What can stable isotope ecologists learn from election polling?

We are preparing for an election here in Ireland—we’ll be voting for our 10th president this day next week—and the pre-election polls, now synonymous with major elections, are landing fast and frequently. Thankfully, the Irish presidential election is not quite the ordeal that the US one is, and the polling is not quite as frantic.

My decade in Canada coincided with several US election cycles, and like many Canadians, I spent far too long focusing on what was likely to happen south of our border. I was one of many people who tuned in to Nate Silver, his 538 website, and the accompanying podcast for information. Preaching to the statistically minded, Nate and colleagues frequently discussed how reliable each new regional poll was and how it accounted for the ‘priors’…

‘Priors’ was a term I had heard in my own world—not as a true statistician, but as someone who could feign at least some understanding of statistical modeling to convince my academic colleagues that I knew what I was talking about (we’ll save the blog on imposter syndrome for another day). As an isotope ecologist, I knew of ‘informative priors’ as something I could include in mixing models to improve their accuracy. But how it worked in practice? Your guess was as good as mine!

Learning more about election polling taught me that Nate was actually speaking to me on a number of different levels (scary, I know!).

This figure, lifted with thanks from The Economist, is a nice way to explain it:

As researchers, our ‘new data’ come from a sample. We have done all we can to make that sample as representative of the whole population (of fish or voters) as possible—but it’s still a sample, and still has biases. The ‘Prior’ is what we know about these biases: have I sampled fish that predominantly use a specific habitat? Or am I polling people in an East-Coast college town that predominantly skews toward one political party?

If I want to relate my sample estimate to the full population (fish or voters), I need to combine these two pieces of information. In other words, I’m viewing my sample data through the informed lens of my ‘prior’ knowledge. The mathematics behind how that works is, sadly, still beyond me—but thankfully, computer processing power is here to lift that load.

So, the next time you’re asked how confident you are in the interpretation of your stable isotope data, remember that it can be every bit as reliable as election polls. And if that doesn’t convince you that it’s worth working with an expert, I don’t know what will!

The importance of ‘informative’ and ‘null’ priors, and how to incorporate them into models, will be covered in my upcoming course, Stable Isotope Data Analysis Through R. Follow me online or join my mailing list to stay up to date.