Manipulating
Underdetermination in
Scientiªc Controversy:
The Case of the
Molecular Clock

Michael R. Dietrich
Dartmouth College

Robert A. Skipper, Jr.
University of Cincinnati

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

Where there are cases of underdetermination in scientiªc controversies, como
the case of the molecular clock, scientists may direct the course and terms of
dispute by playing off the multidimensional framework of theory evaluation.
This is because assessment strategies themselves are underdetermined. Within
the framework of assessment, there are a variety of trade-offs between differ-
ent strategies as well as shifting emphases as speciªc strategies are given more
or less weight in assessment situations. When a strategy is underdetermined,
scientists can change the dynamics of a controversy by making assessments
using different combinations of evaluation strategies and/or weighting what-
ever strategies are in play in different ways. Following an underdetermina-
tion strategy does not end or resolve a scientiªc dispute. Como consecuencia, manip-
ulating underdetermination is a feature of controversy dynamics and not
controversy closure.

1. Introducción
Underdetermination is common in scientiªc practice. It is widely recog-
nized by scientists even if they do not discuss it in terms frequently used
by philosophers. We argue that in the course of scientiªc controversies un-
derdetermination can be and often is manipulated by scientists to direct
the terms and course of dispute. During a dispute, scientists use a wide ar-
ray of different strategies in different combinations in their comparative

Thanks to Dick Burian, Roberta Millstein, and two anonymous reviewers for Perspectives on
Science for comments on earlier drafts of this paper. RASjr also acknowledges the support of
the Charles P. Taft Research Center for their support in the form of a Fellowship during the
time this paper was written.

Perspectives on Science 2007, volumen. 15, No. 3
©2007 by The Massachusetts Institute of Technology

295

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

296

Manipulating Underdetermination

assessment of theories. This assessment process involves trade-offs between
different strategies as well as shifting emphases as different strategies are
given more or less weight in an assessment situation. The multidimen-
sional framework of theory assessment provides the resources for the ma-
nipulation of underdetermination. Manipulating underdetermination is
not a feature of controversy closure in that it does not end or resolve dis-
pute. Bastante, it is a feature of a continuing controversy and so of contro-
versy dynamics.

We elaborate and illustrate our view with a case study from molecular
evolution, es decir., the controversy over the molecular clock. As molecular bi-
ology emerged as a discipline in the 1960s, some biologists began to con-
sider how molecules themselves had evolved. Two startling and controver-
sial concepts emerged from this early work on molecular evolution. En el
early 1960s, Emile Zuckerkandl, Linus Pauling, and others recognized
that they could use similarities and differences between molecules to infer
their degree of evolutionary relatedness. When Zuckerkandl and Pauling
compared protein sequences, sin embargo, they discovered that molecules like
hemoglobin seemed to be changing at a constant rate. They called this ap-
proximate rate constancy the molecular clock. For evolutionary biologists
used to thinking in terms of natural selection, a constant rate of evolution
was extremely unlikely since selection depended on interactions with the
environment which was understood to be continuously changing. The rate
of evolution under selection should also change or ºuctuate. The molecu-
lar clock was made even more controversial by its association with a new
theory of molecular evolution. En 1968 Motoo Kimura argued that most
detected changes in molecules were not due to the inºuence of natural se-
lection. En cambio, they were governed by genetic drift and were therefore
considered neutral with respect to selection. The neutral theory was per-
ceived as a direct threat to selectionism then dominating organismic evo-
lution (Dietrich 1994). Since the early 1970s the molecular clock has been
a very important part of the neutral theory of molecular evolution. En efecto,
an important part of the appeal of the neutral theory has been that it pro-
vides a very elegant explanation for constant rates of substitution. Nosotros estafamos-
sider efforts to argue for a selectionist alternative for the molecular clock
and so to undermine the clock’s support for the neutralist position. Mientras
the neutralist and selectionist positions are often represented as polar op-
posites, the efforts to propose a selectionist clock do not proceed by simply
asserting the correctness of the selectionist clock and the failure of the
neutralist clock. Instead selectionists have proposed a new set of argu-
ments through the manipulation of underdetermination. These arguments
attempt to put the neutralist and selectionist interpretations of the clock
on equal evidentiary footing, but do not resolve the controversy.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Perspectives on Science

297

We begin in section (2) with a discussion of theory assessment in bio-
logical science that emphasizes the numerous standards, categories, y
dimensions of evaluation. En la sección (3), we explore the connection be-
tween theory evaluation and underdetermination. With these apparatuses
in place, in section (4) we show, by way of an analysis of the controversy
over the molecular clock, precisely what we think manipulation of under-
determination consists in and how we think it works. Sección (5) is re-
served for discussion and concluding remarks.

2. Theory Evaluation.
In the assessment of scientiªc theories, a number of methodological prin-
ciples (epistemic, pragmatic, and social) may be at work. Por ejemplo,
Thomas Kuhn offered a list of ªve criteria for good theories, viz., exactitud,
consistencia, broad scope, simplicity, and fruitfulness (Kuhn 1977, 322),
and W. h. Newton-Smith has offered a list of eight “good-making fea-
tures of theories,” viz., observational nesting, fertility, track record, enterrar-
theory support, smoothness, internal consistency, compatibility with well-
grounded metaphysical beliefs, and simplicity (Newton-Smith 1981,
226–230). There are other such lists. Además, these lists could be ex-
panded to include pragmatic criteria, such as cost and tractability, y entonces-
cial criteria, such as opportunism in context. Some of these criteria, semejante
as simplicity, are considered epistemic by some philosophers (Laudan
1981, 196) and social by some sociologists (Bloor 1981, 201). En efecto, el
debate between philosophers and sociologists is often construed in terms
of the epistemic and the social, but it need not be. Bruno Latour, for in-
postura, advocates a move beyond the social and epistemic, beyond relativ-
ism and realism (Latour 1992; cf. Longino 1990; Solomon 2001).

Biologists have used an identiªable constellation of standards in their
comparative evaluation of most theories, and philosophers of biology have
done considerable work to understand the ways in which the process
obras. Three main approaches may be identiªed: conªrmation, epistemol-
ogy of experiment, and strategies for generating scientiªc change. Cada uno de
these approaches has been championed in the philosophy of biology:
conªrmation by Elisabeth Lloyd (1987; 1988), epistemology of experi-
ment by David Rudge (1998, 2001), and assessment during cycles of
scientiªc change by Lindley Darden (1991). En otra parte, Skipper (2000)
has argued that there is no good reason to think that these three ap-
proaches to theory assessment are mutually exclusive. En efecto, based on a
critical historical analysis of a persistent controversy in evolutionary ge-
netics, Skipper developed a comprehensive framework of theory evaluation
highlighting the ways in which the three approaches complement each
otro. We think this framework holds considerable promise for under-

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

298

Manipulating Underdetermination

standing theory assessment in the biological sciences. En particular, nosotros
think it is a useful framework for understanding the kinds of standards bi-
ologists appeal to in making theory choices in underdetermination situa-
ciones. Herein, we review this multidimensional framework for theory as-
sessment, applying it and elaborating on how it can be used to make
theory choices in situations of underdetermination in subsequent sections
of the essay. The framework is summarized in Figure 1 abajo; discussion
of the approaches in which they are embedded follows.1

Many in the philosophy of science are convinced that the core of theory
assessment is the examination of the very speciªc relationship between hy-
potheses and the data adduced for them, es decir., the conªrmation relation.
This view may well be right. Sin embargo, let us be clear that conªrmation is
not all there is to theory assessment. Sin embargo, let us start there. En
philosophy of biology, Lisa Lloyd (1987,1988) has developed a well-
conocido, qualitative account of conªrmation of evolutionary and ecological
models drawn from a series of case studies in population genetics and ecol-
ogia.

Lloyd has argued that conªrmation of evolutionary and ecological mod-
els can be practically speciªed by three criteria, viz., ªt between theoreti-
cal model and data, independent support for aspects of a particular theo-
retical model, and variety of evidence (Cifra 1: B1–B3). Variety of
evidencia (Cifra 1: B3) is at the core of Lloyd’s account. Variety of evi-
dence as a strategy itself is at bottom iterations and/or combinations of her
other two strategies, es decir., ªt between model and data and independent
support for aspects of a model (Cifra 1: B2–B3). Lloyd’s intuition behind
variety of evidence is that numerous and varied demonstrations of, p.ej., ªt,
can increase the strength of the hypothesis that a theoretical model corre-
sponds to the natural system modeled. A similar claim can be made for in-
dependent support. The more times, the more different ways and types of
maneras, and the more assumptions are tested independently, the greater the
empirical support for a hypothesis made about a model that appeals to
that assumption. Variety of types of support is a mixture of variety of in-
stances of ªt and variety of independent support. So, a hypothesis made
about a theoretical model that has a variety of instances of ªt along with a
variety of instances of independent support for its assumptions is in a dif-
ferent situation regarding its conªrmation than a theory with only one in-
stance of ªt.

1. The sheer number of strategies, 30 if we are counting, makes it impractical to expli-
cate each one in the present paper. Como consecuencia, only a few will be discussed. See Darden
1991, lloyd 1987, 1988, Rudge 1998, and Skipper 2000, 2002, 2004a, 2004b for a thor-
ough explanation and justiªcation of each.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Perspectives on Science

299

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

Cifra 1. Categories and strategies for theory assessment. There are five main cate-
gories listed across the top row, viz., (A)-(mi). Strategies from the approaches dis-
cussed in the text are represented. We do not take the list to be exhaustive.

