Log-normal variation belts for growth curves.

Area of Science:

• Biology
• Statistics
• Bioinformatics

Background:

• Prediction and tolerance belts combine sample uncertainty with individual variation estimates.
• Variation belts offer a more direct representation of individual variation by substituting population parameters with sample estimates.
• Variation belts can visually assess the fit of error models.

Purpose of the Study:

• To address limitations of existing methods for fitting multiplicative error models, particularly in biological growth studies.
• To introduce and validate a new method, constrained iteratively reweighted multiplicative least squares (CIRMLS), for improved model fitting.
• To demonstrate the utility of variation belts in evaluating error models for biological data.

Main Methods:

• Comparison of prediction belts and variation belts for representing biological variation.
• Application of iteratively reweighted multiplicative least squares (IRMLS) for heteroscedastic multiplicative error models.
• Development and implementation of constrained iteratively reweighted multiplicative least squares (CIRMLS) to overcome IRMLS limitations.

Main Results:

• Standard multiplicative least-squares (MLS) methods are inadequate for heteroscedastic data.
• IRMLS can yield unacceptable estimates such as negative residual variance.
• CIRMLS effectively prevents issues like negative variance estimates and excessively wide variation belts, ensuring more reliable model fitting.
• Successful application of CIRMLS to diverse biological datasets including metabolic allometry, somatic growth, and population growth.

Conclusions:

• Variation belts provide a superior graphical tool for assessing error model fit in biological studies.
• CIRMLS is a robust and reliable method for fitting heteroscedastic multiplicative error models in biological growth analysis.
• The presented method enhances the accuracy and interpretability of analyses involving biological variation and growth patterns.