I'm sorry to bog down this discussion with another long post, but I think there is an enormous potential benefit if philosophers actually understood more about how the theoretical apparatus of evolutionary genetics is brought into play, because my fellow evolutionary biologists don't seem to care very much if their understanding of causation makes any sense at all, whereas you folks do care about such things.  

Darwin's original theory has no explicit formalization and is incompatible with genetics as we know it, though it persists as a folk theory. Thus, its a bit absurd to talk about a "population genetics" of "Darwinian selection".  The Duke of Argyll said, in an 1864 speech, that "Strictly speaking, therefore, Mr. Darwin’s theory is not a theory of the Origin of Species at all, but only a theory on the causes which lead to the relative success and failure of such new forms as may be born into the world", and Darwin responded in a letter to Lyell saying "That may be a very good theory, but it is not mine" (this exchange is related by Poulton, 1909).  

To understand Darwin's view, one must think of heredity and variation as processes mediated by *fluids*, not discrete particles.  The Duke and others mistakenly try to discretize Darwin's world, but Darwin was deeply committed to a doctrine of natura non facit salta, and this was entangled in his view of heredity and variation.  In Darwin's view, the variations manifested in an individual emerge because the underlying hereditary substances FLUCTUATE continuously in potency on the scale of one or a few generations.  Due to the fluidity of inheritance, these fluctuations BLEND during reproduction.  Infinitesimal fluctuations emerge (roughly) in every character in every generation, especially under altered "conditions of life".  It is these relative fluctuations, NOT mutant particles, that provide the leverage for selection in Darwin's view. 

The idea that a single variant individual (a "mutant" we would say) could be the start of something new in evolution is our view today, and it is a view suggested by many of Darwin's critics, and quite a few of his friends (e.g., Galton, Huxley), but it was NOT Darwin's view.  

Thus, "Darwinian selection" is alien to population genetics.  One may attach a selection coefficient to a Mendelian allele, but in Darwin's theory, there is NO constant informational or hereditary unit to which one may attach a selection coefficient. 

Instead, a properly constructed Darwinian theory for the movement of an evolving system in morphospace would yield an equation for the instantaneous rate of flow as a function of direction (in morphospace), conditioned on the underlying flow of variation and the selection differential.  A key problem with this view is that the evolving system has no inertia-- given the high rate of fluctuation, it would flow all over the place if it were not for selection keeping it in place.  Its no wonder that Darwin and his advocates placed so much faith in selection.  In their view, the evolving system was like a ball placed on an uneven surface, ready to roll in any direction and come to rest in a local minimum.

I wish that some theoretician or philosopher would develop a formalism for this. Perhaps it corresponds to the quantitative evolutionary genetics, a formalism that does not address genes explicitly, but simply allows for some heritability of the variation in a quantitative trait. The master equation (2002) is
 


where delta z, the "response to selection", is the change in values for a vector of trait-values z (and G/P is a heritability matrix, and S is selection differentials).  This framework corresponds most closely to the folk theory of Darwinism, because it assumes abundant infinitesimal variation.  However, it can be justified under Mendelian inheritance given extreme assumptions (no drift, infinitely many loci each with alleles with infinitesimal effects on the trait, and no linkage, etc).  Steppan, et al. say, in regard to this "quantitative genetics" framework,

"If stochastic events, such as genetic drift, fluctuating adaptive landscapes and rare mutations, are more important, then quantitative genetics might not be informative and macroevolution might be decoupled from microevolution.  Resolution of this issue is crucial to evolutionary biology as a whole" (p. 322)

Once we take an explicitly genetic view, everything changes.  Historically, there are two Mendelian views (2006).  For some purposes, its convenient to think of them as extremes on a continuum defined by the availability of variation, from low to high.  

1. At one extreme, there is the "neo-Darwinian" or "Modern Synthesis" view that allelic variation is always abundantly available, "selection" (reified) never waits for a new mutation, and evolution consists of shifting the relative frequencies of Mendelian alleles so that the multi-locus frequency distribution collectively reaches an optimum.  This view re-defines "evolution" as "shifting gene frequencies".  This view is easy to fit with the "forces" concept, so long as we make it deterministic.  But we could make it probabilistic and then it doesn't fit as well.

2. The other extreme is the canonical "mutationist" view in which evolution is a two-step process of mutation and acceptance. That is, a mutation arises, and its subject to acceptance or rejection. This view does *not* fit the force model in any way at all.   The probability of acceptance reflects any effects on fitness, i.e., what we might call "selection", although its a different kind of thing.  It corresponds roughly to the folk theory of causes used in molecular evolution, where evolutionists tend to think of evolutionary changes as "mutations" that have been "accepted" either by "drift" or by "selection". 

Arlin 

Poulton, E. B. 1909. Fifty Years of Darwinism. Pp. 8-56. Fifty Years of Darwinism: Modern Aspects of Evolution. Henry Holt and Company, New York.