Note: This post is old wine in a new bottle. It basically repeats some old posts, just in a slightly different way. I’m only doing it because the comment threads on those old posts are really good, but I felt like they petered out a bit too soon.* This is my attempt to revive them. So if you’ve been reading my old posts on macroecology and the associated comments and thinking “More please!”, this post is for you. But if, as is more likely, you’ve been reading those old posts and thinking, “Jeebus, doesn’t he have anything new to say?”, just skip this one.
Note: Man, this post ended up way longer than I originally intended! Sorry about that. Maybe everybody should skip it. Basically, all the post does is argue that macroecology is not at all like astronomy, except in a couple of superficial ways. If you really care why I argue that, and can’t guess it from having read old posts, read on…
I write a lot on this blog about how difficult it is to infer process from pattern, mechanism from observation, and causation from correlation, even tentatively. In response, ‘macroecological’ colleagues, whose work emphasizes the documentation and interpretation of observed ‘large-scale’ ecological patterns, often point to the example of astronomy. The example of astronomy, they say, illustrates how observational science can be successful science, and even causation-inferring science.
But unless I’ve missed it (which is quite possible), they never keep following that line of thought, at least not in any detail. I wish they would. Yes, absolutely, astronomy is a successful, causation-inferring science, and it does so without being able to manipulate stars or galaxies or whatever. But precisely how does it achieve its successes? After all, there are also unsuccessful observational sciences. Macroeconomists, for instance, infamously remain in vociferous disagreement on even the most basic points. So what makes astronomy successful, and can macroecology emulate it? I emphasize that in asking this question, I really don’t know the answer and I’m genuinely curious. I’ve asked this question in the comments on previous posts, but never gotten a reply. So I decided to do a post on it, in the hopes of smoking out Brian McGill and Ethan White’s inner astronomers. 😉
Just so I don’t come off as totally lazy (“Hey readers, teach me astronomy!”), I did do a bit of background research**, the fruits of which are below.
Origins of the macroecology-astronomy analogy
The macroecology-astronomy analogy seems to originate from a comment by Robert MacArthur to Jim Brown, reported in Brown’s Macroecology (p. 21):
“Astronomy was a respected, rigorous science long before ecology was, but Copernicus and Galileo never moved a star.”
Brown goes on to cite geology, specifically the theory of plate tectonics, as a second example. That mechanistic theory is well-confirmed even though geologists have never experimentally manipulated entire tectonic plates or the mantle plumes that move them around.
Brian Maurer, in his Untangling Ecological Complexity: The Macroscopic Perspective (p. 112-113) pursues this analogy a little bit further (in fact, as far as I’ve ever seen it pursued):
“Explaining these patterns [in assemblages of species] is difficult because the standard tools of science for developing mechanistic explanations cannot be used. With few exceptions, there is no way to manipulate the biodiversity of a large geographic region. Another problem is that often the same pattern might be consistent with more than one process…[These limitations] also apply to astronomy, but this has not prevented astronomers from learning a great deal about stars and galaxies. Astronomy has a strong, quantitative theoretical foundation in physics, and this foundation allows fairly precise predictions to be made about what patterns can be expected in complex systems like galaxies.”
I can totally see why early macroecological texts pushed this analogy. It’s an effective response to anyone who would claim that, if you’re not doing manipulative experiments, you can’t possibly be doing science. But while the analogy establishes the possibility of a successful, astronomy-like macroecology, I don’t know that it establishes the actuality. Indeed, based on what little I know about astronomy, I’m skeptical of the ability of macroecology to emulate astronomy. I don’t think this means macroecology can’t be successful or infer causality–but I do think its methods aren’t really analogous to those used in astronomy. But these thoughts are tentative. I hope that people who know more than I do about astronomy (or macroecology!) will chime in.
I emphasize that I’m not actually saying that any macroecologist takes the astronomy analogy as seriously as I’m about to try to take it. Nor am I saying that they should have done so–I’m not accusing them of failing to follow their own argument to its logical conclusion or anything like that. I’m actually not out to criticize their use of the analogy at all. I’m just intrigued and curious. I want to see what we can learn by pushing the analogy as far as it will go, even if that means pushing it further than macroecologists have ever taken it.
