1. Surprising Announcements

From the 1960s through the early 1990s, substantive and far-reaching changes in population genetics occurred. The debates concerned the units of selection and changed the nature and scope of evolutionary explanations, turning attention from genotypic models to models of structured populations. Furthermore, the relation of a trait to its environment was explored in both models and empirical investigations.Footnote ¹ E.g., Maynard Smith's early theoretical work on group selection provoked theoretical debate, while Michael Wade and colleagues produced empirical evidence for the efficacy of group selection as an evolutionary component. Wade's (Reference Wade1978) paper, in particular, criticized the ill-conceived, oversold, and rigid theoretical requirements for the efficacy of group selection in nature. The result was a blossoming in a corner of population biology—both in theoretical genetics and in empirical studies—of investigations into the possibility of higher-level interactors such as those that had been suggested in the house mouse by Lewontin and Dunn Reference Lewontin and Dunn1960 and Lewontin Reference Lewontin1962.Footnote ² Various attempts had been made to expand population genetics beyond the organismically-focused efforts of the pre-1960s, but the work of the 1980s is distinguished by its achievement of consensus about many fundamentals of hierarchical population genetics and by its empirical substantiation of these ideas. Philosophers of biology soon joined the revolution, attempting to sort out the various disputes and definitions, analyzing the variety of evolutionary models, focusing in on what they saw as the key issue of contention: How could the various methods and principles be used for sorting out the key ‘units of selection’ (the objects directly involved in selection processes determining the success of genes), thus engaging in the ‘interactor debates’.

In the midst of this scientific abundance, some philosophers, calling themselves ‘genic pluralists’, arose and attacked the new population genetics, its models, and reinterpreted its empirical findings.

In 1990, Philip Kitcher, Kim Sterelny, and Ken Waters (henceforth `KSW') announced that no one had to concern themselves with the units of selection problem anymore: “Once the possibility of many, equally adequate, representations of evolutionary processes has been recognized, philosophers and biologists can turn their attention to more serious projects than that of quibbling about the real unit of selection” (Reference Kitcher, Sterelny and Waters1990, 161). Yet, the debates continue. Concerns about units of selection in biology and philosophy have not been settled by claims about representational equivalence.

I argue that three of the chief claims of the genic pluralists are defective, and that their overall argument is unsupported. First, there is the above-quoted remark about philosophers and biologists who quibble over the units of selection. Second, Sterelny and Kitcher (henceforth `S&K') also claim that there are no targets of selection (interactors). Third, KSW claim that they have a concept of genic causation that gives independent genic causal accounts of all selection processes. I argue that each of these claims is either false or misleading. My approach to this literature is unusual; it doesn't take up the proffered notion of genic cause directly, nor does it defend another view of cause. Rather, my approach is indirect, proceeding through a fuller understanding of the role of interactors in selection theory. KSW's claim, that debate over units of selection becomes superfluous once the existence of equivalent (read: genic) representations has been established, is shown to have things exactly backwards: Problems of units of selection must be overcome in order to generate the adequate genic-level theory that they take as one of their “many, equally adequate, representations” (KSW Reference Kitcher, Sterelny and Waters1990, 161). Debates over interactors, therefore, are not ‘pseudoproblems’, and this fact ultimately has fatal consequences for their claims that there are independent genic causal stories for any case of selection (KSW Reference Kitcher, Sterelny and Waters1990, 161).

The burden of proof lies with the pluralists to show that any such genic-level causal accounts exist at all. I will not defend any particular hierarchical view in the paper, as these have been defended elsewhere.Footnote ³

2. The Genic Pluralist Challenge

‘Pluralist genic selectionists’ (such as S&K) believe that “there are often alternative, equally adequate representations of selection processes and that for any selection process, there is a maximally adequate representation which attributes causal efficacy to genic properties” (Reference Sterelny and Kitcher1988, 358; emphasis added). This sort of pluralism is peculiarly weak: It is simply an equivalence condition. Not only that, the arguments, as given by S&K, entail genic reductionism; an ironic twist, given that pluralism is usually an antireductionist position.Footnote ⁴

