- Random regression models are important statistical tools in animal breeding and quantitative genetics. They are used to study genetic and environmental changes in traits that vary over time, age, or other continuous factors. Unlike traditional statistical models that estimate a single breeding value or genetic effect for an animal, random regression models can describe how these effects change throughout an animal’s life. This makes them particularly useful for evaluating traits such as milk production across lactation, body weight during growth, feed intake at different ages, and reproductive performance over time. By providing a more detailed understanding of genetic variation, random regression models help breeders make more accurate selection decisions and improve the long-term productivity of livestock populations.
- In animal breeding, many economically important traits are not measured only once. Dairy cows, for example, produce milk repeatedly throughout lactation, and their daily milk yields change as lactation progresses. Similarly, the body weight of cattle, sheep, pigs, and other livestock changes as they grow. These repeated measurements contain valuable information about an animal’s genetic potential, but they also create statistical challenges. Measurements collected from the same animal are usually correlated because they share common genetic and environmental influences. A random regression model accounts for this relationship while allowing genetic and environmental effects to change according to the age or stage at which each measurement is recorded.
- Traditional animal evaluation models often use a single trait value, such as total milk yield during lactation or body weight at a particular age. Although these approaches can be effective, they may not fully describe the biological pattern of a trait over time. For example, two dairy cows may produce the same total amount of milk during lactation, yet one may produce more milk early in lactation while the other maintains production more consistently. These differences can have implications for feed requirements, health, fertility, and management. Random regression models can capture such differences by estimating individual production curves and describing how animals vary genetically throughout the period being studied.
- The basic principle of a random regression model is to represent the relationship between a trait and a continuous variable, such as age or time, using a mathematical function. The model includes fixed regression coefficients that describe the average pattern of the trait in the population and random regression coefficients that describe how individual animals differ from that average. These random coefficients can be associated with genetic effects, permanent environmental effects, or other sources of variation. By combining these components, the model estimates how an animal’s performance changes across the observed period and how much of that change is attributable to genetic differences.
- A simplified random regression model can be expressed as:
- yit=xit’+k=0pk(t)aik+k=0qk(t)pik+eit
- In this equation, yit represents the observed performance of animal i at time t. The term xit’ represents fixed effects, such as herd, year, season, parity, or other systematic influences. The functions k(t) describe the selected regression basis across time. The coefficients aik represent the animal’s random genetic regression coefficients, while pik represent random permanent environmental regression coefficients. Finally, eit is the residual error associated with the observation. The precise structure of a model depends on the trait, the data available, and the objectives of the genetic evaluation.
- One of the most important features of random regression models is their ability to estimate genetic effects across an entire trajectory rather than at a single point. The genetic effect of an animal can be represented as a curve, with its shape determined by the animal’s random regression coefficients. These coefficients describe characteristics such as the overall genetic level, the rate of change, and the curvature of the trait over time. For example, in dairy cattle, an animal may have a high genetic potential for milk production early in lactation but a less persistent production pattern later. Another animal may have a lower initial production level but maintain its yield more consistently. Random regression models allow these genetic patterns to be evaluated separately.
- The choice of regression function is an important part of model development. Polynomial functions are commonly used because they can describe smooth changes over time. Legendre polynomials are especially popular in animal breeding because they provide a flexible and mathematically convenient way to represent trajectories over a standardized range. Other functions, including splines and alternative basis functions, may also be used depending on the shape of the trait and the required flexibility. A model with too few regression coefficients may fail to represent important changes in the trait, whereas a model that is excessively complex may estimate unnecessary fluctuations and increase computational demands. Therefore, selecting an appropriate function and polynomial order is essential for reliable genetic evaluation.
- Random regression models also provide a framework for estimating genetic and environmental covariance structures. The genetic covariance between two measurements depends on the time points at which they were recorded and on the covariance matrix of the genetic regression coefficients. This allows the model to estimate genetic variances at different ages or stages, as well as genetic correlations between measurements taken at different times. For example, the genetic correlation between milk yield early and late in lactation may differ from the correlation between milk yield at two closely spaced stages. Such information helps researchers understand whether the same genes influence performance consistently throughout a production period or whether genetic rankings change as animals develop.
- Heritability can also vary over time, and random regression models can be used to estimate this variation. Heritability is the proportion of phenotypic variance attributable to additive genetic variance under a specified population and set of conditions. In a longitudinal trait, both genetic and phenotypic variances may change with age or stage of production. As a result, heritability estimated at one point may not accurately represent the genetic contribution at another point. Random regression models allow researchers to calculate age-specific or time-specific heritabilities and investigate when genetic differences are most important for the trait being studied.
- Another important component is the permanent environmental effect. Repeated records from the same animal are influenced not only by its genes but also by environmental conditions that persist across measurements. These may include early-life nutrition, long-term housing conditions, udder development, and other lasting influences. If these effects are not appropriately modeled, some environmental variation may be incorrectly attributed to genetic differences. Random regression models can represent permanent environmental effects as trajectories, helping separate persistent non-genetic influences from inherited genetic effects. This distinction is especially important when animals have many repeated records.
