Sheep, lupines, pattern, and process

Photo: Lupines below Öræfajökull, and sheep grazing at Sandfell, Iceland (S. Heard)

Last summer, we were driving around southern Iceland, admiring the fields of lupines (beautiful, even though they’re invasive) and the gamboling sheep (also invasive, at least to the extent they’re allowed to graze free). Before long, we noticed an interesting pattern: we saw dense fields of lupines, without sheep; and we saw thousands upon thousands of sheep, in fields without lupines – but we drove for days without ever seeing sheep and lupines together.

Being a nerd scientist, I came up with a hypothesis to explain this pattern: lupines are toxic, and sheep avoid them. That hypothesis was lonely, so I came up with another one to keep it company: lupines are delectable, and sheep graze them down wherever they can.

Perhaps you noticed that my two hypotheses assert completely opposite processes. And yet they predict exactly the same pattern: the checkerboard* of sheep and lupines observed separately, never together. At least in ecology, it’s often easy enough to come up with hypothesis sets like this.  The converse is doable too: one process, but predictions of totally opposite patterns (in phylogenetic community ecology, for example, it’s been claimed that competition can generate both phylogenetic clustering and phylogenetic overdispersion).

It’s also why a good number of senior ecologists** crankily proclaim that “you can’t infer process from pattern”. I have been guilty of crankily proclaiming this myself (for example, in this paper). But of course, this proclamation is nonsense. If it were really true that you can’t infer process from pattern, we’d have no science of astronomy, no deep-earth geology, no particle physics, and no phylogenetics (just for starters).

My inclusion of particle physics on that list, by the way, should make it clear that the process/pattern issue is not the same thing as the contrast between experimental and observational data. It’s easy to slip and think it is – to think that observational data are pattern, but experiments measure process. This misapprehension is one reason that many biologists fetishize experiments over observation.   Actually, it’s very common for us to run experiments in which we still observe and interpret a resulting pattern, rather than the process we’re interested in. When high-energy protons collide in accelerators, for example, we don’t observe the collisions themselves, or even the short-lived exotic particles formed in those collisions, but rather the spray patterns of slightly-less-exotic particles arising from cascades of particle decay. In fact, it’s an interesting philosophical question whether we ever observe process directly (but not one I’ll tackle today).

So how do we infer process from pattern, if the world is full of logical booby-traps like my sheep and lupines? It’s both simple and complex: it involves inferring process from pattern with care, identifying patterns that are predicted by one hypothesized process, but by no other. This is far from trivial, because the set of processes we could hypothesize may be very large, and thus it’s hard to know whether a pattern is uniquely predicted by a particular hypothesized process, or whether we just haven’t thought hard enough yet. Such uniquely predicted patterns may rarely be accessed by a single study; instead, they may be compounded outcomes of many different observations and experiments. This is one reason that conclusions in science are tentative, and that confidence in a hypothesis tends to come after the consilience of evidence from many studies and many perspectives (as, for example, with natural selection as the major mechanism of evolution).

So “you can’t infer process from pattern” is just one of those things people like to say because they think it makes them sound rigorous and clever. It’s a slogan. Politicians like to bandy these about, and sometimes, we scientists do too. Real rigour and cleverness don’t lie in slogans; they lie in careful thought that recognizes the complexity of nature.

Oh, and about the sheep. Later we saw a lupine field with a remarkably straight edge – which turned out to be one sheep’s-neck length beyond a fence. Now that’s a pattern predicted by one process (sheep eat lupines) but not the other (lupines poison sheep). So until further evidence accumulates, I’ll infer that sheep love lupines.

© Stephen Heard (sheard@unb.ca) October 26, 2015

Related posts:


*Checkerboard patterns were much discussed in community ecology in the 1970s and 1980s, as evidence (to some) of competition, and as evidence (to others) of sloppy thinking about competition.

**I name no names, but it won’t take much literature searching to find some. Or you can buy me a beer and ask me about my PhD supervisory committee.

Advertisements

16 thoughts on “Sheep, lupines, pattern, and process

  1. Jeremy Fox

    The problem with the argument “of *course* you can infer process from pattern; just look at astronomy!” argument is that it doesn’t go nearly far enough. Astronomers are able to infer process from pattern because of specific features of the data they can collect and the (*very* precise and *very* well-validated) background physical theories of which they can avail themselves:

    https://dynamicecology.wordpress.com/2012/06/07/is-macroecology-like-astronomy/

    I’d say that ecology rarely if ever can be done as astronomy is done.