Lloyd sees herself as describing at least part of the set of techniques evo-
lutionary biologists use to relate theoretical models to natural systems.
More speciªcally, she sees her three conªrmation criteria as concretely ar-
ticulating techniques that are part and parcel of an experimental model
(following Suppes 1962). The experimental strategies Rudge (1998,
2001) delineates are also part and parcel of experimental models in much
of biology. Rudge (1998, 2001) considers the evaluation of experimental
results made via observational and experimental procedures in the context
of detecting natural selection and includes such strategies as calibration
(of apparatuses), artifact reproduction, experimental interventions, y entonces
en (ver figura 1: C1–C11 for all of Rudge’s strategies). Rudge’s account is
based directly on Allan Franklin’s (1986, 1990) epistemology of experi-
mento.

Via historical case studies from high energy physics, Franklin intro-
duced a set of experimental strategies he packaged as an epistemology of
experimento. Franklin’s broad aim was to demonstrate that claims based on
experimental results are epistemically warranted, pace the then key social
constructivists of science, such as Harry Collins (1985). Franklin offered

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

300

Manipulating Underdetermination

nine experimental strategies in his earlier work. To demonstrate the rea-
sonableness of the strategies, Franklin justiªes each using Bayesian episte-
mology. franklin (2002) has continued and expanded his work on episte-
mology of experiment. Rudge (1998) takes Franklin’s epistemology and
critically examines it against a handful of detailed historical case studies,
most notably H. B. D. Kettlewell’s studies of industrial melanism in the
peppered moth, Biston betularia. Rudge found that Franklin’s experimen-
tal strategies were used in the Kettlewell work and, further, that addi-
tional Franklin-like strategies could be drawn from that work, como el
use of experimental controls (Cifra 1: C11). Más, like Franklin,
Rudge “epistemologizes” his experimental strategies using the Bayesian
epistemology.2

Rudge’s experimental strategies are meant as standards for evaluating
experimental or observational data, whereas Lloyd’s conªrmation criteria
are meant as standards for assessing the relationship between that data and
speciªc theoretical hypotheses. En efecto, looking at the relationship be-
tween the practical procedures used to produce data and the formal tools
for evaluating the relationship between hypotheses and data is looking at
the opposite sides of the same coin. Llevar, por ejemplo, variety of evidence
o, more speciªcally, variety of instances of ªt between model and data.
Replication of experimental results is one way of concretely specifying
that variety of evidence. And in so doing, we can assess the techniques of
replication used as a way of qualitatively evaluating the variety of in-
stances of ªt between model and data. And in fact this kind of assessment
happens often in comparative assessment situations (p.ej., Skipper 2002).
Other examples of the ways in which experimental strategies may elabo-
rate conªrmation criteria are available. Recientemente, por ejemplo, Skipper
(2004b) showed how calibrating laboratory populations using data from
natural populations can be used to assesses the relative strength of claims
that there is independent support for aspects of a model. The important
point here is that experimental evaluation strategies are available for criti-
cally examining the strength of empirical claims made about models.
Además, it should clear that different combinations of experimental and
conªrmational criteria may be used.

Now consider Darden’s account of strategies for theory assessment dur-
ing cycles of scientiªc change. Darden draws many of her strategies out of

2. Recientemente, Rudge 2001 has compared his Bayesian analysis of the Kettlewell work
with an analysis of that work using Deborah Mayo’s 1996 error statistical epistemology of
experimento. Rudge now thinks that Mayo’s epistemology of experiment better captures
the epistemology of the Kettlewell experimental work. But the issue exercising Rudge
here concerns Bayesian vs. error statistical approaches to understanding the growth of
conocimiento, not whether the experimental strategies identiªed in Rudge 1998 apply.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Perspectives on Science

301

a substantial case-study in the history of Mendelian genetics (Darden
1991). Other of her strategies come from a survey of the philosophical lit-
erature. For Darden, scientiªc change proceeds through interacting cycles
of discovery, evaluación, and revision, and her task is to delineate the con-
stellation of reasoning strategies scientists use (or might use) to catalyze
change during these cycles. Darden approaches theory evaluation by con-
sidering how theories are evaluated on four of the ªve categories of theory
evaluation in Figure 1 (columns A1–A6, B4–B8, D1–D2, E1–E2; ver
Skipper 2000, 2004a on A7–A8). En efecto, the categories of theory evalua-
tion we delineate in Figure 1 are based on Darden’s organization of her
own evaluation strategies (Darden 1991, 258).

In terms of categories of theory evaluation, we have so far explored the
related categories of the relationship between hypotheses (or models or
teorías) to data and the relationship between data and the experimental
techniques used to generate them (columns B and C in Figure 1). Among
the strategies Darden delineates, several are meant as standards for assess-
ing the relationship between theory and evidence. En efecto, we think that
such strategies Darden discusses, including explanatory adequacy, predic-
tive adequacy, and scope/generality can be particularized by Lloyd’s
conªrmational criteria. Consider Darden’s strategy of assessing explana-
tory adequacy (Cifra 1: B4). How might a scientist determine whether
some explanation is adequate? Darden does not give many details. Nosotros
think that Lloyd’s conªrmational criteria (as well as Rudge’s experimental
estrategias) can play this role in a straightforward way. Además, Nosotros pensamos
that information gleaned from appealing to criteria of conªrmation and
more speciªc assessment of experimental results may be used by scientists
to judge claims about predictive adequacy, scope/generality, and lack of ad
hocness (Cifra 1: B5–B7).

But Darden’s strategies for theory evaluation go beyond just the rela-
tionship between theories and evidence to include the assessment of the
formal elements of theories (Cifra 1, column A), the assessment of theo-
ries in relation to other theories (column D), and the future prospects of
teorías (column E). Eso es, in addition to empirical strategies for theory
evaluación, Darden includes non- or extra-empirical ones. Ahora, some phi-
losophers argue that non- or extra-empirical assessment standards are not
epistemic but are instead pragmatic (p.ej., Sober 1994). We do not wish to
enter into this debate here and, anyway, have already committed ourselves
to the view that theory evaluation standards cut across epistemic, prag-
matic, and social dimensions. In that case, it is fair to say that Darden’s
framework for theory assessment is broader than either Lloyd’s or Rudge’s
(which is not really an account of theory assessment, but can be under-
stood to contribute to Lloyd’s). Y, importantly, it seems clear that

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

302

Manipulating Underdetermination

Darden’s framework exempliªes the multidimensionality of theory evalua-
ción.

The central point of the examination of theory evaluation in this sec-
tion is this: Biologists, and surely scientists more generally, appeal to a
broad array of theory assessment strategies that cut across epistemic, prag-
matic, and social dimensions. Además, these strategies may be used in
various combinations. Además, we will see that assessment strategies
are assigned different weights in different combinations during compara-
tive evaluation of theories in situations of underdetermination. And in
that context, it will become clear that scientists are able to shift the
grounds and terms of dispute by deploying the multidimensional frame-
work of theory assessment which we call “manipulating underdetermina-
tion” in different ways. Let us now turn to the relationship between com-
parative theory assessment, or theory choice, and underdetermination.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

3. Underdetermination and Theory Choice.
In its strongest form, the Duhem–Quine thesis claims that a single theo-
retical hypothesis cannot be conclusively falsiªed, so “any statement can
be held to be true come what may if we make drastic enough adjustments
elsewhere in the system” (Quine 1953, 43).3 In effect, this form of the
Duhem–Quine thesis claims that a very strong type of underdetermina-
tion follows from scientiªc holism: when faced with contravening evi-
dencia, it is always possible in principle to save some favored hypothesis by
making enough revisions in other parts of the system associated with it. En
otras palabras, any number of theories can be generated which agree with
the evidence presented. Philosophers and sociologists often take Quine at
his word when he claims that “any statement can be held to be true come
what may if we make drastic enough adjustments elsewhere in the sys-
tem” (Quine 1953, 43).4 Interpreting Quine’s famous claim, sin embargo, es
anything but straightforward (Laudan 1990).

When philosophers and sociologists talk about underdetermination,
the usual case concerns the underdetermination of theory by evidence.
From this pattern of usage, it would seem then that, en general, underde-
termination is a two part relation: {X1, X2, . . . Xn} is underdetermined by
{Y1, Y2, . . . Ym}, where the members of the X-set are usually theoretical
sistemas, theoretical models, or theories and members of the Y-set are usu-
ally statements of experimental results or models of data. De hecho, sin embargo,

3. Quine is referring to a system of beliefs composed of an array of more or less theoreti-

cal and observational propositions.

4. On the limits of interpretive ºexibility and the debate over underdetermination be-

tween philosophers and sociologists, see Dietrich 1993.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Perspectives on Science

303

underdetermination is not so much about theories, as it is about the
grounds for judging between theories. Put another way, theories are not
underdetermined; choices between theories are underdetermined.

Given this distinction, underdetermination can be deªned as follows: A
choice between members of the set {X1, X2, . . . Xn} based on the set {Y1,
Y2, . . . Ym} is underdetermined if and only if: (1) there is a set {X1, X2, . . .
Xn}, (2) each X in {X1, X2, . . . Xn} is pair-wise contrary with every other
member of {X1, X2, . . . Xn}, (3) there is a set {Y1, Y2, . . . Ym}, (4) hay
a relation C such that for each member X of {X1, X2, . . . Xn}, X stands in
the relation C to {Y1, Y2, . . . Ym}, where C (the underdetermination rela-
ción) is some relation concerning compatibility which should be spe-
ciªcally deªned in each case. Underdetermination, entonces, is always relative
to some basis for choice used by scientists in their deliberations over alter-
native theories. We will articulate this basis for choice in terms of evalua-
tion standards. De este modo, underdetermination is relative to the grounds for
claims made by way of the assessment standards in Figure 1. Sin embargo, el
basis for choice need not be articulated in terms of evaluation standards.

