Is Natural Selection a Population-Level Causal Process? - Mechanism and Causality in Biology and Economics - page 165

According to Brandon, random drift is any deviation from expected result due to
sampling error. 1 Now, sampling error occurs whenever the sampling process is
unrepresentative. Thus, any deviation from the expected result is a proof that the
sampling process is unrepresentative; conversely, only deviation can prove that the
sampling process is unrepresentative. On this account, deviation and sampling
error/unrepresentative sampling are, so to speak, two sides of one coin. And that
is why, on Brandon's view, drift as process (viz., sampling error/unrepresentative
sampling) cannot be defined independently of drift as outcome (viz., deviation from
expected result). 2 For Brandon, deviation and only deviation indicates a sampling
process is not representative, which in turn requires that the effective population
size be finite. 3
Not surprisingly, Millstein disagrees with Brandon's account of drift. As men-
tioned above, the idea of “fitness” on which Brandon heavily relies to make his case
is highly controversial, and Millstein does not need it to make the points she wants
to make. In my view, one can appreciate Brandon's account without drawing on the
notion “fitness.” And that is why, as shown in the preceding paragraph, I rephrased
his remarks without using “fitness.” To parallel Millstein's process-oriented
account, I would say what is essential to Brandon's view is that drift is an
unrepresentative sampling process.
Millstein ( 2005 ) points out a second problem with Brandon's view: granting that
natural selection is a probabilistic sampling process, any result, with or without
deviation, is no proof of whether natural selection as process has occurred. And
since drift is also a probabilistic sampling process, the same can be said of drift as
process. So contrary to Brandon's view, outcome is neither sufficient nor necessary
to distinguish conceptually natural selection from drift.
Granting that discriminate sampling process is probabilistic and that indiscrimi-
nate sampling process is probabilistic as well, I think Millstein is right that process
alone suffices to make a conceptual distinction between them. But it does not follow
that process alone suffices to make a conceptual distinction between natural selec-
tion and drift. Unless it is the case that natural selection surely is a discriminate
sampling process and that drift surely is an indiscriminate sampling process,
Millstein would not be justified in inferring that outcome is neither sufficient nor
necessary to distinguish conceptually natural selection from drift. As it turns out,
the crucial issue is whether drift surely is an indiscriminate sampling process.

1

As an anonymous referee points out, Brandon's views on drift may have changed since 2005.
Because my paper focuses on Brandon's ( 2005 ) argument where he responds directly to Millstein,
I will not refer to Brandon's ( 2006 ) “The Principle of Drift: Biology's First Law,” where he seems
to offer a fleshed out alternative view of drift.
2 Note that Brandon deliberately separates deviation from sampling error in using the expression
“due to,” which suggests, among others, that he views deviation as outcome and unrepresentative
sampling/sampling error as process.
3 In an infinite population, it is extremely unlikely that deviation arises. Alternatively speaking, it
is extremely likely that sampling process is representative.