as an outlier.

Coefficient of Variation

To compare spreads on different scales, divide SD by the mean: CV = s / x̄. Player A’s KDA has CV ≈ 9%; Player B’s is ≈ 74%. CV lets you ask “is my GPM more or less consistent than my KDA?” even though the units differ.

z-Scores

How unusual was a single game? Convert it to a z-score:

z = (x − x̄) / s

The 520 GPM game: (520 − 442.9) / 39.2 ≈ 1.97, almost 2 standard deviations above your average. z-scores will come back in Module 4 as the basis of hypothesis testing.

🎮 Rift Lab: Match History Explorer

Pick a dataset or paste your own numbers (comma or space separated). The histogram marks the mean, the median, and a ±1 standard deviation band.

Games (n)—
Mean—
Median—
Mode—
Std. deviation—
IQR (Q1–Q3)—
Min – Max—
Coeff. of variation—
Distribution
Count per binMeanMedian±1 SD
KDA spread: same average, different player
Player APlayer BBars show mean ± 1 SD

Reading a Histogram

A histogram groups values into bins and shows how many fall in each. Look for:

  • Shape: symmetric, skewed right (long tail of high values), skewed left, or bimodal (two peaks)
  • Center: where the bulk of the data sits
  • Spread: how wide the bulk is
  • Outliers: isolated bars far from the rest

A bimodal GPM histogram is a clue that two different things are going on, for example games on a farming champion vs. games on a roaming one. When you see two peaks, split the data before summarizing it.

⚠️ Common Pitfall: Averaging Ratios Like KDA

Common mistake: averaging per-game KDA values. KDA is (Kills + Assists) / Deaths, and a 10/0/8 game is usually reported as 18 (dividing by 1 instead of 0), which can single-handedly drag the mean upward. A more stable season KDA divides total kills plus assists by total deaths. The same issue shows up whenever you average ratios, like click-through rates or batting averages: decide whether you want the average of ratios or the ratio of totals, and say which one you used.

🧠 Knowledge Check

1. Your GPM is skewed right because of a few stomp games. Which statistic best describes a typical game?

2. Two players both average 3.0 KDA. Player A's SD is 0.3, Player B's is 2.3. What can you conclude?

3. Your mean damage is 18,000 with SD 3,000. A game with 24,000 damage has what z-score?