Different versions of underdetermination will be a result of differences
in the X-set and Y-set and/or in the deªnition of the underdetermination
relation. The range of the Y-set is always speciªed and relative to a partic-
ular judgment at a particular time. The range of the X-set varies, pero
will always include at least two members. Using this general schema, uno
can clarify exactly what is meant by various claims of underdetermination.
In contrast to general and vague claims about underdetermination, el
schema advocated here provides the basis for what we will call speciªed
underdetermination.

Consider for instance the general and familiar claim that a theory is
underdetermined by evidence. In order to clarify this claim the nature of
the relation between the theories and evidence must be articulated; en
terms of the scheme presented here, the underdetermination relation must
be speciªed. If we deªne C, the underdetermination relation, as logical en-
tailment, then we have what Larry Laudan calls deductive underdetermin-
ación. In his words, deductive underdetermination can be formulated as a
claim similar to the following one: “For any ªnite body of evidence, allá
are indeªnitely many mutually contrary theories, each of which logically
entails the evidence” (Laudan 1990, 269).

Notice that deªning C in terms of logical entailment restricts the ways
the theories (the X-set) and the evidence (the Y-set) can be formulated—
they must be formulated such that they can stand in a relation of logical
entailment to each other, p.ej., as theoretical and observational statements.
Deductive underdetermination produces the familiar result that deductive
logic alone is not capable of determining whether or not an empirical the-

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

304

Manipulating Underdetermination

ory is true with certainty, regardless of the evidence. Sin embargo, deductive
underdetermination alone does not mean that theory choice is completely
underdetermined, but only that deductive logic is not going to be the way
the choice is made. Deductive underdetermination does not rule out the
possibility that other grounds for theory choice could be fully determinate
(Laudan 1990, 270).

A more typical formulation of the underdetermination relation found
in claims about the underdetermination of theory by evidence involves
some form of epistemic relation. We will call an underdetermination rela-
tion epistemic if it is chieºy concerned with some type of empirical sup-
port or the relation of the theories to the world (Laudan 1990, 271,
McMullin 1982, 18, van Fraassen 1980, 88).5 Por ejemplo, Laudan distin-
guishes three possible epistemic relations: (1) any theory in the set of rival
theories can be made logically compatible with the evidence; (2) any theory
in the set of rival theories can explain the evidence; y (3) any theory in
the set of rival theories can be empirically supported by the evidence (Laudan
1990, 275). Note that implicit in these underdetermination relations are
theory evaluation standards. So, por ejemplo, the standard “prefer theories
with greater explanatory adequacy” (Cifra 1: B4) is implicit in Laudan’s
second alternative listed above, while the standard “prefer theories which
are empirically supported by all the presented evidence” (Cifra 1: B1) es
implicit in the ªrst alternative. It should be evident that numerous ways
of specifying epistemic underdetermination relations (C) are possible.

Underdetermination need not be limited to grounds for theory assess-
ment that appeal to empirical support; it can be extended to cover almost
any assessment strategy used, p.ej., pragmatic or sociological. Consider the
sociological standard “prefer theories which when implemented in labora-
tory practice will have the lowest reasonable cost.” If in some speciªc case
the Y-set is just the lowest reasonable cost in monetary terms, the X-set is
a set of rival implementation plans for related theories, and C is deªned as
“is the cost of,” then we can make the underdetermination claim that any
implementation plan in the set of rival implementation plans will cost the
same—they all ªt equally well with the lowest reasonable cost. For any
strategy of assessment in the multidimensional framework of Figure 1,
there is plausibly some underdetermination relation that can apply to it
provided the Y-set includes the appropriate entities.

As we have seen, on our view any given theory assessment involves a
constellation of evaluation strategies. Because in any given situation a
number of different standards may be at work, the degree of underdeterm-

5. Laudan refers to these epistemic relations as ampliative relations. We prefer the more

general term “epistemic.”

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Perspectives on Science

305

ination must be speciªed. The degree of underdetermination refers to the
number of standards that are shown to be underdetermined relative to
some particular theory. The degree of underdetermination can be weak
when only one or two of the standards are underdetermined or it can be
very strong when many of the strategies are shown to be underdetermined.
We might say, por ejemplo, that some X-set is underdetermined with re-
spect to explanatory adequacy (Cifra 1: B4); this is a low degree of
underdetermination. O, we might say that some X-set is underdeterm-
ined with respect to several forms of variety of evidence (Cifra 1: B3);
this is a higher degree of underdetermination. Specifying the degree of un-
derdetermination requires that the strategies which could possibly be
underdetermined in the case at hand be articulated. For the moment we
are not assuming that any kind of strategy is more important than any
other in theory choice or controversy resolution. Pero, as we will see in our
discussion of the controversy over the molecular clock in the next section,
differential weighting of standards could affect the signiªcance attached
to different underdetermination claims.

Another important way in which underdetermination claims ought to
be speciªed involves the range of underdetermination, where that refers to
the number of members of the X-set. Por ejemplo, Laudan’s formulation
of deductive underdetermination, arriba, allowed the X-set to range over
indeªnitely many theories, but it could, and often does, have a much
smaller range, p.ej., two genuine rivals. Differences in the range of the X-
set provide further grounds for distinguishing between different types of
underdetermination. Differences in the range of the X-set, por ejemplo, lie
behind Laudan’s distinction between what he calls the nonuniqueness the-
sis and the egalitarian thesis. Put in terms of theoretical systems, el
nonuniqueness thesis is the thesis that:

For any theory, t, embedded in a system, S, and any body of evi-
dencia, mi, there will be at least one other system, S’ (containing a ri-
val to T), such that S’ is as well supported by e as S is (Laudan
1990, 292).

Similarly stated, the egalitarian thesis is the thesis that:

For any theory, t, embedded in a system, S, and any body of evi-
dencia, mi, there will be systems, S1, S2, . . . , Sn, each containing a
different rival to T, such that each is as well supported as T (Laudan
1990, 292).

The basic difference here is that the egalitarian thesis indicates that there
are a number of equally supported rivals and the nonuniqueness thesis
claims that there is at least one equally supported rival. The egalitarian

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

306

Manipulating Underdetermination

thesis is, de este modo, stronger than the nonuniqueness thesis. Since the differ-
ence between nonunique forms of underdetermination and egalitarian ver-
sions of underdetermination is a matter of the number of theories in-
volved, we will refer to the range of underdetermination rather than to
nonunique or egalitarian underdetermination. The question of what the
range of underdetermination is in a speciªc case is empirical. If one wishes
to claim that an indeªnite number of theories are compatible with the evi-
dencia, then one must provide some reason for advocating such an ex-
tended range.

But suppose that there is a strong degree of underdetermination of an
indeªnite range of rival theories by some set of data. It would follow then
that in a Quinean fashion one could conclude that any theory may be held
regardless of the evidence it faces. As Laudan notes, this claim can be in-
terpreted descriptively or normatively (Laudan 1990, 272). Si, for in-
postura, Quine is just claiming that it is possible to retain beliefs despite
contravening evidence, then he is making a purely descriptive claim about
underdetermination. If he is claiming it is rational to hold on to any state-
ment regardless of the evidence facing it, then he is making a normative
claim about underdetermination. Laudan argues that if underdetermina-
tion is to have any implications for normative epistemology, then it must
be construed normatively or in terms of rationality (Laudan 1990, 272).6
Differently put, descriptive underdetermination refers to all the possible
courses of action available, while normative underdetermination refers to
only the rational courses of action. In most cases, the set of rational courses
of action will be a subset of the set of possible courses of action. So, in ef-
fect, considerations of rationality or reasonableness constrain the courses of
action that are available. Por ejemplo, Laudan considers construing under-
determination in terms of theories being equally compatible with the
evidencia. Logical compatibility can be achieved quite easily by simply
eliminating without replacement those parts of the theory that make it in-
compatible with the data. This kind of elimination, sin embargo, comes with
a price; eliminating large chunks of a theory may drastically and nega-
tively affect that theory’s explanatory adequacy (Cifra 1: B4). If explana-
tory adequacy is valued in theory assessment, then it may not be reason-
able to sacriªce explanatory adequacy for compatibility. So, En el caso de
compatibility, the question is: Is it rational to preserve a theory by making
it compatible with the evidence if this action is going to harm the theory’s
explanatory adequacy? Similar tradeoffs are not hard to imagine. The bot-

6. Rationality is understood by Laudan to be instrumental rationality. The sense of

normativity advocated here need not be limited to instrumental rationality.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Perspectives on Science

307

tom line here is that if underdetermination claims are to have any
epistemic bearing, their advocates must attend to the question of the ra-
tionality of a course of action, not just its possibility (Laudan 1990, 276).
We have argued for a schema that suggests that underdetermination
must be speciªed so that an underdetermination relation, es decir., epistemic,
pragmatic, sociological, must be articulated. Además, the view re-
quires speciªcation on a case-by-case basis the relevant strategies for eval-
uating alternative theories, the degree of underdetermination, the range of
underdetermination, and the normative status of the underdetermination
afirmar (Dietrich 1993). The controversy over the molecular clock in molec-
ular evolution, discussed in the next section, illustrates these facets of
underdetermination. As we will see, the case establishes a claim of epi-
stemic underdetermination, making use of some of the strategies from the
framework of theory evaluation delineated in Figure 1. Más importante,
the following discussion of the molecular clock controversy illustrates this
paper’s main, and novel, afirmar: During the course of scientiªc controver-
sies underdetermination of the strategies for theory assessment is manipu-
lated by the scientists to direct the course and terms of dispute. En efecto,
the case presented below makes manifest one example of the way in which
this manipulation, or “underdetermination strategy,” has been used.