More importantly, the peculiar sort of pluralism offered by S&K stands in contrast with philosophically significant, stronger forms of pluralism. There are three such forms. First, one in which each level of description is understood as indispensable and independent, either locally or globally, for describing a given phenomenon (Dupré Reference Dupré1993). This type of pluralism offers an important type of unity. Each level of description is insufficient, and connecting models are needed for the purposes at hand; this type of pluralism can be used to support a global disunity of science claim. A second important, and stronger, form of pluralism claims that either some or all the different descriptions are jointly relevant to the phenomenon under consideration; this is especially important when the relevant information cuts across different levels or different kinds of description (Darden and Maull Reference Darden and Maull1977; Dupré Reference Dupré1993; Cat Reference Cat1998, Reference Cat2000, Reference Cat2004). A third significant type of pluralism concerns unity and pluralism at the level of criteria, independent of any specific theory or application. An example would be multiple causal criteria, wherein criteria of causality above any specific theory contribute a new dimension to the question of unity and pluralism independent of the issue of the connection among theoretical facts (Cat Reference Cat2000, Reference Cat2004). S&K's pluralism (further details of which are worked out in Section 5) exemplifies the close relationship between pluralism as equivalence and pluralism as reduction, both of which are in tension with the general treatment of pluralism in the wider literature, as is made clear in the conclusion.

KSW see themselves as attacking “some biologists” who “think … there is a unique account that will identify the level of selection” (Reference Kitcher, Sterelny and Waters1990, 159; emphasis added). This question of levels is the question of identifying the interactors for that process. But, they proclaim, “We believe that asking about the real unit [level] of selection is an exercise in muddled metaphysics” (KSW Reference Kitcher, Sterelny and Waters1990, 159). This is because, according to KSW, the gene itself can always be construed as the interactor in a selection process—the entity that directly interacts with its environment such that replication is differential—once ‘environment’ is construed in terms of the allelic perspective (Waters Reference Waters1991, 554, 571; S&K Reference Sterelny and Kitcher1988, 339, 341, 348). Thus, “[hierarchical selectionists] err … in claiming that selection processes must be described in a particular way, and their error involves them in positing entities, ‘targets of selection’, that do not exist” (S&K Reference Sterelny and Kitcher1988, 359).Footnote ⁵

Thus, the pluralists’ claim that there is a causally adequate, general evolutionary theory purely at the genic level—one that does not require any appeal to higher-level causal interactors. This is not so. By helping themselves to the necessary higher-level (interactor) information, the pluralists make it appear that the hard-won methods for obtaining and incorporating such higher-level causal information (which is represented in the mathematical structure of the models as a whole) are irrelevant; or that debates over these methods have been resolved or overcome by applying the genic approach. They have not. One of the primary claims of my paper is that, contrary to its proponents’ claims, the genic account does not give us a theory independent of individuating causal interactions at various levels of the biological hierarchy, nor does it solve or dissolve the problem of how to individuate those very interactions.

3. The Basics

In Lloyd Reference Lloyd, Keller and Lloyd1992, Reference Lloyd, Singh, Krimbas, Paul and Beatty2001, I analyzed the units of selection problem into four distinct questions, and argued that much confusion had arisen because participants in the debates were arguing at cross-purposes.

The four questions I delineated as distinct ‘units of selection’ questions were the replicator question, the interactor question, the beneficiary question, and the manifestor of adaptation question. The replicator question concerns which entity passes on its structure directly in replication (usually, but not only, genes) (Dawkins Reference Dawkins1982; Hull Reference Hull1980; but see Griesemer Reference Griesemer2000). I shall expand on the second, interactor question, below, since it is the focus of this paper. The beneficiary question involves which entity benefits, in the long term, from the evolution by selection process, while the question of which entity manifests adaptations plays a central role in determining which entities have ‘engineering’ adaptations at a given level of organization (see Lloyd Reference Lloyd, Singh, Krimbas, Paul and Beatty2001; Williams Reference Williams1966). Using this framework, I analyzed many of the major positions in the units of selection controversies, and concluded that numerous players were mixing and matching the various questions in developing their requirements for what it takes to be a real unit of selection.

