ON CONFIRMATORY VERSUS EXPLORATORY RESEARCH

We discuss the difference between data that are gathered to test a priori alternative hypotheses (confirmatory research) and those gathered to determine whether any interesting a posteriori hypotheses might be generated from the data set (exploratory research). Confirmatory studies should be designed so as to minimize type I statistical errors while exploratory studies should be designed to minimize type II errors, which may require different approaches to statistical analyses. We suggest that authors clearly state whether they are reporting confirmatory or exploratory research in manuscripts, especially for experimental studies, because evaluations by referees and readers often hinge on such a distinction. Key tvorcts: Confirmatory research; Exploratory research; Experimentation; Type I error; Type II error As editor of Herpetologica (Jaeger) and former editor of Animal Behaviour (Halliday), we have observed instances where a manuscript received unfavorable reviews because referees were not aware of the intentions of the author. This sometimes is caused by the author failing to state whether the research being reported is confirmatory or exploratory in concept. Confirmatory research proceeds from a series of alternative, a priori hypotheses concerning some topic of interest, followed by the development of a research design (often experimental) to test those hypotheses, the gathering of data, analyses of the data, and ending with the researcher's inductive inferences. Because most research programs must rely on inductive (rather than deductive) logic (e.g., Popper, 1968), none of the alternative hypotheses can be proven to be true; the hypotheses can only be refuted or not refuted. Failing to refute one or more of the alternative hypotheses leads the researcher, then, to gain some measure of confidence in the validity of those hypotheses. This procedure closely follows what Platt (1964) called "strong inference": making a "scientific step forward" by invalidating certain alternative hypotheses. We have observed that many readers tend to assume that a manuscript attempts to achieve strong inference, a testimony to the influence that Platt's paper has had in biology. Sometimes, though, an author is not attempting to achieve strong inference; the attempt is to determine what novel hypotheses might be generated from a previously unexplored biological situation: i.e., exploratory research. For example, the researcher might be interested in employing signal honesty theory (e.g., Dawkins and Guilford, 1991) using a species the agonistic and courtship behaviors of which are poorly known. Explicit hypotheses tested with confirmatory research usually do not spring from an intellectual void but instead are often gained through exploratory research. Thus exploratory approaches to research can be used to generate hypotheses that later can be tested with confirmatory approaches. While confirmatory research is often experimental (Platt, 1964), exploratory research may be either experimental or observational. The end goal of exploratory research, though, is to gain new insights, from which new hypotheses might be developed. The inductive logic of confirmatory and exploratory studies was discussed by James and McCulloch (1985), and we defer to that paper for a more comprehensive treatment (philosophical and statistical) of the subject. However, we believe that the difference between these approaches to research should be made obvious in the way that a manuscript is written, and thus later perceived by readers. We have ob-