Data transformation and model selection in bivariate allometry

Area of Science:

• Biological sciences
• Ecology
• Evolutionary biology

Background:

• Biological allometry traditionally uses logarithmic transformations to linearize data.
• This method assumes log-linearity, which is often not met in contemporary studies.
• Back-transformation of linear models can lead to inaccurate power functions.

Purpose of the Study:

• To propose an alternative method for bivariate allometry analysis.
• To address limitations of logarithmic transformations in allometric studies.
• To improve the accuracy of describing biological scaling patterns.

Main Methods:

• Fit multiple nonlinear regression models to untransformed data.
• Utilize nonlinear regression to avoid data transformation.
• Employ maximum likelihood-based model selection procedures.

Main Results:

• Nonlinear regression on untransformed data provides a more robust analysis.
• This approach accommodates diverse functional forms and error structures.
• Avoids potential misinterpretations from back-transformed power functions.

Conclusions:

• Foregoing logarithmic transformations and using nonlinear regression is superior for bivariate allometry.
• Newer statistical methods offer greater power and versatility in studying allometric variation.
• Direct analysis of original data is recommended whenever feasible.