The following text has been automatically reproduced by an Optical Character Recognition (OCR) algorithm. It may not have been checked over by human eyes. For matters of precision please consult the original pdf.

Realism and Social Science

Realism and Social Science
Some Comments on Roy Bhaskar’s ‘The Possibility
of Naturalism’

Ted Benton

1 Introduction

An increasing body of philosophical work l
is now available which (a) presents a
‘realist’ alternative to the hitherto predominant ‘positivist’ and ‘conventionalist’

currents in the philosophy of science and
(b) attempts to use this realist account of
science in the analysis of social scientific
practice. In general the objective of this
analysis has been to transcend the polar
opposition, which has always characterised
debate in the philosophy of the social
sciences, between positivism and ‘humanist’,
‘hermeneutic’, or ‘neo-Kantian’ dualisms.

Commonly the outcome of this work has been
to sustain the explanatory procedures of
historical materialism, in one reading or
another, as compatible with realist philosophy. Further, elements of a realist
epistemology have also been attributed to
Marx, Engels and other Marxists in their
philosophical writings. What is remarkable,
though, has been the great diversity of
readings of Marxism – ranging from Critical
Theory to Althusserian structuralism which seem to be indifferently assimilable
to the realist defence.

Since, though, the new ‘transcendental’

realism is concerned solely with the general
conditions of possibility of a number of
characteristic forms of scientific activity
(experiment, scientific education, etc.), it
is neither surprising nor worrying to discover that it is equally compatible with
several different, even mutually incompatible substantive attempts at explanation
within a particular science. What might be
more worrying, though, is that it appears
to be compatible with more than one of a
number of conflicting philosophical reflections on those scientific traditions. In
part, I shall argue, this difficulty derives
from the reliance of the most influential
realist account of the natural sciences on
consideration of a narrow and inappropriate
range of these sciences. The application of~
the resulting model of natural scientific
activity to the social sciences has been
problematic in such a way as to reproduce

some of the familiar characteristics of the
positivist/dualist opposition.

The influential work i$ question is that
by Roy Bhaskar. His first book, A Realist
Theory of Science eRTS from now on), made an
lmmense contrihution-In establishing and
systematising transcendental realism as a
coherent and well-articulated alternative to
the established traditions in the philosophy
of science. These rival accounts of science,
characterised as ’empirical realism’ and
‘transcendental idealism’, were subjected to
formidable critiques, but almost wholly in
relation to their accounts of the natural
sciences. In RTS the question of the possibility of naturalistic social and psychological sciences is posed, but not systematically dealt with. Roy Bhaskar’s second
book, The Possibility of Naturalism (hereafter PN), takes up thlS challenge, arguing
for: ‘a qualified anti-positivist naturalism,
based on an essentially realist view of
science. Such a naturalism holds that
it is possible to give an account of
science under which the proper and moreor-less specific methods of both the
natural and social sciences can fall.

But it does not deny that there are
significant differences in these methods
grounded in real differences in their
subject matters and in the relationships
in which their sciences stand to them. ,2
Stated in these general terms, I am in broad
sympathy with Roy Bhaskar’s project, but on
the nature of the differences which he identifies, and their significance, I shall take
issue. In particular, I propose to argue
that the extent and significance of the
natural science/social science asymmetries
which Roy Bhaskar claims to identify would
justify description of his position as a
form of anti-naturalism, rather than as a
‘qualified naturalism’. It would follow
from this that his intended transcendence
of the positivism/hermeneutics polarity is
not entirely successful. The failure in
this respect derives from the reproduction
in Roy Bhaskar’s work of the very dualist
ontology of a natural/human opposition which

is the basis of hermeneutic and neo-Kantian
forms of anti-positivism. This ontology is,
in turn, sustained by an unnecessarily
restricted conception of the natural
sciences. This excludes, or under-represents, the philosophical and methodological
characteristics of a range of historical and
life-sciences whose bearing on the social
sciences, both philosophically and substantively, is direct and most pertinent to Roy
Bhaskar’s philosophical project.

2 The Argument of RTS

It will be remembered that RTS poses in
relation to a number of characteristic natural scientific practices the transcendental
question, ‘what are the conditions of possibility (presuppositions) of these activities
(or their rationality, or intelligibility)?’

