Quite some time ago I blogged about a ‘new’ book published by Oxford University Press and edited by Navjot Sodhi and Paul Ehrlich called *Conservation Biology for All *in which Barry Brook and I wrote a chapter entitled *The conservation biologist’s toolbox – principles for the design and analysis of conservation studies*.

More recently, I attended the 2010 International Meeting of the Association for Tropical Biology and Conservation (ATBC) in Bali where I gave a 30-minute talk about the chapter, and I was overwhelmed with positive responses from the audience. The only problem was that 30 minutes wasn’t even remotely long enough to talk about all the topics we covered in the chapter, and I had to skip over a lot of material.

So…, I’ve blogged about the book, and now I thought I’d blog about the chapter.

The topics we cover are varied, but we really only deal with the ‘biological’ part of conservation biology, even though the field incorporates many other disciplines. Indeed, we write:

“Conservation biology” is an integrative branch of biological science in its own right; yet, it borrows from most disciplines in ecology and Earth systems science; it also embraces genetics, dabbles in physiology and links to veterinary science and human medicine. It is also a mathematical science because nearly all measures are quantified and must be analyzed mathematically to tease out pattern from chaos; probability theory is one of the dominant mathematical disciplines conservation biologists regularly use. As rapid human-induced global climate change becomes one of the principal concerns for all biologists charged with securing and restoring biodiversity, climatology is now playing a greater role. Conservation biology is also a social science, touching on everything from anthropology, psychology, sociology, environmental policy, geography, political science, and resource management. Because conservation biology deals primarily with conserving life in the face of anthropogenically induced changes to the biosphere, it also contains an element of economic decision making.”

And we didn’t really cover any issues in the discipline of conservation planning (that is a big topic indeed and a good starting point for this can be found by perusing The Ecology Centre‘s website). So what did we cover? The following main headings give the general flavour:

- Measuring and comparing biodiversity
- Mensurative and manipulative experimental design
- Abundance time series
- Predicting risk
- Genetic Principles and tools

The first section covers biodiversity indices, ecological scale, biological surrogates, similarity/clustering techniques & multivariate approaches. The second deals with hypothesis testing, sample size issues, replication and control, and random sampling. The fourth section covers cross-taxa approaches and population viability analyses.

We also included a number of ‘boxes’ by us and other authors:

- Cost-effectiveness of biodiversity monitoring
- Working across cultures
- Multiple working hypotheses
- Bayesian inference
- Functional genetics & genomics
- Useful textbook guides

Yes, it’s a lot to cover but what I really want to highlight here is something that continues to distress me – why are the majority of conservation biologists (and other conservation scientists) still holding on stubbornly to an archaic, sub-standard, incomplete and often misunderstood statistical paradigm to provide some semblance of objectivity to their observations? I’m talking about the Neyman–Pearson Null Hypothesis Testing (NHT) paradigm where a single ‘null’ hypothesis is rejected or not, based on an arbitrary probability of observing the metric of choice as extreme as the one observed.

It’s important to remember that the hallowed ‘P’ value in the NHT paradigm simply refers to an arbitrary threshold below which we say there is a relationship (i.e., reject the null hypothesis). Let me wax lyrical a little on this little P value to which so many biologists cling desperately – a value of 0.05 (1 chance in 20) has no inherent meaning *per se*, and is in fact a holdover from the days when statistical tables had to be printed in the back of textbooks. There was traditionally insufficient space to write all manually calculated rejection probabilities for distribution-specific NH tests, so they were often truncated at 0.05. Why therefore, in the age of advanced computing do we still lean on this broken crutch as if it actually meant something? Why 0.05 and not 0.04 or 0.06?

NHT approaches have many other problems. They DO NOT nor CANNOT simultaneously consider other dimensions of the problem, namely, evaluating the relative statistical support for an alternative model. Neither can NH tests evaluate Type II errors (the probability of making an error when failing to reject the null hypothesis).

If you’ve always been confused by the non-intuitive language of the NHT paradigm (e.g., failing to reject the null…, etc.), really want to evaluate multiple models simultaneously, and seek to compare particular models based on relative (bias-corrected) statistical support, then you should be wholeheartedly embracing an approach that’s been around since 1890 – *multiple working hypotheses *(MWH).

Instead of considering a single (null) hypothesis and testing whether the data can falsify it in favour of some alternative (which is not directly tested), MWH does not restrict the number of models considered. In fact, MHW can specifically accommodate the simultaneous comparison of hypotheses in systems where it is common to find multiple factors influencing the observations made (sounds like most questions in conservation biology to me).

The basic approach is to construct models (mathematical abstractions of complex systems) that represent combinations of hypotheses constructed to explain variation in the metric of interest. Models are then ranked based on relative evidential support using methods that tend to reinforce the principle of parsimony (the simplest combination of factors providing the strongest explanatory power) via their bias correction terms. Many people have heard of Akaike’s or the Bayesian information criterion (AIC or BIC), and these are probably some of the more common ways to compare models. Obtaining bias-corrected model weights even allows the construction of the *evidence ratio*, which is the relative bias-corrected statistical evidence of one model compared to another (which, as mentioned, NHT cannot do).