4. The Case of the Molecular Clock: Illustrating the Underdermination
Strategy.
Although they had articulated the idea earlier, the molecular clock was
christened in 1965 by Emile Zuckerkandl and Linus Pauling (Zucker-
kandl and Pauling 1965, morgan 1998). The idea behind the molecular
clock was that the observed changes in the amino acid sequences of a
protein from different species should be “approximately proportional in
number to evolutionary time” (Zuckerkandl and Pauling 1965, 148). El
differences observed across species could then serve as an evolutionary
timescale. As a timekeeping device, the molecular clock was inherently
stochastic. The “ticks” of the clock were not uniform, but were best un-
derstood as a regular statistical process producing a distribution of rates of
substitution for the species and molecule under consideration. The molec-
ular clock immediately attracted interest and controversy (Dietrich 1998,
2006).

4.1. The Neutralist Approach.
In their initial attempts to explain the mechanism for the clock,
Zuckerkandl and Pauling invoked both natural selection and random drift
to explain the clock’s apparent constancy (morgan 1998). Después 1968,

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

308

Manipulating Underdetermination

Motoo Kimura, Allan Wilson and others used the neutral theory of molec-
ular evolution to explain the mechanism of the clock. En efecto, Motoo
Kimura’s original arguments in 1968 for the signiªcance of neutral muta-
tions and random drift were based in part on the hemoglobin data pre-
sented by Zuckerkandl and Pauling in 1965. Por 1971, Tomoko Ohta and
Motoo Kimura asserted that the “remarkable constancy of the rate of
amino acid substitutions in each protein over a vast period of geologic
time constitutes so far the strongest evidence for the theory (Kimura
1968, King and Jukes 1969) that the major cause of molecular evolution
is random ªxation of selectively neutral or nearly neutral mutations”
(Ohta and Kimura 1971, 18). The reason the constancy provided such
strong support is that the neutral theory provided a mechanism for that
constancy—the prediction of rate constancy followed easily from basic
theoretical commitments of Kimura and the neutralists. According to the
neutralists, the rate per generation of mutant substitutions in a population
(k) is equal to the mutation rate per gamete (v):

k (cid:2) v.

[1]

The result in [1] is derived as follows. In a population of actual size N,
there are 2Nv new mutations produced in the entire population per gener-
ación. Only a certain fraction of these new mutations will become estab-
lished in the population; eso es, only a certain fraction will reach ªxation.
Let u represent the probability that a new mutation will reach ªxation.
Entonces, in Kimura’s (1979, pag. 108) palabras, “in a steady state in which the
process of substitution goes on for a very long time, the rate k of mutant
substitution per unit time is given by the equation

k (cid:2) 2Nvu.”

[2]

For selectively neutral mutations, sin embargo, the probability of ªxation (tu)
is equal to 1/(2norte), because “any one of the 2N genes in the population is as
likely as any other to be ªxed, and so the probability that the new mutant
will be the lucky one is 1/(2norte)" (Kimura 1979, pag. 108). Substituting 1/
(2norte) for u in equation [2] yields equation [1]. It is important to note here
that the rate of mutant substitution is independent of population size.

The rate of evolution for selectively advantageous mutants, in contrast
to neutral mutants, is dependent on both population size and selection
pressure. Kimura (1979, pag. 110) uses a lengthy derivation to show that for
selectively advantageous mutants the equation for the rate of evolution is

k (cid:2) 4Nsv.

[3]

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Perspectives on Science

309

De este modo, in order for a selectionist model to account for the constancy of the
rates of evolution it must show how constancy is possible when the rates
are strongly dependent on the environment, as represented by the selec-
tion coefªcient s and the measure of population size N, which can be quite
variable. Under the selectionist model, the rates of molecular evolution
should show nearly the same variability as the rates of phenotypic evolu-
ción. Unfortunately for the neutralists, the rates of molecular evolution
were found to be far from uniform.

The rate of amino acid substitution was known from the beginning
(1965) to vary among different proteins. The neutralists explained this
difference in terms of different proteins having different fractions of neu-
tral mutants; the number of neutral mutants depends on the functional
constraints for each protein. So, por ejemplo, ªbrinopeptide A has a much
higher rate of substitution than histone IV, which is highly constrained
(King and Jukes 1969, pag. 792). But even within protein families variation
was observed. So, por ejemplo, insulins in the line leading to guinea pigs
seem to have evolved faster than insulins in other lines (King and Jukes
1969; Ohta and Kimura 1971, 19). The neutralists needed a way to ex-
plain these deviations from the intrinsic rate of molecular evolution.

En 1971, Tomoko Ohta and Motoo Kimura analyzed these variations in
proteins statistically. Claims about rates of evolution are based on the
number of amino acid substitutions that actually occurred during the
course of evolution and an accurate estimate of the time of divergence for
the proteins being compared. The fraction of amino acid differences be-
tween two protein sequences (pd) is equal to the number of differences in
the two sequences (daa) divided by the total number of amino acid sites in
the sequences (naa); entonces, says Kimura (1969, pag. 1182),

pd (cid:2) daa/naa.

[4]

In the course of evolution, sin embargo, a given amino acid site may have
changed more than once; it may have had multiple hits. To correct for
multiple hits, Zuckerkandl and Pauling and then Kimura assumed that
the process of amino acid substitution could be modeled by a simple Pois-
son process.7 With this assumption, the probability of no substitution oc-
curring at a given site is e K aa
, where Kaa is the average number of amino

(cid:2)

7. A Poisson process is a purely random stationary process. In a purely random process,
the behavior of the process in the past has no inºuence on future behavior (Lindley 1965,
67). A process is stationary if the probability of the incidents in the process are not
inºuenced by temporal order (en otras palabras, if it is invariant with respect to time)
(Lindley 1965, 68). The classic example of a Poisson process is radioactive decay.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

310

Manipulating Underdetermination

acid substitutions per site for the protein sequences being compared. Él
follows then that

y eso

1 (cid:3) pd (cid:2) e K aa

(cid:2)

Kaa (cid:2) (cid:3)ln(1 (cid:3) pd).

The rate of amino acid substitution per site per year is

kaa (cid:2) Kaa/(2t),

[5]

[6]

[7]

where T is the number of years since divergence. From here the expected
variance of kaa can be computed and compared to its observed variance.
When Ohta and Kimura did this for different alpha and beta hemo-
globins, and for cytochrome c, they found that observed variance in the
beta hemoglobin and the cytochrome c were signiªcantly larger than ex-
pectado. From this they concluded that “the variations in evolutionary rates
among highly evolved animals are larger than expected from chance”
(Ohta and Kimura 1971, 21). Ohta and Kimura did not take this as a rea-
son to give up the neutral theory. The increased variance in substitution
rates was chalked up to a small fraction of advantageous mutations that af-
fect the molecule’s function but do not interfere with the constancy of the
overall rate of substitution (Ohta and Kimura 1971, 23).

After Ohta and Kimura’s paper in 1971, a tremendous amount of em-
pirical research was done on the molecular clock (Wilson et al. 1977).
Charles Langley and Walter Fitch, por ejemplo, used a procedure based on
minimum phyletic distances to test whether or not the process of nucleo-
tide substitution was a constant Poisson process. They concluded that “it
is clear that the total rate of substitution (as observed through the mini-
mum phyletic distance procedure) varies markedly in geological time and
among divergent lines of descent” (Langley and Fitch 1974, 174). Similar
conclusions stressing the non-uniformity of rates in hemoglobin and a
slow down of rates in primate lineages were offered by Goodman, moore,
and Matsuda (1975). Even Kimura himself admitted that the rate of mo-
lecular evolution is not perfectly uniform (Kimura 1983, 79), but in his
opinión, “emphasizing local ºuctuations as evidence against the neutral
theory, while neglecting to inquire why the overall rate is intrinsically so
regular or constant is picayunish. It is a classic case of ‘not seeing the forest
for the trees’” (Kimura 1983, 85).