The practices investigated in this way include experimentation, the application of
scientific knowledge in ‘open’ systems,
scientific perception, scientific education,
and change and development in science.

The~e are, unfortunately, some ambiguities
in Roy Bhaskar’s posing of these questions,
however, which have implications for the
status of the answers he gives. Some of
these ambiguities, and possible sources of
misunderstanding, are cleared up in Chapter
1 of PN, but some are persistent. Most significant are ambiguities surrounding the 3
premises of the transcendental deductions.

Are we to take as a premise the existence of
a scientific practice, such as experlment,
or, rather, its intelligibility, or, yet
again, its rationality (in the sense of
‘rational justifiability’)? It CGuld well
be argued, of course, that, since experiment
is a symbolically meaningful cognitive practice, it could hardly be said to exist unless it were intelligible. But there
remains an important difference between
accepting as a premise the intelligibility
of scientific experiment and accepting it
as rationally justifiable. It seems to me
that the strong ontological conclusions of
the transcendental deduction follow only
from the latter version of the premise, and
not the former.

In other words, it is legitimate to argue from the intelligibilitr of
scientific experiment to the presuEPosltion
that the world has such-and-such c aracteristics (i.e. that scientists who conduct
experiments are thereby committed to the
existence of a world with these characteristics) but that the world really does
possess those characteristics follows only
from the premise that experimentation is
rationally justified. It is, however, my
view that these difficulties of articulation
can be resolved, and, in any case, they are
not centrally involved in this paper’s concern with the application of the transcendental realist model of science to the
social sciences.

In RTS, then, transcendental arguments
are adduced to demonstrate the general
characteristics which must be possessed by
the world if it is to be a possible object
of scientific knowledge, and by society if
knowledge, as a species of social practice,
is to be sustained. These ‘conditions of
possibility’ of science can be grouped as
belonging to two ‘dimensions’, a ‘transit14

ive’ and an ‘intransitive’ dimension, which
are characterised as follows:

‘ … a transitive dimension, in which
the object is the material cause or
antecedently established knowledge
which is used to generate the new
knowledge, and an intransitive dimension, in which the object is the real
structure or mechanism which exists
and acts quite independently of men
and the conditions which allow men
access to it. ‘4
In the intransitive dimension, transcendental
deduction yields the conclusion that the
world is both structured and differentiated.

That is to say, that the world (unlike the
world of em~irical realist epistemology)
has ontologlcal depth. It is constituted by
mechanisms whose tendencies and powers may
or may not be exercised. When exercised,
the powers of real mechanisms may not be
‘realised’, and even when realised, the
resulting event-sequences may not be detected by ‘man’. The world is differentiated in
the sense that mechanisms may exist and
operate either in closed systems, where
‘constant-conjunction’ event-sequences do
occur, or in open systems where the outcomes
of the operation of a multiplicity of mechanisms are such that constant conjunctions
do occur. Characteristically mechanisms in
nature operate in open systems: usually,
though not always, closure is artificial,
the achievement of experimental practice.

Laws are ‘normic’ statements concerning the
tendencies or powers of things, which are
manifested in the form of constant conjunctions under conditions of closure, but
which must be supposed also to exist and be
exercised in open systems, where no constant
conjunction is manifest, because of codetermination of outcomes by other mechanisms.

In the transitive dimension, RTS concludes, society must be an ‘ensemble of
powers irreducible to, but present only in
the intentional actions of men’S who must,
in turn, be causal agents, capable of intentionally acting on the world, monitoring
this activity, and second-order monitoring
of this. In the transitive dimension, the
‘object’ is antecedently established knowledge which is transformed to produce new

Now, it follows directly from this that,
since social and psychological mechanisms
and structures clearly cannot exist and act
‘quite independently of men’, they are not
possible intransitive objects of scientific
knowledge. It may be that certain of their
general characteristics may be derived by
transcendental deductions of the conditions
of possibility of natural scientific practices, but here they figure as conditions, in
the transitive dimension, of scientific
knowledge of nature only, and as objects of
philosophical, rather than scientific,

