Lipton’s (1) argument for greater support from successful prediction,
relative to accommodation, departs from a conventional Popperian account
in which support depends on surviving tests based on predictions and
possible falsification. Lipton argues instead that the accommodation case
suffers because it is not so improbable that the fit of the hypothesis to
the observations is due to “fudging” (which “competes with the truth
explanation”).
I use the word “improbable” because, phrased this way, Lipton’s
argument recalls an under-appreciated (2) aspect of Popper’s philosophy:
“corroboration”, or test “severity”. Popper (3) argues: "we demand
intuitively that only severe tests should count, and that the more severe
they are, the more they should count. But this is the same as to demand
that [the evidence] should be improbable on our background knowledge." Background knowledge not only may suggest that fudging “competes” as an
explanation for fit-as-evidence, but also may suggest other factors, such
as chance effects, as alternative explanations of evidence. For example,
Popper (3) argues that the "predictions which lead to the discovery of
Neptune, were such a wonderful corroboration of Newton's theory because of
the exceeding improbability that an as yet unobserved planet would, by
sheer accident, be found in that small region of the sky where their
calculations had placed it" (4).
Popperian corroboration provides general support for Lipton’s
arguments, but also suggests that there is no simple support-distinction
between accommodation and prediction-based evidence. A successful
prediction might provide very little corroboration [as for Popper’s
“soothsayers” (3)], while accommodation might provide lots [noting that
evidence for an hypothesis need not be based on prediction, (3)].
In phylogenetic inference (2), we adopt the phylogenetic hypothesis
that best fits observed character data. This accommodation nevertheless
provides good corroboration when the resulting degree of fit is so good
that it cannot easily be explained away by “chance” or other factors.
Contrast this with weak predictions; we start with a favorite
phylogenetic hypothesis and predict observation of a character state
shared by two hypothetical sister taxa –- but observing this, among a
multitude of observed characters, means little (2).
Corroboration therefore suggests guidelines in phylogenetics (and
elsewhere) for prediction and accommodation. Just as predictions should be
improbable given only background knowledge, accommodations may seek
improbability. For example, we may accumulate those taxonomic character
sets (e.g., particular gene sequence data) that have strong phylogenetic
signal. We don’t know which hypothesis will accommodate these observations
-- but we can expect that the strong signal will mean improbably good fit,
and so corroboration, for that best-fit hypothesis.
References and Notes
1. P. Lipton, Science 307, 219-221.
2. D. P. Faith, Aus. Syst. Bot. 17, 1-16 (2004). Available at:
http://www.amonline.net.au/systematics/pdf/sb03017.pdf
3. K. Popper, "Realism and the aim of science." in Postscript to
the Logic of Scientific Discovery, W.W. Bartley, Ed., (1983) (Reprinted
in 1992 by Routledge: London, UK); K. Popper, Conjectures and Refutations
(Harper and Row, New York, NY, 1968).
4. Past discussions [e.g., (5)] of Popper’s Neptune example have
focused on the associated difficulties in falsification, and ignored the
successful corroboration, highlighting the under-appreciation of
corroboration and improbability of evidence.
5. S. Thornton, in The Stanford Encyclopedia of Philosophy, E. N.
Zalta, Ed., (Winter 2002 edition). Available at:
http://plato.stanford.edu/archives/sum2003/entries/popper/
E-Letters: ![[Read E-Letter]](/icons/misc/sectionDOWN.gif){kind=icon}
