REGRESSION ANALYSIS OF NATURAL SELECTION: STATISTICAL INFERENCE AND BIOLOGICAL INTERPRETATION

Area of Science:

• Quantitative genetics
• Evolutionary biology
• Ecological research

Background:

• Recent theoretical advances highlight the need to measure natural selection in wild populations.
• The Lande and Arnold (1983) multiple-regression approach is a key tool for this measurement.
• Application of this method faces statistical and biological challenges.

Purpose of the Study:

• To discuss statistical and biological issues in applying multiple-regression methods to measure natural selection.
• To identify common problems, such as multicollinearity and unmeasured environmental factors, that complicate selection inference.
• To propose solutions, including jackknife tests for significance and manipulative experiments for causal inference.

Main Methods:

• Review of statistical assumptions in regression analysis for quantitative genetics.
• Analysis of multicollinearity and extrinsic factors impacting fitness estimation.
• Application of jackknife tests for significance when normality assumptions are violated.

Main Results:

• Violation of regression assumptions, particularly multicollinearity and unmeasured environmental variables, can seriously impede inference on natural selection.
• Non-normality of residuals invalidates standard significance tests.
• Jackknife tests offer a viable alternative for significance testing in such cases.

Conclusions:

• While fitness regression aids in predicting selection response, manipulative experiments are superior for inferring causal relationships between traits and fitness.
• Careful consideration of statistical assumptions and potential confounding factors is crucial for accurate measurement of natural selection.
• Alternative definitions of stabilizing and disruptive selection are also discussed.