The answer, I think, must be a clear and resounding “no.” The two alternatives represent strikingly different views about the nature of reality and causality. We all agree that we need to know causes — and natural selection is a causal process. Gene selection confuses bookkeeping (properly done at the genic level) with causality (a question of evolutionary individuals plurifying [Page 656] differentially, based on interaction of their phenotypes with the environment). If we dissolve interactors into an overall “environment” of the genes, and then average a gene's fitness across all environments — the procedure of gene selectionism — then we lose causality.
Wilson and Sober (1994, p. 642) also reject the purely pluralist, or Necker Cube view: “There is no room for pluralism on these substantive empirical issues ... Group-level adaptations can be represented at the individual [organism] and gene level by averaging the fitness of lower level units across higher level units. Gene- and individual-level adaptations cannot be interpreted as group adaptations without committing the errors of naive group selection, but the gene's-eye view and the individual's-eye view cannot deny the existence of group-level adaptations (when groups are vehicles of selection) without being just plain wrong.”
Arnold and Fristrup (1982, p. 115) make the same point for the intrinsic reality — and not just preferential status vs. other equivalent representations — of species selection: “The characters that increase individual [organismic] fitness do not necessarily cause speciation or prevent extinction. Thus, it is misleading to adopt the convention of expressing all higher level trends in terms of individual [organism] level fitness.”
For all these reasons, I strongly advocate that we define higher-level selection as the differential proliferation of relevant evolutionary individuals based on causal interaction of their properties with surrounding environments — rather than by representing the effect of higher-level membership on the fitness of a designated lower-level individual. Only in this way will we avoid a set of confusions, and two pitfalls that easily follow, one after the other, with the first as a kindly delusion, and the second as an outright error: first, a falsely pluralistic belief in the equivalency of alternative representations at different levels; and, second, the siren song of gene selection as defining the only legitimate level of causal analysis in evolution. Only in this way will we achieve a clear and unified view that treats each level in the same manner, and approaches each evolutionary individual with the same set of questions. With this apparatus of analysis, we can attain a coherent and comprehensive theory of hierarchical selection — the most potentially fruitful, promising, and proper expansion of the Darwinian research program now before us.
Shall emergent characters or emergent fitnesses define the
operation of species selection?
Once we agree to define higher-level selection by differential proliferation of relevant units based on interaction between their traits and the environment, then we must (above all) develop clear criteria for the definition and recognition of traits in the unfamiliar world of higher-level individuals. Since we encounter enough trouble in trying to define and parse traits for the kind of individuals we know best — integral, complex, and continuous organisms like ourselves — we should not be surprised that this issue becomes particularly refractory at higher levels, and thus acts as a considerable impediment to the [Page 657] development of a rigorous theory of hierarchical selection. In particular, what should count, for purposes of defining evolutionary interaction with the environment, as a trait of a species?
The developing literature on this subject has featured a rich and interesting debate between two quite different approaches that, nonetheless, can be united in a coherent way to form the basis of a unified macroevolutionary theory of selection: the “emergent character” approach, as particularly championed by Elizabeth Vrba (1983, 1984b, 1989; Vrba and Eldredge, 1984; Vrba and Gould, 1986); and the “emergent fitness” approach inherent in the classic paper of Lewontin (1970), developed and defended in the important work of Arnold and Fristrup (1982), given further mathematical form in Damuth (1985), and Damuth and Heisler (1988), and most fully codified and expressed by Lloyd (1988 — see also Lloyd and Gould, 1993; and Gould and Lloyd, 1999).
Grantham (1995), in an excellent review of hierarchical theories of macroevolution, has christened this discussion “The Lloyd-Vrba Debate,” so the issue has now even acquired a proper name. The codification makes me feel a bit strange, since I have written papers on the subject with both protagonists (Gould and Vrba, 1982; Vrba and Gould, 1986; Lloyd and Gould, 1993; Gould and Lloyd, 1999), and do not view the issue as dichotomous; though the two viewpoints are surely distinct, and I have changed my mind — as a former supporter of Vrba's “strict construction,” who became convinced that Lloyd's more inclusive formulation forges a better match with conventional definitions of selection, and provides more promise for constructing an operational theory. But Lloyd does not disprove Vrba; rather, Vrba's exclusive domain becomes a subset of “best cases” in Lloyd's formulation. In this crucial sense, the theories sensibly intermesh.
