Weekend before last I spent a day in Milan to see an old colleague (I am on leave in Zurich just now so miraculous things like a dayhop to Milan are feasible). He used to be a research technician in my lab where he did the classical ecology lab tasks of things like identifying and tagging trees in forests, making thousands of observations of leaf unfolding in a growth chamber experiment (on paper! Though that part was neither my idea nor his) and nudged the lab forward through working on automating photo capture of leaf unfolding. The resulting images and time lapse videos were gorgeous, but they didn’t change how I did science.
But since then my Bayesian models have become far more generative—many thanks to my Stan collaborators–and I have started to realize some sad hard truths about my data and my science. The first is that you need a lot more data than I often have to fit some of the models I think underlie the processes I am studying. I work on winegrapes because they have way more data than other systems I am interested in (such as forests, which critically store carbon) and when I don’t have enough data to fit a temperature response curve for winegrapes I can skip trying it on more `natural systems.’ The second is that we also need way better data. A stats-department colleague said to me this year, ‘it’s not like you don’t have a lot of data, it’s like the data you have are quarters [and you need higher denominations].’ By later in the day when he next picked up the metaphor, but was now calling my data pennies. (Sigh.)
He’s not wrong. Ecology has a history of valuing what you can learn staring intently at a backyards pond or an artificial pool near Palermo full of water bugs. We need to understand the ‘natural history’ well to understand systems. Very true, but I sometimes wonder how we advance. A massive NSF project to collect lots of large-scale, NEON, has not revolutionized much. For my own research, I think we need to cover greater temperature variation — and work harder to know what happens at the extremes (you have to wait a long time for plants to do something at 5C, but maybe we need to wait for that; at the other end many warming chambers don’t function about 30 C well, but that’s well below what’s too hot for most plants), we need more replicates and we need to collect better data, at finer scales.
Which is part of why I went to Milan. To pick my colleague’s brain about how my lab can best break out. He’s now building up new FabLab for his company’s new North American headquarters in Chicago, and had some useful ideas.
At the end of our meeting he asked if he should go to grad school, which struck me. It struck me for a lot of reasons, one is that some undergrads in my lab, who I think could bridge new technologies to ecology, are getting scooped up by start-ups digitizing agriculture, and putting their undergrad degrees on a possibly never-ending pause.
The same weekend I saw a colleague from my long-lost NCEAS days. In between nerd-crushing/raving about our colleague Jim Regetz, we discussed the apparent disconnect between the number of PhDs being awarded (erm, not sure about that verb) and the number of job openings where a PhD is critical. Some of his former postdocs were starting a new company, trying to make theoretical ecological models more useful to the point of underpinning a for-profit company.
I imagine academia often feels to be falling behind, but this weekend I felt it a little more acutely. We’re supposed to have the freedom and metaphorical space to be racing ahead. But it doesn’t feel that way when I can easily see why students in my lab would ‘pause’ undergrad to race around North America to improve how we harvest wheat, when we churn out publications faster and faster at the expense of the time it takes to really advance science (it’s so much quicker to grab a p-value than to develop a model with parameters you care about, then step back and gape at that the uncertainty around the estimates; p-values are so happy to hide your meaningful parameters and their uncertainty from you), or similarly churn out PhDs without a clear idea of their job prospects (hello Canada’s `HQP’). I am not so worried about folks in my lab, we train strongly in computational methods and how to design and answer useful questions—skills industry and beyond needs, but I worry about the future, and I could certainly train in this area better if there was more pressure, recognition, and support for it in ecology.
On the good news side, I enjoyed my take-out pizza from Milan for two glorious dinners!
Indeed, if you try to do something like model carcinogenesis you soon discover we lack some really basic info (eg, rate of cell division in different tissues at various ages… or even just the number of cells).
That type of info is the “dollar” the science machine takes, it doesn’t take these A is higher than B pennies. If one is sufficiently clever and persistent, it is sometimes possible to convert some pennies into a dollar, but why not just collect the dollars to begin with?
> how my lab can best break out
What does a lab breaking out mean?
If you have undergrad students being picked out of your lab and hired for things that they go on to enjoy, that seems pretty successful in its own terms. But maybe there is some other goal.
I was glad to read “The first is that you need a lot more data than I often have to fit some of the models I think underlie the processes I am studying”: I too often tell colleagues that we need much larger sample sizes to parameterize population models. Of course these ideas will go nowhere until we collect data with drones or robots (which seems within reach for some plants).
But this raises a bigger point: individual-level processes are the foundation of much ecological theory, but it is macroecological studies that seem to gain traction these days. At least some of this increased interest comes from the abundance of distribution data. This is a problem – that is, an opportunity once we perfect automated data collection! – for the whole discipline.