Furthermore, since Roy Bhaskar’s central
arguments have been concerned with the implications of experimental activity, since
experimental activity presupposes the possibility of closed systems, and since we are
told that social and psychological mechanisms occur only in open systems, there
follows a further epistemolo·gical obstacle
to naturalistic social and psychological

sciences: the absence of experimental

Strictly speaking, then, Roy Bhaskar’s
position in RTS commits him to a radical
dualism of tne-natural and human domains,
which further commits him to an epistemological dualism with respect to the possibility of knowledge of these domains:



2 Predictive science
3 Experimental
practice sustained
4 Intransitive objects
of scientific

2 Predictive science
3 No experimental
4 Transitive condition
of scientific
knowledge only





The outcome of the position adopted in RTS,
then, seems to be a dualist anti-naturalism,
so far as the human sciences are concerned.

But this is not a conclusion which Roy
Bhaskar is readily prepared to accept.

Though apparently already ruled out by
definitional fiat, the possibility of a
naturalistic scientific knowledge of social
and psychological mechanisms does get discussed in RTS. Roy Bhaskar recognises that
so far his central argument has ‘turned on
the possibility of experimental activity’ ,6
so either some analogue of this in the human
sciences must be found, or we must ‘appreciate the great gulf that must separate them
from the sciences of nature,.7
Throughout the discussion there appears
to be’ an assumed correspondence of experimental sciences with natural sciences, on
the one hand, and non-experimental with
human, on the other, though this is neither
explicitly stated nor defended.

Fortunately, there is an analogue of experimentation in the social sciences. It is
that the theories which become embodied in
social practices may come to be seen by
participant social actors themselves as incapable of non-ad hoc explanation of significant phenomena (e.g. Neo-Classical economics and the 1930s depression). However, the
characterisation of this experiment-analogue
in RTS is very brief and sketchy. It also
seems to be rather unpromising for any proponent of a naturalistic approach in the
social sciences. Society itself is to be
understood as a colossal self-constructed
and self-interpreted experiment. There
seems to be no room for social science as a
distinct cognitive practice, with distinctive methods, and autonomous theory, as is
the case with the natural sciences. The
conception also is comparable in several

respects to the Popperian notion of ‘social
engineering’ as the social science analogue
of experiment, and is susceptible to broadly
similar objections. S
However, leaving aside the question of
the adequacy of this proposed experimentanalogue, it is important to recognise that
the very speculation which gives rise to it
– that naturalistic social science might be
possible – entails a revision in the definition of the transitive/intransitive boundary. If it is possible even to consider
that there may be scientific knowledge of
social and psychological mechanisms, then
it follows that it must be possible to consider person-dependent mechanisms as potential intransitive objects of knowledge.

Since this is ruled out by Roy Bhaskar’s
original definition of the intransitive
dimension, then it follows that a revision

of this definition is required if consistency is to be restored and the possibility
of naturalism explored.

3 The Argument of PN and Some Criticisms

A necessary condition of Roy Bhaskar’s
project in PN, then, is some revision of the
transitive/Intransitive distinction, and
consequent dispersal of the natural/human
opposition. Without this, the impossibility
of naturalism follows directly. The first
revision of the distinction comes in Chapter
1, where the mark of intransitive objects of
knowledge becomes that they ‘exist and act
independently of the knowledge of which they
are the objects,.9 This revision does allow
for the possibility that social and psychological mechanisms, processes etc., at least
under some characterisations of them, might
be possible intransitive objects of knowledge. It does, however, seem to rule out
the possibility in the case of one class of
such mechanisms and processes, namely those
which constitute knowledge. This problem of
the partial identity of subject and object
of knowledge is, indeed, a general difficulty for the maintenance of the transitive/
intransitive distinction in the human
sciences, and Roy Bhaskar later 10 produces
a further revision in the distinction to
take account of it. We can distinguish
between existential and causal independence:

such social relationships are existentially
independent of knowledge of, but causally interdependent with it. For the
social and human sciences, their intransitive objects are existentially but not causally independent of the processes by which
they are known.


But of course, to remove one obstacle to
the consideration of the possibility of
naturalism is not the same thing as to establish its possibility. It is to Roy
Bhaskar’s attempt to argue this that I shall
now turn, focussing on his argument as it
affects specifically social, as distinct
from psychological sciences. The main burden of the argument with which I shall be
concerned is given in Chapter 2 of PN.

