Can ornithology advance as a science relying on significance testing? A literature review in search of a consensus

Full article

Published:

Volume 52(2), December 2005. Pages 377-387.

Language:

English

Keywords:

alternatives, Bayesian statistics, confidence intervals, decision-making, knowledge accumulation, null hypothesis, power analysis, P-values, significance testing

Abstract:

There is a sense of frustration among ornithologists, especially among those dealing with conservation biology, regarding the usefulness of research results for decision-making. Similar problems affect the social and behavioural human sciences and their failures have been partially linked with the strong dependence of data analysis on the statistical testing of null hypothesis. We review the literature on reasoning against and in favour of null hypothesis testing, in search of some consensus procedure for proper data analysis, so that knowledge acquisition is feasible and decision-making policies are useful. We conclude that a consensus on appropriate data analysis could be reached if we focus on precision of the estimates and biological relevance instead of on significance. This implies an important change in our way of thinking, promoting confidence intervals for the difference of effects rather than p-values. The alternative approach of Bayesian statistics can also be of great help in the decision-making process typical of applied ornithology as it provides a measure of the evidence of the effects and can solve complex models with straightforward procedures. We suggest that the slow progress of theory accumulation in ornithology and the low reliability of results for decision making may be due, to a large extent, to the way we analyze data rather than to the nature of the topics approached by ornithologists.