Determine the basis and applicability of conventional ΔBIC significance thresholds

Determine the derivation and validity of the conventional ΔBIC rule-of-thumb thresholds for assessing model-comparison evidence, particularly whether those thresholds remain meaningful when ΔBIC values are hundreds or thousands and the dataset contains thousands of spectroscopic measurements.

Background

The paper evaluates ammonia and hydrogen sulfide abundance perturbations using differences in the Bayesian Information Criterion (ΔBIC). Conventional guidance classifies ΔBIC values from 0–2 as weak evidence, 2–6 as moderate evidence, 6–10 as strong evidence, and values above 10 as very strong evidence. However, the analysis obtains ΔBIC values ranging from hundreds to thousands, far outside the range in which these rules of thumb are ordinarily discussed. The authors explicitly question the origin of the conventional thresholds and, consequently, their legitimacy for interpreting the molecular-abundance comparisons in this high-dimensional spectroscopic setting.

References

A common rule of thumb given to evaluate the significance of $\Delta$BIC is that values from 0 to 2 are weak evidence, 2 to 6 is moderate evidence, 6 to 10 is strong evidence, and that greater than 10 is very strong evidence (See page 26 of ), however it is not entirely clear where exactly these thresholds are derived from.

— The Dyn-Atmo Survey: JWST/NIRSpec spectroscopy of dynamical benchmark GJ 758 B  (2609.11828 - Madurowicz et al., 10 Sep 2026) in Section 5, subsection “Trace molecules”