The New Statistics

What's new in ecological statistics?

Not so long ago, biostatistics textbooks were entirely devoted to hypothesis testing and every paper published in ecological journals had fists-full of p-values.

But alternatives to null hypothesis significance testing (NHST) have emerged and are increasingly being integrated into mainstream ecology (see What's wrong with NHST). In a recent (January 2006) special section in Ecological Applications, these were dubbed "new statistics" and editors are clearly seeking to promote these approaches.

The following are key features of the "new" approaches:

• a focus on estimating effect sizes and selecting suitable models for ecological processes;
   
• explicit treatment of uncertainty, rather than categorizing propositions as 'true' or 'false';
   
• use of computationally-intensive numerical methods, which were not possible before the advent of modern computers and the necessary software;
   
• in wildlife studies, recognition that not all animals present are detected, and that detection probability can be estimated and must be taken into account.

Most of the principles involved are straightforward and have a long history. The theory of likelihood was developed by R. A. Fisher in the 1920's, while Bayes' Theorem dates to the late 18th century. Hypothesis testing came to dominate biometrics only in the latter part of the 20th century, so 'traditional statistics' is a misnomer: maybe 'frequentist' is a better term.

The concept of likelihood is central to these new approaches and is used in Bayesian methods. In fact, it underpins much of frequentist statistics. However, most standard frequentist methods do not involve explicit consideration of likelihoods and it barely appears in older textbooks, because calculating likelihoods for large data sets was not feasible before computers became ubiquitous.

Butcher et al (2007) drew attention to the narrowness of existing academic courses, which often focus exclusively on NHST, and they ask that "academic supervisors...become familiar with the statistical approaches available to wildlife scientists and ecologists...".