How Survival Analysis Reveals Player Churn in Naraka: Bladepoint

At GDC 2022, Lei Huo UX data analyst Bo Yue delivered a talk titled “All Roads Lead to Rome: Analyzing Churn in Naraka: Bladepoint Using Interdisciplinary Methods,” providing a full transcript of the presentation.

About Naraka: Bladepoint – a 2021‑released battle‑royale game by NetEase’s 24 Entertainment where 60 players fight on an island, collecting equipment to survive until the end.

Why churn analysis matters – excessive matchmaking wait times increase churn, and high churn can cause streamers to abandon the game, further accelerating player loss.

Types of churn – (1) Systemic churn: dissatisfaction with core game mechanics; (2) Event‑driven churn: reactions to recent game changes; (3) Trend‑driven churn, which is harder to address and is not the primary focus.

1. Interdisciplinary Methods

Survival analysis (biostatistics) – suitable for time‑to‑event data, offering richer insights than binary classification and better interpretability compared to black‑box machine‑learning models.

Epidemiology experimental design – randomized controlled trials (RCTs) to compare intervention vs. control groups, cohort studies for observational tracking, and case‑control studies to examine outcomes after they occur.

Econometrics causal inference – Propensity Score Matching (PSM) to mimic randomization in non‑experimental data, and Difference‑in‑Differences (DID) to estimate causal effects by comparing pre‑ and post‑intervention differences between groups.

2. Applying the Methods to Churn Analysis

Using Kaplan‑Meier (KM) estimation, retention curves were plotted for players on SSD, HDD, and unknown storage devices. The SSD group showed significantly higher retention, confirmed by a Log‑rank test with a negligible p‑value.

PSM was employed to compare churn among players preferring different heroes while controlling for other features (games played, weapon usage, mode preference, device, version, etc.). This enabled a fair assessment of whether hero preference influences retention.

3. Conclusion

Behavioral data should be leveraged to understand player preferences, detect anomalies early, and address systemic and event‑driven churn. Combining causal inference with AB testing and rigorous experimental design yields more scientific insights, even though perfect experiments are challenging.