Estimation of genetic and environmental parameters for two binary traits with an underlying continuous scale : a simulation study

Abstract EN

The study objective was to compare heritability and correlations estimated from linear model with those estimated from threshold model.

Two simulation programs were used, one to simulate two continuous traits with the same level of heritability (h2) of 0.17 and another to modify these traits into binary traits.

Fifteen replicates were generated with two classes of fixed effect.

Each replicate was analyzed twice, once by GS algorithm and another by MTDFREML algorithm using animal model.

The concerned results from each analysis were heritability and genetic and residual correlations.

Bias and mean squared errors (MSE) of h2 and genetic and residual correlations estimates were used to assess the quality of h2 and genetic and residual correlations estimates obtained by different algorithms.

For the two traits, the statistical model used included type of algorithm (MTDFREML vs Gibbs sampling), type of trait (continuous trait vs binary trait) on the bias of heritability and genetic and residual estimates plus the 2-way interaction.

For bias of heritability, all these effects were significant (p<0.01).

For genetic correlation estimates, none of the effects and interaction were significant (p>0.05) while for residual correlation estimates, all these effects were significant (p<0.0001).

In estimation of heritabilities for continuous traits, the GS algorithm is equivalent to MTDFREML algorithm.

For binary traits, the GS algorithm is the best.

GS estimates of genetic correlations were similar to MTDFREML estimates while for the residual correlation, the marginal maximum likelihood (GS) estimates are superior to the estimates from the linear model (MTDFREML).