Identification of the unit in experiments on supplementary feeding of beef cattle grazing native pasture.

Abstract

Animal Production in Australia IDENTIFICATION OF THE UNIT IN EXPERIMENTS ON SUPPLEMENTARY FEEDING OF BEEF CATTLE GRAZING NATIVE PASTURE. G.W. BLIGHT* and P.M. PEPPER** SUMMARY Published results of similar grazing experiments reveal inconsistencies about whether the animal or paddock variation is the appropriate estimate of experimental error. To look at the problem of identifying the experimental unit (EU) in supplementary feeding trials, we present results from the analysis of 24 experiments with growing beef cattle grazing native pasture and covering a range of environmental and management conditions in Queensland. Our investigation showed that in many cases the individual animal could be regarded as the EU and animal variation gave a good estimate of random error; but this could not be recommended universally. The difficulties of obtaining uniform replicates resulted in significant interactions in experiments from one site and demonstrated that paddock replication was essential in all experiments. INTRODUCTION In the five distinct types of grazing experiment in Table 1, the nature of the trleatments dictates whether the experimental animals may graze together as one herd, or need to be grouped in separate treatment paddocks. The correct identification of the experimental unit (EU) is the first step in any design. Text blook definitions aimed at covering experimental design in any field are 'the unit of material to which one application of a treatment is applied' (Steel and Torrie 1960), and 'the unit corresponds to the smallest division of the experimental material such that any two units may receive different treatments in the experiment' (Cox 1958). With respect to grazing experiments these definitions are incomplete - it remains to define 'material'. Beattie and Alexander (1973) give unequivocal advice to experimenters in the choice of unit for a number of types of grazing experiment with beef cattle. In general, their EU is the paddock for the first two types of experiment in Table 1, but when the animal 'carries' its treatment with it, the EU is the * QDPI, Biometry Branch, Rockhampton, Qld 4700 ** QDPI, Biometry Branch, Brisbane, Qld 4001 297 Animal Production in Australia animal. Supplementary feeding experiments are a special case of the animal as the EU since for practical reasons the supplement is paddock fed; our study is restricted to this third type. Animals may also be grouped for veterinary comparisons where treated and untreated animal groups have to be isolated to avoid contamination. With breed comparisons the animals may graze together as one herd. Whenever animals are grouped the paddocks need to be replicated or paddock differences would be completely confounded with treatment effects. Defining the individual animal to be the EU determines that animal variance is the experimental error. The paddock replication x treatment interaction would be used to test the effect of treatments only if replicates are taken to be a random effect (Henderson 1959) i.e. the replicate sets of paddocks are sited at randomly (or objectively) selected sites in the region so as to broaden the applicability of results. We examine whether within paddock variation is an appropriate estimate of animal variance and whether it can be used to test the effect of treatments. This approach has attracted two major criticisms in the past (i) the within paddock variation may seriously underestimate or overestimate animal variance because of group feeding or competition effects, respectively and (ii) that the experimental error should contain both pasture and animal variation. As Morley and Spedding (1968) note the problem merits investigation. MATERIALS and METHODS Weexaminedthe analysis of variance results from 24 supplementary feeding experiments, which were carried out between 1968 and 1979 by QDPI officers at two sites at 'Swan's Lagoon' near Ayr, and at one site at 'Brian Pastures' near Gayndah. All experiments involved growing beef cattle grazing native pasture (NP) of mainly speargrass (Heteropogon contortus). The range in experimental and management conditions is summarised in Table 2. TABLE 2 Experimental and management conditions All experiments were stocked at a heavier rate than the district average for animals of the same age grazing native pasture - the rates varied from 25% higher to four times the average for the region. During the feeding period the growth rate of the unsupplemented NP groups varied from -234 to 332 g/h/d. The experimental supplement treatments were mainly based on molasses and/or urea, with particular treatments comparing either mineral additives or level of feeding; in two experiments urea/molasses were compared with a standing legume supplement fed in sub-paddocks. An unsupplemented NP treatment was included in all experiments. All urea/molasses based supplements were paddock fed