The impact of experimental design on the application of grazing research results: an exposition.

Funding Bmitations often restrict pasture replication in grazing research on rangeland. Consequently, subsample error has been used to estimate treatment effects or characterize populations. Assumptions associated with experimental designs which utilize subsample error to make inferences are discussed and an example evaluated. The appropriate experImentaI unit for inferential grazing research is the pasture. Animals or vegetation aampIing within pastures must be considered as subsamples in inferential grazing research. Pasture replication must be used in intensive grazing trials to establish treatment dffferences or provide adequate characterization. Following intensive trials, extensive, unrepIicated trials implemented by private producers can be effective in estabIishIng broad-based applicability. Unreplicated pasture trIaIs may also be used for screening several treatments. Shrinking research budgets, increased competition for research dollars, and increased complexity of grazing research have contributed to efforts to enhance the efficiency of the research process. Additionally, range and pasture research competes with other research for land in organizations with finite land resources. Differentiating between efficiency of experimental design and the ability to make the proper characterization or inference is particularly important in range science. Minimizing cost in the experimental design process often is done at the expense of inference. Most books on experimental design or statistics provide few examples representing experimentation at the population, community, or ecosystem levels of organization (Hurlbert 1984). The principles of design for grazing studies are most often violated. It is the purpose of this discussion to explore some aspects of experimental design critical to evaluating efficiency and cost versus interpretive credibility of grazing research. Hurlbert (1984) defined experimental research as mensurative (sampling studies) or manipulative (external treatment). Mensurative experiments involve only the making of measurements at one or more points in space or time; space or time is the only experimental variable or “treatment”. Cochran (1977) provided an excellent review of sampling techniques for sampling studies. Manipulative experiments always involve two or more treatments and have as goals making one or more comparisons. Several references provide valuable information on the design of manipulative experiments (Federer 1955, Cochran and Cox 1957, Cox 1958). Measuring attributes of interest is important in both mensurative and manipulative experiments. There are several references available on measurement techniques (Brown 1954, U.S. Forest Service 1963, Greig-Smith 1964, Mueller-Dombois and Ellenberg 1974, ‘t Mennetge 1978, Stubbendieck and Schacht 1984, Cook and Stubbendieck 1986). Experiments can be inferential, providing information about a population; or they can be descriptive, providing information about specific individuals within a population. Inferential Grazing Trials Population of Inference The focal point of inferential research is the characterization of or inference about some population. Steel and Torrie (1980) defined a population as all possible values of a variable. The Authors are research animal scientist, USDA, ARS, Rt. 2, Box 144A, Booneville, Ark. 72927; professor, Department of Agronomy, University of Nebraska-Lincoln, Lincoln 685834915. Manuscript accepted 8 August 1985. JOURNAL OF RANGE MANAGEMENT 39(3), May 1996 selected population of interest becomes the population of inference from which individuals are selected for experimentation. The population could be all native pastures in eastern Nebraska for a grazing study or all silty range sites in Vegetative Zone IV of Nebraska for a vegetative survey. Explicit definition of the population of inference must be made prior to the selection of the experimental material. If the population is well defined; means, variances, covariances, probabilities and other statistics generated from a research project can be properly interpreted. Once the population is defined, sample units can be selected within the appropriate experimental material. Sample units could be termed observational units in descriptive research and experimental units in experimental research. Experimental Unit A proper experimental design can be destroyed by failing to recognize what constitutes the experimental unit (Nelson and Rawlings 1983). According to Cox (1958) the experimental unit corresponds to the smallest division of experimental material such that any two units may (independently) receive different treatments in the actual experiment. Because responses are not constant in biological research, sample units should be chosen at all levels which can affect characterizations or responses. For example, treatment, pasture, animal, year of measurement, year of treatment, the failure of main effects to respond consistently within the grazing season, and appropriate interactions are all potential sources of variability in a manipulative grazing trial. Animals can be considered experimental units in mensurative and manipulative research trials in which the forage resource has no differential effect on the measured response. Breed comparisons, reproductive physiology, insect or parasite studies are examples of such studies. Prior knowledge is the basis for the assumption that pasture effects are negligible or of no interest. In identifying the animal as the experimental unit and not replicating pastures, the researcher foregoes the opportunity to evaluate potential pasture X treatment interactions. Animals in manipulative grazing research on rangeland are generally used as a treatment (e.g., stocking rate study) or as measures of treatment effect (e.g., gain/ head, gain/area). Freegrazing animals in range situations can seldom be considered as experimental units. Each animal must receive a treatment to be considered an experimental unit in manipulative research. In addition, each animal must be independent of other animals for the response measured. If forage availability or selection are important factors in the response variable, animals within a pasture cannot be considered independent. As an example, forage consumed by one animal cannot be consumed by another, implying a dependency. Thus, animals are actually repeated measures on the same experimental unit. The experimental unit of a production system in the measurement of animal production from grasslands must consist of an area of land and the animals grazing on it as well as auxiliary facilities for such management activities as supplemental feeding (Morley 1978). Pasture size and number is a critical consideration for research facilities with limited land resources. Pastures should be large enough to support adequate animal numbers to provide the precision in response measurements (Peterson and Lucas 1960). However, small pastures are more desirable than large, production-size pastures for intensive grazing trials since land is restricted.