Quality of heritability estimates as affected by level of heritability, number of progeny persire, type of algorithm, type of model and type of trait

الملخص EN

Two simulation programs were used in this study, one to simulate a continuous trait and another to modify this trait into a binary trait.

Twelve populations were created (three levels of heritability (h2), 0.10, 0.25 and 0.50; four levels of number of progeny per sire, 5, 10, 15 and 20), each with three parities as the only fixed effect.

Twenty replicates were generated for each population.

Each replicate was analyzed twice, once with sire model and another with animal model, using two algorithms for each model (MTDFREML or Gibbs Sampling (GS)).

Bias and mean squared errors (MSE) of heritability estimates were used to assess the quality of heritability estimates obtained by different models and different algorithms.

The effect of h2 level, number of progeny per sire, type of algorithm, type of model, type of trait and the interactions on the bias and MSE were examined.

All main effects were highly significant (p < 0.0001).

For estimating variance components, for a continuous trait, the animal model was the best in the case of using MTDFREML and GS at all levels of h2.

Also, at all levels of h2, the GS was the best algorithm in the analysis of a binary trait.

For a binary trait within GS, the sire model was the best at h2 equals to 0.1 with number of progeny more than 5 whereas, at h2 equals to 0.25 or 0.5 with 20 progeny per sire, the use of animal model was equivalent to the use of sire model.

At all levels of h2, the 20 progeny per sire had the lower MSE for heritability.