Here, the argument is that there are fundamental differences between natural and
social objects of knowledge, which constitute ‘limits’ to naturalism in the social
sciences, but that these differences are
themselves conditions of possibility of
social scientific knowledge, in the same
sense, but not achieved in the same way as
natural scientific knowledge.

It might seem that, in investigating the
conditions of possibility, and the question
of their satisfaction, of social scientific
knowledge, the most obvious method would be
for a transcendental realist to apply the
procedures of RTS to this new domain. Social
scientific practlces would be identified,
and a transcendental deduction of their
conditions of possibility attempted. But,
as Roy Bhaskar rightly points out, what is
at issue here is precisely the question
whether there are any social scientific
practices, and:-rf there are, which they are.

The extension of~he method of~would
simply beg the question in favour, not just
of the possibility but the actuality, of

But the method adopted in PN as an alternative isn’t entirely clear. There are conflicting accounts of it, and the practice of
it doesn’t appear to be entirely consistent
with any of these accounts. My reconstruction of the argument is, therefore, rather
tentative. The argument appears to have
three main phases. First, the a priori
deduction of certain general propertles of
societies (and persons). Second, a comparison of these with those general properties
of natural objects in virtue of which they
are possible objects of natural scientific
knowledge. This comparison yields a series
of epistemologically significant ontological
differences. Third, the attempted demonstration that scientific knowledge of social
objects is possible, notwithstanding, or
rather, because of, these differences.

I shall deal with each of these three
phases of the argument in turn. The first
phase, the a priori demonstration of the
relevant emergent properties of societies,
is problematic in several respects. Sometimes the claim is that this demonstration
consists in an analysis of the necessary
conditions for any form of social life,ll
whereas elsewhere it is presented as a
derivation from the analysis of a number of
characteristic types of human activity
(‘ saying’, ‘doing’, ‘making’) .12 The principal argument, however, seems to be one
which takes the existence of intentional
activity as such as its premise. 13 On all
three of these characterisations the argument is a transcendental one – what must be
the case if ‘a’ (activity etc) is poSSI1)le.

If we take Roy Bhaskar’s argument that the
pre-existence of social forms is necessary
for intentional action, for example, this
is clearly a transcendental argument. But

there seems to be nothing, except, perhaps,
its greater generality, to distinguish it
from other uses of transcendental argumentforms in substantive social scientific
research. 14 Its status as a specifically
philosophical argument is in doubt. Its
content and plausibility rely on the acceptability of the ‘transformational’ model of
human practice which is introduced along
with it, and on a specific characterisation
of intentional action which is subject to
controversy among the different sociological
research traditions.

Now, the significance of this criticism
is not simply that Roy Bhaskar fails to sustain a distinction between philosophical and
substantive enquiry in the social sciences.

I am not sure that I would wish to place too
much weight on this distinction, in any
case, though ad hominem the argument must
have some force, Slnce Roy himself devotes
considerable space and ingenuity in the
attempt to preserve the distinction. lS

Rather, the significance of this criticism
is that the procedure adopted in this first
phase of the argument involves Roy Bhaskar,
after all, in siding with certain substantive research traditions within the social
sciences (specifically, Durkheimian, and
Marxian, or, rather, some versions of these)
against others, and not just in his conclusions, but in his very premise: the characterisation of intentional action. In short
this procedure is question-begging just as
much as would have been a direct application
of the method of RTS. There are, indeed,
systematic links between disputes over the
proper characterisation of intentional
action and disputes over what is and what is
not a properly ‘scientific’ approach to
social scientific investigation. Similar
remarks could be made about Roy Bhaskar’s
use of Durkheim’s conception of the ‘coercive power’ of society to demonstrate its
sui ~eneris reality.

T e second phase of the argument – the
comparison of the general properties of
societies with those of the objects of the
natural sciences with a view to their epistemological significance – is no less problematic. Of course, strictly speaking, if
the first phase of the argument fails, then
so does the second, but I propose to treat
the comparison of natural and social objects
in abstraction from the methodological difficulties involved in independently establishing the epistemologically significant properties of social objects. This is partly



because Roy Bhaskar’s argument has a great
deal of intrinsic interest, and partly
because I am in broad sympathy with some of
the most important features of his characterisation of social objects, despite my
reservations both about his methods of demonstrating them and about his ways of representing those methods.

