Design: a Critical Need in Pest-damage Control Experiments

The manner in which an experiment is conducted determines the inferences that can be made from the results of the analysis of the experiment. This paper emphasizes the critical need in pest-damage control (PDC) experiments for a detailed planning process (i.e., the design of experiments) by exampling improper designs that prohibit a researcher from making valid inferences about his hypotheses of interest. Emphasis is placed on identification of experimental units, determination of restrictions on the randomization procedure, and specification of treatment forms of pest control materials. A list of some specific actions to strengthen PDC experiments is given. INTRODUCTION Design of experiments is the planning process that allows researchers to efficiently and objectively direct their efforts toward gathering information pertinent to the hypothesis under test. Unfortunately, many authors of texts on statistical design and analysis concentrate on analysis at the expense of design. These authors unwittingly encourage analysis by rote. Vague terminology (e.g., "cell" and "cross classification") often obscures the manner in which the experiment was conducted. Users of statistics obtain the erroneous impression that, to analyze an experiment correctly, they must first arrange their data in a standard tabular format. A similar tabular format is then observed in a statistical analysis text and the analysis performed on the researcher’s data is the analysis associated with tabular format. Consequently, the inferences made from the results of many pest-damage control (hereafter referred to as PDC) experiments are incorrect because the researcher, as well as the statistician, have failed to recognize the hypothesis under test. Recognition is the key to design of any experiment. Unless the researcher recognizes the hypothesis under consideration, he cannot select treatments that address this hypothesis. Unless he recognizes the sources of variability present in the experiment, he cannot select an experimental design that will increase the efficiency and sensitivity of his treatment comparisons. However, recognition also is the key to the analysis of an experiment and to an understanding of the inferences that can be drawn from the results of an experiment. In this paper, I attempt to show how improper design can prohibit a researcher from making valid inferences about the hypotheses of interest. The approach is somewhat backward because the ideal procedure is to adequately design an experiment and then to verify the validity of the experiment through the use of a mathematical model and an outline of the analysis. Unfortunately, there are many cases in PDC research where a less than perfect design is forced on the researcher or where a good design is inadvertently modified in the field. In these circumstances, it is the responsibility of the statistician and the researcher to identify the deficiencies associated with the experiment and to determine how these imperfections could cloud inferences. The researcher has one additional responsibility; he must assess the biological importance of the imperfections. Three topics in the design of PDC experiments will be considered in this paper. The first, recognition of the randomization procedure, is presented through incomplete examples of the type that appear in many statistical texts. The purpose of this presentation is to inform the reader that, while the mechanical computations involved in obtaining sums of squares and mean squares for an analysis of variance table are identical, the inferences that can be made concerning the effects of interest are highly dependent on the randomization procedure. Second, we consider specification of treatment forms of a pest control material (hereafter called PCM) investigated in PDC experiments. The purpose of this discussion is to define the parameters of a treatment form and to illustrate the need for well-defined commercially-realistic treatment forms. Finally, we deviate from the general approach and list specific actions to strengthen PDC experiments.