Master’s thesis · MSc in progress
A champion-level evaluation of balance design across rank, role, and match phase
Overall win rate can hide whether a balance change met its stated goal. I took twelve changes from Riot’s notes and asked what each one should look like in a match. Then I compared patches 25.20 and 25.21 by rank, role and match phase. A change aimed at Jungle, for instance, needs a measure that shows what happens in Jungle.
A champion’s win rate is only one outcome. I first identify what the patch says it is trying to change, then choose observable measures that match that intention, such as jungle clear output or objective damage.
I compare those measures across the two patches within rank, role and match-phase groups. The classification follows a descriptive procedure. It does not estimate a causal patch effect or establish statistical significance.
I collected 75,833 unique matches and retained 68,372 under the 20-minute duration rule. Performance analysis then excludes Gwen and observations without a recognised rank, leaving 2,885 champion appearances in 2,486 matches. Accounts and matches can appear more than once among those observations.
Diana’s stated intention concerned jungle efficiency in Iron through Diamond II. The earlier exploratory figure below compares neutral minions killed at 10 and 30 minutes and end-of-match objective damage, split by rank. The neutral-minion results are mixed, while objective damage is lower. The choice of measure and checkpoint matters to the interpretation.
Six of the twelve changes met my criteria for matching Riot’s stated intention. One met them in part and five missed. That describes what I found in the retained matches across these two patches.
Context changed the interpretation. Qiyana’s damage rose in the targeted Jungle groups and fell in Middle, a difference an overall rate could not show. My main takeaway is to specify the expected outcome, target group and measurement time before analysing the results.
This is one patch transition with unequal sample sizes and changing role composition. Rank was recorded during collection, which may differ from rank at match time. Some checkpoints were selected after inspecting the data.
Available match and timeline fields only approximate parts of the design intention. Other changes between patches can affect the same metrics. An alignment label is therefore a description under this procedure, not a verdict on whether a balance decision worked.
My MSc is in progress. The counts and classifications here follow the October manuscript. The retained exploratory figure is labelled separately.
More research and background