Reasons why macroecology is not like astronomy
1. Ecologists can do experiments! Ok, it’s true that you can’t manipulate, say, the biodiversity of an entire continent. But you certainly can do smaller-scale experiments that are directly relevant to interpreting large-scale patterns. One example that I’ve raised before, and which I’ll raise again because it’s such a great example***, is the work of Jon Shurin on local-regional richness relationships in freshwater zooplankton communities (Shurin 2000, Shurin et al. 2000). After correcting for an artifact to do with variation in spatial extent of different regions, lake zooplankton appear to exhibit a linear relationship between local (within-lake) species richness, and regional richness (total richness of all lakes in a large region). This has been taken to indicate that local diversity just reflects dispersal limitation, a sort of “passive sampling” from the regional “species pool”, meaning that local communities are open to colonization by whatever species happen to arrive. But here’s the thing: local zooplankton communities aren’t open. You can directly test that by trying to invade lakes with species not currently present. It turns out that invasions mostly fail, unless you first drastically reduce the densities of resident species, thereby eliminating competition from residents. Far from being open to whatever colonists happen to arrive, lakes are almost completely “saturated”. This means that linear local-regional richness relationships had been misinterpreted. As is now well-established, linear local-regional richness relationships are one of those patterns that are, as Maurer says, “consistent with more than one process”. Small-scale experiments like Jon’s were key to establishing that point in the context of local-regional richness relationships.
So macroecologists can’t “manipulate the stars” or “manipulate tectonic plates”–but they certainly can do experiments that provide information directly relevant to the macroecological equivalents of the heliocentric theory or continental drift. In this respect, macroecologists are actually in a better position than astronomers or geologists when it comes to inferring causality. They have more weapons in their arsenal.
This is one way in which I think the analogy with astronomy might actually be holding macroecology back a bit. Their emphasis on the impossibility of large-scale, “manipulate the stars”-type experiments sometimes seems to cause them to downplay the relevance of other sorts of manipulations.**** Just because the pattern is large-scale doesn’t mean the only relevant experiments are large-scale. The underlying processes that putatively generated the pattern typically operate everywhere, at all times, and so can be tested for anywhere, at any time.
As a second example, Jon Levine and others have powerfully combined small-scale experiments with larger-scale observations to show that the large-scale correlation between native and non-native species richness is down to the fact that the same environmental conditions that promote native diversity promote non-native diversity (e.g., high rates of propagule supply). This effect swamps the fact that, all else being equal, more species-rich communities are more resistant to colonization by species not already present (Levine 2000, Levine 2001 Oikos)
Now in fairness, Ethan White has argued in the comments on old posts that all the best macroecology actually recognizes this and is based on synthesis of all relevant information, including small-scale experiments. I wish I fully shared his confidence that this kind of work is what everybody is out to do, at least ideally. I’m torn between taking his word for it (since he knows the literature far better than me), and my own admittedly-limited experience as a reviewer of macroecological papers which too often neglect directly-relevant experimental work.
2. Astronomy is based on the quantitative estimation of different effects, using well-developed and -validated physical theory. My point here is basically an expansion on Brian Maurer’s brief remark about astronomy’s “strong, quantitative foundation in theoretical physics.” What that foundation allows astronomers to do is to estimate and subtract out from their observations the effects of all sorts of “nuisance” factors and sources of error, leaving them with accurate, precise estimates of the quantities of interest. For instance, see here, here, and here for layman-level discussion of how “transits”, such as the recent transit of Venus between the Sun and the Earth, allow astronomers to accurately and precisely estimate quantities like the mass of the Sun, the absolute (not just relative) distances from the Earth to other objects in the solar system, and the chemical composition of the atmospheres of other planets. See here for the estimation of stellar parallax (a very subtle effect). And see here for discussion of how subtle deviations in the orbits of planets from what would be predicted based on Newtonian mechanics and the masses of known planets led to the discovery of new planets. Note in every case that getting a good estimate does not involve just accumulating a big sample size and then averaging away the “noise”, thereby allowing the desired “signal” to reveal itself. Rather, getting a good estimate involves using well-established,quantitative background knowledge to precisely quantify, say, the Doppler shift in the spectrogram of the atmosphere of Venus due to the Earth’s movement around the Sun.
