<oologualJournal o f f h e Linnean SocieQ (1982), 74: 305328. With 2 figures
The use of parsimony in testing
phylogenetic hypotheses
A. L. PANCHEN
Department of <oology, The UniversiQ, Newcastle upon Tyne
Accepted f o r publication
3une 1981
With the advance of cladistic theory differencesin principle between it and other systematic techniques
are few but of fundamental importance. In the mechanics ofclassification they are confined to ranking
and the rejection of paraphyletic taxa. In cladistic analysis, leading to cladograms, trees and phylogeny
reconstruction, inconsistencies in apparent synapomorphies are said to be resolved using Popper’s
hypothetico-deductive method together with the principle ofparsimony. However, not only do cladists
not use Popper’s methodology, which is inconsistent with parsimony, but their use of parsimony is
invalid as a test of phylogeny. The only accepted extrinsic test of a classification is that enunciated by
John Stuart Mill. It has been claimed that cladistic classificationsyield the best results when judged by
Mill’s criteria, but this is only possibly the case with analytic classifications produced by numerical
techniques. No satisfactory test exists in normal (synthetic) cladism for distinguishing synapomorphy
from homoplasy. The effects of this are particularly dire in cladograms and classifications involving
fossils in which a Shfenreihe arrangement is adopted.
KEY WORDS:--cladistics - homoplasy - hypothetico-deductive - parsimony - stufenreihe.
CONTENTS
Introduction . . . . .
Cladograms and classification .
Cladistic analysis . . . .
The hypothetico-deductive method
The use of parsimony . . .
A natural classification? . .
Conclusions-Does it matter? .
Acknowledgements. . . .
References. . . . . .
Addendum
. . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. . . . . . . . . . . . . .
305
307
308
309
314
317
319
324
324
326
INTRODUCTION
J
The most remarkable feature in the history of systematic biology in the last two
decades has been the rise of ‘Phylogenetic Systematics’ associated with the name of
W. Hennig. Hennig’s (1950) original account of his systematic methodology
attracted little attention outside his native Germany but the translation of a revised
manuscript by Davis and Zangerl (Hennig, 1966) resulted in an extraordinary
surge not only of interest in, but also advocacy of, Hennig’s system by some
taxonomists and, in reaction, its equally emphatic rejection by others. The spread
+
0024-4074/82/030305 24 $02.00/0
305
0 1982 The Linnean Society of London
306
A. L PANCHEN
of cladism, as it is now almost universally termed, has been marked by much
irrationality and considerable acrimony on both sides, which has frequently
obscured the scientific issues involved. The controversy is a depressing instance of
the inability of two groups having different paradigms to communicate with each
other (Kuhn, 1970).
Many of the differences between cladists and non-cladists have now been
resolved b) the advance ( o r retreat) of cladist theory and others concern nonessential con\.entions rather than points of principle. Common criticisms are
summarized and discussed by Mayr (1974). Notably irritant was Hennig’s
supposed attitude to speciation. Firstly, it was asserted that he believed that
speciation was invariably dichotomous. In fact Hennig (1966) was quite clear that
a dichotomous differentiation of the phylogenetic tree . . . “is primarily no more
than a methodological principle” (p. 2 10, see also immediately following
discussion\. The idea that dichotomy represented the normal, or even only, mode
of speciation seems to ha\re been introduced by Brundin (e.g. 1968) and supported
by a number ofcladists, e.g. more recently, Eldredge & Tattersall (1975), although
it is rejected by Eldredge 11979). Fortunately, it is unlikely that it is supported as a
dogma by any cladist today.
Another side issue did originate with Hennig: this is his curious convention that
e\en if an ecologically isolated population gains genetic isolation from its parent
stock, both should be regarded as new species (Hennig, 1966: 56-65) which seems
at odds with his own ‘dexriation rule’ (1966: 59, 207; Brundin, 1968) that this is the
normal mode of speciation. However, the consensus now seems to be that
“Hennig’s views on limiting species at branch points are irrelevant to cladistic
practice” (Platnick, 1980).
Also contentious has been, and still is, Hennig’s definition of monophyly
(Hennig, 1966: 72-73; Nelson, 1971 ; t\shlock, 1971, 1972, 1980; Farris, 1974;
Platnick, 1977). Cladists claim, correctly, that they have a strict and unambiguous
definition ofa monophyletic group as one containing all the known descendants of a
single ancestral species and including that species. Others, notably Ashlock and also
Mayr (1974) accuse the cladists, again correctly, of using an established term in a
novel sense. For most workers interested in phylogeny the concept of a unique
common ancestral species is associated with monophyly (Mayr, 1969) in contrast
to Simpson’s (1961) far too accommodating definition, usually characterized as
‘minimal monophyly’. However, the non-cladist concept includes what the cladists
exclude as paraphyletic groups, those taken to be derived from a unique ancestor
but not containing all its descendants: Reptilia has been the favourite example
from Hennig onwards. Ashlock provides a sensible compromise. Hennigian
monophyly is distinguished as holophyly, which together with paraphyly
comprises monophyly in its original sense. Holophyly is then unambiguous, as is
paraphyly, and both may be contrasted with polyphyly. A polyphyletic group is
one in which the membership is not uniquely derived.
I n the sections which follow I propose to identify the irreducible minimum of
significant differences between cladism and traditional classification. These
include ranking and the attitude to paraphyletic groups. I then go on to examine
the claims of cladists that their methodology includes the use of the hypotheticodeductive method, which leads to what I hope to demonstrate is a fatal
methodological flaw in their system-their
eccentric use of the criterion of
parsimony.
USE OF PARSIMONY
307
CLADOGRAMS AND CLASSIFICATION
One of the most important distinctions that has emerged in recent cladistic
discussion is that between cladograms and trees. Eldredge (1979) defines a
cladogram as “a branching diagram depicting the pattern of shared similarities
thought to be evolutionary novelties (‘synapomorphies’) among a series of taxa”
and a phylogenetic tree (phylogram) as “a diagram (not necessarily branching!)
depicting the actual pattern of ancestry and descent among a series of taxa”.
The distinction between the two was made explicit by Tattersall & Eldredge
(1977) following an unpublished manuscript by Dr G . Nelson. Before this many
cladists claimed that their cladograms represented phylogeny directly (or at least
the cladogenetic component of it) and also that a classification should be a
redundant image of the cladogram which is logically prior to it. Now the emphasis
has changed and cladograms and the classifications derived from them represent
“not . . . the order of branching of sister-groups, but the order of emergence of
unique derived characters, whether or not the development of these characters
happens to coincide with speciation events” (Hull, 1980). Thus the emphasis of
cladistics has moved away from the reconstruction of phylogeny (not all or perhaps
even a majority of cladists are interested in trees) to a hierarchical clustering of
synapomorphies for its own sake, with the assumption that the regularities exposed
are inherent in nature (Platnick, 1980).
With the development of this attitude to the significance of cladograms, cladists
seem to be losing or to have lost any interest in the mechanism or mode of
evolution (Hull, 1980). Platnick (1980) notes that cladistic analysis can be used to
demonstrate relationships between any entities which change by modification
through descent. Platnick & Cameron (1977) have shown the applicability of the
technique to linguistics and to the texts of ancient manuscripts. The ultimate, or
perhaps penultimate, position seems to have been reached by Nelson (1979: 8) to
whom “a cladogram is an atemporal concept .. . a synapomorphy scheme”.
However, Nelson still appears to retain an interest in phylogeny, which has been
lost by some other cladists.
An inherent technical difficulty in cladistic classification concerns the matter of
ranking. The problem with respect to the incorporation or fossil taxa of low rank as
sister-groups of taxa (with extant members) of high rank has been ably discussed
by Patterson & Rosen (1977). They propose the adoption of two conventions; (1)
sequencing, as proposed by Nelson (1972, 1974), such that a series of fossil taxa of
the same rank within a taxon of higher rank is arranged so that each taxon is the
sister group of all those succeeding it; (2) the use of the term ‘plesion’, in addition
to the normal category name from species up, to avoid the necessity of (e.g.) a fossil
species being the sole representative of a monotypic genus, family, order (and
intermediate ranks) etc. The alternative is exemplified by McKenna’s (1975)
classification of mammals, without (as Patterson & Rosen point out) the mammallike reptiles which together with the mammals form the clade Theropsida. An
enormous number of ranks and thus categories was required. McKenna formulates
this number as lying between log,n and n-1 for n species of organisms: he
optimistically suggests that the number (r) is nearer the former than the latter. I
had independently reached the same formulation ( 1, if the terminal taxa, species
in this case, are counted). r = 1 +log,n derives from a ‘perfect’ cladogram in which
there are exactly 2“ species (where x is any whole number) and every branch has a
+
308
A. L. PANCHEN
dichotomy at every rank (Fig. 1A). The addition of one species to such a perfectly
symmetrical cladogram requires an additional rank ! The worst case (r = n) applies
to th; ‘Hennigian comb’ (Fig. lB), in which each terminal taxon is the sister group
to all those to its (conventional) right. This type, or something close to it (e.g.
Patterson & Rosen, 1977; Gaffney, 1979), has become almost standard for cladistic
palaeontologists and, unless the autapomorphies of the terminal taxa are
emphasized, comes remarkably close to the scorned ancestor-descendent sequences
of unreconstructed palaeontologists. It is, if considered equivalent to a ‘tree’, also
identical to the concept of ‘Stufenreihe’ of Abel (1929; Simpson, 1953).
For something over a million known species of living animals (e.g. Shorrocks,
1978), the minimum conceivable number of ranks required for a complete cladist
classification of the animal kingdom is something over 20 ( r = 1 logp) (as noted
by McKenna), while the maximum is of course something over a million! I feel
much less sanguine than McKenna that the true number would be near the
smaller figure, but the pressure to approach the less extravagant formulation, even
for much more modest taxonomic enterprises, probably affects the judgement of
neontological cladists (see Conclusions below).