One of the easy cases for my Reference Lloyd, Singh, Krimbas, Paul and Beatty2001 paper to sort out was the debate between Dawkins and the group selectionists. Dawkins makes clear that by a ‘unit of selection’, he means a replicator, not a vehicle (interactor).Footnote ⁶ Since he categorizes not only organisms but also groups as vehicles, groups cannot, by definition, be units of selection (Reference Dawkins1982, 115). In contrast, the genic pluralists have developed arguments, driven by new approaches to interactors,Footnote ⁷ which pit genic selectionists directly against group selectionists.Footnote ⁸ While numerous authors since 1988 have attacked these genetic pluralist views (originating in their present form with Waters Reference Waters1986), none of these criticisms has been particularly successful in convincing readers exactly what is wrong with the position.Footnote ⁹

I argue that the basic problem lies in the pluralists’ understandings of the role of interactors in models. Their genic models are explicitly derived from causal models involving higher-level interactors, as I review in Section 5. Any genic causal account is thus derivative from an interactor causal account, and is not independent at the genic level, since it incorporates these higher-level causes. This result undermines their claim to have established independent causal genic selection models at all. Before presenting the details of this argument, let me return to the crucial interactor question.

4. Interactors

In its traditional guise, the interactor question asks, What units are being directly selected in a process of natural selection? An interactor may be at any level of biological organization, including group, kin-group, organism, chromosome, gene, or even parts of genes. The interplay between an interactor and its environment is mediated by traits that affect the interactor's effects on genic success. Some portion of the expected fitness of the interactor is directly correlated with the value of the trait in question. Finally, the expected fitness of the interactor is commonly expressed in terms of fitness parameters, that is, in terms of the fitness of replicators; hence, interactor success is most often reflected in and counted through, replicator success.Footnote ¹⁰

At what levels of biological organization do interactions occur that make a difference to replicator success? There are a number of ways to study this question, including modeling, experimentation, and fieldwork,Footnote ¹¹ and various methods have been proposed over the years for identifying interactors.Footnote ¹² These have been used primarily in theoretical and philosophical discussions. At the same time, population biologists (sometimes in interaction with philosophers) have developed a wide variety of technical definitions.Footnote ¹³ There has been much argumentation over which approach is best.Footnote ¹⁴ The issue as it stands is undecided, although various research groups using different (but closely related) approaches, each with their various strengths, have made advances.

The emphasis in biological discussions of interactors is on getting the statistical and causal information that will make the model empirically adequate to the phenomena. For example, Heisler and Damuth's popular contextual analysis approach (Multilevel Selection I) to discovering and isolating interactors has three goals: the measurement of relationships between characters and fitness; the location of the level of biological organization at which these relationships occur; and the evaluation of causal models of selection that are proposed on the basis of prior research (Heisler and Damuth Reference Heisler and Damuth1987, 594–595). The struggles in the literature involve the best ways of doing so. Some of the empirical workFootnote ¹⁵ involves demonstrating the efficacy of various hierarchical models in accurately modeling an empirical system (Goodnight and Stevens Reference Goodnight and Stevens1997). Thus, constructing models with information at the higher-level is at the heart of the empirical and causal explanatory uses of hierarchical models. Dugatkin and Reeve emphasize explanatory (Reference Dugatkin and Reeve1994, 123–124) and research (Reference Dugatkin and Reeve1994, 126–129) successes of hierarchical models as against ‘broad individual’ models. Sober and Wilson, using the trait group approach to modeling hierarchical selection, review many case studies in which higher-level information makes the difference between building empirically, explanatorily, and causally adequate and inadequate models (Reference Sober and Wilson1998). Griesemer and Wade (Reference Griesemer and Wade1988) discuss nontrait-group hierarchical models.Footnote ¹⁶ Despite the variety of models and the disagreements regarding the success of particular models, the take-home lesson here is clear. Empirical adequacy, explanatory sufficiency, and research promise are all evaluated to determine the appropriateness of interactor models at distinct levels of the biological hierarchy.