Having introduced a limited dispersal of
the human/natural opposition (by means of
the revision of the transitive/intransitive
boundary) as a condition of even posing the
question of the ‘possibility of naturalism’,
Roy Bhaskar now proceeds to re-consolidate
that opposition in the forms of a series of
ontological, epistemological and relational
‘limits to naturalism’. The first ontological difference between natural and
social structures, which constitutes a limit
to the possibility of naturalism, is that
social structures do not, whereas natural
ones do, exist inde~endently of ‘the activities they govern’ .10 Now, this supposed
dis-analogy is imprecisely expressed, and,
moreover, does not appear to have been
established in phase one of the argument.

It is introduced, rather, as if selfevidently true. However, on the most
obvious interpretation of ‘activities they
govern’, it simply is not true that the
existence of social structures depends on
these activities. For example, the concept
of a power-structure required in empirical
sociological research must enable the investigator to identify power-relations where
powers are not, in fact, exercised, though
they continue to be possessed.1 7 In such
cases, the activities constituting the
exercise of powers (; governed by the powerstructure?) are not necessary to the existence of the power-structure (though other
activities may well be). The full coercive
power of the state, for example, may continue to be possessed without being exercised,
though such activities as the raising of
taxes, the recruiting, training, and equipping of armed personnel may well be necessary to the maintenance of that structure of
power-relations. This is entirely comparable with many natural mechanisms. An organism may, for example, never engage in reproductive activity, but yet retain its reproductive system and powers. However, some
activities of the organism (such as nutrltion) would be necessary to the retention of
these powers, but not the ones directly
governed by the reproductive system itself.

Elsewhere, Roy Bhaskar offers, possibly
as a general proposition including the
above, the characterisation of social
structures as not existing independently of
their effects: they (social structures) are
present only in and through the activities
of human agents. 18 It follows from this,
then, that in the social domain, all activities are activities of human agents. But,
to sustain the sui ieneris character of
social structures, It is necessary to distinguish between those activities of agents
which are exercises of their own intrinsic
powers, and those activities which are
really exercises of powers which reside in
social structures, but operate through the
activities of human agents. Surely, though,
if any person ‘A’ is the agent of an activity, ‘a’, then ‘A’ must be the possessor of
the power of which ‘a’ is the exercise. If

this is accepted then it follows that, at
best, we can distinguish only between powers
of agents possessed in virtue of ‘ their
intrinsic natures, and powers of agents
possessed in virtue of their relational
properties. Roy’s conception of social
structures does not, after all, sustain them
as autonomous possessors of causal powers,
or, therefore, as stii ieneris realities.

Roy Bhaskar is, it seems, committed to a
variant form of individualism in social
science. 19
A second ontological limit to naturalism
is that social structures do not, whereas
natural structures do, exist independently
of agents’ conceptions of what they are
doing. This thesis of the concept-dependence of social structures plays a large part
in the argument of PN, as well as in other
anti-naturalist works,20 but is subject to
varying interpretations which radically
affect its epistmological significance. Is
the thesis that, in general, social structures exist only if agents have some conception of what they are doing? N~it seems
to me hard to sustain the concept of an
agent at all without the notion of concept~
ualisation of activity, so that in so far as
human agents are a necessary condition for
the existence of social structures (and this
is hardly disputable) then the thesis is
sustained. But, as it stands, it seems to
me that it has little or no epistemological
significance. Certainly, it suggests that,
once established, scientific conceptions may
be in competition with pre-existing agents’

conceptions of the same activities. A
series of political consequences and problem~
flow from this, but no special” epistemological ones, vis-a-vis the natural sciences,
where similar disparities between science
and ‘common-sense’ persist.

At the opposite extreme, the thesis of
concept-dependence may be to the effect that
the existence of social structures depends
upon agents’ having the particular conception they do have of what they are doing.

Some relationships are, indeed, like this
(e.g. friendship).

If each party to the
relationship changes his or her conception
of what the relationship is, then the
relationship ipso facto ceases to exist.

But many, perhaps most, and certainly the
most sociologically significant, social
relationships are not like this at all.

