Maximum likelihood outperforms binning methods for detecting differences in abundance size spectra across...

Area of Science:

• Ecology
• Quantitative Biology
• Statistical Modeling

Background:

• Individual body size distributions (ISD) are consistent across scales and described by power-law size spectra.
• Estimating the size spectra exponent (λ) is key for ecological studies, especially for detecting anthropogenic impacts.
• Traditional methods often use data binning and ordinary least squares regression, but may introduce bias.

Purpose of the Study:

• To compare the accuracy of maximum likelihood estimation (MLE) against two binning methods for estimating size spectra exponents (λ).
• To evaluate how estimation method bias affects the detection of changes in λ across environmental gradients.
• To assess the impact of estimation methods on the reliability of ecological findings.

Main Methods:

• Simulations were used to compare MLE and two normalized binning methods (equal logarithmic and log2 bins).
• Methods were tested for their ability to recapture known λ values and regression parameters across simulated gradients.
• The methods were also applied to two real-world datasets examining body size variation along temperature and pollution gradients.

Main Results:

• MLE consistently outperformed binning methods, showing less bias in λ estimation.
• Bias in binning methods propagated to regression analyses, reducing accuracy.
• MLE yielded significantly lower variance in estimates compared to binning methods.
• Binning-induced errors can be comparable in magnitude to previously reported ecological effect sizes.

Conclusions:

• Maximum likelihood estimation (MLE) is a more reliable method for estimating size spectra exponents (λ) than traditional binning.
• Using MLE enhances the accuracy of detecting ecological changes across environmental gradients.
• The findings question the effect sizes of previous studies relying on binning methods, advocating for MLE adoption.