In Hull's analysis, selection must be understood as two distinct causal subprocesses: replication and interaction. In the genic selectionism of KSW (derived, as Waters acknowledges, from Williams Reference Williams1966),Footnote ¹⁷ the same processes are involved; it's just that both take place at the genic level. The fundamental claim here is that genes are the ultimate interactors. Whenever other structures can claim to be interactors, genes can claim to be interactors, too, and the genic claims are more ‘general’ and ‘unified’.

I now examine how the pluralist genic selectionists manage to formulate genes as interactors.

5. Claims of Equal Representation

The fundamental claim of the pluralists is that anything that a hierarchical selection model can do, a genic selection model can do just as well. Thus, much attention is paid to showing that the two types of models can represent certain patterns of selection equally well, especially those that are conventionally considered hierarchical selection exemplars.

6. The General Pluralist Argument

So far, I have focused on how the genic pluralists’ models are derivative from higher-level models, in particular their dependence upon prior solutions to the units of selection problem (in this case, the interactor question). This might seem to leave their general arguments about the existence of alternate models, and the seemingly arbitrary nature of choice among them, untouched. But, this is not so. The issues of derivativeness, dependence, and the general pluralist argument are intimately connected. The basic structure of the argument for pluralism, along with evaluations of its premises, are set out below.

Analysis. Derivativeness implies that these alternative representations are not genuine alternatives. Rather, they are semantic reinterpretations dependent on the empirical and explanatory adequacy of higher-level models. Investigators into higher-level interactors incorporate environmental factors, trait-bearing entities, and demonstrations of correlations with fitness components, etc., (as reviewed in Section 4) into their models. This results in mathematical structures, (however represented) key aspects of which are required for the empirical adequacy of the structures. Renaming parts of these structures does not change conclusions drawn about other parts of these structures, particularly the level(s) at which entities interact with their environments through their traits (traditionally, the definition of an interactor or unit of selection). Premise 1, therefore, is undermined by the derivative nature of the multienvironmental allelic structures, and especially by the derivative nature of any causal claims. There are no alternative causes, nor operative levels of selection; there are simply renamed structures in derived models.

Analysis. This premise, a special case of Premise 1, is undermined by Lewontin's argument, which shows the dynamical insufficiency of presenting selection processes purely in terms of genic causes in genic state space and reconstructed allelic environments. The premise is already undermined by the derivativeness discussed in the analysis of Premise 1. Lewontin gives an in-principle argument against the possibility of such allelic models. (See Section 8.)

Analysis. This premise appears to result from metaphysical assumptions involving agents. Specifically, once a model is reformulated in terms of a particular state-space in which alleles are the state variables, there seems to be a tendency to identify a distinct causal agent. An example of this metaphysical move occurs in Waters’ account of sickle-cell anemia, wherein the ‘agent’ in the model to which we attribute ‘causal power’, as well as a different ‘level of selection’, is the allele, because it is the state space of the model (Waters Reference Waters1991, 564). Waters gives no defense of this position. Williams defends it by appealing to what I have analyzed as the ‘beneficiary question’: What entity is around in the evolutionary long run to benefit from an evolution by selection process? In sexually reproducing species, the answer must be the allele; hence, according to Williams (Reference Williams1966), the allele is the ‘unit of selection’.

