Mark Whittingham & His Colleagues on Stepwise Analyses
Fast Moving Front Commentary, January 2012
|
Article: Why do we still use stepwise modelling in ecology and behaviour?
Authors: Whittingham, MJ;Stephens, PA;Bradbury,
RB;Freckleton, RP |
Mark J. Whittingham, Philip A. Stephens, Richard B. Bradbury, & Robert P. Freckleton talk with ScienceWatch.com and answer a few questions about this month's Fast Moving Fronts paper in the field of Plant & Animal Science.
Why do you think your paper is highly
cited?
The paper is highly cited within the ecological and behavioral science literature (for which it was aimed) but also in a wide range of other disciplines. We think the most likely reason for this is that: (a) we demonstrate that the method we criticize (stepwise modeling) is prevalent in the scientific literature; (b) we demonstrate the underlying issues with this approach including worked examples; and (c) crucially we also provide suggestions to get around the problem.
Given that stepwise modeling has long been a primary method of analysis in a whole range of disciplines, it is unsurprising that researchers in those other disciplines are using our paper in support of a change of analytical approach.
Does it describe a new discovery, methodology, or
synthesis of knowledge?
It is a methods paper but it does also synthesize knowledge across different disciplines such as statistical literature, ecology, and behavior.
Mark Whittingham (second from right), with one of his
post-docs, two of his Ph.D. students, and one of his collaborators at the
IOC in Brazil in 2010.
How did you become involved in this research, and
how would you describe the particular challenges, setbacks, and
successes that you've encountered along the way?
The first author became involved in this study by having a cup of tea with the last author. The latter informed the former about the problems of stepwise modeling which the former had learnt to use through statistical courses and text books.
Some of the authors were surprised when the results of the literature review revealed how prevalent the technique was in the literature. Judging by our experiences of reviewing papers the problem is still prevalent in the literature but its use has declined.
Where do you see your research leading in the
future?
Our study points at other ways to solve complex multi-predictor models. These include Information Theoretic models and the use of full models (including all predictors). The use of Information Theory in the biological sciences has been strongly promoted by Burnham and Anderson (2002). Since then, Information Theory has been an active area of research that has led to a range of papers exploring the techniques involved, including a range of papers written by us.
Do you foresee any social or political
implications for your research?
Policy, by governments or others, is often influenced and informed by a range of sources. The most rigorous of those sources are the results of peer-reviewed literature. By changing methods of analysis to acknowledge uncertainty in our findings—and, especially, uncertainty in our beliefs about cause and effect (whether in the social realm or in the natural world)—we can ensure that policy is founded on the best possible interpretation of available data.
Mark Whittingham
Newcastle University
Newcastle Upon Tyne, UK
Rob Freckleton
Sheffield University
South Yorkshire, UK
Phil Stephens
Durham University
Durham, UK
Richard Bradbury
Royal Society for the Protection of Birds
Bedfordshire, UK
KEYWORDS: ECOLOGICAL MODELLING, HABITAT SELECTION, MINIMUM ADEQUATE MODEL, MULTIVARIATE STATISTICAL ANALYSIS, STATISTICAL BIAS, INFORMATION THEORY, LOWLAND FARMLAND, SPATIAL SCALES, SELECTION, REGRESSION, YELLOWHAMMERS.