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Perspectives on Science

311

Kimura backs up his account of rate constancy in his 1983 libro, El
Neutral Theory of Molecular Evolution, by providing a more thorough statis-
tical treatment of the variations in the rate of molecular evolution. kim-
ura’s analysis uses what has since been called a star phylogeny (Gillespie
1984a). The lineages in a star phylogeny are taken to have all diverged
from a common ancestor in a relatively short period of time. Kimura con-
siders the case of six mammals; humanos, mice, rabbits, perros, caballos, y
cattle, which diverged from each other about 80 million years ago
(Kimura 1983, 76). What Kimura wants to know is “whether the intrin-
sic rates of amino acid substitutions among the six lineages are equal and
whether variation of the observed numbers of substitutions as shown in
Mesa 1 below lie within the limits of normal statistical ºuctuations”
(Kimura 1983, 76–77). Kimura’s method is as follows for the case of alpha
hemoglobin (ver tabla 1): The number of amino acids (naa) for alpha he-
moglobin is 141. Using the information in Table 1 and equation [6], el
number of substitutions corrected for multiple hits can be calculated and
then used to calculate the mean and variance of the number of amino acid
substitutions. In order to see if the variation in amino acid substitutions
among lineages is larger than expected, the value of R is then determined,
where R is the ratio of the observed variance to the expected variance. Pero,
for Poisson processes, the mean is equivalent to the expected variance, so R
can be readily computed as the ratio of the observed variance to the mean.
For the case of alpha hemoglobin, Kimura calculated that R (cid:2) 1.26. In or-
der to ªnd out if R’s deviation from the expected value of 1 is signiªcant,
Kimura argued that “the statistic (norte (cid:3) 1)R should follow the X2 (chi-
square) distribution with n (cid:3) 1 degrees of freedom” (Kimura 1983, 77).
For the alpha hemoglobin case, the table of X2 values shows that the devi-
ation is not larger than that expected by chance, so Kimura infers that the
hypothesis, es decir., that the intrinsic rate of evolution in these six lineages is
lo mismo, stands. Kimura then extended his analysis to include four other
proteínas, beta hemoglobin, myoglobin, ribonuclease, and cytochrome c.
Of these additional proteins, myoglobin and ribonuclease showed no
signiªcant variation, but beta hemoglobin and cytochrome c showed sig-
niªcantly higher variation than expected. In Kimura’s words, “these re-
sults suggest that although the strict constancy may not hold, yet a rough
constancy of the evolutionary rate for each molecule among various lin-
eages is a rule rather than an exception” (Kimura 1983, 79). Además,
Kimura notes that the average value of R for the ªve molecules used is 2.6,
which is consistent with earlier results showing that observed variances up
a 2.5 times larger than expected are allowable if the variation is a result of
chance alone (Kimura 1983, pag. 79; Ohta and Kimura 1971). So, en el

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

312

Manipulating Underdetermination

Mesa 1. Observed numbers of amino acid differences between six mammals when
their hemoglobin alpha-chains are compared. The ratio of the observed to the ex-
pected variances turns out to be as follows: R (cid:2) 1.26. (Adapted from Table 4.3 en
Kimura 1983.)

Humano

Mouse

Rabbit

Dog

Horse

Bovine

Humano
Mouse
Rabbit
Dog
Horse
Bovine

18

25
27

23
25
28

18
24
25
27

17
19
25
28
18

end, Kimura admits that an approximate rate constancy holds as a rule,
but he also admits that there may be deviations from the rule; hence, su
admonition about not seeing the forest for the trees, as mentioned above.

4.2. The Selectionist Alternative and Critique of the Neutralist
Acercarse.
Soon after Kimura published his account of variations in the rate of evolu-
ción, John Gillespie used the same data that Kimura had used to propose a
rival interpretation that called for an episodic molecular clock (Gillespie
1984a). Where Kimura emphasized the underlying constant rate, Gilles-
pie argued that the signiªcance of the two proteins with larger than ex-
pected variances was “by no means obvious” (Gillespie 1984a, 8010).

Gillespie and Kimura were not new opponents in 1984; Gillespie had
proposed a natural selection model for molecular evolution in 1978
(Gillespie 1978) and had questioned rate constancy as early as 1979 (Gil-
lespie and Langley 1979). En 1984, Gillespie also published a negative re-
view of The Neutral Theory of Molecular Evolution that acknowledged the
achievements of the neutral theory but questioned Kimura’s style of advo-
cacy (Gillespie 1984, 733). In effect, Kimura and Gillespie represent
opposite poles in the molecular evolution community. Kimura was the se-
nior advocate of the neutral theory. As head of the department of popula-
tion genetics at Japan’s National Institute of Genetics, he had gathered
around him a team of scientists, incluido, por ejemplo, Tomoko Ohta and
Naoyuki Takahata, who developed the neutral theory and explored its im-
plications. Gillespie was a senior professor at the University of California
at Davis. He made his career devising and promoting sophisticated selec-
tionist approaches to molecular evolution.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Perspectives on Science

313

The debate over rates of molecular evolution, as Gillespie saw it, era
concerned with estimates of “the ratio of the variance to the mean of the
number of substitutions per lineage,” what Gillespie calls “R(t)" (Gil-
lespie 1986a, 140). According to Gillespie, the neutralists need to explain
why the value of R(t) is larger than unity in many cases, and the selection-
ist needs to explain why the value of R(t) is not estimated as being any
mayor que 3.4. Gillespie, por supuesto, takes up the challenge for the selec-
tionist position.

Gillespie’s main argument for an episodic molecular clock is that “the
variation in the numbers of substitutions per lineage must ultimately be
attributable to variation in the rates of substitutions” (1986b). Gillespie
cashes out his position by proposing a model of the molecular clock based
on a doubly stochastic Poisson process. This doubly stochastic Poisson
process is “a Poisson process for which the rate of the process itself is a sta-
tionary stochastic process” (Gillespie 1984a, 8011). An important feature
of this kind of model is the relation between the clocks in different lin-
eages. Gillespie assumes that the clocks have “equal tick rates at the time
of radiation” and that “the correlation in tick rates between lineages drops
off at the same rate as the correlation within a lineage” (1984a, 8011). So,
the rates can change randomly within lineages. Gillespie then assumes
that the clock changes rapidly, o, as he says, “the rate of change of the rate
of molecular evolution is assumed to occur on a time scale that is much
shorter than the lengths of the lineages under study” (1986a, 141). Gilles-
pie then says, “this assumption is motivated by the fact that major envi-
ronmental changes, such as the recent ice ages, occur on a time scale of
thousands to tens of thousands of years while the time between substitu-
tions is typically on the order of millions of years” (1986a, 146). With this
assumption Gillespie inferred that in order to ªt the parameters of his sta-
tistical model the clock should be episodic, es decir., it should reºect a substi-
tution process where there are long periods with no substitutions broken
up with occasional short bursts of substitutions (Gillespie 1986a, 141).

There are two basic versions of the episodic clock: one is the two-state
clock and the other is the gamma clock. The two-state clock is based on a
two-state Markov process. According to Gillespie, “the process remains in
state zero for an exponentially distributed time with mean 1/u0 and then
jumps to the value 1/(u0t), where it remains for an exponentially distrib-
uted time with mean t before returning to zero, and so forth” (Gillespie
1986a, 147–8). The gamma clock jumps between a gamma-distributed
height and zero and has a substantially different mathematical basis from
the two-state clock. Despite these differences, the two clocks produce sim-
ilar estimates of both the mean number of substitutions per episode and
the mean number of episodes. According to Gillespie, “given the very dif-

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

314

Manipulating Underdetermination

ferent characters of the two clocks, this suggests that the inferred episodic
structure is robust to the assumptions made about the clock, at least for
the restricted range of values of R(t) for the currently available data”
(1986a, 149). This demonstration of robustness puts the episodic clock on
ªrmer footing by making it more resilient to attacks on its assumptions.
In this way it also becomes harder to dismiss as non-genuine.

The episodic clock has a number of advantages from Gillespie’s selec-
tionist perspective. Primero, the episodic nature of the clock agrees well with
the selectionist model of molecular evolution developed by Gillespie
(1984b). Gillespie’s (1986a, 141) model has three components:

1) a changing environment,
2) an epistatic scheme in which each environmental change presents a
challenge to the species that may be met by substitutions at one of
several nearly equivalent loci, y

3) a mutational landscape that makes it unlikely that any particular
locus will experience the substitution of an allele that is two muta-
tional steps away from the allele that is currently ªxed in the popu-
lación.

This model assumes that both the environmental changes and the number
of loci available to respond are large. Segundo, the episodic clock ªts well
with the data. Gillespie’s model of the episodic clock predicts that the val-
ues of R(t) should be in the range of 1.0 (cid:4) R(t) (cid:4) 3.5, if “the number of
loci that can respond to a particular environmental change is large”
(Gillespie 1986a, 152–3). This was exactly the range of values observed at
el tiempo. De este modo, in Gillespie’s words, “from the perspective of purely sta-
tistical considerations, our model of evolution by natural selection actually
seems to ªt the data better than does the neutral model” (Gillespie 1986a,
153). Given this agreement with observed values of R(t) and the agree-
ment between the episodic dynamics of the model and the episodic struc-
ture of the data, Gillespie concludes that he has shown that “a very plausi-
ble model of molecular evolution by natural selection ªts the sequence of
data just as well as does the neutral allele model” (Gillespie 1986a, 141).
Notice that what Gillespie has done is establish a case of epistemic
underdetermination that renders the ability to generate predictions that
agree with the observed values of R(t) insufªcient grounds for choice be-
tween existing neutralist and selectionist models of the rate of molecular
evolution. Both models ªt the data and are predictively accurate (Cifra 1:
B1, B5), so that the degree of underdetermination is two, even though
Gillespie’s may be a slightly better ªt. But that is not enough for Gilles-
pie, who shifts the grounds and terms of assessment off the underdeterm-
ined assessment standards of ªt and predictive accuracy and emphasizes

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Perspectives on Science

315

the independent support of the underlying selectionist assumptions of
his model (Cifra 1: B2), es decir., by producing what he says is “a very plausi-
ble model of molecular evolution” (Gillespie 1986a, pag. 141). He still
wants to charge that the predictions of the neutral model do not ªt well
with the large values of R(t). And the neutralists want to explain these
large values away as exceptions. The type of underdetermination argued
for by Gillespie has the effect, por lo tanto, of partially shifting the grounds
of the controversy away from emphasizing ªt between model and data and
the ability to predict certain values of R(t) (Cifra 1: B1, B5), given cer-
tain theoretical commitments, to emphasizing other features of each of the
modelos, particularly independent support for underlying assumptions
(Cifra 1: B2), así como, simplicity and lack of ad hoc theoretical as-
sumptions (Cifra 1: A6, B7) as we shall see below. Differently put, Gil-
lespie manipulates the underdetermination situation, changing and re-
weighting the combination of epistemic standards for evaluation. Este
underdetermination strategy has the effect of shifting both the grounds
and the course of comparative assessment.