A final point about classification concerns a substantial and important difference
between cladists and their opponents. The aim of cladistic analysis is the
production of holophyletic taxa ; paraphyletic taxa are rejected. Traditional
taxonomy accepts paraphyletic taxa if these result from the effective removal from
a holophyletic taxon of a clade or clades which have reached a significantly higher
grade. The standard example, as noted above, is the paraphyletic class Reptilia
from which the holophyletic taxon (traditionally class) Aves has been in effect
removed (possibly together with the class Mammalia, depending on one’s opinion
of the status of the ‘mammal-like reptiles’ and origin of the Theropsida (Panchen,
1972; Kemp, 1980)). In other words, while only cladistic criteria are used by
followers of Hennig, yielding (they hope) holophyletic groups, traditional taxa are
monophyletic in Ashlock’s sense but may be paraphyletic if the separation of a
descendant clade is dictated by phenetic criteria (Mayr, 1974). It is worth noting
that this is equivalent to the use of significant autopomorphies to separate and rank
taxa (leaving behind paraphyletic groups) by traditional taxonomists, while
cladists reject the use of autapomorphy in this way.
+
CLADISTIC ANALYSIS
Cladistic analysis consists of the recognition, amongst taxonomic characters, of
Figure 1 . The two extreme forms of cladogram, shown for eight hypothetical terminal taxa. A,
Symmetrical cladogram, giving the minimum number of ranks per number of terminal taxa:
r = I +log,n; B, ‘Hennigian comb’ (Shrfmrcihc) cladogram, as often p d u c e d by palaeontologists,
giving the maximum number of ranks: r=n. For further explanation see text.
USE OF PARSIMONY
309
autapomorphies, synapomorphies and symplesiomorphies, preliminary to the
construction of a cladogram. Autapomorphies are diagnostic characters unique to
a taxon of a given rank; synapomorphies are unique to a pair of taxa (‘sistergroups’) of that rank, and symplesiomorphies extend beyond the sister group
within an inclusive taxon of higher rank. As has been repeatedly emphasized by
cladists a symplesiomorphy at one rank may be a synapomorphy at the next higher
rank, while a synapomorphy at one rank is always a symplesiomorphy at the rank
below: thus the cladogram resulting from cladistic analysis is a nested set of
synapomorphies uniting sister groups.
It seems to be agreed by most, if not all, cladists that the cladograms that they
produce represent hypotheses (e.g. Cracraft, 1979; Gaffney, 1979; Eldredge,
1979). Those cladists who still hold that they are entitled to call themselves
‘phylogenetic systematists’ in the literal sense assert that a cladogram represents
phylogeny, via one of the possible trees that can be constructed from it.
Synapomorphies are then taken to be evidence of genuine relationship, with the
underlying acceptance of the axiom that evolution has occurred. Nonphylogenetic cladists, who do not insist on that axiom, nevertheless presumably all
accept that the hierarchy of synapomorphies, and the homologies they represent,
correspond to something in nature and not just a solipsist ordering of sense-data.
The most obvious ‘something’, is the pattern of ontogenies of organisms which can
be arranged in a hierarchical system according to the principle of divergence in
ontogeny enunciated by von Baer (1828; Gould, 1977). A classification based on
homology due to the pattern of ontogenies, to which we have direct access, need
not presuppose evolution (Patterson, 1982).
Given that cladograms represent hypotheses, most cladists would like to gather
their methodology within the fold of empirical science by asserting that cladograms
can be tested. They further assert that this can be done using the hypotheticodeductive method.
THE HYPOTHETICO-DEDUCTIVE METHOD
This is invariably associated, and overtly by cladists, with the methodology
enunciated by Popper (1934, 1959, 1963). Platnick & Gaffney (1977, 1978a’b)
have provided useful reviews of Popper’s work, specifically aimed at their
taxonomist colleagues. Popper established a criterion of demarcation of empirical
(scientific) theories, separating them from metaphysical ones by potential
falsifiability. Thus, if scientific theories are in the form of strictly universal
statements, as Popper thinks they should be, then each is potentially falsifiable by
one or more singular statements, whereas a universal statement is not logically
verifiable. However, ‘numerically universal’ and ‘existential’ statements (“all
human beings are less than 9ft tall”, “there are black ravens”, respectively, are
examples of Popper’s), are not empirical theories and are potentially subject to
verification.
Popper’s system therefore consists of the proposal of a theory, cast as a universal
proposition, in explanation of the phenomena under study. After testing for
internal consistency, non-tautology, and against rival theories for explanatory
power an attempt is made to falsify it. Thus, new empirical predictions are made
from the theory and an attempt made to falsify these by observation and
experiment. The observation or experiment may yield a singular statement which
contradicts a prediction of the theory and thus falsifies the theory. Every time
710
A. L PANCHEN
falsification fails the theory is corroborated, but it can never be proved.
Unfortunately it was soon spotted that there is a logical flaw in this scenario (e.g.
Achinstein, 1968). Using the logician’s modus tollens (of which more below) a
universal proposition can only be refuted by a contradictory particular proposition
or ifide Popper) basic statement. However, these basic statements, derived from,
but not identical to, the results of observation or experiment, cannot be verified.
They have empirical content, or are “facts interpreted in the light of theories; they
are soaked in theory, as it were” and contain universal terms denoting “physical
bodies which exhibit a certain law-like behauiour” (Popper, 1963: 387). Thus basic
statements, being unverifiable, cannot irrevocably refute universal propositions.
Popper was aware of this difficulty, but his discussion of it is perfunctory (e.g.
Popper. 1963: 41 i. As a result both critics and followers of Popper, including a
number of distinguished scientists (e.g. Bondi & Kilmister, 1959; Medawar, 1967 :
144J concluded that he held that irrevocable falsification was possible and also that
it was characteristic of the best science. Meanwhile, however, a number of
scientists and philosophers pointed out that, whatever the logical status of
falsification and despite its excellence as a moral precept for the working scientist,
it was not the way in which scientists (even physicists) usually operated
(ilchinstein, 1968; Kuhn, 1962; Lakatos, 1970; Harris, 1972; Brush, 1974).
Lakatos ( 19701 in fact tried to save Popper from himself by asserting that he
\Popper) never was a ‘dogmatic falsificationist’ (except perhaps in a prepublication phase). Lakatos further distinguished two other stances which he
claimed characterize Popper’s writings, those of ‘naive’ and ‘sophisticated
methodological falsificationism’. The former is distinguishable from dogmatic
falsificationism only in that one admits the logical fallacy of irrevocable
falsification and then proceeds as if that fallacy did not exist. The latter is a very
different kettle of fish: it is not clear that Popper accepted it before Lakatos (1970),
but he certainly appears to have done so afterwards (Popper, 1972), despite his
strictures on Lakatos’ interpretation of his work (Popper, 1974).
’Sophisticated methodological falsificationism’ is characterized by Lakatos
( 1970: 1 1 6 132) “. . . a scientific theory 7 is falsified if and only if another theory
7‘ has been proposed with the following characteristics: (1) T’ has excess
empirical content over T : that is it predicts novel facts, that is, facts improbable in
the light of, or even forbidden by T ; (2) T‘ explains the previous success of T, that
is, all the unrefuted content of 7’ is included (within the limits of observational
error) in the content of T and (3) some ofthe excess content of T is corroborated.”
Thus, the emphasis shifts from the confrontation of theories by data to the
necessary competition between rival theories. The sophisticated methodological
falsificationist would never accept the falsification of any theory by observation or
experiment unless a rival theory having Lakatos’ three characteristics is on hand to
take over : “no experiment, experimental report, observation statement or well-corroborated
lozdevel f a l s i f i n g hypothesis alone can lead to falsification” (Lakatos, 1970: 119).
If, as seems to be the case, Popper (1972) assents to all this it seems to me to
constitute a U-turn of governmental completeness. There is, however, a verbal
differences between Popper and Lakatos. Popper (1974: 1009) would assert that a
theory may be falszjied in the absence of a more successful rival, but not necessarih
rejected.
One other feature of Lakatos’ formulation must be noted. To him scientific
progress proceeded by the progress from one theory to another and so on:
USE OF PARSIMONY
311
. . . If this progress resulted in increasing empirical content at each stage
the ‘problem-shift’ was progressive, if not, degenerating.
The idea of using Popper’s methodology to test taxonomic hypothesis seems to
have originated with Bock (1974), an emphatic non-cladist, and Miles (1973),
then an incipient cladist. Despite his stance, Bock concludes (also Bock, 1977) that
the most severe tests in classification are cladistic tests involving falsification.
Cladists rapidly accepted this thesis (Wiley, 1975; Gaffney, 1975) and in the last
few years an enormous and repetitive literature on the use of the hypotheticodeductive method in testing cladograms and/or classifications has sprung up (the
pages of Systematic <0010gy give an all-too-generous sample of this).
Two themes are prominent, firstly, whether the method is applicable to
classification at all, and, secondly, instructions for its use. Works of actual
classification purporting to use the hypothetico-deductive method are relatively
(but perhaps not surprisingly) rare.
Seminal in exploring the former theme was a paper by Kitts (1977). Firstly he
pointed out that the diagnoses of taxa are not empirical hypotheses potentially
subject to falsification, but nominal definitions (subject to acceptance, rejection or
modification in the light of increasing knowledge) simply because we cannot
determine if an organism belongs to a given taxon by reference to anything other
than the sort of character states embodied in diagnoses. Secondly, he concluded
that the hypotheses of taxonomists, such as the denotation of a given taxon, are
arrived at inductively and thus correspond to Popper’s ‘numerically universal
statements’. As such they cannot be subjected to testing by the hypotheticodeductive method, which deals only with ‘strictly universal statements’, although
in theory they are both verifiable and falsifiable by enumeration. This was further
developed by Fox (1978) in an unpublished paper in which he pointed out the
temporal and spatial limitations of any hypothesis of relationships, and hence
descent, based on a cladogram. In response to Kitts, Cracraft (1978) played down
the importance of Popper’s distinction: Patterson (1978) on the other hand
suggested that, philosophically, holophyletic taxa are individuals to be recognized
rather than tested, acknowledging the priority of Griffiths and Hull. H e further
rejected the idea that the principle ofparsimony constitutes a test in Popper’s sense,
with which I strongly agree (see below). The idea that taxa are individuals
parallels the philosophical theory, independently reached by Putnam ( 1975) and
Kripke (1980)’ that terms denoting ‘natural kinds’ are proper names.
