Hawariat, Girma Wolde. Alternative statistical models with emphasis on characterization of efficiency of weight gain of bulls from a modified five breed diallel. Doctoral Dissertation.
Thesis
Data on 167 bulls from a modified five breed diallel involving Angus, Brahman, Hereford, Holstein and Jersey were used to compare alternative statistical models for estimation and characterization of various breed types of bulls for efficiency of feed conversion. Due to the nature of the data set, regular and generalized least squares methods of analyses could not be used. The alternative method of analysis used here utilized weight gain as the dependent variable and feed and other relevant variables as independent variables in a covariance model (first differences approach) and interpreted feed efficiency directly as a coefficient of feed consumption. Two models containing cumulative feed intake and weight as dependent variables and relevant independent variables were used to obtain residuals so that the magnitude of correlations among residuals from these models and first difference approach model could be compared. In addition, efficiency was determined as ratios of weight gain to feed intake from predicted cumulative feed intake and predicted weight and from weight and feed intake for comparison with the first differences approach. A regression model that included weight gain as a dependent and feed intake per day and weight to .75 power as independent variables was identified as the most relevant within breed type model. This model was used to tune up the data set. A covariance model that included weight gain as a dependent and breed, season, month within season, feed/day, assignment weight, weight('.75), breed X feed/day and season X feed/day effects as independent variables was fitted to determine efficiency as the sum of the coefficient of feed consumption and the coefficient of breed X feed/day. The correlations between residuals using the first difference model were lower (.199) than those using the cumulative models (.809). Average efficiency as determined by ratios from predicted values (.1324) and simple ratios (.1140) were higher compared to that determined by the first differences approach (.0817). The first differences approach is the most relevant model to determine efficiency because of its biological and statistical relevance to this type of data set. There was no significant average heterosis effect indicating that crossbreeding to utilize the effect of hybrid vigour may not significantly improve efficiency of feed conversion. None of the various categories of breed types contrasted indicated a significant (P (LESSTHEQ) .05) difference using the first differences approach.