S&K, in contrast, do give arguments for the differences in causal attribution of the hierarchical and genic-level models, but the arguments they appeal to are Dawkins’, nearly all of which are aimed at organismic selectionists, which are not under consideration here. (See Lloyd Reference Lloyd[1988] 1994, Reference Lloyd, Singh, Krimbas, Paul and Beatty2001 for discussion.) Dawkins’ arguments apply to the replicator question and not to the interactor debate, and hence are misapplied by S&K.Footnote ²¹ Dawkins’ own argument is also based fundamentally on his interest in the ‘units of selection’ as beneficiaries, following the lines of Williams, and fares no better against those who are engaged in the interactor debates (Lloyd Reference Lloyd, Singh, Krimbas, Paul and Beatty2001). The problem with focusing on the ‘units of selection’ as beneficiaries question in the late 1980s is that all of the debate in population genetics, and the vast majority of the debate in philosophy, focused clearly on either the ‘interactor’ question, or a combination of the ‘interactor’ question with the ‘manifestor of adaptation’ question. Thus, KSW, in addressing the ‘unit of selection quibbles’ relating to population genetics, can only plausibly be read as addressing the genetics arguments and related philosophical arguments concerning specific disagreements about how interactors or interactors-and-manifestors-of-adaptations are defined and identified. Any appeal to the beneficiary question would be, at best, off-point—a metaphysical sideshow.

But perhaps we do not have to appeal to metaphysics at all. There is a precedent in scientific practice for changing causal accounts when the state space is changed. As Rob Cummins (personal communication, 2005) has noted:

These cases, however, are purely empirically driven.

With the genic pluralists, in contrast, we change the state space, but not the empirical content of the theory at all, thus undermining the justification for accompanying a change in state space by a change in causal structure. The principle stated above holds where there are changes in empirical content.Footnote ²² Thus, the pluralists are trying to have it both ways. An unwary reader, however, may easily be misled by this particular move made by KSW.

Thus, we are back to the metaphysical claims discussed above. In sum, much of the genic pluralists’ view rests on implicit, heavily loaded metaphysical assumptions in which causal agency is assigned according to the state space in the model: If you have a model in which the allele is the entity with a fitness parameter, then assume that the allele is the ‘actor’, ‘agent’, or ‘cause’ of the selection scenario.

Analysis. The corollaries of the conclusion have been shown to be incorrect (see Sections 4 and 5).

7. Pragmatic Virtues

In presenting their claims for the importance of a completely genic-level evolutionary theory, S&K contrast ‘two images’ of evolution by selection. The two images they have in mind are organismic and genic selectionism. They state that the relative worth of the two images turns on two theoretical claims in evolutionary biology:

1. 1. Candidate units of selection must have systematic causal consequences. If Xs are selected for, then X must have a systematic effect on its expected representation in future generations.

2. 2. Dawkins’ genic selectionism offers ‘a more general theory of evolution’, one that can handle organismic selection cases, but also other cases that are problems for organismic selection. (Reference Sterelny and Kitcher1988, 340)

Challenges to pluralist genic selectionist accounts typically assert that pluralism fails to meet Condition 1. Pluralists’ arguments purportedly showing the equivalence of genic and higher-level models are aimed at rejecting this critique. As I have shown, this defense comes at a very high price, namely, an explicit appeal (via adoption of hierarchical methods and models) to higher-level causal consequences that the genic pluralists claim to avoid.

Additionally, S&K's arguments for the superior generality of the genic approach, aim at the wrong target, organismic selection, since none of the authors they attack actually defend that view, as Sober pointed out (Reference Sober1990, 152). When S&K do finally confront the real contender, hierarchical selection theory, they seem to misunderstand it (Reference Sober and Wilson1988, 359). Their mistakes here are identical to their mistakes in the arguments for the other putative virtue of the genic approach, its unity, so we will skip ahead to that virtue, since at least it's aimed at an appropriate target.

Genic selectionism's other ostensible virtue is that it is more unified than the hierarchical view, according to Dawkins and KSW. With genes as both replicators and interactors, the entire theory of evolution by natural selection is unified and all selection phenomena can be taken into account. Against this they compare the hierarchical account, where “processes are diverse in the kinds of representation they demand” (Reference Sterelny and Kitcher1988, 354). Thus, it appears that you have need for one type of representation for group selection, another for kin selection, another for genic selection, and so on. Hierarchical selection looks like a big mess. But this reveals a deep misunderstanding of hierarchical theory, which was unified in its initial form (ironically) by Dawkins, and then in its current form by Hull.

