Any model of real phenomena has to state a null hypothesis, namely, the description of a state in which there is nothing to explain, compared to which the actual state will have to be explained. Many radical changes in scientific thought, be they called "revolutions'' in the Kuhnian sense, or more modestly "shifts,'' consist in new definitions of the null hypothesis. For instance, Galilean physics began by conceiving of the rectilinear uniform motion as the "null hypothesis'' (instead of rest), pointing out acceleration or trajectory changes as the right explanandum. And such a definition of the null hypothesis has been called "the principle of inertia.''
So, the words themselves suggest that phylogenetic inertia is the null hypothesis in evolutionary theory. In any population, traits have to be explained if they are not obviously the result of descent, e.g., if they are not homologous of traits in the ancestor species. Of course, the determination of the traits as homologous or not depends on the set of species that are to be compared. Thus, the preliminary definition of this set of compared species in order to account for traits in a given species is an absolute condition for applying the PNS. Wrongly determining homologous traits by inadequate specification of the initial set of related species to which the explanandum species has to be compared immediately entails false results (Sober and Orzack 2001). The "just so stories'' stigmatized by Gould and Lewontin as unfalsifiable and abusive applications of the PNS often stem in the absence of data from such misunderstanding of the right null hypothesis.
However, methodologically, for a set of species, the relationship of homology and homoplasy implied by the statement of a null hypothesis is epistemologically related to the more fundamental principle of parsimony.32 It can easily be seen that
32 See this volume Chapter 5 by Folinsbee KE, Evans DC, Frobisch J, Tsuji LA and Brooks DR. For a general philosophical account see Sober (1981, 1988b).
the more we judge there to be homologies, the less evolutionary lineages we have to draw on: this is a kind of parsimony, so Hennig's auxiliary hypothesis can be called upon if one subscribes to epistemological parsimony. But the stronger, ontological claim of parsimony also supposes this way of defining the null hypothesis.
Phylogenetic inertia, however, is not incompatible with selection. We have to distinguish between the question of the origin of traits and their presence. When the traits exist by phylogenetic inheritance but decrease in fitness in the new environment and the new species, selection is likely to suppress them or render them vestigial. In the reptilian family, this was probably so in the case of the four legs when it came to the snakes. So origin and presence are two distinct topics. If the inherited traits are still present, one is allowed to postulate no negative selection, but a positive fitness value may promote stabilizing selection to keep them. So selection and inertia are not two competing hypotheses but are sometimes distinct explanans for distinct explananda, and sometimes complementary explanatory resources. In this regard, the idea of a null hypothesis in the question of the maintenance of traits has even been challenged (Reeve and Sherman 2001).
As early as the 1930s, Wright forcefully emphasized the evolutionary role of random processes (Wright 1932) such as genetic drift. The smaller a population is, the more powerful those kinds of processes are. This is a rather simple idea, since the same phenomenon is illustrated by the toss of a coin: a small sample is more likely to show a random bias (for example, seven heads versus three tails), than a large sample, which will show a half/half distribution of tails/heads, according to the law of large numbers in probability theory. So, in small populations, some genes, whose fitness is either equal or lower than other alleles, can go to fixation.
However, the concept of drift is not only a negative one, if it is connected with Wright's other concept, the adaptive landscapes. The fact is that in a gene pool some combinations are local optima, and if a genotype is on the slope of this kind of local optimum, selection will lead it toward this peak.33 But there may be a fitness valley which separates it from another, higher, fitness peak so that its fitness will have to decrease in order to get it onto the other fitness peak. For this reason, only random drift, provided that population is small, can lead it through decreasing its fitness across the fitness valley toward another hill so that selection can, after that, lead it toward the global fitness peak. Then, through migration, the new genotype can spread. In this model, drift helps to increase fitness by moving genotypes to global fitness peaks. Drift is, then, together with natural
33 For the several interpretations of the adaptive landscapes see Gilchrist and Kingsolver (2001).
selection, the other process accounting for the evolution of species, modeled by the travel of genotypes across fitness valleys and hill climbing. Wright named this schema the "shifting balance theory,'' and empirical evidence for its generality is sometimes given but is not generally persuasive.34
Of course, this goes against Fisher's formulation that average fitness is always increasing; but the conditions of the two assumptions are not the same, since Fisher's theorem speaks about large, theoretically infinite, populations. Hence, evaluating the conflict between Wright's view of the role of random drift and Fisher's claim of an overall selectionist view, according to which the fittest always invades the gene pool, entails a decision on whether large or small effective populations are mostly to be found in nature.
But random drift raises some epistemological questions (and a more metaphysical one that I will address in the next section). The issue is stating the difference between drift and selection: are they two competing hypotheses? A first model, explored in detail in Sober (1984), takes drift and selection as two kinds of forces acting on an equilibrium model formulated by the Hardy-Weinberg law (together with the forces of mutation and migration, which I do not consider here). If equilibrium is changed, then selection is acting; when the fitter allele is not fixed, then random drift must have perturbed selection. Outcomes are the result of the addition of selection and drift in an analogous manner to summation of forces in Newtonian mechanics. However, this model has been challenged in two papers by Walsh, Ariew, Lewens, and Matthen (2003, 2002). They argue that selection and drift are not equivalent forces since they are not as comparable as two directional forces in dynamics. They do not compete at the same level, because "natural selection'' is not exactly a force like the sum of selection pressures, but a sampling effect, supervenient on the real selective forces (fight, predators, mate choice, foraging, reproduction...) in the same way that entropy supervenes on microphysical states of molecules. On the other hand, drift is another kind of sampling, in the manner of a sampling error (compared to the fitness coefficients). Since summation of forces presupposes that their effects are additive, therefore, are acting at the same level, one cannot logically treat the state of a gene pool as the composed effect of selection and drift; and, finally, to talk of forces proves in general to be misleading, even for selection.
The question is not an empirical one but concerns the logical types of those population-level theoretical entities or processes that are selection and drift. Epistemologically, this means that there may be such a gap between drift and selection that the model of composition of forces has to be replaced by a
thermodynamical model of macroscopic effects of the statistical accumulation of heterogeneous microphenomena. The analogy of selection is no longer gravity but entropy, and we know that entropy as a variable bears no causal effect. Whether or not Walsh, Ariew, Lewens and Matthen's challenge is right, the point is that considering drift leads to no obvious model of selectionist explanation, since when one is about to derive empirical content from mathematical models of population genetics and the PNS, one has no sure principles for conferring an epistemological status to the process of selection. This does not affect our study of phylogenies or our making and evaluating models for its mechanisms, but the interpretations of those models, hence of the very nature of the mechanisms, are certainly at stake. If drift and selection are not to be compared as two different forces like electromagnetism and gravity in physics, Darwin's statement about the composed nature of the processes of evolution, and the subsequent agenda of weighting the components, has to be qualified.
Was this article helpful?