Nelson (1978) asserted that a hierarchical pattern resulting from cladistic
analysis (and thus expressible as a cladogram) is testable. He cites the hierarchy
(the letters on each line representing the included terminal taxa in a taxon of
higher rank) :
T,-T,-T,
ABC
A
BC
B
C
falsified by additional information giving :
ABC
AB
A
B
C
A. L. PANCHEN
312
. . . “under the assumption that the AB component is a shared derived character
[synapomorphy) and not a mistake” (my italics). But the identification of such
‘mistakes’ depends ultimately on the principle of parsimony.
A work frequently cited with approval by cladists (e.g. Bonde, 1977) as an
example of the use of the hypothetico-deductive method is Miles’ (1975) discussion
of the relationship of the Dipnoi (lungfish). Miles commences his study by
presentation of a cladogram of the major groups of gnathostome vertebrates, but
omitting the Dipnoi, as a framework for discussing the position of the latter on the
cladogram. Various theories of the relationship of the dipnoans to different groups
in the cladogram are then discussed seriatim. In each case, common characters
(putative synapomorphies) cited by previous authors are listed as ‘tests’ of
alternative proposed relationships and those characters discussed and identified as
symplesiomorphies, synapomorphies, autapomorphies, convergent and parallel
characters and non-diagnostic ones (‘chance’), before being used to test those
relations hips.
Miles’ paper is written with exemplary clarity and has an original format which
enhances that clarity. However, his testing technique is not that of the hypotheticodeductive method. This is principally because he makes no potentially falsifiable
predictions, bold or otherwise, from the various hypotheses of relationship that he
presents. Instead, he presents in effect a sort of matrix of incompatible theories of
dipnoan relationship each with the set of characters cited by their partisan authors.
It is in many respects an admirable study but save for some characteristic jargon
has little to do with Popperian method or philosophy.
It seems to me that Miles and other cladists may well be confusing Popperian
falsification (the hypothetico-deductive method) with the logician’s modus tollem,
which is merely the logical form in which it is cast. Modus tollens ( I f p then q, not q,
:. not p) may be characterized by the following valid argument whose
propositions (in agreement with Gardiner, 1980) I believe to be true:
(1’1 r f the porolepiforms are the sister-group of some or all tetrapod groups, then
they will be among those fish known to be choanate (if p then q ) .
( 2 : Porolepiforms are not among those fish known to be choanate (not q ) .
(3) Therefore porolepiforms are not the sister-group of any tetrapod group (.’. not
P).
Unfortunately, in this case the consequent ( q ) does not constitute a new
prediction from the antecedent (p) (the choanate or non-choanate status of the
porolepiforms is taken to be already known). This is an exact analogy of the state of
affairs in Miles’ (1975) discussion. It could of course be the case that the
consequent represents a prediction from the antecedent: “let us see whether these
fish (none of whose snouts are yet described) are choanate, as a test of their status as
close relatives of at least some tetrapods.” However, again (and again
unfortunately) this would still not be a case of the hypothetico-deductive method
according to Popper, despite the logical validity of modus tollens and the truth of the
propositions. This is because each of those propositions is either contingent,i.e. not
a strictly universal statement but one that described a spatio-temporally limited
state of affairs, or (as might be the case with q ) a matter of definition. This is exactly
the point made by Kitts (1977). As he also points out, both types of proposition
are arrived at inductively, whatever the validity of that particular procedure.
Gaffney (1979) presents a useful didactic account of cladistic method
particularly with reference to the cladist’s attempted use of Popper’s methodology
USE OF PARSIMONY
313
and of parsimony. He further presents a tentative ‘theory of relationships’ of
‘rhipidistian’ fishes and tetrapods in the form of eight sets of nested
synapomorphies and the ‘Hennigian comb’ (Stufenreihe) type of cladogram. It is
perhaps significant that the autapomorphies of his nine terminal taxa are not
listed (see above, p. 308).
More importantly, however, Gaffney sets out very clearly the cladist approach
to falsification. H e considers the cladists’ paradigm case, the ‘three-taxon
statement’. Three taxa, thought to constitute a monophyletic group, are tested by
apparent synapomorphies to yield the conclusion that of the three (say X, Y, Z), Y
and Z are more closely related to one another than either is to X. Consideration of
only one apparent synapomorphy, shared by two of the taxa but not the third, will
yield this apparently unambiguous result ; but more than one may give equivocal
results ‘falsifying’ the original one. Assuming that the original result is ‘correct’ this
can happen for two reasons, parallel evolution, or misidentifications of a character
in two taxa as homologous when it is not (convergent evolution-Nelson’s
‘mistakes’). However, Gaffney (1979: 96-97) continues “When multiple series of
characters are used, it is often the case that all three possible hypotheses relating
three taxa will appear to be falsified by one or more character distributions. In this
case, the least-rejected hypothesis (the onefalsijied thefewest number of times) is the one
incorporated into the internesting system of hypotheses .. . The principle of
parsimony is widely invoked for this decision” (my italics).
This seems to me to be a bizarre use of the idea of falsification, being neither
‘dogmatic’, nor either version of ‘methodological falsification’ and certainly not
what either Popper or Lakatos meant by the term. Thus, while an accurate
representation of what cladists mean by falsification, it is not an application of the
hypothetico-deductive method at all.
All this, of course, need not matter: the fact that cladists have misunderstood
Popper does not necessarily invalidate their methodology. Nor does the fact that
they have apparently overlooked Popper’s strictures on the use of parsimony.
These are implicit in Popper (1959) and explicit in Popper (1972). In The Logic of
Scientijic Discovery, Popper (1959 : 78-84, 134-145) explores the relationship of
‘Simplicity’ to ‘Conventionalism’. The stance of conventionalism, which Popper
rejects, is the adoption of the convention, in the absence of a corresponding theory
of truth, that all ‘laws of nature’ shall be simple and that ‘simple’ laws shall be
protected from falsification by ad hoc auxiliary hypotheses. Conventionalism seems
to me to be in exactly the same spirit as the cladist’s acceptance of the phylogeny
derived from the cladogram which implies the least parallel and convergent
evolution and the fewest reversals, “ . . . not because nature is parsimonious, but because
only parsimonious hypotheses can be defended by the investigator without
resorting to authoritarianism or apriorism” (Wiley, 1975 : 23-uoted
with
approval by Cracraft 1979: 34; my italics). Admittedly, Popper (1959: 145) uses
the word ‘parsimony’ with approval, but here he is advocating not the
conventionalist appeal to simplicity, but the expression of hypotheses in a form in
which an attempt at falsification can be most easily made; not a choice amongst
hypotheses (cladograms), all of which have already been ‘falsified’ at least once,
but a choice among new hypotheses none of which has yet been tested.
Popper’s overt stigmatization of conventionalist parsimony ( 1972 : 294-295)
follows a discussion of reductions (e.g. of chemistry to physics, or biology or
psychology to physics): “I call bad reduction or ad hoc reduction the method of
314
A. L. PANCHEN
reduction by merely linguistic devices ; for example, the method of physicalism
which suggests that we postulate ad hoc the existence of physiological states to
explain behaviour which we previously explained by postulating (though not by
postulating ad hocj mental states . . . This second kind of reduction or the use of
Ockham’s razor is bad, because it prevents us from seeing the problem. I n the
picturesque as well as hard-hitting terminology of Imre Lakatos, it is a disastrous
case of a ‘degenerating problem shift’ [see above, p. 31 1 ; and it may prevent either a
good reduction, or the study of emergence, or both.” In the last clause I would
suggest the applicability of substituting “the study of parallel and convergent
evolution” for “the study of emergence’’ to make my point!
I have laboured my rejection of the cladists claim to be using the hypotheticodeductive method at some length, because implicit in that claim is one to a type of
scientific respectability which they deny their rivals. I n fact cladist methodology,
at least in the construction of cladograms, rests on only two principles, the search
for synapomorphies and the principle of parsimony. The former is undoubtedly
Hennig’s greatest contribution to systematics. It is not of course original, but most
non-cladist systematists who have commented on Hennig acknowledge the
importance of his insistence on it, (e.g. Mayr, 1974; Bock, 1977; Simpson, 1975),
and the word ‘synapomorphy’ has entered, albeit against some resistance, the
vocabulary of many non-cladists (including mine). However, in the construction of
dichotomous cladograms, Hennigians would be helpless without the principle of
parsimony. Cladism stands or falls (except for the use of one atrocious immunizing
strategy to be discussed below) on that principle.
THE USE OF PARSIMONY
The use of parsimony in classification and in phylogeny reconstruction was not
invented by cladists. Indeed, while used informally before, it was introduced as a
distinct technique into phylogeny reconstruction by phenetic (‘numerical’)
taxonomists including Camin & Sokal ( 1965), who incidentally referred to their
dendrograms of phylogeny based on phenetic principles as ‘cladograms’, a term
also suggested (in parallel!) by Mayr (1965). Sneath & Sokal (1973 : 321) cite
three types of ‘minimum evolution’ reconstructions of trees based on parsimony.
These are ( 1 ) a minimum number of evolutionary steps (Camin & Sokal, 1965) ;
( 2 ; a minimum number of mutational steps (Fitch & Margoliash, 1967), or ( 3 ) a
minimum length tree (Edwards L?L Cavalli-Sforza, 1964; Kluge & Farris, 1969).
The justification of all three types of reconstruction is the principle of parsimony
and that alone.