According to Hull, at each level of the biological hierarchy where there is selection going on, there is an interactor at that level and a replicator at that level, or more usually at a lower level. The biology of the situation determines which entities play the functional roles of interactors and replicators in a given case (Reference Hull1980). The hierarchical theory is thus a highly unified theory, capable of accounting for all cases of selection, including genic selection.Footnote ²⁴ The formal models for hierarchical selection illustrate this unification beautifully.Footnote ²⁵

8. Conclusion

In sum, the genic view does not have the big advantage claimed; it does not avoid the unit of selection ‘quibbles’ after all. But, the causal questions concerning which entities are interacting with their environments in a way that affects replicator success—including questions concerning how to know when to take such interactors into account and how to detect them—are crucial for the genic approach, as the genic pluralists themselves have shown. The causal equivalence of the models is better thought of as derivative; higher-level causes are reformulated down in model translation and imported into the genic models, not changing one whit the level at which the identified interaction between trait and environment that makes the difference to genic success (the interactor) actually appears in the overall mathematical structure from which the genic model is derived.

This means that the crucial question regarding the levels of interaction necessary to model a given selection process is still outstanding, whether it is put in hierarchical terms or genic ones. Worse, the pluralists we examined attempt to sidestep this question, but cannot; they end up taking sides inadvertently in the interactor debate, all the while claiming they have overcome just that debate. Neither philosophers nor biologists can ignore the debate about interactors in selection; the outcomes make a difference to the empirical, explanatory, and causal success of the models produced.

Pluralist genic selectionists also claim that once genic and other alternative representations of selection processes are available, there is no scientific way (except perhaps the reductionist one) to choose among the various alternative models (S&K Reference Sterelny and Kitcher1988, 358).Footnote ²⁶ In addition to the derivativeness argument, the additional flaw with this argument is that Lewontin has shown the dynamical insufficiency of their approach to genic models, wherein the allelic frequencies and the layered allelic environments of the alleles determine fitnesses. Even if they had an independent causal definition at the genic level (which they do not), the models would not be the promised “[presentations of] a selection process in terms of the causal efficacy of genes” (S&K Reference Sterelny and Kitcher1988, 358). Thus, there can be no ‘causal equivalences’ of the sort they claim to have shown.

Let me spell this out in more detail. Perhaps S&K, especially, have simply chosen badly in illustrating their point, and perhaps the revamping of Williams that Waters is suggesting could do the trick, without relying on the derivation from hierarchical models. Not so, says Lewontin. There is something fatally wrong with any model that attempts to model a selection process using genic selection parameters and associated allelic environments. Indeed, there is also something faulty in using purely genotypic state space models to represent selection, since, at every generation, we need allelic-level information about segregation and usually, recombination, in order to produce a pool of gametes. Thus, properly conceived, population genetics models need, at a minimum, information from both the allelic and the genotypic space.

Since all regular population genetic models require both allelic and genotypic level models to produce dynamically workable models, Lewontin argues the following. Take an allelic-level model in which we are using allelic-environments to determine allelic-level frequencies (as all the models suggested by the pluralists require). At each generation, we must readjust the selection coefficient according to the frequencies of the different alleles. How is this done? The obvious way is to take a weighted average of the fitnesses in the different allelic ‘environments’, weighted by how frequent the different ‘environments’ are (as suggested by S&K Reference Sterelny and Kitcher1988, 346). But this will only work if the alleles find themselves in the different environments randomly, i.e., if we are in Hardy Weinberg proportions. So we are required to determine how frequent the combinations of alleles and ‘environments’ are, i.e., what the frequencies of different diploid genotypes are. That is, we are required to go up to the higher level; it can't all be done at the allelic level, any more than genotypic calculations can all be done at the genotypic level alone—they are reliant on the allelic level.