Gillespie ºeshes out his attack by claiming that the neutral theory’s
model of the molecular clock, as found in Kimura’s 1983 presentación, es
at a disadvantage because of the artiªciality of its assumptions. In Kim-
ura’s model the rates of evolution are ªxed at the origin of each lineage and
each lineage’s rate is independent of the others. Pero, “why,” Gillespie asks,
“should the rates of mutation be altered only at the time of origin of the
orders of mammals and then remain unaltered for the next 60 Myr even
though other lineages are branching off to form families, genera, and spe-
cíes?" (Gillespie 1986a, 153). This artiªciality leads Gillespie to assert
eso

[oh]ne conclusion is clear: the neutral theory cannot, Actualmente, ac-
count for the statistical patterns in the sequence data without the
use of a very artiªcial model of rate variation. While the theory
could be modiªed to be in better agreement with the data, semejante
modiªcations will make the theory less compelling as a uniquely
parsimonious explanation for molecular evolution (Gillespie 1986a,
153).

The formal or mathematical simplicity (Cifra 1: A6) of the neutral the-
ory is taken by Gillespie and others to be a major point in its favor, Alabama-
though not enough of an advantage as it stands to withstand his
underdetermination strategy. Indeed Gillespie believes that the good
standing of the neutral theory is partly a result of the difªculties inherent
in producing a relatively simple selectionist alternative and is partly a re-
sult of the possibility of doing “a little rearranging” and having the theory

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

316

Manipulating Underdetermination

“emerge as strong as ever” (Gillespie 1987, 33–4). In the end, if the neu-
tral theory is to emerge from his challenge, Gillespie predicts that “salva-
tion will most likely come with a redeªnition of the neutral model as
exempliªed in the recent work of Takahata (1987)" (Gillespie 1987, 33).

4.3. The Neutralist Response.
The controversy over the molecular clock became fully articulated when
Gillespie introduced an alternative account of the molecular clock that
challenged the mathematics, suposiciones, and mechanisms of the neutral-
ist’s interpretation of the clock. In the face of Gillespie’s challenge that the
molecular clock is episodic, Naoyuki Takahata proposed a number of pos-
sibilities to try to diffuse the negative implications that an episodic clock
has for the neutral theory. Takahata’s account is particularly revealing be-
cause he offers three ways to bring the observation of large values for R(t)
in line with the neutral theory, but rejects two of them as implausible. So,
Takahata’s strategy is to produce a neutralist model that can account for
the entire range of observed values of R(t) from small to large.

Gillespie’s claims, according to Takahata, are that rate constancy is an
artifact of a slow rate of amino acid substitution, that “molecular evolu-
tion may be episodic, with bursts of rapid evolution separated by periods
of very slow evolution,” and that directional selection may explain the epi-
sodic nature of the clock (Takahata 1987, 170). Given these claims,
Takahata admits that “it seems certain that some, if not all, genes evolve
with sloppier clocks than those of radioactive materials” (Takahata 1987,
170). En otras palabras, the rate of molecular evolution shows signiªcant de-
partures from the expectations of a simple Poisson process model, cual
can usually be exempliªed by the process of radioactive decay. Takahata’s
goal is to propose possible alternatives that are compatible with the neu-
tral theory.

In order to account for values of R(t) greater than unity, Takahata ar-
gues that we need to account for values of the dispersion index, I(t), cual
are greater than unity. I(t) is easier to manipulate than R(t), but is related
to R(t).8 Takahata proposes three ways of generating large values of I(t):

1) Multiple substitutions. The idea behind this option is that multiple
substitutions at one time would produce a larger than expected vari-
ance. Takahata lists a number of authors who recognize that “some,
if not many, substitutions in a gene might not have occurred singly

8. The difference between R(t) y yo(t) is that R(t) is an actual estimate of the ratio of
the sample variance to the sample mean for the number of substitutions among lineages,
while I(t) is a ratio of the expectation of the sample variance to the expectation of the sam-
ple mean.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Perspectives on Science

317

because of highly specialized intramolecular interactions” (Takahata
1987, pag. 171). According to this model, in order for the index of
dispersión, I(t), to be much larger than unity, substitutions would
have to have both a high multiplicity and rate of occurrence.

2) Fluctuating substitution rate. This option employs a doubly stochastic
Poisson process, as Gillespie did, but tries to reach a different con-
clusion. In order to explain the bursts of changes and long periods of
stasis that follow from this model, Takahata appeals to higher rates
of occurrence and accumulation of deleterious mutations coupled
with a reduction of population size (Takahata 1987, 173). With the
help of a computer simulation, Takahata concludes that,

[i]f there are very severe bottlenecks whose mode of occur-
rences differs from lineage to lineage and deleterious mu-
tation rates are much higher than neutral mutation rates,
then there would be bursts of substitutions in each lineage
and at the same time it can be expected that I(t) (cid:5) 1
(Takahata 1987, pag. 174).

3) Fluctuating neutral space. This option is based on a time-dependent
renewal process that assumes that “the substitution rate ºuctuates
through changes of selective constraints as new substitutions take
place one after another” (Takahata 1987, 174). The degree of selec-
tive constraint is what Takahata calls the neutral space. If changes
only occur at the time of substitution, then this renewal model pre-
dicts that if the neutral space ºuctuates to a large extent, I(t) can be
fairly large (Takahata 1987, 175). Fluctuations in the neutral space
are supported by research on opossum alpha hemoglobin and guinea
pig insulin. Both of these molecules have shown large changes in se-
lective constraints as a result of intragenomic causes, such as the loss
of a previously invariant amino acid (Takahata 1987, 176).

Takahata’s alternatives are meant to demonstrate that the neutral the-
ory is compatible with large values of R(t) which are held as evidence
against it by Gillespie. So, Takahata accepts the reliability of the experi-
mental result; a saber, that the values of R(t) are signiªcantly larger than
the expected value of 1. Takahata’s commitment to the neutral theory thus
demands that he ªnd components of the old theoretical system proposed
by Kimura that can be revised or replaced such that the neutral theory
will then be compatible with the experimental results. In his response to
Gillespie, Takahata is following his own strategy of underdetermination;
he is striving to formulate models that will generate high values of I(t),
and hence R(t). In this way he can bring the neutral theory better in line

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

318

Manipulating Underdetermination

with the data. Eso es, he wants to reªne the ªt and predictive accuracy of
speciªc assumptions of the neutral theory (Cifra 1: B1, B5) to gain im-
proved overall ªt with the data (Cifra 1: B1). But notice that observed
values of R(t) greater than unity will not decide between the rival models
proposed because of Gillespie’s underdetermination strategy; en cambio, ser-
ing able to generate an expected value of R(t) (cid:5) 1 is a minimal require-
ment for the adequacy of a proposed model. Takahata places greater em-
phasis on the predictive accuracy of speciªc assumptions that achieve
greater ªt than on overall ªt. Being able to generate a high value of R(t),
sin embargo, is not sufªcient for a model to be considered choice-worthy, como
can be seen by Takahata’s own evaluation of the models he proposes.

Takahata’s ªrst alternative pushed for a different assumption about how
substitutions occurred. If multiple substitutions occurred, then Takahata
could retain a Poisson process model, although it would be compound,
not simple. The rates of substitution and the multiplicity of substitutions
observado, sin embargo, were not as high as required for this model to produce
signiªcantly large values of I(t). This option was thus rendered “unlikely”
as a result of a poor empirical support for the speciªc assumption about
multiple substitutions (Cifra 1: B2). This explains why the multiple
substitution model was not pursued, but not why the revision was made
in the ªrst place. An explanation and justiªcation for this is suggested by
Takahata’s citing of six papers ranging from 1968 a 1985 that substitu-
tions may be inºuenced by intramolecular interactions (Takahata 1987,
171). Multiple substitutions are a critical part of Kimura’s original model
as evidenced by the concern for estimating multiple substitutions. If sev-
eral authorities suggest that perhaps our ideas about substitutions need re-
visión, then the importance of this element warrants its attempted revi-
sión.

Takahata’s second alternative follows Gillespie’s lead in using a doubly
Poisson process model, but adds to it a neutralist twist borrowing from
work by Tomoko Ohta and Motoo Kimura on slightly deleterious muta-
tions and on the effects of population bottlenecks (Ohta 1973, 1977;
Kimura 1979). The emphasis this model places on deleterious mutations
that arise during bottlenecks in the population makes the model depend-
ent on the periods that each lineage spends in a bottleneck phase. For bot-
tleneck events to cause signiªcant variation in the rate among lineages,
they would have to “vary considerably from lineage to lineage” (Takahata
1987, 176). Sin embargo, Takahata, like Kimura, has assumed that all the lin-
eages under study follow common rules, so mutation rates themselves are
not allowed to differ among lineages. Takahata’s emphasis on bottlenecks
in this model also has the result that it can account for the large values of
R(t) only by creating another problem of explaining why bottlenecks af-

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Perspectives on Science

319

fect only some protein lineages when they all experience the bottleneck.
These constraints and problems raised by the neutralist interpretation of
the doubly stochastic Poisson model argue against its plausibility, accord-
ing to Takahata.

