Earlier this month I finally got to attend the ASN standalone meeting in Asilomar, which I’ve been dying to attend since they started doing them. This was the first time the meeting didn’t overlap with the first week of the Calgary winter term. It was a great meeting, both fun and productive. Here are a bunch of hopefully-interesting thoughts:
I had several really interesting conversations:
- One was about whether debates at scientific meetings serve any scientific purpose, as opposed to being (at best) mere entertainment and (at worst) encouraging ecologists’ over-developed tendencies to get into polarized unproductive arguments. I think debates can serve a useful purpose: clarifying the key issues for bystanders. But for debates to serve that purpose, I think you need the right debaters–people who are happy to engage in scientific give-and-take without taking it personally. In support of this view, I submit the first ASN standalone meeting debate, which I hear was great and which led to a great and well-cited pair of papers. And no, I don’t think “find the right debaters” is such a difficult task as to be effectively impossible. But there is definitely scope for reasonable disagreement with me on this!
- Another was about when it’s useful to revisit the foundations of some line of research, with a view towards rebuilding it from the ground up if necessary. Obviously, we’d never get anywhere if we were constantly setting aside everything we think we know and questioning the framing of every question we’re asking. But equally obviously, the optimal frequency of revisiting the foundations isn’t zero either. So are there any broadly-applicable “rules of thumb” for scientists to recognize when they need to go back to square one? One possibility: if well-informed outsiders from “adjacent” fields have fundamental questions about your research program that you can’t answer to their satisfaction, that’s one sign that maybe your research program needs to go back to square one. Part of this conversation was about the use and misuse of simple “baseline” or “null” models to identify and frame good questions. In evolutionary biology in particular, a common way to identify a question is to set up some simple evolutionary model (e.g., an optimization model) as a baseline, and then treat observed deviations from that simple baseline expectation as “puzzles” that need to be explained. The most common criticism of that approach is that it’s strawmanning, that the solutions to the purported puzzles are too numerous or obvious. But a deeper problem might be that the approach sometimes leads, not to questions with overly numerous or overly obvious answers, but to questions that have no answers at all because they don’t make sense. Because the organisms aren’t actually solving the optimization problem that the researchers assume they’re solving. It’s hard to recognize when you’re asking a question that doesn’t make sense, because no data will reveal that your question doesn’t make sense. Because for any data you collect, you’ll be able to spin some post-hoc interpretation of those data in light of the (nonsensical) question you’ve framed. I don’t have any answers here, though I’m confident the answer isn’t “never set up a simple baseline expectation and then seek to explain deviations from it”. I don’t think we can or should completely do without baseline expectations, or completely get away from treating deviations from our baseline expectations as puzzles in need of solving.
- A third interesting conversation was about partitions. I’m a fan of partitions, such as modern coexistence theory, but it wasn’t until this meeting that I’d fully appreciated other people’s deep objections to them. One objection is that partitions draw a boundary between the things they apply to, and the things they don’t apply to. So if you’re interested in the links between things to which some partition applies, and the things to which it doesn’t apply, you might well see that partition as getting in the way.
Speaking of modern coexistence theory…I continue to follow the emerging technical critiques of modern coexistence theory with great interest. I think the critics have some good points, though I disagree with some of them and I don’t know enough math to fully grasp others. I think for those critiques to gain traction among non-experts (who are by far the largest audience for modern coexistence theory, or indeed for any scientific idea), the critics are going to need to (i) make their criticisms more intuitive to non-experts (which is hard), and (ii) show how their criticisms connect to empirical data. After all, widespread interest in modern coexistence theory itself only took off after Chesson (2000) and other papers made the core ideas intuitive to non-experts, and a bunch of people started applying the theory to data. Now that there’s widespread interest in modern coexistence theory, there’s a pre-existing audience for critiques and modifications of it, and for alternative approaches that achieve the same broad goals. I hope the critics will see that pre-existing audience of people interested in modern coexistence theory as an opportunity rather than an obstacle.