Camin & Sokal’s minimum evolutionary step method bears superficial
resemblance to the cladists’ method using parsimony in that they ( C & S) are using
a heterogeneous collection of characters in at least some of which it is possible to
identify a primitive and one or a series of derived character states. The phenetic
cladogram is then constructed (with hypothetical ancestors, ‘HTS’s, for the
terminal taxa, ‘OTU’sj such that the minimum number of evolutionary steps, i.e.
total of changes (within character and summing characters) from less to more
derived states, is involved. The method differs from that of cladists in that (1) there
is no identification of symplesiomorphies and synapomorphies at each rank; (2) all
character state changes (positive and apparent negative) are summed to give the
USE OF PARSIMONY
315
mean minimum number of steps, so that there is no concept of ‘falsification’ and
( 3 ) the maximum number of characters that can be specified is used. The method
is in fact an extension of phenetic classification in which ‘parsimony’ in the sense of
the ratio of like to unlike character states is of the essence. The best classification is
regarded as the one using the largest number of characters (originally unweighted)
and in cluster analysis ‘relationship’ (phenetic resemblance) is merely represented
by that ratio, without any phylogenetic or essentialist connotation.
The ‘minimum mutation’ use of parsimony is used by those reconstructing
phylogeny using differences between different organisms in the molecules of
homologous proteins (e.g. cytochrome c, myoglobin) . The method is explained by
Fitch & Margoliash (1967).A mutation distance between two homologous proteins is
defined “as the number of nucleotides [in the coding DNA] that would need to be
altered for the gene for one [protein] to code for the other”. This is derived from
the number of different homologous amino-acid radicals between the two
molecules and the number of nucleotide changes required to change from one
homologous amino-acid to the other in each case. A tree is then constructed by
associating the pair of organisms, represented by their homologous proteins, with
the minimum mutation distance. The pair is then averaged and the procedure
repeated until all the organisms are associated in a dichotomous tree in which the
length of the branches between nodes represents the mutation distance. Various
correction techniques are employed to allow for the bias in the order of clustering.
This technique is perhaps the most easily justified form of parsimony. Within the
system, using one series of homologous proteins and thus the codon they represent,
it is unknowable how many back-mutations or roundabout mutational routes there
have been (except perhaps by adding more organisms) so parsimony is the only
refuge and at least like is being compared with like.
However, when a tree so produced is compared with an agreed phylogeny from
normal anatomical characters there is not a one-to-one correspondance. Both the
pattern of dichotomies and the internode distances fail to match, if the latter are
compared to probable time distances estimated from the fossil record (RomeroHerrera et al., 1973, 1978). Molecular evolution is not parsimonious.
The minimum tree-length method derives from another technique of phenetic
classification in which characters are represented by axes in multi-dimensional
space so that each organism is represented in that hyperspace by a point plotting its
position on the axes of all its characters (Edwards & Cavalli-Sforza, 1964). A
phylogeny is then reconstructed as a network of minimum distances between the
points. The parsimonious assumption here is that evolution has covered the
minimum ‘phenetic distance’ to arrive at the terminal taxa. Philosophically this
assumption is similar to that of the ‘minimum step’ method.
The cladists use of parsimony contrasts strongly with all three of these phenetic
techniques. It should be particularly noted that cladists use parsimony in the
construction of cladograms not trees. When cladograms were taken to be direct
reconstructions of phylogeny (or of cladogenesis, which is what was meant by
‘phylogeny’), then the purpose of parsimony was the same as that of phenetic or
molecular taxonomists engaged on phylogeny reconstruction. This is still the case
with those cladists who assert that cladograms (via trees) represent phylogeny. For
those who are unwilling to make this assertion, or regard it as unnecessary
(Patterson, 1982), then parsimony, apart from being a methodological principle, is
presumably a test of homology.
316
A. L. PANCHEN
Two cladist uses of parsimony may be distinguished. The first is in the ‘threetaxon statement’ described by Gaffney (1979; and numerous earlier papers by
other cladists). This method is used to resolve inconsistent apparent
synapomorphies. The second is at first glance characterized by the example of
‘falsification’ cited by Nelson (1978) and noted above. This involves the
confrontation between conflicting cladograms.
The first use of parsimony is related to the cladist concept of ‘falsification’ noted
in the last section. The procedure is firstly to recognize symplesiomorphies present
in each of the three taxa (XYZ) being considered. The normal technique here is
out-group comparison (e.g. Gaffney, 1979 : 92-94). Secondly, autapomorphies of
each of the three taxa are used to diagnose those respective taxa but not to ‘test’
relationship. The remaining significant characters recognized by the cladist should
by synapomorphies of the presumed monophyletic group comprising all three taxa
and apparent synapomorphies uniting two of the taxa to the exclusion of the third.
If a number of the latter type of characters is considered then inconsistencies will
almost certainly arise due to parallel and convergent evolution, as noted above.
The cladist use of parsimony at its crudest then dictates that a simple majority vote
is taken: the two taxa united by the largest number of apparent synapomorphies
are declared ‘sister’ qroiips, while the ‘synapomorphies’ apparently producing the
two other possible combinations are pronounced false and thus due to parallel or
convergent evolution. 4 s I have pointed out before (Panchen, 1979), such a
conclusion could only be valid if ( a ) the organisms comprising the three taxa could
potentially be atomized into a series of discrete diagnostic characters of equal
taxonomic weight, such that the state of each character was in no way correlated
with the state of any other (or alternatively that only characters having these
features were used); and also (bj if a competent cladistic taxonomist could be
reasonably sure that in each taxon, the relatively tiny number of apparent
synapomorphies that he uses in his test has a similar ratio of true to ‘false’
synapomorphies as does the class of all its characters.
The first proviso introduces the vexed problem of weighting and character
correlation which has an extensive literature. It is worth noting that cladists
appear to he embarking on the same agonizing consideration of weighing and
correlation (Gaffney, 1979: 99-101; Eldredge, 1979: 172-1733 as did the
pheneticists perhaps a decade previously (compare Sokal & Sneath, 1963 :
118-120: Sneath & Sokal, 1973: 109-113; also Burtt, 1964). Both Gaffney and
Eldredge quote Hecht (1976) and Hecht & Edwards (1976) who rank kinds of
morphological features on a scale of significance of one to five or inversely (five to
one) of probability of parallelism. Their reactions, however, are different. Gaffney
agrees ‘wholeheartedly’ with Hecht & Edwards’ (1977) dictum that “From the
viewpoint of lineage detection, i t is more important to use a few well-analysed
morphoclines than many poorly or improperly analysed ones” ; while Eldredge
concludes that “However risky raw parsimony may be, we are still better off if that
is our primary criterion”.
The difficulties inherent in this first proviso are of course inherent in the
pheneticists’ methodology as a whole, in which a ‘head-count’ of characters is
taken in estimating phenetic distance, and thus a fortiori in their uses of parsimony
(minimum evolutionary steps and minimum length tree: see above). ‘The
stringent, in fact unrealizable, conditions of the second proviso are however unique
to the cladist use of the parsimony criterion.
USE OF PARSIMONY
317
Thus, even fz we allow that the course of evolution is parsimonious, which we
know it not to be (Romero-Herrera et al., 1973, 1978; Friday, this volume), and I f
we further allow that evolution is normally divergent, so that true synapomorphies
outnumber false synapomorphies, which in many cases appears to be untrue (e.g.
Butler: 1982) ; the first cladist use of parsimony (in the ‘three-taxon statement’) is
still illegitimate.
This is more emphatically the case because, unlike pheneticists, cladists do not
aim to use the maximum number of separable characters in assessing ‘relationship’.
Their use of parsimony is thus analogous to predicting the results of a competitive
election with a large electorate by asking the voting intentions of the first dozen
people one happens to meet.
The second possible cladist use of parsimony, involving the confrontation
between several incompatible cladograms, is even worse than the first. The
hypothetical case cited by Nelson (1978) and noted above (p. 3 11) might appear to
fall within this second category. He gives a classificatory hierarchy (in effect a
cladogram) falsified by another resulting from ‘additional information’. However,
in Nelson’s example the established division of the major taxon (ABC) is given as
(A) and (BC) while the new data are supposed to give (AB) and (C).The example
therefore resolves itself into a simple three-taxon statement. However, a possible
system would be to produce a series of cladograms, each based on a, different suite
of characters and then to accept the pattern to which the majority of cladograms
conform. This is perhaps an improbable exercise for one individual starting from
scratch, but a probable one for a cladist reviewing (e.g.) the number of
incompatible cladograms depicting the relationships of the major groups of fishes.
In this case, unless this is specified, even the number of ‘synapomorphies’ uniting
groups at the same rank may not be the same, so the use of parsimony is even less
advisable.
A NATURAL CLASSIFICATION?
The ideal of ‘natural classification’ has probably existed as long as naturalists
have tried to group the organisms in which they are interested into a hierarchical
system. In fact the ‘species problem’ that Darwin wrestled with and that the young
Alfred Russel Wallace and Henry Walter Bates optimistically set out for Amazonia
to solve (e.g. McKinney, 1972), was based on the concept of ‘natural
classification’. Evolution was the hypothesis explaining the seeming true
relationships displayed by classification. The species problem was the problem of
how the evolution suggested by classification occurred.
Thus, while most cladists and traditional taxonomists agree that a natural
classification should reflect phylogeny and thus evolution (and even pheneticists
agree that it should be possible to reconstruct phylogeny from classification), the
concept of natural classification is logically as well as historically prior to that of
phylogeny. Phylogeny then cannot be used to test classifications and some extrinsic
criterion must be used to evaluate them if, as I have tried to show, all intrinsic tests
are invalid.
The only extrinsic criterion for judging the naturalness of a classification which
has gained any currency is that formulated by John Stuart Mill (1872: 1974 edn:
714):
“The ends of scientific classification are best answered, when the objects are
318
A. L. P.4NCHEN
formed into groups respecting which a greater number of general propositions
can be made, and those propositions more important, than could be made
respecting any other groups into which the same things could be distributed
. . . A classification thus formed is properly scientific or philosophical, and is
commonly called a Natural, in contradistinction to a Technical or Artificial,
classification or arrangement.”
