Avaliação de um modelo de regressão não-linear para a estimação da variância genética em populações naturais

Abstract

This work had the aim of evaluating the non-linear regression model proposed by Shrikande (1957), to estimate genetic variance in natural populations. It was used experimental data from progeny evaluation with Eucalyptus, and simulated data, by generating populations varying spatial autocorrelation among individuais (p) and broad sense heritability (h2). Experimental data were adjusted to Shrikande's model using the methods of ordinary andweighted least squares. To each combination of p andh2, simulated data were adjusted using the methods above, and also generalized least squares. Estimates of genetic variance yielded by Shrikande's model were compared to those obtainedby momentmethods, considering the experimental designs ofthe Eucalyptus trials. Simulated data allowed obtaining empirical sampling distributions of the estimators of genetic and environmental variances, which were evaluated as regard to bias and precision. With simulated data, it could be seen that, with high autocorrelation and heritability, good adjustments were obtained, specially with generalized least squares, which yielded the most precise estimates. When even autocorrelation and heritability were low, it has not been observed reasonable adjustments, whatever the method of estimation employed.