Vrba's “emergent character” approach requires that a trait functioning in species selection be emergent at the species level — basically defined as origin by non-additive interaction among lower-level constituents. Since all science works within particular sociological and historical circumstances, we must understand that the greatest appeal of this strict criterion lies in its ability to “fend off” the conventional objection to species selection in a Darwinian and reductionistic world — namely, that the trait in question, although describable as characterizing a species, “really” belongs to the constituent, lower-level parts — and that the causal process therefore reduces to ordinary Darwinian natural selection on organisms or genes. For, when Vrba's criterion of emergence holds, one can't, in principle, ascribe the trait in question to lower levels. The trait, after all, does not exist at these lower levels. It makes a “first appearance” at the species level, for the trait arises through non-additive interaction of component lower-level parts or influences. If one species proliferates differentially within a clade by higher rates of speciation based upon such populational traits as geographic range, or density of packing among organisms, then we cannot ascribe selection to the organismic level — for organisms, by the logic of definition, cannot possess a population density, while the geographic range of a species need not correlate at all, or in any [Page 658] simple way, with the size of an organism's personal territory during its lifetime.
The strength of the “emergent character” criterion lies in its ability to identify a set of hard-line, unambiguous cases for species selection. For we must speak of selection among species if the relevant trait not only doesn't exist at any lower level, but can't even be represented as a linear combination of lower-level parts — for the nonadditive interactions that build the populational trait only arise within the population, and make no sense outside such an aggregation.
But we soon begin to worry that such a criterion may be too restrictive in eliminating a wide variety of traits that we intuitively view as features of populations, but that do not arise by nonlinear interaction of subparts, and do not therefore qualify as emergent by Vrba's criterion (which also matches the standard definition of the important concept of emergence in philosophy). Species and other higher-level individuals also develop features that seem to “belong” to them as an entity, but that arise additively as “aggregate” or “sum-of-the-parts” characters. Consider the mean value of a trait? This figure belongs to no individual and becomes, in this legitimate sense, a character of the population. But a mean value doesn't “emerge” as a functional “organ” of the population by nonlinear interactions among organisms. A mean value represents an aggregate character, calculated by simple summation, followed by division.
And how shall we treat variability — an even more “intuitive” candidate for a species-level characte
r that may be important in survival and proliferation of species? An individual organism doesn't possess variability, so the property belongs to the species. But variability also represents an aggregate character — another average of sum-of-the-parts. Do we not want to talk about species selection when species B dies because constituent organisms show no variation for a trait that has become strongly inadaptive in the face of environmental change — while species A lives and later multiplies because the same trait varies widely, and includes some states that can prosper in the new circumstances? Yes, species B dies because each of its parts (organisms) expires. In this sense, we can represent extinction as a summation of deaths for organismal reasons. But don't we also want to say that A survived by virtue of greater variability — a trait that does not exist at the organismal level, but that surely interacted with the new environment to preserve the species?
Vrba's solution, which I greatly respect but now regard as less useful than the alternative formulation, requires that we not designate differential proliferation of species based on aggregate characters of populations as species selection — but rather that we interpret such cases as upward causation from the traditional organismal level. Vrba (1980 et seq.) has coined and developed the term “effect hypothesis” for such situations — since the differential proliferation of species A vs. species B arises as an effect of organismal properties (of the individuals in species A that vary in the “right” direction), resulting in the survival of species A. [Page 659]
Vrba, and (I think) all other major workers in this area, have always regarded the effect hypothesis as a macroevolutionary theory because, in a heuristic and descriptive sense, one must apply the notion to species considered as items of evolutionary history. But events under the effect hypothesis are causally reducible to the traditional organismic level. (This kind of situation represents the minimal claim for an independent macroevolutionary theory — the need for descriptive engagement at the level of species, even if no distinct causality emerges at this higher level. This book defends the stronger claim for important causal uniqueness at the species level and above. Vrba, of course, also advocates this stronger version because she argues that some cases of differential species proliferation arise by the effect hypothesis, while others occur by true species selection based on emergent characters. I advocate a much larger role for causal uniqueness by defending the emergent fitness approach, a criterion that greatly expands the frequency and importance of species selection.)
To facilitate this distinction, Vrba and I developed a terminology to resolve a common confusion in evolutionary theory between the simple, and purely descriptive, observation of differential reproductive success — which we named “sorting” — and the causal claim — always and properly called “selection” — that observed success arises from interaction between properties of the relevant evolutionary individual and its environment (see Vrba and Gould, 1986). Evolutionary biology needs this distinction because students of the field have often — with misplaced confidence in selection's ubiquity and exclusivity — made a case for selection based on nothing more than an observation of differential reproductive success (sorting), without any attempt to elucidate the cause of such sorting. A leading textbook, for example, proclaimed that “selection ... is differential survival and reproduction — and no more” (Futuyma, 1979, p. 292).
