Genetic parameters of litter size (LS) and number of kids born alive (NBA) were estimated across parities in Florida and Payoya goat breeds using a threshold random regression model (RRM). We analysed the reproductive records for the Florida and Payoya breeds separately, and a total of 130.849 and 67.478 reproductive records from the first to the seventh parities from Florida and Payoya, respectively, were included in the analysis. Random regressions on Legendre polynomials of standardised parity were included for permanent environmental and additive genetic effects. We based our estimation of all the covariance components on Bayesian inference under categorical distribution and considered heterogeneous residual variances in the model. The estimates of heritabilities ranged from null to 0.37 for LS and from 0.02 to 0.20 for NBA, while the repeatability estimates were between 0.008-0.60 and 0.03-0.44 for LS and NBA, respectively. Phenotypic correlations within-trait along parities ranged widely from À0.31 to 0.97, while the genetic correlations within-trait through parities showed a wider range and magnitude of values, from À0.98 to 0.99 for LS and from À0.72 to 0.99 for NBA. The estimates of genetic correlations between LS and NBA in the different parities were positive in the first two parities and negative in the later ones. Due to the noticeable changes in variance and the incomplete genetic correlations, the use of a random regression model to analyse LS and NBA is recommended in goat breeds.
HIGHLIGHTSA bivariate threshold random regression model has been used for the first time to estimate genetic parameters of litter size and number of kids born alive across parities in Spanish goat breeds. The evolution of genetic parameters along parities is different between Florida and Payoya that have different production systems, being more intensive in the case of the Florida breed. Estimates of variance were not constant across parities for both traits and breeds. Genetic correlations within-trait across parities were different from unity. Thus, LS and NBA should be treated as different traits in successive parities. Given that the estimates of variance and covariances are not constant across parities for both traits and breeds, the use of RRM is highly recommended.