I would go so far as to say that there are absolutely no situations in conservation biology where classic NHT ‘significance’ tests are justified – we have much better techniques now. I will admit though that simulation using resampling can provide probabilities of deriving the pattern (metric, etc.) at random, but in these cases the ‘P’ value actually has a specific meaning that doesn’t convolute Type I and II errors as do NHT approaches. Some good references that provide the gory detail our chapter just didn’t have the space to cover include:

- Burnham & Anderson (2002)
- Burnham & Anderson (2004)
- Link & Barker (2006)
- Elliott & Brook (2007)
- Lukacs et al. (2007)

One last word on this issue. The above-mentioned methods can only provide information on the *strength of statistical evidence *for a pattern or relationship – they do not tell us anything about the *strength *or *magnitude *of the effect. You really should be talking about the amount of variation in the response (the thing you’re trying to explain) each model (and each component variable therein) describes. Nothing irks me more (philosophically) when I read something along the lines of: “…we found a strong relationship between *x *and *y* (P < 0.01)” – the ‘P’ here says NOTHING of the relationship’s strength!

Well, that’s my little statistical diatribe out of the way – hopefully I’m not either just preaching to the converted or having my words fall on deaf ears. This is extremely important stuff because I firmly believe it makes an important difference in the magnitude and even direction of reported trends and patterns in conservation.

Want a copy of the chapter? Send me a message using the available form and I’ll email you a PDF copy.

Time to put significance out of its misery | ConservationBytes.com(21:59:32) :[…] rationale as to why we should also ignore the word’s statistical meaning. While I’ve explained this before, it bears […]

LikeLike

perissodactyla(17:19:41) :Reblogged this on Perissodactyla and commented:

My list of the problem with a P-centric world – 10 disadvantage sof null-hypothisis testing:

1. The alternative hypotheis is not evaluated

2. The null hypothesis is uninformative, often ridiculous e.g., males and females are different – duh!

3. It assumes asymptotic distributions

4. It assumes thresholds of ‘evidence’, but

5. A P-value is not evidential

6. A P-value is not a strength of evidence

7. The world is not explained by single models

8. A null-hypothesis test has no ability to add and quantify model selection uncertainty, and

9. No ability to deal with large systems or datasets, and lastly

10. It has proven to be unuseful in a court of law!

LikeLike

CJAB(15:34:21) :This chapter (as well as all others in the book) is now available free to download here.

LikeLike

No (statistical) warming since 1995? Wrong « BraveNewClimate(20:25:27) :[…] Alternatively, and more sensibly, we can fit two models: a ‘null’ with no slope, and a TREND model with a slope, and then compare how well they fit the data (after bias corrections). A useful way to do this comparison is via the Akaike Information Criterion – in particular, the AICc evidence ratio (ER). The ER is the model probability of slope model divided by that of the intercept-only model, and is, in concept, akin to Bayesian odds ratios. The ER is preferable to a classic null-hypothesis significance test because the likelihood of the alternative model is explicitly evaluated (not just the null). Read more about it in this free chapter that Corey Bradshaw and I wrote. […]

LikeLike

They always whinge about the maths « ConservationBytes.com(03:02:28) :[…] rising negative influences on biodiversity, to a more detailed synopsis of our chapter, The Conservation Biologist’s Toolbox, and I’ve reproduced a review printed in Trends in Ecology and […]

LikeLike

Joel Snodgrass(03:32:34) :Corey,

This is truly an uphill battle that takes most students by surprise as they move through grad school. Please send a pdf of the chapter; might be ideal additional reading for conservation biology class. Thanks-Joel

LikeLike

CJAB(10:25:28) :Agreed, Joel. The problem is that most conservation biology programs (or general ecology) have insufficient prereqs for mathematics, so the students aren’t prepared from the outset. Mathematics needs to be placed front and centre in the undergraduate and even high school years.

LikeLike

Scientia Pro Publica: Answers to 28 popular and not-so-popular questions(22:03:35) :[…] What toolkit does conservation biology have to study and hopefully intervene in all this […]

LikeLike

Francisco Viddi(12:34:43) :Hi Corey,

great article, as many other articles I have been reading in your blog.

It was great to see you at the conference in Edmonton, I hope all is well in Aussie.

I was wondering if I could have a PDF copy of your chapter “The conservation biologist’s toolbox – principles for the design and analysis of conservation studies”

Thanks!

Chico

LikeLike

CJAB(13:07:11) :Absolutely, Chico. It’s in the e-mail. Thanks.

LikeLike

Tom Keen(00:16:07) :What an utterly interesting article to read, particularly on the same day as sitting through a(nother) 2 hour tutorial on these methods (not including MWH).

I won’t, as an undergrad, neglect the relative importance of learning the traditional classroom statistical methods (though you may disagree), but I’m glad I’m reading this now. If MWH are as useful as you say, I’m going to learn this stuff so I can use it.

Cheers Corey, this site really is an invaluable resource.

Oh, and could I also grab a copy of that PDF? T.A.!

LikeLike

CJAB(09:28:02) :Thanks, Tom. PDF on its way. Yes, my advice is to embrace MWH as soon as possible, and get a good training in Bayesian methods as well.

LikeLike

Claire Treilibs(13:17:35) :Hi Corey,

I found your ‘diatribe’ very useful! Would love a pdf copy of the chapter if you’re offering.

Thanks,

Claire

LikeLike

CJAB(13:23:05) :Dear Claire,

Do come and see me to discuss options – always open to potential PhD ideas.

PDF is ‘in the mail’.

LikeLike