Where society surrounds and sustains a relationship with sanctions, including coercive
powers, social relationships can be, and
are, sustained across great diversity of and
through immense changes in participating
actors’ conceptions of what they are doing
(employer/employee relationships, imperial
domination, and marriage are three clear
examples of such social structures).

Alternatively, the thesis of conceptdependence may be taken as specifying a
causal relationship between actors’ conceptions and the character of social structures,
such that changes in actors’ conceptions of
what they are doing are among the causes of
structural change. Such changes mayor may
not be in line with the intentions of the
actors whose conceptions change. Again, it
seems to me that this thesis is not obviously wrong. On the other hand, it hardly
counts as an a priori demonstrable truth
about society as such. Questions as to the

causal relationships between social structures of various types, and actors’ conceptions of them are open questions, whose
answers require empirical and theoretical
research. There is no reason to suppose
that any answer universalisable across all
types of social structure will be forthcoming. Furthermore, on this version, too,
there seem to be no serious epistemological
difficulties for the possibility of a social
science arising from the thesis of conceptdependence.

The third supposed ontological limitation
on the possibility of naturalism is that
‘social structures, unlike natural structures, may be only relatively enduring (so
that the tendencies they ground may not be
universal in the sense of space-time invariant),.21 It is, of course, true that
social structures may be in fact instantiated for historically limited periods of
time, and within geographically restricted
areas, but this is quite consistent with
their tendencies and powers being universal
wherever the appropriate structures are
instantiated. This is space-time invariance
in the required sense– i.e. spatio-temporal
locations are not in themselves causal
factors. Oddly, Roy Bhaskar himself seems
to recognise this when, later on, he says
that social laws may be universal within
their range, though restricted in their
scope. 22 But precisely the same is true of
the laws and structures of the natural
world. As Engels argued, the discovery of
historicity in nature was a distinctively
nineteenth-century achievement, culminating
in Lyell’s geology and Darwin’s evolutionary
biology.23 Natural mechanisms, like social
ones, are not eternal, but have definite
conditions of existence which mayor may not
be present at any point in space or time.

If we take into account Roy Bhaskar’s later
qualification of his position with respect
to the space-time variance of social structure, then he is committed to a denial of
historicity in nature. This would, indeed,
constitute a limit to naturalism in the
social sciences. Fortunately, we do not
have to agree that natural mechanisms are
not historical in character.

There is one respect, though, in which
the historicity of the social presents disanalogies with the historicity of natural
mechanisms which might be held to have epistemological consequences. It is generally
the case that the historical changes which
require basic conceptual distinctions in
their science (i.e. ‘qualitative’ changes,
in some uses of this term), have a temporal
periodicity which is very great in relation
to the periodicity of conceptual change in
science itself. In all cognitively relevant
senses, then, it can be said that the world
which is grasped through the categories of a
science following a revolution in that
science is the same world as was grasped,
perhaps with less penetration, by the superseded categories of the science. No major
new division of living organisms was, for
example, emerging contemporaneously with the
Darwin/Wallace production of the concept of
natural selection and which itself rendered
earlier theories obsolete. Now, scientific
revolutions and cognitive advances generally
are social processes. When they take social
processes as their objects, too, their ob18

jects have a temporal periodicity of change
which is of the same order as the periodicity of change in the knowledge-process
itself. Now this certainly can give rise to
methodological problems in the social
sciences – particularly with respect to lon~
term historical prediction. But epistemologically speaking, the situation is quite
comparable with the natural sciences. On
the very much greater time-scale of biologi~
al geological and cosmological change the
comparable long-term historical prediction
is equally suspect. There would be a distinctive epistemological problem for the
social sciences only if there were some
mechanism which ensured a necessary correspondence between cognitive and broader
social change. Such a mechanism is, indeed,
suggested in RTS and is a familiar feature
of some historICist Marxisms. Such a
necessary correspondence is, however, quite
incompatible with a conception of science
as a distinct and relatively autonomous
cognitive social practice which Roy Bhaskar
(most of the time) and myself, too, would
wish to sustain.