Mill’s criterion of naturalness was adopted by Gilmour (1961 ; Gilmour &
LYalters, 1963) and has been commended by pheneticists (Sneath, 1961 ; Sneath &
Sokal, 1973 : 24-25), traditional taxonomists (Mayr, 1974 : 96) and by Farris
( 1977, 1979, 1980), a “mathematical cladist”. Two related concepts appear to
derive from the development of Mill’s criterion. T h e first concerns the distribution
ofcharacters within and between taxa. Sneath (1961) points out that Mammalia is
a natural taxon because it corresponds to the distribution of a large number of
correlated features, but that a taxon confined to horses, mice and rabbits is not a
natural taxon because most of the features which would characterize such a group
are not exclusive to it but are common to other mammals (or at least placentals).
Sokal (1977, quoted but not cited by Farris (1980), further writes “Natural
polythetic classifications permit two types of predictions concerning character
states. These states should be homogeneous within taxa and heterogeneous
between them.” This introduces the second concept, the idea of the predictive
power of classifications, characterized by Fitch (1979) : “The essence of
predictivity in the sense used here is the degree to which a specific classification
agrees with characters not used in the formulation of that classification.” A
different formulation is to suggest that the most natural classification is the one that
yields the most predictions about a newly discovered organism assigned to a taxon
within it. This in the case of the discovery of an incomplete fossil specimen
representing a new species, might actually be tested (but inductively!) by
discovery of a second more complete specimen.
Farris (1977) set out to show that ‘phylogenetic systematics’ (in fact his version
of cladism could give more natural classifications, in the Mill-Gilmour sense, than
phenetic classification, defined as clustering by overall similarity. Using a
hypothetical data matrix of 14 characters in eight taxa, he showed that the
criterion of naturalness was correlated with the co-phenetic correlation coefficient
used by pheneticists, with a maximum of 1. The states for each of his 14 characters
were coded on a binary system as 0 ( =symplesiomorphy) and 1 (=synapomorphy
when shared b) two or more taxa within the table, autapomorphy if unique). By
an ingenious trick, Farris then demonstrated that under certain circumstances
clustering by o\ era11 similarity, as recommended by pheneticists, gave a coefficient
of 0, while clustering by special similarity (said to be equivalent to clustering by
synapomorphy) still gave a coefficient of 1. I n his original matrix both techniques
gave the same (‘natural’) result, but by replicating the distribution ofstates of some
of the characters a specified number of times his special similarity clustering
emerged unscathed.
Farris’ intention was to represent the replication as several characters sharing
the same 0 or 1 distribution in all the taxa, but Janowitz (1979) pointed out that it
could also be regarded as the weighting of a series of single characters, which
would amount to the demonstration of something very different. Furthermore, Dr
Garth Underwood pointed out to me at the present symposium (Underwood, this
USE OF PARSIMONY
319
volume) that Farris’ 0 and 1 character states need not necessarily represent
plesiomorphy and apomorphy respectively, but that the 1 states could be merely
minority states of a character. Farris’ demonstration, therefore, is of the superiority
of clustering by special similarity but not necessarily by Synapomorphy.
However, Farris (1977) also claims to have applied the same techniques to “50
sets of data on real organisms” and Mickevich (1978) demonstrates the superiority
of classification using clustering by synapomorphy, using stability as a criterion
(contrast Gaffney 1979 : 103-104: “stability is ignorance”). What Mickevich
means by stability is the congruence of classifications of the same organisms based
on different sets of characters, and she shows effectively that using the Wagner tree
method developed by Farris and his colleagues such congruence is significantly
better than for a number of other phenetic techniques. The method automatically
involves the use of parsimony by minimizing hypotheses of ‘homoplasy’ (parallel
and convergent evolution).
The mathematical techniques employed by Farris and his colleagues do appear
to result in more ‘Gilmour-natural’ classification than phenetic techniques using
overall similarity. I n the case of grouping by ‘special similarity’ as demonstrated by
Farris (1977) this is obviously only the case if the plesiomorph or apomorph status
of each character stated has already been correctly arrived at in advance, but there
is a more important point to be made. The computations of ‘mathematical cladists’
are based on pre-existing data sets that can be expressed as a matrix of taxa
plotted against character states and parsimony is handled much as pheneticists do,
being inherent in the matrix. I n a sense one could refer to the extraction of a
classification, or cladogram (and thus phylogeny), from such a matrix as analytic
(or divisive: Sneath & Sokal, 1973 : 202-203). Non-mathematical cladists on the
other hand tend to use a classificatory method which in the same sense is synthetic
(agglomerative: Sneath & Sokal) (see next section).
CONCLUSIONS-DOES
IT MATTER?
In summarizing my conclusions from this rather lengthy review one point must
be made very clearly. In criticizing cladism I am not thereby advocating any other
system of classification or phylogeny reconstruction. Traditional taxonomy
(‘evolutionary’ or ‘eclectic’ or ‘syncretistic’ classification) has always been
somewhat ad hoc in its procedure, yielding good results with competent taxonomists
and bad with incompetent ones. The standard warks on procedure (Simpson,
1961 ; Mayr, 1969) are to some extent rationalizations of a tradition that is too
largely intuitive.
Further, I am willing to accept that the mathematical cladism of Farris and his
colleagues yields more ‘Gilmour-natural’ classifications, greater congruence
between classifications of the same organisms based on different data and possibly
(although this is unknowable) truer phylogenies, but only when pre-existing (and
unbiased and accurate) data are used and only when the data refer to living
organisms, or fossils which are sufficiently complete for comparison of homologous
characters to be spread right across the series. It is also the case that if phylogeny
were entirely divergent (or, to include the views of non-phylogenetic cladists,
homoplasy did not occur) cladistic analysis would by the nature of the definitions
of plesiomorphy and apomorphy yield the best classification, even if done by rule
of thumb (and by people who collected their own data) rather than by
computation.
320
A. L. PANCHEN
However, in the real world parallel evolution is common and in some organ
systems in some taxa is known to be the dominant mode of evolution (Butler,
1982). Moreover, convergent evolution is frequently detected and is inherent in all
controversies about the taxonomic position of major groups. All these things being
so it is probable that the greatest danger in cladistic classification is the
unassailable conviction of most cladists that their system is not just the best, but the
only valid one, that the true “cladogram ‘falls out’ of the analysis in much the same
way as a phenogram is the direct output of some computer program in numerical
taxonomy” (Eldredge, 1979 : 170). Eldredge, while still a cladist, obviously sees the
dangers more clearly than many of his colleagues when he adds a footnote on the
same page:
“If; as is nearly always the case when more than one character is examined,
more than one possible pattern of synapomorphy is evident, it is clear that
only one can be “correct”. Other apparent patterns of synapomorphy are
therefore false, and the similarities upon which they are based are, by
definition, parallelisms. In the case where the investigator is making
conscious evaluations as to which character states are primitive and which
derived, conflicting cladograms imply actual analytic error, as well as the fact
of parallelism. As a result of the relatively recent production of a plethora of
cladograms, parallelism turns out to be a far more common evolutionary
phenomenon than even most of its more ardent aficionados had thought.”
T o be more specific I believe that overconfidence, combined with the inherent
difficulties in the use of parsimony when homoplasy is common and, a fortiori, with
the cladists’ invalid use ofparsimony, are together sufficient to cast grave doubt on,
or even invalidate the method. it‘hile some of the dangers are present in all nonmathematical cladistic analysis, there are others characteristic of neontological and
palaeontological cladism respectively.
The first obvious fault of cladism in general is perhaps a temperamental and,
one may hope, a temporary one. This is the temptation to produce new and
revolutionary groupings of organisms, or to resurrect old and discredited ones, just
to demonstrate that cladism is getting results. I suspect that several regroupings of
major vertebrate taxa now in preparation or in press fall into this category but it
would be invidious to cite them without closer study of the results.
Unfortunately this kind of exercise, often but perhaps wrongly interpreted by its
critics as professional frivolity, is all too easy with the cladists misuse of parsimony.
Starting with Hennig’s (1966: 1391 most important operational dictum-“Once a
monophyletic group has been recognized the next task of phylogenetic systematics
is always to search for its sister group,” the cladist, being human, will be anxious to
reinforce the pairing he has ‘discovered’. He will therefore, and rightly, search for
as many synapomorphies between his sister-groups as he can; but there are no
methodological constraints on this process. The cladist can always win his parsimony vote
by searching for more real or apparent synapomorphies to the neglect of apparent
synapomorphies uniting either of his sister groups to a third, at the same time
reassuring himself that he is subjecting his hypothesis to falsification and to
Ockham’s razor. This attempt at inductive confirmation exactly parallels the case
cited by Bonde (1975: 308-310), a committed cladist, concerning the argument
about the diphyletic origin of tetrapods. Jarvik, in a long series ofpapers (summary
in Jarvik, 1968) claimed that, while the Urodela were derived from one
USE OF PARSIMONY
32 1
group of crossopterygian fishes-the Porolepiformes, the Anura, and possibly all
other tetrapods, were derived from another-the Osteolepiformes. His conclusions
were based initially on very detailed anatomical studies of the snout of a restricted
number of species in the four groups principally involved and then extended, in an
attempt at inductive confirmation, to other regions of the head. Attempts at
refutation were rather unconvincing, as Bonde notes, until Jurgens (1971)
demonstrated that Jarvik’s urodele snout characters also occurred in primitive
Anura. This has been generally accepted as a refutation of Jarvik’s improbable
theory, but it would still be open to Jarvik to defend himself using the cladist type
of parsimony argument, although I am not aware that he has done so.
It may be argued (correctly) that biased defence of one’s own theories is
characteristic not only of cladistic groupings, but also of those traditional
taxonomy, and in fact of all science. However, because cladists group by
synapomorphy rather than by overall phenetic resemblance, there seems to me to
be a greater danger that they will dismiss obvious common features which conflict
with groupings, as due to symplesiomorphy or convergence.