Takahata’s third alternative, the ºuctuating neutral space model, era
developed using a previously unused type of mathematical model (un momento
dependent renewal process), but it was able to account for the troublesome
evidence using intragenomic causes sympathetic to the neutralist view of
molecular evolution—changes in selective or functional constraints after
substitutions. The idea behind this revision, sin embargo, seems to be closely
tied to Gillespie’s argument that it is changes in the action of natural se-
lection that accounts for the higher rates.

Takahata clearly favors the ºuctuating neutral space model, but admits
that a satisfactory description of the mechanisms underlying the statistical
models is still lacking. The heart of the difference between Takahata and
Gillespie is in their basic assumptions about molecular evolution; el
ºuctuating neutral space model “emphasizes intragenic or intragenomic
causes” while the episodic model “stresses external or environmental
causes.” Mathematically, the two models show different evolutionary pat-
charranes. In Takahata’s words,

. . . in the ºuctuating neutral space model, some lineages fortu-
itously undergo more substitutions than do others and it is this
variation that inºates R(t). This pattern of evolution may be
exempliªed by insulin genes of guinea pig and coypu (Kimura
1987). In the episodic clock model, por otro lado, the rate
drastically changes many times along each lineage and it is this
change that increases R(t). Direct evidence for this pattern cur-
rently appears to be nil (Takahata 1988, 388).

So, according to Takahata, Gillespie has no empirical support for his
assumption that there are rate changes in each lineage. Takahata’s attack
on Gillespie’s assumptions is particularly important here, because the
strength or plausibility of underlying assumptions is taken by both Taka-
hata and Gillespie to be an important basis for choice among rival models.
Remember that Takahata ruled out two possible neutralist models in part
on the basis of the quality of their assumptions. Más, recall that
Gillespie has shifted the grounds of assessment away from ªt and toward
independent support. Por supuesto, Takahata used his own underdetermina-
tion strategy, but not by ignoring Gillespie’s.

From the standpoint of manipulating underdetermination, Takahata’s
paper is particularly illuminating because he actually goes through a list
of rivals to Gillespie’s model. Each of Takahata’s models can generate pre-

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

320

Manipulating Underdetermination

dictions of high values of I(t) and so put the neutralist position on equal
footing with regard to ªt between model and data for values of I(t). Para
each model, sin embargo, Takahata examines the price to be exacted, y solo
in the case of the ºuctuating neutral space model decides that the price is
not too high. The grounds for this decision, improved ªt and independent
support for assumptions, are thus revealed as open routes for choice be-
tween his and Gillespie’s models. Takahata’s underdetermination strategy
shifts the grounds of assessment away from emphasizing overall ªt with
and predictive accuracy of values of R(t) (Cifra 1: B1, B5), the grounds
on which Gillespie launched his attack, to the speciªc predictions of I(t)
generated by the kinds of assumptions and mechanisms embodied in each
of the rival models resulting in a differently weighted combination of ªt
and independent support for assumptions (Cifra 1: B1, B2, B5).

The outcome of the debate between Takahata and Gillespie was not to
settle any part of the larger neutralist selectionist controversy. If anything,
the manipulation of underdetermination in this debate strategically lev-
eled the playing ªeld by equating neutralist and selectionist models in
terms of their ability to explain high values of I(t). Given that neutralist
models were judged to have an advantage in terms of the explaining rate
constancy in a very parsimonious way, Gillespie’s use of underdetermina-
tion represents a modest success for the selectionist perspective.9 Gilles-
pie’s manipulation of underdetermination to shift the grounds of the con-
troversy did not resolve the dispute; bastante, it used new phenomena
(overdispersion) to drive a search for new models, but most importantly it
shifted the standards by which those models would be evaluated.

5. Discusión y conclusión.
The controversy between Gillespie and Takahata over the molecular clock
reveals that scientists can manipulate underdetermination, or deploy an
underdetermination strategy. By drawing on the multidimensional frame-
work of theory assessment, scientists can re-combine and re-weight assess-
ment strategies in response to underdetermination so that the grounds
of assessment, terms, and course of dispute change. Gillespie began by
claiming that both Kimura’s neutral theory and his own selectionist
model ªt the relevant molecular data (acerca de) equally well and are both
predictively accurate (Cifra 1: B1, B5). He has identiªed a case of epi-
stemic underdetermination in which two assessment strategies are under-
determined for two models. The consequence of Gillespie’s underdeterm-

9. Throughout the 1990s neutralists, such as Tomoko Ohta, developed more sophisti-
cated models of nearly neutral or slightly deleterious mutations to try to capture the
overdispersion in the molecular clock. See Cutler 2000 para una revisión.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Perspectives on Science

321

ination claim is that any choice between the two models must be made on
other grounds. Gillespie begins the assessment process by giving Kimura’s
neutral theory an advantage; not only does it ªt the data, it is also parsi-
monious (Cifra 1: A6). And he admits that it is difªcult to generate a
parsimonious alternative of his selectionist model. But the neutral theory’s
advantage does not last long. Gillespie argues for the superiority of his
selectionist model essentially by arguing that the theoretical assumptions
of the neutral theory are ºawed and that the selectionist assumptions of
his own model have independent support (Cifra 1: B2). For Gillespie,
the independent support for underlying assumptions of his selectionist
model trumps the neutral theory’s parsimony; independent support for as-
pects of a model (Cifra 1: B2) is given more weight than simplicity (Higo-
ura 1: A6).

When Takahata responds to Gillespie, he agrees that ªt between model
and data is important and also, given epistemic underdetermination, eso
the grounds of assessment must change. There is no question for Takahata
that he will stay true to at least the basic framework of the neutral theory.
So he needs a way of countering the independent support for the underly-
ing selectionist assumptions claimed as an advantage for Gillespie’s selec-
tion model. Al hacerlo, he shifts the grounds of assessment off “inde-
pendent support vs. parsimony.” When Takahata revises the neutral theory
he shifts the grounds for assessment to ªt between its assumptions and
values of I(t), and better overall ªt with values of R(t). Takahata generates
three options in revising the neutral theory’s assumptions and chooses the
tercero, es decir., the ºuctuating neutral space option. En efecto, Takahata argues
that Gillespie lacks evidence for his claim that rates ºuctuate in each lin-
eage and pushes the point that there is independent support for the re-
vised assumptions of the neutral theory (Mesa 1: B2). According to Taka-
hata, the support for the revised assumptions is stronger than the support
for Gillespie’s selectionist assumptions. Now what carries the most weight
is the ability for the models under scrutiny to ªt and accurately predict
values of I(t) in a way that generates a better overall ªt of the model with
the data for R(t). Both the (revisado) neutral theory and Gillespie’s selec-
tionist model are in line with the data. And Gillespie’s selectionist
assumptions have independent support. But the neutral theory has the ad-
vantage that its assumptions ªt the data for values of I(t) and that ªt im-
proves overall ªt of the theory with the data for values of R(t). For Taka-
hata, that combination trumps Gillespie’s independent support for the
underlying assumptions of his selectionist model.

Gillespie and Takahata manipulate underdetermination by arguing
that different combinations of differently weighted assessment strategies
both create and break the “tie” the two models have regarding overall ªt

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

322

Manipulating Underdetermination

with model and data in different ways. Gillespie and Takahata were able
to construct (apparently quite good) arguments in favor of their respective
models of the clock by manipulating the combination of assessment strat-
egies and the weights of the speciªc strategies in those combinations.
They adopted speciªc underdetermination strategies. En efecto, this point is
precisely what we argued for at the outset of the paper. Where there are
cases of underdetermination, such as the case of the molecular clock, científico-
tists may direct the course and terms of dispute by playing off the multi-
dimensional
framework of theory evaluation in the ways we have
descrito. This is because assessment strategies themselves are under-
determined. Within the framework of assessment, there are a variety of
trade-offs between different strategies as well as shifting emphases on
those strategies as different ones are given more or less weight in assess-
ment situations. Our analysis focused on a small set of assessment strate-
gies. Numerous other strategies delineated in Figure 1 may also be in-
volved in the manipulation of underdetermination. If we are right that the
underdetermination relation can be speciªed for epistemic, pragmatic, o
sociological grounds for choice, then the sheer number of assessment strat-
egies biologists have recourse to, the permutations of trade-offs, combina-
ciones, and weightings within combinations is enormous, and that makes
for a broad range of underdetermination strategies. The case discussed
here presents a very speciªc controversy illustrating a very speciªc case of
scientists following underdetermination strategies. While we are con-
ªdent that the history and sociology of science affords ample data for the
thesis of this paper, it is a massive empirical enterprise to demonstrate the
multitude of ways in which underdetermination may be manipulated.

We claim that when a strategy is underdetermined, scientists can
change the dynamics of a controversy by making assessments using differ-
ent combinations of evaluation strategies and/or weighting whatever
strategies are in play in different ways. Following an underdetermination
strategy does not end or resolve a scientiªc dispute. Como consecuencia, manip-
ulating underdetermination is a feature of controversy dynamics and not
controversy closure.

Referencias
Bloor, David. 1981. “The Strengths of the Strong Programme.” Philosophy

of the Social Sciences 11: 199–213.

collins, Harry. 1985. Changing Order. Londres: Sage Publications.
Cutler, David J. 2000. “Understanding the Overdispersed Molecular

Clock.” Genetics 154: 1403–1417.