- Dairy cattle breeding is one of the best-known applications of random regression models. Milk production records collected throughout lactation can be analyzed to estimate individual lactation curves, genetic differences in production, and the persistence of milk yield. These models can use test-day records instead of relying exclusively on total lactation yield. Test-day evaluations can account for changes in production throughout lactation and make better use of records from cows with incomplete lactations. They can also support more detailed genetic evaluations of traits associated with milk composition, such as fat and protein yield, when the model is designed to accommodate those traits.
- Random regression models are also useful in beef cattle, sheep, and other livestock species where growth changes continuously with age. Body weight measurements collected at birth, weaning, yearling age, and later stages can be used to estimate genetic growth trajectories. Some animals may grow rapidly at an early age, while others may show greater genetic potential for later growth. A model that evaluates the complete growth trajectory can help breeders identify animals that match specific production objectives. For example, a breeding program may aim to improve early growth while avoiding excessive mature body size because larger animals can require more maintenance feed. Random regression models help investigate these trade-offs by describing genetic effects at different ages.
- In pig and poultry breeding, repeated measurements of body weight, feed intake, growth rate, or other performance traits can also be analyzed with random regression approaches. When records are collected at multiple ages, the model can identify changes in genetic variance and correlations over the growth period. This is useful for designing selection programs that balance rapid growth, efficient feed utilization, appropriate mature size, and animal health. The usefulness of the model depends on the quality and frequency of the measurements, as well as the biological relevance of the traits being evaluated.
- Random regression models can also be applied to functional traits, including fertility, health, and behavioral measurements, when these traits are recorded repeatedly or change meaningfully over time. For example, researchers may investigate how genetic differences in feed intake develop as animals mature or how a physiological trait changes across lactation. However, not every trait is best analyzed with a random regression model. Traits recorded only once, or traits that are inherently categorical without a meaningful continuous trajectory, may require other statistical methods. The choice of model should be guided by the structure of the data and the biological question rather than by the availability of a sophisticated statistical technique.
- An important advantage of random regression models is that they can make effective use of incomplete and irregularly spaced records. In livestock production, animals do not always have measurements collected at identical ages or intervals. Some animals may leave a herd before completing a production cycle, while others may have missing records because of illness, management changes, or recording limitations. Random regression models can accommodate measurements taken at different time points without requiring every animal to have the same number of records. Nevertheless, missing data and selective recording can still introduce bias, so the assumptions underlying the model and the data collection process must be carefully evaluated.
- These models are generally fitted using mixed-model methodology, often through restricted maximum likelihood estimation to estimate variance components. Once the relevant variance and covariance structures have been estimated, genetic effects can be predicted using methods such as Best Linear Unbiased Prediction (BLUP). The estimated genetic regression coefficients are combined with the chosen basis functions to obtain predicted genetic effects at specific ages or stages. These predictions can then be used to calculate estimated breeding values for traits at particular time points or to summarize performance over a defined period. In practical genetic evaluation systems, the results may contribute to selection indexes that combine multiple traits according to their economic importance.
- Despite their advantages, random regression models present several challenges. They can be computationally demanding, particularly when large populations, long recording periods, and complex covariance structures are involved. The number of parameters increases as the regression order and the number of modeled effects increase. Poorly chosen regression functions can lead to unstable estimates, unrealistic curves, or excessive fluctuations near the boundaries of the observation period. Researchers must therefore assess model fit, convergence, prediction accuracy, and the biological plausibility of the estimated trajectories. Adequate data coverage across the full period is also important, because estimates at the beginning or end of a trajectory may be less reliable when relatively few observations are available.
- The accuracy of random regression models depends on the quality of pedigree information, genomic information when available, environmental records, and repeated performance measurements. Incorrect pedigree relationships, systematic recording errors, or insufficient representation of particular ages and management groups can weaken the resulting genetic evaluation. Contemporary groups and other relevant fixed effects must be specified appropriately so that systematic environmental differences are not mistaken for genetic effects. In genomic evaluations, genomic relationships may be incorporated into the model to improve the prediction of genetic trajectories, provided that the data and evaluation framework support this approach.
- Random regression models can be extended to analyze multiple traits simultaneously. For example, milk yield, fat yield, and protein yield may be modeled together to investigate how genetic relationships among these traits change throughout lactation. Similarly, body weight and feed intake may be evaluated jointly across growth. Multivariate random regression models can provide useful information for selecting animals that perform well across several traits and stages of life. However, these models may require substantial computing resources and large datasets to estimate complex covariance structures reliably.
- The use of random regression models has strengthened modern animal breeding by moving genetic evaluation beyond single records and fixed-age measurements. They allow breeders to study how genetic potential develops over time, estimate age-specific genetic parameters, and identify animals whose performance patterns are suited to particular production systems. This is particularly important when breeding goals involve not only maximum production but also persistence, efficiency, health, and adaptability. An animal’s value may depend on the consistency of its performance across a production cycle rather than on its performance at one isolated point.
- In conclusion, random regression models are powerful tools in quantitative genetics because they describe genetic and environmental effects as functions of time, age, or another continuous variable. They are especially valuable for repeated traits such as milk production, body weight, feed intake, and other longitudinal performance records. By estimating genetic trajectories and changes in variance and covariance over time, these models support more detailed breeding value estimation and better-informed selection decisions. Although they require careful model specification, sufficient data, and appropriate computational methods, random regression models remain an important part of modern livestock genetic evaluation and the development of more productive, efficient, and sustainable animal populations.