The extreme case of this is what I referred to above (p. 314) as a n “atrocious
immunizing strategy” (against supposed ‘falsification’). ‘l’his is the onesynapomorphy grouping. Mayr, ( 1974: 1 19- 120) comments unfavourably on an
example from the Diptera cited by Hennig. A notorious example is Kuhne’s (1973)
attempt to demonstrate, against the consensus, that monotreme and marsupial
mammals are sister-groups. I quote his summary in its entirety: “By
reinterpretation of the tooth formul of Ornithorhynchus as dp/PM,-, a
synapomorphy with Marsupials has been discovered. Monotremes and Marsupials
are sister groups, together derived from Theria of the Middle-Cretaceous.” I also
quote from the body of the paper: “if only one synapomorphy is found and
recognized as such, and generally acknowledged, the problem is solved, in spite of a
number of other characters, which are not synapomorphic but plesiomorphic,
dubious or-under
the present state of knowledge-cannot
rationally be
discussed” (Kuhne, 1973: 61). “An essential point: a proven synapomorphy
cannot be counterbalanced by other arguments; if it cannot be refuted it has to be
accepted as evidence.” Emphasis or comment on my part would be superfluous.
Cladistic classification of living organisms, particularly if the terminal taxa are
species, is probably less deserving of criticism than that of major taxa which
include fossils. However, I believe there to be one factor which may bias the result,
of which the operator may be only dimly conscious. This is the temptation to
produce the most parsimonious cladogram. ‘Parsimonious’ this time, not in the
sense of minimizing homoplasy, but in minimizing the number of ranks. If all the
terminal taxa can be neatly paired 0% to the exclusion of monotypic taxa at any
rank, then the ‘perfect cladogram’ referred to above (p. 307) can be approached.
Thus, despite the absence of any other particular prejudice on the part of the
operator, the classification produced will deviate from true phylogeny. This is
more strongly the case again because cladists do not take overall phenetic
resemblance or difference into account.
The tradition of cladogram type amongst palaeontological cladists is quite
other. Here the ‘Hennigian comb’ (Stufeenrezhe) type of cladogram prevails. The
reasons why this is so are interesting. They are explained by Patterson & Rosen
(1977 : 128, 158-159) following a convention suggested by Hennig (1969). Firstly,
it was necessary to agree that fossil species or higher taxa, if they could be
322
.4. L. PANCHEN
sufficiently well characterized, could appear as terminal taxa together with living
species or groups. This was not then a matter of general agreement amongst
cladists, but a paper by Schaeffer, Hecht & Eldredge (1972) gave powerful
support. Secondly, the position of the fossils is established by reference to a
previously agreed pair of living sister groups, in Patterson & Rosen’s case
represented by the primithve living (‘holostean’) fish Amia and the extant
Teleostei. It always seems to transpire that one sister group, in this case Amia, is
more ‘primitive’ (,has more plesiomorph features) than the other, following
Hennig’s (1966) deviation rule. A search is then conducted amongst appropriate
fossils for groups, species, or occasionally single specimens which have one or more
character states corresponding to the autapomorphies of the more advanced living
group (extant teleosts). The fossil taxa are then ranged on a scala naturae of
cumulative synapomorph characters until t h e full total is reached at the extant
(Teleostei ) . Thus, the former autapomorphies of living teleosts become an internested series of synapomorphies of Teleostei more widely defined. This wider
Teleostei is then the sister group of the group represented by Amia and any fossil
relatives. However, the emphasis is always on the more apomorph extant group
and the Stufenreihe leading to it: hence the ‘Hennigian comb’. In many cases the
plesiomorph living taxon is not included in the cladogram, or even specified at all
(e.g. Gaffney, 1979 for the major taxa of tetrapods: Fig. 2). Gaffney presents an
anal>,sisof his cladogram as a series of internested three-taxon statements, but it is
clear from his list of cumulative synapomorphies that it was conceived as a
Stgfenwihe. It is also worth noting that some of his proposed taxa are polythetic
isneath Br Sokal, 1973): e.g. “bipartite braincase ossifications” is cited as a
synapomorphy of all Choanata (crossopterygian fishes plus tetrapods) while “at
least partial braincase fLision” is taken to be characteristic of the contained taxon
‘Iet ra pod a .
However, the worst special feature of classification by Hennigian comb, in
addition to those inherent in all forms ofcladism, is the even stronger temptation to
ignore contrary e\ridence, because even evidence that might suggest a symmetrical
cladogram is ignored or rejected. Again, perhaps invidiously as it is only presented
as a methodological example, we may cite Gaffney’s cladogram. H e gives as two of
his three autapomorphies of amniotes, the presence of an astragalus and a ‘closed’
\i.e.absent) otic notch. The second feature characterises all Microsauria and the first
much more distinctive one occurs in at least two families of that group (Carroll &
Gaskill, 1978 ) . However, Gaffney places the microsabrs (within the polyphyletic
group ’Lepospondyli’” as two steps more primitive (or two ranks up).
Thus, cladistic practice when dealing with taxa including fossils is only in
practice distinguishable from those old-fashioned ancestor-descendent sequences of
palaeontologists which terminated in a living taxon (e.g. Watson, 1940) by the
more careful specification of the h c t that the members of the sequence have their
own autapomorphies iif these are cited) and by the repudiation of stratigraphic
e\idence. However, most pre-Hennigian palaeontologists were well aware, at least
as far as their more speculative phylogenies are concerned, that their sequences did
not literally represent ancestral and descendent species. The rejection of
stratigraphy is a more important point. Schaeffer et al. (1972) quite rightly remind
their colleagues of the dangers of a too literal interpretation of stratigraphy and
also suggest that classification of fossil or fossil plus recent organisms should be
based entirely on morphology (and presumably other innate features) but that
USE OF PARSIMONY
323
CHO AN ATA
A
r
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
7
I
-I
I
I
TETRAPODA
I
I
I
*
L
f
I
1
I
N EOT ETR APODA
I
I
I
I
I
I
I
I
I
I
I
I
I
I
*
f
!
--
-4
I
AMNIOTA
I
I
I
I
I
I
I
I
I
8
I
I
I
I
I
I
I
I
SAUROPSIDA
’
DlAPSlDA
r----4
2
3
8
P
P
I
Figure 2. Stufenreiha-type of cladogram of the relationship of tetrapods and ‘rhipidistian’ fishes.
Asterisks refer to unnamed groups. Numbers refer to Gaffney’s list of synapomorphies. (After Gaffney,
1979.)
stratigraphic evidence should be used as a guide in reconstructing trees. I would
strongly suggest that stratigraphic evidence can also be used as a test (but not a
falsifier) of cladograms. A Hennigian comb whose dichotomies run in the reverse
direction to an established stratigraphic sequence (e.g. Watson’s, 1919, 1951,
‘trends’ in the evolution of temnospondyl Amphibia) needs more than the invalid
use of ‘parsimony’ for its defence.
In conclusion, it is fair to note the enormous distance that cladism has come in
its short career. Like the early years of ‘numerical taxonomy’ before it, it passed, or
is passing, through a phase of arrogant and repellent dogmatism which likewise
unleashed a storm of often ill-informed criticism. Its theoretical structure has also
changed enormously, to the point where certainly for some of its adherents
‘phylogenetic systematics’ is a total misnomer. Before its methodology can become
trustworthy, however, it will have to abandon its pretence to the hypotheticodeductive method and to parsimony as anything other than ad hoc method of
comparing alternative groupings. It will also have to come to terms with
traditional taxonomy and the problems of weighting, character correlation
(Riedl’s 1979 concept of ‘Burden’ is a particularly valuable contribution here) and
the functional significance of characters. I t may also be the case that a cladistic
I5
A. L. PANCHEN
324
approach may be more suitable in some cases and a more orthodox one in another.
I cite Fortey (3r Jefferies i1982) in theirjoint enterprise which has all the boldness of
the peace mo\.ement in Yorthern Ireland. Finally, and perhaps the bitterest pill of
all, the cladists may ha\re to admit the persistent intuitive element in classification,
as in all the best science. Only then will cladism, if still distinct and recognizable,
be (in Hennig’s oft-repeated phrase) “the general reference system of biology”.
,ACKNOM’LEDGEMESTS
I t is a pleasure to record the stimulus of discussions at the meeting of which this
\.olume is a record. I would particularly like to acknowledge helpful debate with
Drs K . A. Joysey, II\. E. Friday, Garth Underwood, Colin Patterson and Professor
P. h4. Butler. Dr Jane Heal, of the Department of Philosophy, University of
Sewcastle very kindly read and commented on the manuscript under great
pressure of time and Mrs E. B. Kinghorn typed the manuscript under the same
constraint.
REFERESCES
.\BEI.. 0..1929. P a h o b d o g i e und S/ammesgesrhir-hte. Jena : Fischer.
ACHINSTEIN. P., 1968. (Review of) Karl R. Popper: Conjectures and Refutations. British Journal JOT the
Philosopiy .fScience, 19: 159--180.
ASHLOCK, P. D., 1971. Monophyly and associated terms. SyshzoliC ~ o o l o g y ,20: 6 3 4 9 .
ASHLOCK, P. D.. 1972. Monophyly again. Sysfmafic ~ o o l o g y 21:
, 43CF438.
.ASHLOCK, P. D..1980. An evolutionar). systematist’s view of classification. $ystonalk <oology, 28: 441-450.
B.\ER. K.E. von. 1828. Eiitwi-tlrn~s~~rsrhirhte
der T h i u e : Beobachtuns und Repexion. Konigsberg : Borntrager.
BOCK. W. J . , 1974. Philosophical foundation of classical evolutionary classifications, Systonatu <oology, 22:
‘575 392.
BOCK. W.J.%
1977. Foundation and methods of evolutionary classification. I n M. K. Hecht, P. C. Goody & B.
X4. Hecht :F.ds‘. Major Pa/terns in lbrfebratr I<ioluiion: 851 895. S e w York: Plenum.