Under Vrba's criterion of emergent characters, differential species proliferation by the effect hypothesis counts only as sorting at the species level — since the characters responsible for selection belong to organisms, but transfer an effect to the species level by upward causation. On the other hand, differential species proliferation based on emergent species characters does count as selection at the species level. However, under the broader criterion of emergent fitness, any species-level trait that imparts an irreducible fitness to species in their interaction with the environment defines a true process of selection at the species level, whether the trait itself is aggregate or emergent.
In the “emergent fitness” approach, we do not inquire into the history of species-level traits that interact with the environment to secure differential proliferation. We do not ask where the traits originated in a structural or temporal sense — that is, whether such traits arose by emergence at the species level, or as aggregate features by summation of properties in component organisms or demes. We only require that these traits characterize the species and influence its differential rate of proliferation in interaction with the environment. In other words, we only demand that aspects of the fitness of the [Page 660] species be emergent and irreducible to the fitnesses of component organisms. For cases where species function as interactors, or potential units of selection, Lloyd and Gould write (1993, pp. 595-596):
Interactors, and hence selection processes themselves, are individuated by the contributions of their traits to fitness values in evolutionary models; the trait itself can be an emergent group property or a simple summation of organismic properties. This definition of an entity undergoing selection is much more inclusive than in the emergent character approach, since an entity might have either aggregate or emergent characters (or both) ... The emergent fitness approach requires only that a trait have a specified relation to fitness in order to support the claim that a selection process is occurring at that level. ... In other words, the interactor's fitness covaries with the trait in question.
In a classic example, much discussed in the literature (Arnold and Fristrup, 1982; Gould, 1982c; Lloyd and Gould, 1993; Grantham, 1995), several clades of Tertiary gastropods show trends to substantial decrease in relative frequency of species with planktotrophic larvae vs. species that brood their young. In one common explanation (by no means universally accepted), this reduction occurs by species sorting based on the lower speciation rate of planktotrophic species — an hypothesized consequence of the lower probability for formation of isolates in species with such widespread and promiscuous larval dispersal. The sorting clearly occurs by selection, since low speciation rate arises as a consequence of interaction between traits of interactors and their environment. But at what level does selection occur?
Under the emergent character approach, the case becomes frustrating and ambiguous. Does the crucial property of “low speciation rate” in planktotrophs result from an emergent species character? In one sense, we are tempted to answer “yes.” Organisms, after all, don't speciate; only populations do — so mustn't the trait be emergent at the population level? But, in another sense, low speciation rate arises as a consequence of population structures induced by planktotrophy, a presumed adaptation at the organismal level — so perhaps the key character can be reduced to simple properties of organisms after all.
I have gone round and round this example for twenty years, often feeling confident that I have finally found a clear resolution, only to recognize that a different (and equally reasonable) formulation yields the opposite interpretation. All other participants in this debate seem equally afflicted by frustration, so perhaps, the fault lies in the concepts, and not in ourselves that we seem to be underlings, unable to achieve closure.
However, if we invoke the broader criterion of emergent fitness, the problem gains a clear resolution in favor of species selection. A structural feature of populations, leading to a low frequency of isolation for new demes, must be treated as a character of populations in any conventional usage of language. As stated above, individual organisms don't speciate; only populations [Page 661] do — so the character belongs to the species. However, the character may represent an aggregate rather than an emergent feature — thus debarring species selection under the emergent character approach. But, under the emergent fitness approach, so long as the character (whether aggregate or emergent) belongs to the species, and so long as the fitness of the species covaries with the character — and no one denies the covariation in this case — we have detected an instance of species selection.
Arnold and Fristrup (1982, p. 114) present this argument in a clear and forceful way:
The critical characters — larval strategies — may well have arisen for reasons that can be seen as adaptive in a traditional Darwinian sense. However, regardless of the mechanism by which they became fixed, these strategies behave as properties of species in that they result in distributions of rates of speciation and extinction within this group ... It might be tempting to assume that there are fewer planktotrophic species because the individuals in these species were somehow less fit than the individuals in non-planktotrophic species. However, it is obvious that the same result could obtain even if planktotrophic and non-planktotrophic individuals [organisms] have equal fitnesses, by virtue of the population structures that are concomitants of these larval strategies. Thus, the observed distribution of species types within these gastropods is not predicted from individuals’ level fitness alone, underscoring the necessity of the higher level of analysis.
In other words, the relative frequency of planktotrophic species falls not because planktotrophic organisms must be less fit (they may, in fact, be more fit on average across the clade), but because a character fixed by organismic selection yields the effect of lowering the speciation rate at a higher level. The population structure produced by planktotrophy may not rank as an emergent character, but does confer an emergent fitness at the species level — a fitness irrelevant to individual organisms, which, to emphasize the obvious point one more time, do not speciate.
The Structure of Evolutionary Theory Page 105
