Bias in genetic variance estimates due to spatial autocorrelation

Area of Science:

• Quantitative genetics
• Agricultural science
• Statistical modeling

Background:

• Distinguishing genetic and environmental effects is crucial in genetic field trials.
• Phenotypic observations in field trials can exhibit spatial correlations, often modeled by autoregressive processes.
• Multi-unit plot designs can exacerbate biases in variance component estimation.

Purpose of the Study:

• To investigate the upward bias in family variance component estimates in genetic field trials under an autoregressive (AR1) process.
• To evaluate the effectiveness of nearest neighbor (NN) adjustment procedures in mitigating this bias.

Main Methods:

• Simulated 768,000 family trials using complete randomized block designs.
• Incorporated a first-order autoregressive (AR1) process to model phenotypic correlations.
• Tested modified Papadakis nearest neighbor (NN) adjustment procedures.

Main Results:

• AR1 processes caused significant upward bias in family variances, additive genetic variance, and narrow-sense heritabilities, increasing exponentially with nearest neighbor correlation (ϱ).
• At ϱ = 0.2, family variance inflation ranged from 48-73%, independent of heritability levels.
• NN-adjustments based on Mead's coefficient and Bartlett's scheme removed up to 97% of the bias, with a slight (5-8%) increase in relative errors.

Conclusions:

• Spatial correlation in phenotypic data from plot designs significantly biases genetic variance estimates in field trials.
• Nearest neighbor adjustment methods are effective in correcting bias caused by AR1 processes.
• While NN adjustments reduce bias, they introduce a minor increase in relative error.