BOSDE, S . . 1975. Origin of “higher groups”: viewpoints of phylogenetic systematics. Colloqun internatzonaux du
Centre natinnal dr la Rcrherche Scimtifiqiie. 218: 29:1- r324.
RONDE, N.. 1977. Cladistic classification as applied t o vertebrates. I n M. K.Hecht. P. C. Goody Br B. M. Hecht
iEds Major Pattermi in Ibrtebrate Eoolu/ion: 741 - 4 M . New York: Plenum.
BONDI, H. & KILMISTER. C. \V.. 1959. The impact of I-ogik der Forschung. British Journal fit the Philosophy of
Srirrur. 10: 55 ~ 3 7 .
BRCSDIN, L.. 1968. .\pplication of phylogenetic principles in systematics and evolutionarv theory. .Vobe/
.Smposium. 4 : 173 495.
BRUSH, S. G., 1974. Should the history of science he rated X ? Science, .V. Y.,183: 1164- 1172.
BCRTT, B. L.. 1964. :\ngiosperm taxonom) in practice. In V. H. Heywood & J. McNeill (Edsj, Phenelic and
Phvlogtnetic Clnssification: 5 -16. London : Systematics Association.
BCTLER. P. bl.. 1982. Directions of e\aoltition in the mammalian dentition. In K.A. Joysey & A. E. Friday
(Edsi, Problems of Phvlogenttu Reconstruction: 235244. London : Academic Press.
C.\AMIS.J . H . Sr SOK.AL, R . R. 1965. .A method fnr deducing branching sequences in phylogeny. khlution, 19:
31 1 :326.
CARROLL, R. L. & CASKILL, P., 1978. The order Microsauria. Memoirs. .4meruan Philosophical Sock&, 126:
Ik211.
CR.\C:RAFI’, .I.. 1978. Science. philosophy and systematics. $rJ.tematu .Soology. 27: 21 3 216.
CR:\C:R.IFT. J.. 1979. Phyloqenetic anal)sis, evolutionary models, and paleontology. I n J. Cracraft Pr N.
Eldredge (Eds,,Phvlogenetu .4nnbxir and Paleonkolou: 7 -39. New York: Columbia University Press.
EDW.ARDS, A. W. F. & CAVALLI-SFORZA, L. O., 1964. Reconstruction of evolutionary trees. I n V. H.
Heywood and J . McSeill (Eds;, Phenetic and Phylogmtir ClasstJication: 67- 76. London: Systematics Association.
ELDREDGF., N.. 1979. Cladism and comnion sense. In J. Cracraft & N. Eldredge (Eds), Phjlogenrtic :lnalvsis and
Palennfology: 165-198. New York: Colunibia Cniversit! Press.
ELDREDGE. 3.& TATTERS.ALL. I., 1975. Evolutionar) models, phylogenetic reconstruction, and another
look at hurninid phylogeny. Cnntributionr in Primatolou. 5: 218-242.
FARRIS. J. S.. 1974. Formal definitions of paraphyly and plyphyly. 5jsfematir ~ a o l o p 23:
.
548 554.
F.ARRIS. J. S . . 1977. O n the phenetic approach to vertebrate classification. In M. K. Hecht, P. C. Goodv p ’
h l . Hecht IEdsi, .Major Putterm in I-brtebrate Emlution: 823-850. New York: Plenum.
8.
USE O F PARSIMONY
325
FARRIS, J. S., 1979. On the naturalness of phylogenetic classification. Syskmatic zoologv, 28: 200-214.
FARRIS, J. S., 1980. The information content of the phylogenetic system. Systematic <mlogv, 28: 483-519.
FITCH, W . M., 1979. Numerical taxonomy: a special project (11). syJfcmatic (oology, 28: 254-255.
FITCH, W. M. & MARGOLIASH, E., 1967. Construction of phylogenetic trees. S c k e , N.Y., 155: 27!+284.
FORTEY, R. A. & JEFFERIES, R. P. S., 1982. Fossils and phylogeny-a compromise approach. In K. A.
Joysey & A. E. Friday (Eds), Problems of Phylogcnetic Reconstruction: 197-234. London: Academic Press.
FOX, R. C., 1978. Cladograms as strictly universal statements. (Privately circulated) 7 pp.
GAFFNEY, E. S., 1975. A phylogeny and classification of higher categories of turtles. Bulletin of fhe American
Museum of .Natural History, 155: 387-436.
GAFFNEY, E. S., 1979. An introduction to the logic of phylogeny reconstruction. In J. Cracraft & N. Eldredge
(Eds), Phylogenetu Analysis and Palaeontolou: 79-1 11. New York: Columbia University Press.
GARDINER, B. G., 1980. Tetrapod ancestry: a reappraisal. In A. L. Panchen (Ed.), The Terrestrial Enuiromnent
and the Origin of Land Vertebrates: 177-185. London: Academic Press.
GILMOUR, J. S. L., 1961. Taxonomy. In A. M. MacLeod & L. S. Cobley (Eds),Confcmporary Botanical Thought:
2745. Edinburgh: Oliver & Boyd.
GILMOUR, J. S. L. & WALTERS, S. M., 1963. Philosophy and classification. In W. B. Tunill (Ed.), Virtas in
Botary, 4 : 1-22. London: Pergamon.
GOULD, S. J., 1977. Ontogeny and Phylogeny. Cambridge, Massachusetts: Harvard University Press.
HARRIS, E. E., 1972. Epicyclic Popperism. British Journal f o r the Philosophy of Science, 23: 55-67.
HECHT, M. K., 1976. Phylogenetic inference and methodology as applied to the vertebrate record. Evolutionarv
Biology, 9: 335-363.
HECHT, M. K. & EDWARDS, J. L., 1976. The determination of parallel or monophyletic relationships: the
proteid salamanders-a test case. American Naturalist, 110: 653-677.
HECHT, M. K. & EDWARDS,J. L., 1977. The methodology of phylogenetic inference above the species level.
In M. K. Hecht, P. C. Goody & B. M. Hecht (Eds), Major Patterns in Vertebrate Evolution: 3-51. New York:
Plenum.
HENNIG, W., 1950. Grundzuge einer Theorie der phylogcnetzschen Systemotik. Berlin : Deutscher Zentralverlag.
HENNIG, W., 1966. Phylogenctk SysUmatics. Urbana: University of Illinois Press. (Reprinted 1979.)
HENNIG, W., 1969. Die Sfummesgeschuhte dcr Insckten (Senckenberg Buch 49). FrankfurtlMain : Kramer.
HULL, D. L., 1980. The limits of cladism. Systematic <mlogy, 28: 416-440.
JANOWITZ, M. F., 1979. A note on phenetic and phylogenetic classifications. Systematu zoo lop^, 28: 197-199.
JARVIK, E., 1968. Aspects of vertebrate phylogeny. Nobel Symposium, 4 : 497-527.
JURGENS, J. D., 1971. The morphology ofthe nasal region ofAmphibia and its bearing on the phylogeny of the
group. Annule uan die Universiteit uon Stellenbosch, 46 (A): 1-146.
KEMP, T . S., 1980. Origin of the mammal-like reptiles. Nature, London, 283: 368-372.
KITTS, D. B., 1977. Karl Popper, verifiability, and systematic zoology. Systematic <oolog, 26: 185-194.
KLUGE, A . G. & FARRIS, J. S., 1969. Quantitative phyletics and the evolution of anurans. Svsfematic <oolopv,
18: 1-32.
KRIPKE, S. A,, 1980. Naming and Necessip (Revised ed.). Oxford: Blackwell.
KUHN, T. S., 1962. The Stnrcturc of Scimtrfi Revolutions. Chicago: Chicago University Press.
KUHN T. S., 1970. The Structure of Scientzz Revolutions (2nd ed.) Chicago: Chicago University Press.
KOHNE, W. G., 1973. The systematic position of monotremes reconsidered (Marnmalia). Zeifschriji f i r
Morphologie der Tiere, 75: 59-64.
LAKATOS, I., 1970. Falsification and the methodology of scientific research programmes. In I. Lakatos & A.
Musgrave (Eds), Cn'tz'cUm and the Growth af Knowledge: 91-196. Cambridge: Cambridge University Press.
McKENNA, M. C., 1975. Towards a phylogenetic classification of the Mammalia. Contribufionsin Primatology, 5:
21-46.
McKINNEY, H. L., 1972. Wallace and Natural Selection. New Haven: Yale University Press.
MAYR, E., 1965. Numerical phenetics and taxonomic theory. Qstematic < o o l o ~ , 14: 73-97.
MAYR, E., 1969. Principles of Systematic < o o l o ~ . New York: McGraw-Hill.
MAYR, E., 1974. Cladistic analysis or cladistic classification. <eitschriff f i r <oologische Svstematik und
Evolutionsforschung: 12: 94-128.
MEDAWAR, P. B., 1967. The Art ofthe Soluble. London: Methuen.
MICKEVICH, M. F., 1978. Taxonomic congruence. Svstematic < o o l o ~ ~27:
. 143-158.
MILES, R. S., 1973. Relationships of acanthodians. In P. H. Greenwood, R. S. Miles & C. Patterson (Eds),
Interrelationships of Fishes: 63-103. London: Academic Press.
MILES, R. S., 1975. The relationships of the Dipnoi. Colloques internafionaux du Centre .National de la Recherche
Scientijque, 218: 133-148.
MILL, J. S., 1872. Book IV: ofoperations subsidiary to induction. A &sum of Lug; Ratiocinative and Inductive (8th
ed.). London: Parker. [in J. M. Robson (Ed.) 1974. John Stuarf Mill: Collected Worhs8.Toronto: University
Press].
NELSON, G., 1970. Outline of a theory of comparative biology. Systematic <oology, 19: 373-384.
NELSON, G., 1971. Paraphyly and plyphyly : redefinitions. Systematic <oology, 20: 471-472.
NELSON, G., 1972. Phylogenetic relationship and classification. Sysfematk <oology, 21: 227-231.
NELSON, G . , 1974. Classification as an expression of phylogenetic relationships. Systematic <oologv, 22: 344-359.
