📝 SummarXiv

Abstract

Quantitative assessment of extinction risk requires not only point estimates but also confidence intervals (CIs) that remain informative with limited data. Their reliability has been debated, as short observation spans can inflate uncertainty and reduce usefulness. I derive new CIs for extinction probability $G$ under the Wiener process with drift, a canonical model of population viability analysis. The method uses correlated noncentral-$t$ distributions for the transformed statistics $\widehat{w}$ and $\widehat{z}$, derived from drift and variance estimators, and constructs CIs of the extinction probability by exploiting the geometric properties of $G(\widehat{w},\widehat{z})$ in parameter space. Monte Carlo experiments show that the proposed intervals attain nominal coverage with narrower widths than common approximate methods, including the delta method, moment-based approaches, and bootstrap. A key result is that even with short time series, extinction probabilities that are very small or very large can be estimated reliably. This resolves a long-standing concern that population viability analysis fails under data scarcity. Applied to three 64-year catch series for Japanese eel (Anguilla japonica), the analysis indicates extinction risk well below the IUCN Criterion E thresholds for Critically Endangered and Endangered, with narrow CIs. These findings demonstrate that extinction-risk CIs can be both statistically rigorous and practical for Red List evaluations, even when data are limited.