Statistical evaluation of in vitro gas production kinetics

Abstract: At the Forage Research Centre, Swedish University of Agricultural Sciences in Umeå a technique has been developed to describe the degradation of feeds in ruminant animals. The development of this technique has been made in collaboration with Dr J.W. Cone, Nutrition and Food, Animal Sciences Group of Wageningen UR, Lelystad, the Netherlands. Experiments have been performed in laboratories where feed samples have been incubated with rumen fluid and the amount of gas produced during the digestion has been measured continuously. These feed samples were analysed at three separate occasions. The purpose of this thesis was to identify and describe statistical procedures for detecting differences between feeds analysed within the same laboratory as well as differences between the same feeds analysed in two different laboratories (in Sweden and the Netherlands). To determine the rate of digestion and to describe to what extent feeds are digested in the rumen a gas production model was fitted, using non linear regression. In order to test whether there are significant differences between the feeds, three methods were applied. For each method, the variances were estimated differently. In the first method, Hotelling’s T2 tests and two sample t tests were performed. From these tests, differences between the feeds that were analysed within the same laboratory were detected whereas no differences between the same feeds analysed in two different laboratories could be found. In the other two methods, tests were performed using an assumption of normality. These two methods detected a larger number of differences between the feeds than the first method, primarily due to extremely underestimated variances. The first method is to be preferred since the estimated variances in this method are unbiased. This causes the result to be more reliable. For future experiments it is recommended that the feed samples are analysed at considerably more than three occasions. This would lead to better estimations in the first method and consequently the result would be enhanced.