Methodology for Community Statistics
Data Coverage Disclaimer
The statistics presented on this website are based exclusively on match data recorded by players who are registered on the site and have voluntarily provided an API key. As a result, only matches that include at least one tracked player can be observed by the system.
Because participation is voluntary, the dataset does not represent the complete population of all matches played in the game. Instead, it reflects only the subset of matches that can be observed through registered users. Consequently, the statistics should be interpreted as describing the observed dataset rather than the entire game population.
This limitation introduces several potential sources of bias, including but not limited to:
- Selection bias: Players who choose to register and provide an API key may differ systematically from the general player population (for example in skill level, engagement, or competitive interest).
- Regional bias: Participation rates may differ between regions, which can affect the regional distribution of recorded matches.
- Rating distribution bias: Certain rating brackets may be overrepresented or underrepresented depending on which players contribute data.
- Partial match observation: Not every player in a match may be tracked, which means some matches are only partially observed.
Several statistical methods on this site attempt to mitigate these issues (for example through match weighting based on observation coverage). However, these adjustments cannot fully eliminate the underlying limitations of incomplete data.
For this reason, all statistics on this site should be interpreted as descriptive insights into the recorded dataset. They are useful for identifying patterns within the observed matches, but they cannot be assumed to perfectly represent the full player population or all matches played in the game.
OCWR (Raw)
Observed Class Win Rate (Raw) measures the percentage of matches won by a profession.
For each match, only one entry per account and match is considered to prevent duplicate observations. Matches are then grouped by profession, region and rating bracket.
The win rate is calculated as:
\(\text{OCWR} = \frac{\text{Wins}}{\text{Total Games}}\)
All matches contribute equally to the result. This makes the metric easy to interpret, but matches with incomplete observations are treated the same as fully observed matches.
Interpretation
- Values above 50% indicate that the profession wins more games than it loses.
- Values below 50% indicate the opposite.
- Results may be slightly biased if match coverage varies.
OCWR (Weighted)
Weighted Observed Class Win Rate adjusts win rates based on how many players in a match were observed.
Matches where more players are tracked provide more reliable information. Each match is therefore weighted according to the fraction of recorded players relative to a full 10-player match.
Weight per match:
\(w = \frac{\text{observed players}}{10}\)
The weighted win rate is calculated as:
\(\text{Weighted OCWR} = \frac{\sum (w \times \text{win})}{\sum w}\)
Matches flagged as irregular are excluded from this calculation. Irregular matches are games that are not counted as a win or loss because they were won or lost with an uneven number of players on the teams.
Interpretation
- Matches with more recorded participants influence the statistic more strongly.
- This reduces bias from partially recorded matches.
- The metric is generally more reliable than the raw win rate when dataset coverage varies.
Confidence Intervals
A 95% confidence interval is computed for each OCWR using the Wilson method.
This method provides a more accurate estimate for proportions, especially when the number of observations is small or the win rate is near 0 or 1.
The interval is calculated from the weighted effective number of observations n_eff and the observed win rate p_hat using:
CI = Wilson(p_hat, n_eff, z=1.96)
The resulting interval indicates the range in which the true win rate of the observed population would fall with 95% probability, reflecting the uncertainty due to limited or unevenly weighted observations.
Global Statistics
These metrics summarize general participation and activity across the season.
- Unique Active Players – Number of distinct accounts that played at least one ranked match.
- Total Games Played – Total number of recorded matches.
- Average Matches per Player – Mean number of matches played per active account.
- Median Matches per Player – The middle value of matches played when all players are ordered by activity.
- Region Distribution (Players) – Percentage of accounts associated with each region.
- Region Distribution (Games) – Percentage of matches associated with each region.
- Average Game Duration – Mean duration of matches based on start and end timestamps.
- Total Time Played – Total cumulative time spent in ranked matches.
These statistics help describe the overall participation level, activity intensity, and geographic distribution of the recorded dataset.
Rating Distribution
Matches are grouped into rating brackets based on the player's recorded rating at the time of the match.
- Legendary (≥ 1800)
- Platinum (1500–1799)
- Gold (1200–1499)
- Silver (900–1199)
- Bronze (< 900)
Each unique account–match combination contributes one observation.
The statistic shows the percentage of matches played within each rating bracket.
Interpretation
- Reflects the overall skill distribution of recorded matches.
- Higher rating brackets usually contain fewer matches due to the smaller player population.
Player Pick Rate
Player Pick Rate measures how frequently each profession is played.
Each player appearance in a match counts as one observation. The pick rate is calculated as the share of all player-game entries represented by each profession.
\(\text{Pick Rate} = \frac{\text{Player Games with Profession}} {\text{Total Player Games}} \times 100\)
Interpretation
- Higher values indicate professions that are played more often.
- Pick rate reflects popularity, not strength.