I saw many excellent talks. Anurag Agrawal’s naturalist’s award talk was fabulous. I think it can be really difficult for someone who’s done a large body of important research on one topic to summarize it all in a single talk in a satisfying way. As opposed to a way that comes off as superficial because the talk skims over so much ground. Anurag did a great job of combining breadth with depth. Christopher Moore’s talk on modeling mutualisms was fabulous too. The historical introduction was not only creative and entertaining, but really drove home the need for new thinking in this area–a need to which the rest of the talk responded in a concrete, compelling way. Dave Armitage’s talk on species coexistence and range limits was very cool, though I only caught the second half. Stephen Munch did a wonderful job conveying the intuitions behind some difficult ideas in nonlinear forecasting. Rachel Germain showed how to experimentally test for character displacement without needing to measure phenotypic traits (!), applied the approach to plants (!), showed that competitive coevolution in plants alters their competitive abilities rather than leading to character displacement (!), and showed that the competitive coevolution does not show up in the plant traits that everybody ordinarily measures (!). Paulina Arancibia showed lovely protist microcosm experiment results on patch arrangement and metapopulation persistence. I hope the non-microcosmologists in the audience fully appreciated just how creative the physical experimental setup was, and what a massive amount of work it was to conduct. And I saw (and missed) many other excellent presentations, apologies for not listing them all.
The one problem that showed up in a few talks was overly long introductions. Maybe Probably this is just me. Maybe I’m just getting impatient and grouchy as I get older. I increasingly find myself appreciating speakers who get right to the point. In a 15 minute talk, I don’t want several minutes of “big picture” introduction, usually centered on some conceptual diagram illustrating the speaker’s worldview/perspective. The best advertisement for your worldview/perspective is your science, not a conceptual diagram. Nor do I want an extended introduction to some basic concept or fact (well, basic for the audience at the ASN standalone meeting). The best talks start telling the audience members things they don’t know as early in the talk as possible. Protip for speakers to help address this: if one or more talks in your session have already introduced the same key concept/fact/perspective you were planning to start with, take the opportunity to skip that bit of your introduction and say more about your science instead. Related old post.
The poster session was small but very good. I liked how most of the posters were very text-light.
My poster went over well, the folks who came by during the poster session all “got” it and asked good questions. One of the best signs that your presentation was good is that you got good questions.
The main problem with presenting a poster is that you can’t walk around and talk to the other poster presenters about their posters.
The venue is old and charming, the location is lovely, and a small meeting at which everyone eats together really encourages interaction. But the venue’s also showing its age; a bunch of people had to be rehoused because of a boiler failure. Any suggestions for other venues?
Disappointed there wasn’t more of interest in the nearby tidepools. I don’t follow the rocky intertidal literature that closely any more, just because my interests have shifted over time. But I had thought the story was that Pisaster was bouncing back? There were no Pisaster in evidence at Asilomar, hope they return soon. Of course, I’m being greedy here, since I’ve actually seen Pisaster a couple of times before, including once as a tourist near Asilomar. I feel bad for anybody who came to the meeting hoping to check Pisaster off their life list and left disappointed.
My team did reasonably well in the natural history trivia quiz by following the highly effective strategy of “let Peter Morin answer all the questions.” Peter is amazing. 🙂
Dynamic Ecology 
The same principle on getting to the point in introductions applies to papers. More like CCR, less like Pink Floyd please.
We should do a whole post comparing scientific papers to pop and rock songs.
Ha! For once in my life, *I* beat *you* to a post idea: https://scientistseessquirrel.wordpress.com/2019/02/19/what-is-sciences-kokomo/
Heh. I had in mind something more like a taxonomy of papers, each matched to an appropriate pop song. For instance, “Paperback Writer” for overlong papers. (“It’s a thousand pages/Will you take a few?”)
So for Watson and Crick, you could use Surfin’ USA, which the Beach Boys recorded without giving credit to Chuck Berry, whose song Sweet Little Sixteen they ripped off.
Apparently I’m just all about Beach Boys songs, which amuses and mystifies me…
I particularly enjoyed “Because for any data you collect, you’ll be able to spin some post-hoc interpretation of those data in light of the (nonsensical) question you’ve framed”
It’s always a combination of humbling and relieving when you catch your own dodgy post-hoc interpretations in time.
