Explain the smaller estimated rating-noise σ0 in Chess 960 versus standard chess

Determine the causes of the observed smaller estimated measurement-error standard deviation (σ0=34) for Glicko‑2 rating differences in Chess 960 compared to standard chess (σ0≈58) under the SIMEX calibration, assessing whether this discrepancy arises from changes in rating computation, the broader time window of data, variant-specific rating dynamics, or other dataset artifacts.

Background

The authors calibrate attenuation from measurement error in Glicko‑2 ratings via SIMEX and estimate σ0 values per variant. For standard chess they estimate σ0≈58, while for Chess 960 they estimate σ0=34.

They explicitly state uncertainty about why Chess 960’s σ0 is much smaller, suggesting potential explanations such as rating-system changes over time or the longer time window used for Chess 960 data, and call for investigation.