Macroecologists certainly try to do this sort of thing. They often include covariates in their statistical analyses to statistically control particular sources of variation, they can use comparisons among different datasets to estimate error sources known to affect only some of those datasets, and they can use randomization-based “null models” to try to figure out what their data would look like in the absence of some particular process like interspecific competition. But are those approaches anything like as precise and well-validated as what astronomers do? Indeed, in the case of randomization-based null models, there are reasonable arguments that they don’t, and can’t, work at all.
Further, macroecologists often deny that we could ever have fully-parameterized models of the “microscale” processes of birth, death, and dispersal that ultimately drive species distribution and abundance. I actually don’t think that’s true, at least not universally, but let’s say for the sake of argument that it is. Doesn’t that amount to a denial that we’ll ever have a “strong, quantitative foundation” that would allow us to estimate and subtract out “nuisance” effects, thereby allowing us to precisely estimate effects of interest? For instance, if you think that it’s impossible to parameterize a many-species competition model for a large, spatially-heterogeneous area, what makes you so sure that just randomizing your observed species x sites matrix while holding the row and column totals constant subtracts out all effects of interspecific competition while leaving effects of all other processes intact? Isn’t that basically like an astronomer saying “I don’t know how to quantify the Doppler shift in my spectrogram, so I’ll just randomize my data with respect to the day of observation and hope that that fixes it?” Doesn’t this lack of what economists call “microfoundations” make macroecology more like macroeconomics than astronomy?
3. Astronomy isn’t really about statistical patterns. In a classic Oikos paper I discussed in one of my first posts, John Lawton (1999) argues that macroecological patterns reflect the fact that, at large scales, the “noise” of species- and system-specific details “averages out”. There are reasons to question whether that’s a good analogy for macroecology, but let’s accept it for the sake of argument. Here’s my question: is astronomy like that? I mean, when astronomers estimate the properties of some object in outer space–its distance from us, its chemical composition, its size, etc.–they’re estimating the properties of that particular object. Yes, they do that repeatedly for lots of objects, and I’m sure there are statistical patterns in the resulting data. For instance, maybe the “species-abundance” distribution of different kinds of stars has some sort of interesting shape? But it’s my impression that most (all?) of the vaunted quantitative, cause-inferring rigor of astronomy comes into play in estimating the properties of individual objects, not in studying the statistical features of collections of objects. Basically, I’m suggesting that, if statistical mechanics is your model for what macroecology is like (as both John Lawton and, in other passages in his book, Brian Maurer suggest), then astronomy is not. Astronomy and statistical mechanics are very different in terms of what they’re aiming to do and how they’re aiming to do it. In his paper, John imagines a fairy who foolishly tries to understand the intractable random movements of individual particles of a gas, totally missing the tractable macroscopic properties of large collections of gas particles. But predicting the highly non-random movements of individual “particles”–planets, say, or comets–is what astronomers live for!
Am I totally wrong about all this? Maybe I’m just focusing on the wrong bits of astronomy, and other bits actually are a great model for how macroecology works, or could work? Or maybe I’ve just shown that the whole macroecology-astronomy analogy can’t be pushed any further than the very short distance that Brown and Maurer push it? (Woohoo! I’ve saved macroecology from a shaky analogy no one would’ve thought of if I hadn’t suggested it!) Or in pushing the analogy further, have we actually highlighted some things that needed highlighting, such as the ability of macroecology to draw on small-scale experimental data? I’m honestly not sure–you tell me.
And then I promise to shut up about macroecology and talk about something else. 😉
*Also, right now I’m a bit low on both time for really substantive posts, and really new things to write substantive posts about. It’s quicker to repeat myself. 😉
**Specifically, I spent 5 minutes googling “astronomy blog” and reading the top hits. 😉
***I did warn you about old wine in a new bottle.
****It also causes them to forget that you can “manipulate the stars” if you create an artificial universe in which the stars are small enough to be manipulable. That is, you can do experimental macroecology in microcosms. Phil Warren and his collaborators have done a lot of nice work on this (e.g., Holt et al. 2002).