Darden, Lindley. 1991. Theory Change in Science: Strategies from Mendelian

Genetics. Nueva York: prensa de la Universidad de Oxford.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Perspectives on Science

323

Dietrich, Miguel. 1993. “Underdetermination and the Limits of Inter-

pretive Flexibility.” Perspectives on Science 1: 109–126.

———. 1994. “The Origins of the Neutral Theory of Molecular Evolu-

tion.” Journal of the History of Biolog, 27: 21–59.

———. 1998. “Paradox and Persuasion: Negotiating the Place of Molec-
ular Evolution Within Evolutionary Biology.” Journal of the History of
Biología 31: 85–111.

———. 2006. “From Mendel to Molecules: A Brief History of Evolution-
ary Genetics.” Pp. 3–13 in Evolutionary Genetics: Concepts and Case
Estudios, eds. Charles W. Fox and Jason B. Lobo. Nueva York, Nueva York: Oxford
Prensa universitaria.

franklin, Allan. 1986. The Neglect of Experiment. Cambridge: Cambridge

Prensa universitaria.

———. 1990. Experimento: Right or Wrong. Cambridge: Cambridge Uni-

versity Press.

———. 2002. Selectivity and Discord: Two Problems of Experiment. Pitts-

burgh, Pensilvania: University of Pittsburgh Press.

Gillespie, John. 1978. “A General Model to Account for Enzyme Varia-
tion in Natural Populations. V. The SAS-CFF Model.” Theoretical Popu-
lation Biology 14: 1–45.

———. 1984a. “The Molecular Clock May Be an Episodic Clock.” Pro-

ceedings of the National Academy of Sciences 81: 8009–8013.

———. 1984b. “Molecular Evolution Over the Mutational Landscape.”

Evolución 38: 1116–1129.

———. 1984C. “The Status of the Neutral Theory.” Science 224: 732–

733.

———. 1986a. “Natural Selection and the Molecular Clock.” Molecular

Biology and Evolution 3: 136–155.

———. 1986b “Variability of Evolutionary Rates of DNA.” Genetics 113:

1077–1091.

———. 1987. “Molecular Evolution and the Neutral Allele Theory.” Ox-

ford Surveys of Evolutionary Biology 4:10–37.

Gillespie, John and Langley, Charles. 1979. “Are Evolutionary Rates

Really Variable?” Journal of Molecular Evolution 13: 27–34.

Buen hombre, METRO., moore, GRAMO., and Matsuda, GRAMO. 1975. “Darwinian Evolution

in the Genealogy of Hemoglobin.” Nature 253: 603–608.

Kimura, Motoo. 1968. “Evolutionary Rate at the Molecular Level.” Nature

217: 624–626.

———. 1969. “The Rate of Molecular Evolution Considered from the
Standpoint of Population Genetics.” Proceedings of the National Academy
of Sciences 63: 1181–1188.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

324

Manipulating Underdetermination

———. 1979. “The Neutral Theory of Molecular Evolution.” Scientiªc

Americano 241: 98–127.

———. 1983. The Neutral Theory of Molecular Evolution. Cambridge:

Prensa de la Universidad de Cambridge.

———. 1987. “Molecular Evolutionary Clock and the Neutral Theory.”

Journal of Molecular Evolution 26: 24–33.

Rey, Jack and Jukes, tomás. 1969. “Non-Darwinian Evolution.” Science

164: 788–798.

Kuhn, tomás. 1977. The Essential Tension. chicago: University of Chi-

cago Press.

Langley, Charles and Fitch, walter. 1974. “An Examination of the Con-
stancy of the Rate of Molecular Evolution.” Journal of Molecular Evolu-
ción 3: 161–177.

Latour, Bruno. 1992. “One More Turn After the Social Turn.” Pp. 272–
294 in E. McMullin ed., The Social Dimensions of Science. Nuestra dama:
University of Notre Dame Press.

Laudan, Laurence 1990. “Demystifying Underdetermination.” W. Savage
ed., Scientiªc Theories. Minnesota Studies in the Philosophy of Science 10.
Mineápolis: University of Minnesota Press.

Lindley, D. 1965. Introduction to Probability and Statistics From a Bayesian
Viewpoint. Parte 1: Probability. Cambridge: Prensa de la Universidad de Cambridge.

Lipton, Peter 1991, Inference to the Best Explanation. Londres: Routledge.
lloyd, Elisabeth A. 1987. “Conªrmation of Evolutionary and Ecological

Models.” Biology and Philosophy 2: 277–293.

———. 1988. The Structure and Conªrmation of Evolutionary Theory. Príncipe-

tonelada: Prensa de la Universidad de Princeton.

Longino, Helen. 1990. Science as Social Knowledge: Values and Objectivity in

Scientiªc Inquiry. Princeton: Prensa de la Universidad de Princeton.

Mayo, Deborah. 1996. Error and the Growth of Experimental Knowledge. Chi-

cago: University of Chicago Press.

McMullin, mi. 1982. “Values in Science.” Pp. 3–28 in PSA 1982, volumen. 2. PAG.
Asquith and T. Nickles ed. East Lansing, MI: Philosophy of Science As-
sociation:

morgan, Gregory. 1998. “Emile Zuckerkandl, Linus Pauling, and the Mo-
lecular Evolutionary Clock, 1959–1965.” Journal of the History of Biology
31: 155–178.

Newton-Smith, William. h. 1981. The Rationality of Science. Oxford: Ox-

prensa de la Universidad de Ford.

Ohta, t. 1977. “Extension to the Neutral Mutation Random Drift Hy-
pothesis”, Molecular Evolution and Polymorphism. Motoo Kimura, ed.
Mishima, Japón: National Institute of Genetics Publications.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Perspectives on Science

325

Ohta, t. and Kimura, METRO. 1971. “On the Constancy of the Evolutionary

Rate of Cistrons.” Journal of Molecular Evolution 1: 18–25.

Pickering, Andrew. 1984. Constructing Quarks. Edimburgo: Edimburgo

Prensa universitaria.

Quine, W.. V. oh. 1953. From a Logical Point of View. 2nd edition. Nueva York:

Harper y fila.

Sarich, V. and Wilson, A. 1967. “Immunological Time Scale for Hominid

Evolution.” Science 158: 1200–1202.

Rudge, David W. 1998. “A Bayesian Analysis of Strategies in Evolution-

ary Biology.” Perspectives on Science 6: 341–360.

———. 2001. “Kettlewell from an Error Statistician’s Point of View.”

Perspectives on Science 9: 59–77.

Schaffner, Kenneth F. 1993. Discovery and Explanation in Biology and Medi-

cine. chicago: University of Chicago Press.

Skipper, Jr., Robert A. 2000. The R. A. Fisher-Sewall Wright Controversy in
Philosophical Focus: Theory Evaluation in Population Genetics. Dissertation:
Universidad de Maryland, parque universitario.

———. 2002. “The Persistence of the R. A. Fisher-Sewall Wright Con-

troversy.” Biology and Philosophy 17: 341–367.

———. 2004a. “The Heuristic Role of Sewall Wright’s 1932 Adaptado

Landscape Diagram.” Philosophy of Science 71: 1176–1188.

———. 2004b. “Calibration of Laboratory Models in Population Genet-

ics.” Perspectives on Science 12: 369–393.

Sober, mi. 1994. “Let’s Razor Ockham’s Razor.” Pp. 136–157 in From a Bi-
ological Point of View. Edited by Elliott Sober. Cambridge, Reino Unido: Leva-
bridge University Press.

Solomon, METRO. 2001. Social Empiricism. Cambridge, MAMÁ: CON prensa.
Suppes, Patrick. 1962. “Models of Data.” Pp. 252–261 in Logic, Methodol-
ogy and Philosophy of Science: Actas de la 1960 International Congress.
Edited by E. Nagel , PAG. Suppes, y un. Tarski. Palo Alto, California: stanford
Prensa universitaria.

Takahata, Naoyuki. 1987. “On the Overdispersed Molecular Clock.” Ge-

netics 116: 169–179.

———. 1988. “More on the Episodic Clock.” Genetics 118: 387–388.
van Fraassen, Bas C. 1980. The Scientiªc Image. Nueva York: Oxford Univer-

sity Press.

wilson, A., Carlson, S. and White, t. 1977. “Biochemical Evolution.”

Annual Review of Biochemistry 46: 573–639.

Wimsatt, William C. 1980. “Reductionistic Research Heuristics and
Their Biases in the Units of Selection Controversy.” Pp. 213–259 in T.
Nickles ed., Scientiªc Discovery, volumen. 2: Case Studies. Dordrecht, El
Países Bajos: Reidel.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

326

Manipulating Underdetermination

Zuckerkandl, mi. and Pauling, l. 1962. “Molecular Disease, Evolución, y
Genetic Heterogeneity.” Horizons in Biochemistry. Kasha, METRO. y B.
Pullman, eds. Nueva York, Nueva York: Prensa académica.

———. 1965. “Evolutionary Divergence and Convergence in Proteins.”
Evolving Genes and Proteins. V. Bryson and H. Vogel, eds. Nueva York,
Nueva York: Prensa académica.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
pag
oh
s
C
/
a
r
t
i
C
mi
–
pag
d

yo

F
/

/

/

/

/

1
5
3
2
9
5
1
7
8
9
4
6
1
pag
oh
s
C
.
2
0
0
7
1
5
3
2
9
5
pag
d

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Descargar PDF