Generalizing in ecology, or any science, requires you to group together different things based on their similarity in some key respect. Debates about generalization in ecology and evolution often are debates about the relative advantages of more vs. less generalization, more aggregated vs. more finely-differentiated groupings. Sometimes literally so, as in taxonomic debates between “lumpers” (who lump distinct subpopulations into the same species) and “splitters” (who split distinct subpopulations into different species). Sometimes only figuratively, as I’ve discussed in previous posts.
But generalizing in ecology doesn’t just require you to decide how much generalization you want. You also have to choose the features on which your generalizations will be based. Taxonomists don’t just debate whether to lump together or split apart distinct subpopulations. They also debate the choice of phenotypic and genetic features on which to base their classifications. The question of whether to lump or split species based on the presence or absence of the hippocampus minor is orthogonal to the question of whether to classify species on the basis of the hippocampus minor or some other feature(s).
I’ve been thinking about this in the context of classifying mechanisms in ecology. Two basic ways to sort mechanisms into groups are via their causes, and via their effects. For instance, consider coexistence mechanisms. One way to classify coexistence mechanisms is in terms of their causes, or more precisely, the conditions required for their operation. For instance, one might classify coexistence mechanisms as variation-dependent vs. variation-independent, depending on whether or not they require, or do not require, spatial and/or temporal variation in order to operate. On this classification, resource partitioning and the storage effect are different, because the former can operate at equilibrium while the latter cannot. Or, one could classify coexistence mechanisms by their effects. For instance, they could be classified as stabilizing vs. equalizing, depending on whether they strengthen intraspecific density dependence relative to interspecific density dependence, or whether they reduce interspecific fitness differences. On this classification, resource partitioning and the storage effect are the same—they’re both stabilizing. As another example, the Price equation in evolution classifies mechanisms of directional evolutionary change via their effects. Any mechanism that generates covariation between parental phenotype and parental relative fitness is classified as “selection”, while any mechanism that generates a deviation between the phenotypes of parents and the phenotypes of their offspring is classified as “transmission bias”.
What are the relative advantages and disadvantages of classifying mechanisms by their causes vs. their effects? I don’t have an answer to this question, but I do think it’s a question worth thinking about. In my experience, people who tend to classify mechanisms by their causes often have difficulty appreciating the value of classifying mechanisms by their effects, and vice-versa. For instance, in my own work applying the Price equation to ecological problems, I classify as “transmission bias” anything that causes some property of a given species (such as its primary productivity, or the temporal variance of its population dynamics) to vary from place to place or time to time (Fox 2010 Oikos, Fox and Kerr in press Oikos). I often run into the complaint that such “transmission bias” has too many possible causes, which are not usefully lumped together, the implication being that we should take those causes as our starting point. Which is fair enough, I suppose, although strangely I never get the same complaint about “selection” even though selection can arise arise from just as many underlying causes as transmission bias. But conversely, what’s the point of classifying mechanisms in terms of their causes if each of those causes gives rise to various different effects? For instance, abiotic environmental variation among sites is one possible reason why a given plant species might vary in productivity among sites (“transmission bias” of productivity). But it’s also a possible reason why different sites might “select” for more or less productive species (“selection” on productivity), and a possible reason why sites might vary in their species richness (an effect with no evolutionary analogue). And the same could be said about any other cause of variation in primary productivity, such as herbivory or competition or etc. Each of those causes will have various, quite distinct effects, which could even cancel one another out. So if you want to generalize about variation in primary productivity, why is a classification scheme that focuses on causes any better (or worse) than one that focuses on effects?
I should note that I ask this as someone who used to be very much a “cause” guy, but who over the years has increasingly become an “effects” guy (while remaining a conceptual “lumper” rather than a “splitter”). But I’m not sure why I shifted. And I don’t think you necessarily have to make a choice here—indeed, classifications in terms of causes vs. effects often can yield complementary insights, as in the case of coexistence mechanisms. But I get the impression that lots of folks don’t think that way—that lots of folks are basically “cause” focused or “effect” focused. Has anyone else noticed this? Or is my own impression just skewed because it’s mostly based on reactions I’ve gotten to my Price equation work (work that might well inspire odd reactions for all kinds of reasons that have nothing to do with the subject of this post…)?
I was just reading about causal inference this morning, in a review by Andrew Gelman (statistician at Columbia U.). Since he is a statistician and political scientist, I am still trying to work through the ideas and figure out what they mean for ecology, but I think that it is very much related to this question of how we see the world (causes versus effects).
It seems that Gelman (at least for social science) is mostly an effect-sider, since the causes can be fairly convoluted/correlated and many potential cause-routes could be true. Gelman says that many social scientists go back and forth, based on what types of questions they want to ask and what types of questions they can feasibly answer. I agree with you that many ecologists seem to be one side or the other. Undoubtedly, if I am giving a presentation on effects, I will get questions about causes (and I’d assume vice versa).
Thanks for the link to the Gelman paper, I look in on his blog from time to time and I should do so more often. Many of the issues he raises in that paper are indeed ones that have been raised in ecology as well (e.g. inference of effects from causes vs. inference of causes from effects, the latter being known in ecology as “the inverse problem”; see for instance recent work by William Nelson & colleagues on statistical “inverse methods”).