A COMPARISON OF PARAMETER ESTIMATES OF NON-LINEAR GROWTH MODEL IN DIFFERENT CHICKEN BREEDS USING BAYESIAN APPROACH WITH NORMAL DISTRIBUTION DATA

Abstract

Bayesian statistics is a system for describing epistemological uncertainty using the mathematical language of probability while growth models provide valuable information for livestock improvement programmes and decision making. In this study, the ability of four non-linear growth models (gompertz, logistic, richards and bertalanffy) to predict the growth of three indigenous breeds (normal feathered, frizzle feathered and naked neck chickens), one locally developed crossbred chicken (FUNAAB Alpha) and three exotic breeds (nera black, white leghorn and giriraja) were evaluated. Data for body weight were collected every week from 993 bird for 20 weeks. The non-linear growth models were used to fit the body weight data using Bayesian approach under normal distribution data. The parameters in the models; asymptotic weight, integration constant, maturing rate, age at inflection point and weight at inflection point for the lLogistic, gompertz, richards and bertalanffy growth models were estimated for each breed using WinBUGS (version 1.4.3) to read the model, priors and posterior data, and run the Markov Chain Monte Carlo simulation (MCMC). The models were adjusted for best fit using Bayesian Information Criterion (BIC). The results showed that bertalanffy model had the lowest maturing rate (0.07, 0.08, 0.08, 0.06, 0.05, 0.07 and 0.08) and predicted the highest asymptotic weight (2104.50g, 1963.00g, 1821.75g, 2668.50g, 3167.00g, 2269.50g and 3326.00g) among the models for naked neck, frizzle feather, normal feather, FUNAAB Alpha, near black, white leghorn and giriraja chickens respectively. The parameter estimates in gompertz and richards models were relatively close within the breed compared to other models with a variation difference of between 1.5 and 2.0% while between 8.3 and 10.0% was recorded within FUNAAB Alpha and nera black. The bertalanffy according to the values obtained in Bayesian information criterion had the least value for all the breeds. In conclusion, Bayesian model should be given consideration in growth decision making and selection program relating to the breeds considered in this study.