So I think the dust has finally settled on Steven Pinker’s Edge article on group selection. It is followed by 20 commentaries from different perspectives (some of which I’ve already linked to on this blog) and Pinker’s reply. Plus another series of replies (some overlapping) on The Social Evolution Forum. There was even a post about Pinker’s article on Andrew Gelman’s statistics blog.
I have noticed that in my blog’s short history I have more posts of criticism than positive contribution. There are many responses I generally agreed with, and I linked to them in earlier posts. But in this one I thought I would highlight responses that I think put the debate in a larger framework.
First, I think David Queller’s response is excellent, short and pithy. He compares kin selection modeling to English and group selection modeling to Russian – with the point that you can often say the same thing in both languages – but sometimes it is easier to say it in one language than another. Queller’s response is particularly interesting because he is a traditional theorhetical evolutionary biologist who does primarily inclusive fitness modeling
D. S. Wilson expands on Queller’s analogy in a post at the Social Evolution Forum called the Clash of Paradigms: Why Proponents of Multilevel Selection Theory and Inclusive Fitness Theory Sometimes (But Not Always) Misunderstand Each Other. What I like about this section is his distinction between “unilinguals” and “bilinguals.” Some quotes:
In short, the current clash between proponents of [Multilevel Selection Theory] and [Inclusive Fitness Theory] is primarily a clash between people who are fluent in one paradigm and confused by the other paradigm, which they falsely attribute to the other paradigm in some absolute sense. This is like saying “Russian is confusing”, rather than “Russian is confusing for a non-Russian speaker such as myself.”
It is possible to be bilingual… Many evolutionists are fluent in both MLST and IFT, a fact that is sometimes obscured by the chest-thumping rhetoric of unilinguals…
For these and many other evolutionists, toggling back and forth between MLST and IFT has become normal science, with no need for chest-thumping rhetoric. The choice of theory is simply a matter of choosing the best tool for the job.
Joe Henrich makes an analogy to an engineer deciding on a coordinate system to use for a problem in orbital mechanics. I like this analogy because it stresses the fact that even if it is possible to solve a problem in two different ways, it is often much more easily done in one This analogy has particular resonance for me because much of my four years of undergraduate engineering instruction seemed largely an exercise in banging my head against the wall for (doing something roughly equivalent to) trying to solve a problem in the wrong coordinate system. For this analogy, your own mileage may vary…
A useful analogy might be the problem that an aerospace engineer faces when trying to model the trajectory of a satellite. A critical first step in solving such a problem is to select a coordinate system and a place to anchor that coordinate system in space (the origin). Among others, one can pick a spherical coordinate system (two angles and a distance) and anchor it to, say, the center of the earth; or, one can pick a Cartesian coordinate system (x, y, z orthogonal dimensions) and anchor it to a passing meteor. It is completely possible to calculate the orbit of a satellite with any number of different coordinate systems including these two, but picking the first system will allow you to easily solve the problem (analytically, using some solid assumptions) while building your intuitions about the movements of earth’s satellites. The second approach will be really hard, and provide you with no new intuitions. So, these are “equivalent” in some sense, but they are not equally useful for any particular problem. And, so it is with evolutionary accounting systems….
Rejecting group selection models is like banning spherical coordinates because you prefer to do your verbal reasoning in Cartesian coordinates.
Richard McElreath (disclosure: one of my committee members) makes a similar analogy to Bayesian vs frequentist statistics in the comments on Andrew Gelman’s blog.
I should mention that both Richard McElreath and David Queller signed the many-authored Nature paper and, from what I have heard, David Queller helped organize it. This should cast doubt on Richard Dawkin’s implication that the signers of the paper were all against group selection. Undoubtedly the co-signers had both bilinguals and unilinguals (and probably some nonlinguals) among them.
So, far these authors see inclusive fitness and multilevel selection as two modeling tools with varying degrees of usefulness. Pinker does not like this line of argument.
If the two theories really are equivalent, then any advantage of group selection… would have to come from the models’ being more convenient, elegant, simple, transparent, explanatory, or mathematically tractable. Yet by stretching the meaning of “group” beyond its ordinary sense, that’s just what they fail to be. According to the old song, “We belong to a mutual admiration society, my baby and me”—the whimsy hinging on the unnaturalness of referring to a pair of individuals as a “society.” The same is true for “group.” While mathematically speaking one can identify a “group” with any arbitrary set, in practice using a single construct for a pair of siblings, a person holding a door open for a stranger, a waitress and a customer, a married couple, a street gang, a traditional band or tribe, a nation, and an empire conceals the significant psychological differences among them.
This line of reasoning is just odd to me. The theorist should define a group in a way that is useful and intuitive to the question under consideration and the model one is building. *shrugs* Just as you can apply inclusive fitness models to different types organisms, you can apply multilevel selection models to different types of groups.
He also does not like the fact that multilevel selection models make simplifying assumptions (comparing them to “spherical cows”). This gives the impression that he does not realize that inclusive fitness models also make simplifying assumptions. In fact, all models make simplifying assumptions – that is why they are models.
To be fair, I generally have a harder time following elaborate word-arguments than more precise math arguments. I do not know if Pinker has actually ever tried his hand at mathematical modeling or has even worked through any of the models he writes about, but he does not give that impression.* But the proof is in the pudding.
Pinker would more easy prove his point to more quantitative naysayers by, as Joe Henrich suggests, taking this MLS model by Sam Bowles and reformulating it as an inclusive fitness model. Joe Henrich:
Of course, it may be possible! It’s also possible to calculate the position of that earth orbiting satellite using the coordinate system attached to the passing meteor. Good luck.
This would be hard enough and if Pinker hasn’t done a lot of modeling this would be a good way to learn. But then, after all the work that task would undoubtedly take, the trick would then be to explain the math both ways to, say, advanced undergrads in evolutionary biology and see which framework is more transparent and intuitive. Good luck!
* Pinker seems, for example, unaware of the folk theorem’s application to reciprocity models – which may be a subject of a future post on this blog.