Competing statistical methods for the fitting of normal species sensitivity distributions: recommendations for...

Area of Science:

• Ecotoxicology
• Environmental Risk Assessment
• Statistical Modeling

Background:

• Species sensitivity distributions (SSDs) are crucial for ecotoxicological risk assessment, modeling chemical toxicity across species.
• Parametric assumptions, like the log-normal distribution, are common but criticized.
• New statistical methods for fitting SSDs have emerged, independent of distributional assumptions.

Purpose of the Study:

• To analyze and compare different statistical methods for fitting SSDs to toxicity data.
• To evaluate estimators for determining the hazardous concentration to a specified percentage of species.
• To identify the most statistically defensible and practical estimator for risk assessment.

Main Methods:

• Analysis of two regression-based estimators using log-transformed toxicity data and probit-transformed plotting positions.
• Comparison with a confidence limit-based estimator.
• Large-scale simulation study to assess estimator performance.

Main Results:

• Regression-based estimators were analyzed.
• The confidence limit-based estimator was found to be more intuitive and statistically defensible.
• Simulation results indicated superior performance of the confidence limit-based estimator.

Conclusions:

• The confidence limit-based estimator is recommended for typical assessments requiring a pointwise hazardous concentration value.
• This estimator offers improved statistical rigor compared to certain regression-based approaches.
• The findings advocate for the adoption of more robust statistical methods in SSD modeling for environmental risk assessment.