Assessing the predictability of racing performance of Thoroughbreds using mixed-effects model

Area of Science:

• Animal Genetics
• Quantitative Genetics
• Equine Science

Background:

• The inheritance of racing performance in Thoroughbred horses is a key interest for breeders and geneticists.
• While genetic parameters have been studied, the predictive accuracy of racehorse performance remains inadequately assessed.
• Understanding performance predictability is vital for selective breeding and performance evaluation.

Purpose of the Study:

• To develop and evaluate mixed-effects models for predicting racing performance in Japanese Thoroughbreds.
• To assess the predictability of racing performance using average velocity as the primary index.
• To identify key factors influencing racing performance and their impact on predictive accuracy.

Main Methods:

• Development of mixed-effects models incorporating race, age, and jockey effects for Japanese Thoroughbreds.
• Utilizing average velocity as the racing performance index, with racecourse and distance treated as distinct traits.
• Model selection via the deviance information criterion to identify significant explanatory variables.
• Prediction of phenotypic values and identification of graded race winners using breeding values from partial datasets.

Main Results:

• Model selection identified race, age, and jockey effects as significant predictors of racing performance.
• Correlation coefficients for predicting phenotypic values ranged from 0.000 to 0.235 (average 0.084 ± 0.066), with higher values for longer distances.
• Area under the curve values for predicting graded race winners ranged from 0.516 to 0.776 (average 0.613 ± 0.073), also showing higher accuracy for longer distances.

Conclusions:

• The study demonstrates a degree of predictability in Thoroughbred racing performance using mixed-effects models.
• Predictive accuracy is influenced by factors such as distance, with longer distances yielding higher correlations.
• Further research is recommended to enhance predictive models by exploring alternative performance indices and refining statistical methodologies.