    I’d also say that an occupational hazard of community ecology in particular is a strong temptation to latch onto unreliable, generally-applicable “off the shelf” methods for inferring process from pattern. Exhibit A: the failed history of attempts to infer interspecific competition from observational data of one sort of another, combined with some vague notion of “limiting similarity” as a background assumption–Hutchinsonian body size ratios; randomization of species x site matrices; indices of phylogenetic community structure…

    Like

    Reply
    1. Jeff Houlahan

      Hi Steve, I’m with you on this one…I always struggle with the “you can’t infer process from pattern” (and its kissing cousin – “correlation doesn’t imply causation”) comments because while they are true they are incomplete. Those comments often seem to imply that there is little or no link between pattern and process (and correlation and cause)…as if pattern and process are essentially uncoupled. This could not be further from the truth. The fact is while pattern may not imply process it is absolutely true that process implies pattern. Pattern follows deductively from process.
      Presumably, the corollary to “process can’t be inferred from pattern” is that process can only be identified through controlled experiments. But, in the vast majority of cases experimental results simply replace one inferential barrier with another. In observational studies, pattern can’t confirm process because of the ‘confound’ problem but in the vast majority of experimental studies there is the extrapolation problem. We are almost always trying to extrapolate from small-scale artificial contexts to patterns that we observe in nature. Why would the process that creates a pattern in a controlled experimental context be the process that causes a roughly similar pattern in nature? I’m not comfortable accepting that small-scale controlled experiments can confirm the processes that are important in nature. Is the confound problem larger than the extrapolation problem? I don’t know.
      I have to confess to knowing little about astronomy and how they get at process but I suspect it is not only about the characteristics of the data or the underlying theory – it is about identifying a reasonably exhaustive list of plausible explanations for a pattern and figuring out where they would make different predictions about related patterns. For example, if we saw a pattern across ecological communities of negative covariance among species there would be several possible explanations (direct interspecific competition, apparent competition, fitness fluctuations as environmental conditions change from year to year). But I suspect creative ecologists could come up with predictions that could distinguish among these explanations. It seems unlikely to me that the process of using pattern to infer process in astronomy is qualitatively different than how we would use it in ecology.
      I think the statement that we can’t infer process from pattern is wrong – truer to say that we can’t confirm process from pattern. It is often very reasonable to infer process from pattern. Doesn’t mean you won’t be wrong – just not as often as if you start from the premise that pattern tells us nothing about process.
      Jeff H

      Like

      Reply
  2. Jeremy Fox

    Following up on myself: Name a case in which ecologists have successfully inferred process from pattern. I actually have a few in mind, but I want to hear your answer–and I’m not going to help you come up with one! 🙂

    No cheating by citing cases in which many different lines of evidence, including both observations and manipulative experiments, were used in combination! And no hedging with that “scientific conclusions are always tentative” stuff! Tentative inferences are a dime a dozen, and “suggestive” observational evidence that’s merely consistent with some hypothesized process is even cheaper. 🙂

    Like

    Reply
    1. ScientistSeesSquirrel Post author

      Well, since you’ve defined all the ways I think we actually make reliable inference in science as “cheating”, a smart man would decline to play the game. 🙂 Now normally, I’d follow a remark like that by playing anyway; but perhaps fortunately I have to go curl. In curling, the scoring is entirely based on pattern, not process; some of my best career shots involve rocks that ended up in the perfect position (pattern) through some never-to-be-repeated fluke of process…

      Like

      Reply
      1. Jeremy Fox

        “Well, since you’ve defined all the ways I think we actually make reliable inference in science as “cheating””

        No I haeven’t, at least not intentionally. For instance, if you hypothesize that pattern X is caused by process Y, and you manipulate process Y and show that pattern X responds to that manipulation in an appropriate way, I think the resulting inference that “process X causes pattern Y” is reliable.

        To help you out, I think there are inferences in population ecology about the processes generating population cycles that are at least pretty close to inferences of process from pattern. We fit alternative process-based models to the population dynamics and find that there’s only one plausible candidate that can fit with data with realistic parameter values. I’m thinking of work from Bruce Kendall, Ed McCauley, Simon Wood, Bill Murdoch, Peter Turchin. Ives et al. 2008 Nature is another good example.

        Sorry, don’t follow the curling analogy…

        Like

        Reply
        1. ScientistSeesSquirrel Post author

          Well, I’m not as subtle as you presume… it wasn’t an analogy, I really couldn’t reply at length because I was headed out the door to go curl. Got a bit cutesy with the rock pattern and I can see why you looked for process behind my writing…

          I’m sure you’re right about the population ecology examples providing pretty good inference (you know that literature much better than I do), except that I’m not sure in any of those cases that we have made solid inference from a single experiment or set of observations. I think it will be exceedingly rare, and possibly non-existent, for an ecological process to be confidently inferred without consilience of evidence from multiple lines of enquiry. (Fisher was even stricter and said that a single significant P-value should never be taken as strong inference). I had offered natural selection in the post as an example, but of course that’s one end of a long continuum; few things are as thoroughly documented as that! I presume that you are looking for the other end of the continuum: processes inferred from one or a few patterns. That is much harder, and I really do conceptualize it as “scientific conclusions start as tentative and gradually achieve more weight as we accumulate evidence from more patterns”. Sorry!