Even if the population is in Hardy Weinberg equilibrium, that must be confirmed, which also requires going up to the genotype level. So, no matter what, in order to calculate the next generation of allele frequencies, we must know the next generation's genotype frequencies, in order to calculate the fitnesses of the alleles in the allelic space. The only alternative is to stay in the allelic space completely, and track empirically the changes in the genic selection coefficients at each generation, in which case, every generation would produce a new model, unlinked to the model of the previous generation. If we were to approach it this way, there would be no science of population genetics, i.e., no way to produce trajectories of gene changes over time, or predictions of changes, or explanations of changes. Consideration of more than one locus for selection (as many of the models touted by the genic selectionists involve) makes the empirical tracking approach mathematically intractable, as well as predictively useless. Thus, there is no way to actually represent a selection process over time with an allelic-level model. They are not dynamically adequate without information from at least the genotypic level, and maybe higher levels, so even setting aside the issue of the origins of the allelic models from hierarchical or genotypic models, the allelic models cannot be used at all without the constant input of genotypic or hierarchical models in order to represent selection processes. There is no equivalence here, only utter dependence.Footnote ²⁷

Obviously, this argument completely destroys Subpremise 1. But even if Lewontin's argument were to fail, and through some other form of modeling, an independent purely genic environmental causal model could be rendered empirically adequate, KSW have certainly not established this. Meanwhile, there are no ‘causal equivalences’, and so no independent genic models. Thus, there are no independent, maximally adequate, causally different accounts of any particular selection episode to choose between: There are no independent genic causal models, and all the other models offered as ‘alternative views’ are derivative. When combined with the problems with Premise 2, the genic pluralists are left in trouble.

Not only have the genic pluralists failed to show that the units of selection question is obsolete, they have failed to show their famous ‘equivalence’ between alternate causal models of selection, particularly the genic ones: Generally, such supposed ‘alternative causal models of selection’ are (1) mere derivations of hierarchically structured models and not causal alternatives as advertised, or (2) as Lewontin shows, nonderivative, but dynamically inadequate.

The only sustainable conclusion of ‘The Return of the Gene’ and ‘Tempered Realism’ turns out to be that accurate, causal, empirically adequate genetics models are precisely the hierarchical structures that KSW have relabeled.

Thus, rather than the very weak (and unworkable) equivalence form of pluralism that KSW find in the units debates (and in the genic reductionism of S&K), I see four factors pointing toward of one of the stronger forms of pluralism mentioned in Section 2: the t-allele case; the methods for isolating and identifying interactors; proven derivativeness; and Lewontin's argument for the unsustainability of an allelic-level model-type. In particular, descriptions of some or all of the different levels of a system are required for an adequate scientific account (see Section 2). This hierarchical form of pluralism instantiates locally, for a given phenomenon, a nonreductive form of unity, because all the different forms of descriptions are brought to bear. Thus, hierarchical selection exemplifies a strong and important form of pluralism, exactly contrary to the claims made by KSW; far from being a form of ‘monism’, as charged, it is a strong form of pluralism, standing in contrast to the weakened (and reductionist, in the hands of S&K) form of pluralism advocated by the genic pluralists.

Sometimes simpler models are useful as tools, as has long been recognized by all geneticists. (For recent discussion, see Michod Reference Michod1999.) But playing with the genetics in such a way that it doesn't improve the empirical adequacy of the models or theoretical understanding of evolutionary change is simply a step backwards to 1966.Footnote ²⁸ Remember, Williams’ (Reference Williams1966) allelic parsimony argument was widely interpreted as a reason to never even take a look at higher-level interactions, interactions that eventually led to empirically more adequate models (Lloyd Reference Lloyd[1988] 1994, 94–96; Williams Reference Williams1966, 5, 66, 93–95; Williams Reference Williams1992). Similarly, the genetic pluralists seem to want to return us to a pre-1980s state of population genetics, one in which structured population genetics and hierarchical selection structures played little or no role, empirically or theoretically.

In the end, confused metaphysics (or twisted scientific practice, take your pick) are insufficient to fill the gaps in the arguments the genic pluralists present. Thus, genic pluralism fails on multiple counts; it serves only to highlight the success of hierarchical approaches to selection processes.