Jeremy, can you explain more about people’s “deep objections” to partitions? Is there any other way to study multiple interactions processes or causes? Or are the objections to particular partitions, rather than to the very concept of partitioning?
I’m not sure, honestly. Some of the objections are to particular partitions. For instance, one worry is that modern coexistence theory draws a line between competitive coexistence (to which it applies), and predator-prey coexistence (to which it doesn’t apply). Thereby interfering with studies of species persistence in food webs apply equally well to competitors, predators, and prey. The argument is that, if you care about the ability of species to increase when rare, well, persistence of predator-prey interactions also depends on their ability to increase when rare. From that perspective, it’s artificial to draw a boundary between “mechanisms that let competitors increase when rare” and “mechanisms that let predators and their prey increase when rare”.
The same worry applies to Vellend’s framework for thinking about community structure within trophic levels. It draws a dividing line between within-trophic level and between-trophic level stuff that some people don’t think should be drawn.
I heard other objections to Chesson’s partition, but some of them sounded to me like objections to any and all partitions rather than to MCT specifically. It’s a bit hard to say because all the discussion was in the context of modern coexistence theory. My own positive feelings about partitions in general are definitely shaped by my fairly substantial experience with several partitions. I wonder a bit if other ecologists’ thinking about partitions in general is overgeneralized from their thinking about whichever one partition they happen to be familiar with.
Interesting. At the end of my book I say that the framework could equally apply to organisms interacting in any way at all (not just on the same trophic level), but that many food web ecologists find that unsatisfying, so sticking to within-trophic-level stuff is enough for now. So, there’s a critique either way, I think. Damned if you do, damned if you don’t.
A further thought: Imagine a world in which within-trophic-level community ecology and among-trophic level community ecology had always been seen as two quite separate fields. I bet that, in that hypothetical world, nobody would have a problem with a partition or other theoretical framework that covered one field but not the other.
That is, I wonder if people object to partitions that seem to draw boundaries where currently no boundaries exist. But don’t object to partitions that reinforce or validate existing boundaries (even though such partitions surely make it harder to cross or erase existing boundaries).
In support of this you sure don’t hear many people object to partitions (I’m thinking simple regression % variance explained partitions now) between biotic and abiotic factors.
To me this example gets to the challenge with partitions – interactions betweeen the parted parts. There are purely biotic factors (so hot or cold you die), but it is the interplay (in statistics interaction term) between biotic and abiotic that really usually matters – sure you can move or use brown fat to keep yourself warm but that takes energy which is scarce due to competition.
Not coincidentally in biotic/abiotic partitions the largest component is usually the part that cannot be identified as solely biotic or abiotic (i.e the interaction part). Which has always made me wonder the point of this particular partition.
“In support of this you sure don’t hear many people object to partitions (I’m thinking simple regression % variance explained…”
Yeah, I was just thinking about that! It’s exactly as you say. Nobody ever complains that ANOVA is a partition of total sum of squares, or complains about the very idea of looking at which predictor variables in a GLM explain which % of variation in the mean of the response variable. But I wonder if that’s because most people don’t think of statistical partitions as partitions at all? At least, not as partitions in the same sense as modern coexistence theory or the Price equation are partitions.
It’s interesting to compare with the Jenny equation in soils:
Soil = f (climate, organisms, relief, parent material, time)
which is really nothing more than a statement that soil is a result of 5 processes or factors – nobody actually tries to write down f. And it has been hugely influential (way more influential than if Jenny had just written a bullet list of 5 factors I think)
That’s how I’ve always thought of partitions as well. As a more formal, mathematically rigorous way of saying X happens as a mix of Y & Z.
I wonder how many people know the same variable can be partitioned multiple ways. And how that would change their perceptions of partitions.
Huh, that’s interesting, I’d never heard of that! And I’m still not quite sure I believe it! You mean, Jenny just used mathematical notation to say “soil results from these 5 things”, and it was hugely influential?!
“I wonder how many people know the same variable can be partitioned multiple ways. And how that would change their perceptions of partitions.”