either by a drum-licker or block. A common feature of the design of all 24 experiments was the use of paddock replication (2, 3 or 4 replicates) in a randomized block 298 Animal Production in Australia layout, with the replicate sets of paddocks being set up at one experimental site; this minimal replication at the same site provides a check on the presence of replication by treatment interaction. Individual animals were allocated to paddocks by stratified randomization based on initial liveweight; a different draft of animals was used in each experiment. We considered three major experimental periods: supplement feeding, postfeeding and total. Individual animal growth rates in the three periods were estimated by average daily gain calculated from full liveweights. For each experiment and period animal variances within treatments groups were tested for homogeneity using Bartlett's test, and paddock variance was compared with animal variance (by F-test). RESULTS AND DISCUSSION TABLE 3 Experimental error mean squares for average daily gain (g/h/d) $ Paddocks error mean squares expressed on a per animal basis. @ Indicates significant differences (PcO.05) by Bartlett's test of homogeneity of within paddocks treatment variances. * Indicates significant differences (PcO.05) by F-test of paddock mean squares versus animal mean squares. ++ Kean squares weighted by degrees of freedom. 299 Animal Production in Australia In general, when the variances between animals within treatments were tested for homongeniety, they were not significantly different (P>O.O5). In particular, there was no indication that unsupplemented animals were more or less variable than supplemented animals. Consider the arguments about group effects influencing the between animals estimate of error. There could be a social affect of animals grazing together which tends to make measurements of animals within a paddock correlated, and so within paddock variation would underestimate true animal variance. On the other hand, animals grazing native pasture at the high stocking rates used in the 24 experiments, could be stressed with perhaps the lighter animals faring better than the heavier animals; this negative correlation would tend to increase animal variation. With supplemented animals, some animals could consume more supplement than others and this could result in larger animal variation in the supplemented groups. We concluded from the homogeneity tests that there are no appreciable group effects on the between animals estimate of error. Estimates of animal variance are reasonably consistent across experiments. For the feeding, post-feeding and total periods, the paddock variation was significantly greater than the animal variation in 15,4 and 9 experiments respectively. A problem in interpretation arises when paddock replication X treatment interaction is significant since one must be more careful in interpreting the main effect obtained for treatments. These interactions occurred most frequently in the analyses of two series of experiments (1.. .4,10...13); both at the same site. Analysis of two years data of a uniformity trial on this site revealed that there were consistent paddock differences and that paddocks were not uniform within a replicate. By using information on paddock differences as a covariate in the analyses of the experiments on that site, the interaction was explained in about half the analyses. Since some interaction effects remain unexplained for site one, our conclusions from this investigation are not clear-cut. In most cases, one can expect the animal variation to be a good estimate of random error. The unexplained replicate X treatment interactions remain a problem; further work is planned to find a suitable measure to explain paddock variation. In addition to uniformity trials, pattern analysis on soil and moisture measurements could help in selecting uniform paddocks for a replicate. The alternative is to increase paddock numbers and estimate error from paddock variation; but there is a difficulty in obtaining uniform paddocks and maintaining sufficient animals in a paddock to simulate a commercial herd. Depending on the number of treatments, a minimum of 20 paddocks are usually necessary to reliably determine experimental error. ACKNOWLEDGEMENTS We thank Mr L. Winks, Mr S. McLennan and Mr A. Foster, of the Qld Dept of Primary Industries who made data available. REFERENCES BEATTIE, A-W., and ALEXANDER, G.I. (1973). In 'Manual of Techniques for Field Investigations with Beef Cattle' p. 8.1, editor G-1. Alexander (CSIRO: Canberra) COX, D.R. (1958). 'Planning of Experiments' p. 2, (Wiley: New York) HENDERSON, C.R. (1959). In 'Techniques and Procedures in Animal Production Research' p. 1, (American Society of Animal Production Monograph: Maryland). MORLEY, F.H.W., and SPEDDING, C.R.W. (1968). Herbage Abstr. 22 p. 279 STEEL, R.G.D., and TORRIE, J.H. (1960). 'Principles and Procedures of Statistics' p. 90, (McGraw-Hill: New York) 300