Author Archives: Mick Keough

Oddities with violin plots

A word of caution about violin plots, that some of you might have encountered already. After casually looking at these (horizontal!) violins to compare their shapes, it struck me that they were misleading – the plots are of reproductive output, and the violins extend into negative values:

Often, we wouldn’t bother checking the scales, if we’re comfortable with the data values. In this case we should, given the impossibility of negative reproductive values.

This issue has been noted before, and it’s because the violins are kernel density estimates, rather than real data values. In a case like this, with negative values impossible, it may be better to use a density plot, i.e. show the actual data. This will work- as long as there aren’t too many data points.

Truncating the scale at 0 isn’t appropriate, because the kernel estimate is independent of the graph scales.

Multivariate analysis with R or Primer?

Interesting post from Marti Anderson explaining similarities and differences between adonis in R and the Primer + Permanova package (beyond cost 😊).

Her take-home message: R and Primer are both great, as long as you think about what you’re doing and understand what your R package is actually doing. We could hardly disagree with that!

Importantly, Marti describes important differences when fitting mixed models to multivariate data sets, with, at the moment, only Primer fitting mixed models appropriately.

More reading…

We recommend several authors for keeping your perspectives broad, including Andrew Gelman and his colleagues. He and Aki Vehtari have just released a companion to their Regression and Other Stories, featuring classroom exercises, stories and demonstrations, practice questions, etc. Active Stories is self-described as “a perfect teaching aid, a perfect adjunct to a self-study program, and an enjoyable bedside read if you already know some statistics”, and Gelman describes it as lots of fun to write. We haven’t read it yet, but we will 😉

If you’re an ecologist, this might be interesting…

Effective EcologySeeking Success in a Hard Science, by Roger Cousens and others, has just been published by Taylor and Francis.

Ecology is one of the most challenging of sciences, with unambiguous knowledge much harder to achieve than it might seem. But it is also one of the most important sciences for the future health of our planet. It is vital that our efforts are as effective as possible at achieving our desired outcomes. This book is intended to help individual ecologists to develop a better vision for their ecology – and the way they can best contribute to science. 
The central premise is that to advance ecology effectively as a discipline, ecologists need to be able to establish conclusive answers to key questions rather than merely proposing plausible explanations for mundane observations. Ecologists need clear and honest understanding of how we have come to do things the way we do them now, the limitations of our approaches, our goals for the future and how we may need to change our approaches if we are to maintain or enhance our relevance and credibility. Readers are taken through examples to show what a critical appraisal can reveal and how this approach can benefit ecology if it is applied more routinely.
Ecological systems are notable for their complexity and their variability. Ecology is, as indicated by the title of this book, a truly difficult science. Ecologists have achieved a great deal, but they can do better. This book aims to encourage early-career researchers to be realistic about their expectations: to question everything, not to take everything for granted, and to make up their own minds.

From the publisher

Two interesting aspects (well, at least one):

  • It’s an interesting publishing model for ECRs and graduate students, with the book available through Routledge but also available freely in bookdown format here. Go for it – it’s interesting!
  • We have a chapter on common data analysis issues