I think some people do know that, and it makes them think less of partitions. They feel like partitions are just arbitrary. That you can always keep subdividing X into smaller and smaller pieces, or slice it up into whatever number and sizes of pieces you want. I don’t feel that way myself. I think there’s usually a natural most-finely-subdivided partition that you should look at first. Then, depending on your goals, you might want to lump together some of the components of that most-finely-subdivided partition into some coarser-grained partition.
“ou mean, Jenny just used mathematical notation to say “soil results from these 5 things”, and it was hugely influential?!”
Yup – its in every soil textbook (and book that covers soil) on the planet (or >90% anyway).
I think the fact there are multiple partitions bothering people reveals they don’t really understand what is going on or their purpose. Partitions are not some natural fault line in a set of processes that is as big as a the grand canyon. They just say you can think about Z being divided up into X and Y and we can show mathematicallly everything about Z can be placed into X or Y.
This is super weird, I was just thinking about this yesterday. Sorry that I’m late to the party.
I think the biggest problem with partitions is how different terms are interpreted. Of course, this is user error, not an intrinsic fault of partitions.
For example, in coexistence theory, people often split up metapopulation growth rates into a “spatial average” of growth rates over patches, and a “fitness-density covariance” term that roughly captures the relative ability of individuals (of a rare species) to end up in locations where they have high fitness. Most people intuitively think that “fitness-density covariance” has something to do with dispersal dynamics. Indeed it does, but the other “spatial average” term contains the effects of dispersal too; it reflects density-dependent growth rates that themselves depend on the local accumulation of individuals.
Thus, the “spatial average” term captures 1) how spatial heterogeneity affects coexistence in ways that are completely analogous to more familiar temporal coexistence mechanisms (e.g. resource competition, relative nonlinearity, the storage effect) in well-mixed populations, 2) the effects of dispersal on coexistence, and 3) potentially the interaction between 1 and 2.
Examining the “fitness-density covariance”, doesn’t allow you to understand the effects of dispersal on coexistence, because it does not contain all of dispersal’s effects. The “spatial average” terms doesn’t really help you understand the effects of dispersal on coexistence because it contains too much: there’s no clear contrast between a system with dispersal and a system without dispersal.
What have we learned from this pedantic example? Well, if you take a mathematical expression and do the most natural thing – here, using the law of total covariance to partition the density-weighted average of growth rates – you might end up with something that is biologically irrelevant, or at least doesn’t lead to understanding. I’m of the opinion that understanding/explanation doesn’t come from deriving an outcome like coexistence from a model (I’m no philosopher, but I think this is in line with the covering law theory of explanation, with the model serving as “a law of nature”). Rather, I think understanding is about weaving a compelling story about causes and effects; the story about the storage effect and coral reef fishes comes to mind. In the abstract, this often requires making contrasts between previously understood systems and the same system with one factor (i.e. ecological process) “switched-on”. This is not actually because theoretical interventions isolate cause-effect relationships through counterfactual reasoning. It is so that our measly brains can isolate the outcome and work backwards to towards the original causal factor. For example, switching on “environmental variation” in some model may give you the storage effect, and while that does mean that environmental variation causes the storage effect in some ultimate sense, it doesn’t mean you understand the storage effect.
The “fitness-density covariance” and “spatial average” decomposition doesn’t make contrasts to previously understood (classes of) models, which may explain why “fitness-density covariance” as a coexistence mechanism hasn’t gotten a lot of attention. By contrast, Chesson’s 1994 decomposition derives the very popular storage effect by comparing population growth rates to previously well-understood ecological processes, like coexistence from specializing on resources and natural enemies (the “\delta C” term), the general superiority of some competitors, and the generally deleterious effects of environmental variation on average growth rates (both shunted into the “\delta E” term).
Basically, I think the value of a partition depends on how closely the terms correspond to what we perceive as distinct and interesting things. This perception depends explanatory power, which in turn depends on a whole bunch of things, including the state of our understanding. Who knows, maybe if Chesson had developed his theory in the 60s or 70s (whenever Paine discovered keystone predation), he would’ve separated the effects of specializing on resources and the effects of being specialized on by predators. I don’t mean to say that his decision was subjective (see Chesson & Kuang, 2008), but that history plays a role.