326
A. L. PANCHEN
NELSON, G.. 1978. Classification and prediction: a reply to Kitts. Systematic <oology, 27: 216-218.
NELSON, G.. 1979. Cladistic analysis and synthesis: principles and definitions, with a historical note on
Adanson’s Fandles dts Plantes i1763 1764). Svstemafic <oologv, 28: 1-2 1.
PANCHEN, .4. L., 1972. The interrelationships of the earliest tetrapods. In K. ,4. Joysey & T. S. Kemp (Eds),
Sfudies in l’eerttbratt Ez,olution: 65 87. Edinburgh: Oliver & Boyd.
PANCHEN, .A. L., 1979. The cladistic debate continued. .Vafure, London, 280: 541.
PATTERSOS, C.. 1978. Verifiability in systematics. Systematic <oology, 27: 218-222.
PA’ITERSON, C., 1982. Morphological characters and homology. In K. A. Joysey & A. E. Friday (Eds),
Problems of Phylogenetu Reconstruction: 21-74. London: Academic Press.
PrZTTERSON, C. & ROSEN, D. E.. 1977. Review of ichthyodectiform and other Mesozoic teleost fishes and the
theory and practice of classifying fossils. Bulletin of the American Museum of .Vatural History,158: 81--172.
PLATNICK, 3.I.. 1977. Paraphyletic and polyphyletic groups. $Wematic <ooLogv, 26: 195-200.
PI,.AT?IICK, 5 . I., 1980. Philosophy and the transformation of cladistics. Systematic <oologv, 28: 537 ~ 5 4 6 .
PLATNICK, X. I . & CAMERON, H. D., 1977. Cladistic methods in textual, linguistic, and phylogenetic
analysis. Systematic Joologp. 26: 380 385.
PLATSICK, N. I . Br GAFFNEY. E. S.. 1977. Systematics: a Popperian perspective. Systematic <oologv, 26:
360 365.
PIATNICK. S . I. & G.AFFNEY, E. S., 1978a. Evolutionary biolog-): a Popperian perspective. Systematic <oologv.
27: 137 141.
PLATNICK. N. I. Br GAFFNEY. E. S., 1978b. Systematics and the Popperian paradigm. Systematic <oo/ogy, 27:
38 1- 388.
POPPER, K.R., 1934. LogzA der Forsrhuq. l‘ienna: Julius Springer.
POPPER, K.R., 1959. The f.ogzr of Srientzfi Dirrozry.. London: Hutchinson.
POPPER, K. R., 1963. Conjectures and Rtfrutations-tht Growth o f & k t z ~Knowledge. London: Routledge & Kegan
PdUl.
POPPER, K. R., 1972. Objectioe Knaudedge--on Evolutionary ,4pproach. Oxford: University Press.
POPPER, K.R.. 1974. Replies to mv critics. In P. A. Schilpp ( t d . ) The Philosophy of Karl Popper: 961-1 197. The
Library of Living Philosophers. 14. La Salle, Ill. : Open Court.
PC‘TSAM, H., 1975. .Wind, f.anguage and Realitr: Philosophiral Papers, Iblume 2. Cambridge: University Press.
RIEDL, R., 1979. Order in Liring Organism. Chichester: J. LViley.
ROMERO-HERRER.4, A. E., LEHMANS. H., JOYSEY, K. A. & FRIDAY. A. E., 1973. Molecular evolution
of myoglobin and the fossil record: a phylogenetic synthesis. .Vature, London, 246: 389-395.
ROMERO-HERRERA. .A. E., LEHMASN, H., JOYSEY. K.A. & FRIDAY, A. E.. 1978. On the evolution of
tnyoglobin. Philosophiral Transactions of the Riyal Socieg of London, ( B ) 283: 61- 163.
SHAEFFER. B.. HECHT, M. K.& ELDREDGE, 5..1972. Phylogeny and paleontology. Evolutionary Biology. 6:
~
3 1 46.
SHORROCKS, B . . 1978. The Genesh ofDizwsi@. London: Hodder & Stoughton.
SIMPSON. G. G., 1953. Tht Major Features of Evolution. New York: Columbia University Press.
SIMPSOS, G . G.. 1961. Prinr$les of Animal TaxonoEv. New York: Columbia University Press.
SIMPSON, G. G.. 1975. Recent advances in methods of phylogenetic inference. Conh’bulions in Primalob& 5:
:i-19.
SNEATH. P. €I. .%.,1961. Recent developments in theoretical and quantitative taxonomy. Systematic.<oologv, 10:
118 139.
SNErlTH, P. H. A. & SOK.AL, R. R., 1973. ,Vumeriral Taxonomy. San Francisco: Freeman.
S O U L , R. R. & SNEATH, P. H . A., 1963. PTinciplcs of Numerical T Q X O San
~ JFrancisco:
.
Freeman.
TATTERS.4LL. I. & ELDREDGE, N.. 1977. Fact, theory and fantasy in human paleontology. American Scientist,
65: 204- 21 1.
WATSON, D. M. S., 1919. The structure, evolution and origin of the Amphibia-The “Orders” Rachitomi and
Stereospondyli. Philosophical Transactions of the Royal SociCg of London, ( E ) 209: 1-73.
WATSON, D. M. S., 1940. The origin of frogs, Transactions ofthe Royal Sock9 ofEdinburgh, 60: 195-231.
WATSON, D. M. S., 1951. Paleonfalagy and Modern Biology. New Haven: Yale University Press.
WILEY, E. 0.. 1975. Karl R. Popper. systematics, and classification: a reply to Walter Bock and other
evolutionary taxonomists. Srstematic <wlogv, 24: 233- 242.
ADDENDUM
.4 recent paper by Rohlf & Sokal (1980) severely criticizes the attempt by
Mickevich ( 1978) to demonstrate the superiority of mathematical cladistics
(‘phylogenetic systematics’) to phenetic techniques (see p. 3 19 above). Mickevich
appears to identify ‘stability’ (as a criterion for judging between taxonomic
methods) with congruence between classifications of the same organisms derived
from different character sets. Rohlf & Sokal refer to four different definitions of
USE OF PARSIMONY
327
stability (the first two from Sneath & Sokal, 1973) none of which coincides with
Mickevich’s. Of these four the first approximates to her definition of congruence,
but whereas she compares classifications based on discrete character sets (e.g.
larval u. adult, male u. female), Sneath & Sokal suggest the criterion that a
classification should change little, if at all, when additional characters (n2) are
considered together with the original set (n,). Rohlf & Sokal thus contend that
Mickevich’s results are not a test of stability but of the largely abandoned nonspecificity hypothesis (Sneath & Sokal, 1973 : 97ff.).
They further charge that she biassed her results by recording the data sets
uniformly for all the taxonomic methods rather than using the coding technique
most suitable to each method, and also that she imposed the inherent restrictions of
her favoured technique (unrooted Wagner trees) on all the others.
However, probably the strongest evidence they cite of apparent bias is in the
presentation of her results. These are summarized (Mickevich, 1978: table 1) as
mean rank, which gives a spurious impression of the degree of superiority of the
Wagner method, particularly over the best phenetic technique (UPGMA). The
mean rank ratio between the two is 2.24, whereas the ratio of their respective mean
consensus information indices is only 1.14. Both these means are tabulated by
Rohlf & Sokal, but only the former by Mickevich. Lastly, as noted by Rohlf &
Sokal, whereas the Wagner method (m.c.i. index 0.593) appears somewhat better
than the best phenetic technique (0.518) two other cladistic techniques (CaminSokal/WISS :0.472 and Le Quesne : 0.459) were decidedly worse, with the latter
the worst of all the eight methods tested.
In a rejoinder Mickevich (1980) attempts to answer all Rohlf & Sokal’s
criticisms. With some justification she rejects their (and Sokal & Sneath’s, 1973)
first definition of stability, suggesting that testing a classification by adding a new
set of characters to the pre-existing set gives a spurious impression of stability by
the conflation of two sets of data. It should be noted, however, that this is precisely
how a non-mathematical cladist confirms his prejudices by the use of parsimony,
whether he introduces a minority of incongruent ‘homoplastic’ characters or
bolsters his parsimony ‘vote’ by seeking additional apparent synapomorphies (see
p. 320). Her answer to the charge of biassing the results by the coding technique
etc. is a simple (or rather complex) denial. However, she does not satisfactorily
answer their criticism of her boosting the apparent significance of her results by the
misleading presentation of mean ranks without mean consensus information
indices or some other valid summary measure. In the light of Rohlf & Sokal’s
criticism and Mickevich’s reply I now regard her attempted demonstration as much
less significant and convincing than I did when the main manuscript of this paper
was completed.
Ghiselin ( 1980) reviews Cracraft & Eldredge (1979), a symposium from which
several of our references are quoted (Cracraft, 1979; Gaffney, 1979; Eldredge,
1979). In his review he comments on the cladist’s proprietorial claim to be using
Popper’s methodology, and notes (as I do) the ad hoc use of the hypothesis of
parallelism as a by-product of their use of parsimony-similarly for convergence :
“If we merely cry ‘Convergence!’ whenever we encounter a contradiction, no
datum under the sun will lead us to correct our errors”.
ADDITIONAL REFERENCES
ROHLF, F. J. & SOKAL, R. R., 1980. Comments on taxonomic congruence. Systemtic &d~gv, 29: 97-101.
A. L. PANCHEN
328
MICKEVICH, M . F., 1980. Taxonomic congruence: Rohlfand Sokal’s misunderstanding. S J J ~ O<oo.bgy,
~ L 29:
162-176.
GHISELIN, M. T., 1980. Phylogenetic mythogenais and paleontology, [review of Phylogmtic Analysis and
PuleontologV . . . (JoelCracraft & Niles Eldrege, Eds). New York: Columbia U.P.]. Evolution, 34: 822-824.
CRACRAFT,J. & ELDREDGE, N. (Eds),Phylogmfic Analysis and Pukontology. New York: Columbia University
Press.