Next, Ray Bhaskar presents, as an epistemological limit to naturalism, the argument,
familiar from RTS, that social mechanisms
exist only in open systems and that, therefore, controlled experiment, prediction and
decisive tests of theory are impossible in
the social sciences. In answer to this, it
is first necessary to ask whether decisive
tests of theory are possible in the natural
sciences either. Even with an experimental
closure of the classic kind, assumptions
have to be made in practice about whether a
closure has, in fact, been obtained (i.e. an
assumption of the non-interference of undetected extrinsic influences on the
instance of the mechanism under investigation). Theoretical assumptions also have to
be made concerning the characterisation of
the mechanism and its activities, as well as
the instrumentation employed. Of course,
Roy Bhaskar is, in other contexts, well
aware of this, but the sharp natural/social
science contrast he draws can only be understood, I think, in terms of a residuum of
the positivist conception of the experiment/
prediction/testing relationship in his

Connectedly, it seems to be presupposed
in Roy’s argument that the constant conjunction of events associated with closure is
necessary for prediction. Why should this
be so? What is to rule out the calculation
of the resultant effects of the joint operation of a plurality of mechanisms? Prediction is always, of course, prediction of
something under some description. Where
very complex systems of interacting mechanisms, operating under conditions and ~
initial states which may be known only
approximately, outcomes may only be predictable as falling within a certain range of
possibilities. Of course, it might be
argued that even assuming a wide sense of
‘prediction’, the outcomes of multi-mechanism (‘open’) systems are only predictable
if it is possible to first isolate each
constituent mechanism to examine its operation independently and its relations with
others. This, of course, is a methodological problem of the social sciences but not,
it seems to me, an epistemological one.

Durkheim, for example, in his classic work
on sUicide,24 uses elementary statistical
comparisons in an effort to demonstrate
that a definite coefficient of preservation
or aggravation is associated with each of
several different religious ways of life.

The purpose of the statistical comparison
in each case is to rule out the possibility
that a given outcome (in this case, suicide
rate), or given contribution to such an
outcome, really is the result of religious
confession, rather than the operation of
some other mechanism (minority status in a
society, persecution etc). Of course,
Durkheim’s implementation is susceptible of
criticism, but the principle is clear, and
more sophisticated (though still, of course,
problematic in various ways) statistical
techniques have since been developed. In
cases like this, the isolation of mechanisms
is achieved theoretically and theory is
corrected on the basis of statistical comparisons of differently constituted systems
which nevertheless have one or more mechanisms in common.

Finally, but most importantly, Roy
B~askar again seems to neglect a range of
natural sciences in which experimental
closure is not an available means of
empirical control on theory. Historical
natural sciences such as geology and evolutionary biology explain phenomena in terms
of the interaction of pluralities of mechanisms in open systems. In each of these
sciences techniques have been developed many of them directly comparable to the uses
of statistics in sociology – for including
an element of empirical control into theoryproduction and theory-correction. The
classic experimental closure is one technique (class of techniques) among many,
which is available in some, but by no means
all, natural sciences. Roy Bhaskar’s
critique of the empirical realist, ‘constant
conjunction’, conception of causal laws is
insufficiently radical in that it retains a
certain paradigm of experimental closure and
its role in the testing of theories in
common with the ‘constant-conjunction’

account. Again, the result of this is an
artificial and unnecessary natural/social

Finally, Roy Bhaskar thinks there is a
‘relational’ limit to naturalism. This
derives from the familiar thesis of the
partial identity of subject and object of
social knowledge. Knowledge is itself a
social practice, so that when it takes socia]
practice as its object, maintenance of the
distinction between transitive and intransitive objects of knowledge is problematic.

As I have already indicated, though,2S Roy
Bhaskar distinguishes existential and causal
independence of the intransitive objects of

In the social sciences, it is
possible to sustain the existential independence of social structures, etc., whilst conceding that there is causal interaction
between subject and object of knowledge. But
the same is true, surely, of the natural
sciences. Experiment, for example, as RTS
well argues, presupposes causal interactl0n
between natural systems and human agents.

If these points are recognised then continued commitment to a natural science/social
science dualism on the basis of the ‘partial
identity’ thesis must derive from some con-

ception of the special or distinctive status
of self-knowledge, such as would be sustained by a residual Cartesian conception of
the subject. This is, for example, the metaphysical basis of Lukacs’s classic formulation of this natural/social science opposition.