Thanks for the great comments! Lots to chew on here.
“The “fitness-density covariance” and “spatial average” decomposition doesn’t make contrasts to previously understood (classes of) models, which may explain why “fitness-density covariance” as a coexistence mechanism hasn’t gotten a lot of attention. By contrast, Chesson’s 1994 decomposition derives the very popular storage effect by comparing population growth rates to previously well-understood ecological processes…Basically, I think the value of a partition depends on how closely the terms correspond to what we perceive as distinct and interesting things. This perception depends explanatory power, which in turn depends on a whole bunch of things, including the state of our understanding.”
I agree, I think that’s very astute. Three anecdotal, non-Chesson illustrations:
1. I have a paper using the Price equation to partition the macroevolutionary drivers of changes in mean mammalian body size during the Paleocene-Eocene Thermal Maximum (PETM; a brief, intense global warming event marking the end of the Paleocene). Mammal at any given temperate site generally got smaller during the PETM. Previous work focused on two non-mutually-exclusive causes: individual species got smaller (e.g., because of selection for smaller body size within species), and small-bodied species that originated in the tropics expanded their ranges into temperate sites and so reduced the average body size at those temperate sites. But nobody had ever done the math to partition the relative importance of those two causes. It just so happens that the Price equation has a term that corresponds exactly to each of those two causes. And the Price equation has a third term that corresponds to a cause previous PETM studies had overlooked: species selection (extinction and/or speciation rates covary with the body size). Along with collaborators from paleontology, we applied the Price equation to some PETM data and partitioned out all three causes of change in mean mammalian body size across the PETM. That paper, Rankin et al. 2015 Proc B, got some of the best reviews I’ve ever gotten (all from paleontologists) and sailed through the review process. I think because the terms in the Price equation mapped onto the question previous researchers had already been asking, and also added something. Our paper basically said “Yes, and” (https://en.wikipedia.org/wiki/Yes,_and…)
2. In contrast, Fox 2006 Ecology applies an extension of the Price equation to partition biodiversity-ecosystem function (BEF) relationships. The terms in that partition map somewhat less well onto the questions BEF researchers were already asking. At a superficial level, it was another “yes, and” situation: two of the terms in the Price equation partition corresponded to mechanisms that previous researchers already cared about (species richness, species composition), plus there was a third term in the partition that previous researchers had overlooked (“context dependence”). But at a deeper level, the Price equation partition defined effects of “species richness” and “species composition” rather differently than any previous researchers had. Some BEF researchers like the way the Price equation carves up effects of biodiversity on ecosystem function. But I know from reviews that other researchers don’t like the way the Price equation carves things up, feeling that its terms don’t map onto or summarize ecological mechanisms that we already care about.
3. One criticism of the Price equation partition in its original domain of microevolution is that it doesn’t correctly partition the effects of selection and transmission bias. Because fitness parameters show up in both the selection term and the transmission bias term. For this reason, Samir Okasha and some others prefer a different version of the Price equation that carves up selection and transmission bias differently, so that fitness parameters only show up in the selection term. (My own view is that which partitioning is preferable depends in part on case-specific empirical details, and is in part a matter of subjective taste…)
Related old post: https://dynamicecology.wordpress.com/2011/07/18/should-we-classify-mechanisms-by-their-causes-or-their-effects/ I think you can loosely divide ecologists into people who prefer to classify mechanisms by their causes, and people who prefer to classify mechanisms by their effects. Partitions like MCT and the Price equation are best thought of as partitions of effects of various unspecified underlying causes, I think (we can also think of those effects as “high level” mechanisms arising from unspecified “low level” mechanisms, but “high level mechanism” is really just another term for “effect”). Partitions of effects tend to catch on when they make everybody happy, because there’s an interpretable mapping from the underlying causes to the effects. Partitions of effects that don’t map neatly onto underlying causes make “effect-focused” people happy, but make “cause-focused” people unhappy.