Inference in Ecology and Evolution

Citation: Philip A. Stephens, Steven W. Buskirk and Carlos Martínez del Rio (2007) Inference in Ecology and Evolution. Trends in Ecology & Evolution Volume 22, Issue 4 (RSS)
Internet Archive Scholar (search for fulltext): Inference in Ecology and Evolution
Download: http://ac.els-cdn.com/S0169534706004009/1-s2.0-S0169534706004009-main.pdf? tid=a5fea692-f59b-11e4-ae11-00000aacb35d&acdnat=1431101138 1e6eb9729c6be45718fa7b3477b2246c
Tagged:

Summary

In this article, the authors point out the limitations of null hypothesis significance tests (NHST), one of main inferential approaches used by ecologists and evolutionary biologists and suggest alternatives. Also they introduce three statistical application and suggested approaches.

NHST are made to determine the probability with which an observed effect would be observed it the true effect is zero. Many researchers have used this approach because of its computational simplicity though there are some limitations due to difficulty of statistical technique or unawareness of the limitations. The main two problems are divided into two categories which are associated with interpretation (e.g. error in equating) and involving deeper philosophical issues (e.g. falsification to acknowledgment of plural hypotheses).

The authors suggest three alternative models, Effect size statistics, Bayesian statistics and Information theoretic model comparison (IT techniques). Effect size statistics can overcome the problems of interpretation of NHST using counternull of NHST. Bayesian statistics focuses on intuitive appeal by rejecting or accepting previous beliefs and assumptions. IT techniques provides an information criterion value for each models and the criterion enables to explain the uncertainty using average of the models and it can overcome the philosophical problem of NHST.

Those four approaches can be used to accomplish the goals of statistical analysis, assessing descriptive findings, generating predictive models and challenging research hypothesis. The authors assert that NHST or IT with effect size statistics, likelihood-based or Bayesian are proper to experimental or exploratory data such as analyzing major factors of change of population (Descriptive or exploratory analyses). IT or Bayesian is reasonable to experimental or observational data such as predicting best habitat for certain species (Fitting predictive models). Only IT is appropriate to challenging research hypotheses which data is rigorous experimental data enabling clear assessment of binary predicted results.