          Liked by 1 person

          Reply
          1. Jeremy Fox

            Ok, I see better what you mean now. But I confess I’m puzzled why you see a connection between “inferring process from pattern” and “making inferences by considering many lines of evidence simultaneously”.

            I agree with you that inferences based on many lines of evidence are stronger than inferences based on just one. And that if all the lines of evidence are observational, this can be a way to strengthen inferences of process from pattern.

            Like

            Reply
    2. Jeff Houlahan

      Jeremy, I would say almost all proposed processes for large-scale patterns have come from observational studies. So, the kind of work that David Currie and others have done identifying energy as a key driver of species richness. Although I haven’t looked at the latitudinal gradient in diversity story in awhile, it seemed to me that using several lines of observational evidence was allowing some pruning of the possible explanations. Jeff

      Liked by 1 person

      Reply
  3. Manu Saunders

    Great post! I think you can infer process from patterns, but of course nothing in nature is ever simple, so most inferences from 1 obs are only half right. Sheep do love lupines, but lupines are also toxic to sheep…in different contexts. In Australia, we grow lupines as stock feed, but under certain conditions sheep (& cows) can die/get very sick – I think when they’re green they’re super-high in alkaloids, and if a certain fungus is growing on the stubble, animals get lupinosis etc. etc. Also, some varieties have higher alkaloid levels than others, and I think wild lupines are more toxic than cultivated varieties. But you would have to see a bunch of dead sheep lying in a wild lupine field *after* you saw the neat sheep-mown line to put those patterns together. It would be interesting to know (on average) how many individual jigsaw pieces you would need to observe before you can infer the whole picture…

    Liked by 1 person

    Reply
    1. ScientistSeesSquirrel Post author

      Wow – one thing about blogging that is both exciting and terrifying is the likelihood that everything I post may be read by someone who knows way more about it than I do. For lupines, it’s you, Manu! Thanks for reading and commenting.

      Your point about ‘how many jigsaw pieces’ is I think exactly what Jeremy and I are “arguing” over up above. I don’t know the answer beyond “at least several”.

      Liked by 1 person

      Reply
    2. Macrobe

      Another bit of information to add to a connection between lupines and sheep: Many Lupinus spp. have high concentations of alkaloids in plant parts that can be toxic to livestock (bovine and ovine have different sensitivities and present different symptoms). Some Lupinus species also sequester soil selenium which, when consumed in high quantities, can directly or indirectly induce toxicity. The latter I learned accidentally while investigating the “mysterious” death of four lambs one year on my sheep ranch in Oregon. Hence the reason why lupines are considered a noxious weed in Oregon, a state where sheep are a major livestock industry. My ranch became a ‘field lab’ for a class studying toxicology in the veterinarian program at the university where I was a faculty member.

      Imagine my reaction when moving to Texas where the sacred and coveted bluebonnet lupine is the state flower. 😉

      Liked by 2 people

      Reply
  4. Brian McGill

    Great post! I 100% agree.

    I would say most of science actually evolves from a sort of two step between
    1) Pattern suggests process
    2) Process suggests multiple patterns
    3) See if multiple patterns confirmed
    Repeat.

    Jeremy asked for an example. I would give island biogeography as an example. In the MacArthur and Wilson book, it was derived from observations about an area-richness correlation and an isolation-richness correlation, one of which suggested extinction and one of which suggested colonization which together suggested dynamic equilibrium. Which led to a general theory which was confirmed in many systems via pattern (yes I know Simberloff did some experiments too, but I think the evidence was pretty convincing with just the observational data).

    Perhaps metapopulations as well. Metapopulations were well accepted long before people were measuring dispersal between patches.

    For that matter Lotka-Volterra competition equations were pretty well established without people having to measure and quantify competition processes. And the original Volterra predator prey equations came from observation of bombing and fish populations if I recall without anybody watching fish die.

    And you could make a case for functional responses a la Holling …

    Like

    Reply
    1. ScientistSeesSquirrel Post author

      Thanks, Brian! I think your 3-step process is bang on, and I wish I’d put it clearly like that. I would, however, argue that your discussion of island biogeography falls into the trap of equating “pattern” with experiments and “process” with experiments. Experiments rarely if ever observe process; what they do is observe pattern in a case where many possible processes are controlled. In ecology, it’s rare for _all_ possible processes except the hypothesized one to be controlled. We are victims of nature’s confounding!

      Liked by 1 person

      Reply
      1. Brian McGill

        You’re more radical at this than I am! But I agree. Simberloff and Wilson mostly observed patterns (e.g. increasing richness until the old level was reached). And lower richness on truncated islands.

        Liked by 1 person

        Reply
  5. Pingback: Recommended reads #63 | Small Pond Science

Comment on this post:

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s