Director’s Blog March 10, 2019
By Ian Billick, PhD

Thanks to a recommendation by Dan Blumstein I recently read “The Book of Why” by Judeah Pearl. It helped me understand both the growing diversity of research occurring at RMBL and helped me think about the general value of place-based research. It also explained why some of the statistics I was taught, and in turn taught, was not helpful. Too much of a focus on the “how” of statistics rather than the “why”. The “why” matters in deep ways.

Pearl made his career in artificial intelligence and the insights are the result of having to understand why computer scientists were/are getting stuck on teaching computers to think. They established (mathematically as well as conceptually) a ladder of causation going from seeing (simple correlation structure which is where most machine learning is today), doing (causal diagrams of which randomized control experiments are a subset), and imagining (think mechanistic models that allow you to probe worlds which may never be).

The book was helpful in understanding some of the methodological tensions we have seen on RMBL’s research committee. To over generalize, classic field ecology (though not so much field evolutionary biology) as funded through NSF and conducted at RMBL has been very focused on using field experiments for hypothesis testing. With an increasing diversity of types of field science (notice I am not using the term ecology, evolutionary biology, or even field biology) being conducted at RMBL, we are seeing more model-based science, including strictly observational studies as well as the use of experiments to understand systems rather than test hypotheses. Coming from a physics background, I appreciate a diversity of approaches. Now I understand better how they formally relate to each other.

The book also deepens my understanding of the ideas Mary Price and I developed in the Ecology of Place as well as some work I did early in my career in modeling indirect effects in community ecology. I hadn’t realized those problems were connected. Wow!

Two insights to whet your appetite in case you are interested in this kind of thing (and there are RMBL scientists already well down these pathways, literally):
1. With the help of causal diagrams you can infer causation from correlation. Sewall Wright and his path diagrams started us down this road, but Fisherian statistics stalled us out. For long-time RMBL’ers, there are some nice references to Ted Porter and his work on Karl Pearson (Pearson correlation coefficient), a historian of science who spent time in Gothic.
2. Not only can you get causation without randomized controlled experiments, but inferring causation from randomized controlled experiments can be problematic, for example with experiments involving multiple
manipulated variables or when chains of causation are involved (e.g., smoking causes cancer because it causes tar build in in the lungs and potentially other mechanisms).

Several caveats if you decide to read the book. It’s written for a general audience, but not really. It’s dense and not easy to follow. I had to reread passages and fill
in logic on my own. I’d highly recommend this book for early career scientists, but for many scientists, especially more senior scientists that have established how they are successful, the book won’t matter-much in the way that Newtonian physics is simpler than Einstein’s work and is often good enough. It may not be worth the effort.

Finally, the book includes ideas that seem very obvious. But the value in the work is the overall framework, how multiple concepts, obvious and not obvious, are merged to create a general inferential framework that leads somewhere new. If you read it, let me know what you think. Especially if you think I’m out in left field!!!