The result of Roy Bhaskar’s comparison of
social and natural objects seems, then, to
be a series of concessions to anti-naturalism, such that his position would be better
described as a form of anti-naturalism,
rather than as a naturalism, however qualified. Nevertheless, he remains committed to
the possibility of a scientific social
science, if not a naturallstlc one. But the
greatest obstacle to even this – the absence
of prediction and experiment in the social
sciences – remains to be removed. As in
RTS, the search is for a social-science analogue of experimentation. This time, it is
the epistemological significance of social
crises which seems to offer promise of a
solution. If it is supposed that during
periods of social crisis, the underlying
generative structures of society become
visible to them, then one result of crisis
will be a transformation of participant
actors’ conceptions of their activities.

These new conceptualisations may now serve
as raw materials in the production of new
knowledge of the social £orm.26
There are, however, some serious difficulties in the way of such a process providing even a partial analogue of scientific
experiment. First, it appears to be a condition of production of new knowledge,
rather than a means of empirical control or
correction. Second, it seems to ‘entail that
social scientific knowledge is possible only
for those societies characterised by periodic crises of the required sort (capitalist
societies?), unless there are yet other
experiment-analogues appropriate to other
types of society. Third, Roy Bhaskar gives
us no theoretical account of the visibility/
invisibility of generative structures and,
surely, even if he could, this would beg the
epistemological question. Finally, one
universal feature of social crisis which is
difficult to reconcile with Roy Bhaskar’s
epistemological requirements is that they
polarise populations ideologically and politically. If actors make sense of the newly
visible generative structures in profoundly
diverse and antagonistic ways, what sense
it is still possible to make of metaphor of
‘visibility’, and how are we to solve the
problem of which actors’ conceptions are
adequate raw materials for scientific

It seems, then, that Roy Bhaskar, having


minimally dispersed the natural/social
opposition as a condition of posing the
question of the possibility of naturalism,
goes on to reconstitute that opposition.

The resulting philosophy of the social
sciences is anti-naturalistic, and seems
incapable of sustaining the possibility of
even a non-naturalistic social science. The
ontological opposition of the natural and
human domain~ continues to affect the epistemological argument throughout The Possibility of Naturalism, determining concessions
to anti-naturalism which are quite unnecessary. The ontology of the natural/human
opposition is itself sustained by the unduly
restricted range of sciences (mainly, though
not exclusively, physics and chemistry) and,
therefore, of scientific practices, which
are paradigmatic for the model of science
constructed in RTS and presupposed in PN.

This model of tne-natural sciences has-rn
common with the logical empiricism which it
so effectively refutes that it underrepresents historicity and development as
epistemologically significant characteristics of the objects of the natural sciences.

Eyolutionary biology, cosmology, geology,
embryology are all natural sciences for
which historicity and qualitative transformation pose epistemological and methodological problems which are in many respects
directly comparable with those encountered
in the sciences of human history, society
and psychology
It is also a characteristic of these
historical natural sciences that the explanatory models they employ designate mechanisms which are not practically isolable in
experimental closures. If the impossibility
of closure is an epistemological obstacle
to a scientific sociology, then it must
also be so for this range of natural
sciences. In fact, a great diversity of
non-experimental means of empirical control
and correction, as well as adaptations of
experimental methods themselves, have been
developed in these sciences. This is true
just as much of the historical social as of
the historical natural sciences. If we consider, for example, the range of empirical
controls involved in the production and
later correction of Darwin’s evolutionary
biology, it is easy to see that these by no
means all fit the paradigm of the classic
‘experiment’. An important raw material
for Darwin, which both establishes the
possibility of organic transformation, and
sets definite limits to the range of possible mechanisms which might be supposed to
bring it about, are rule-of-thumb generalisations derived from stock-breeders and
gardeners. These are forms of reflection
on non-experimental human interventions in
nature, which rule out certain theoretical
explanatory possibilities, and set definite
target-requirements for theoretical reasoning.

Another important set of empirical
controls was the range of theoretically
informed observations of the geographical
distribution of living forms, together with
palaeontological evidence of their historical succession, and the geographical distribution of irelated’ forms. Again these
are evidences from a non-experimental source
which tell against the idea of special creation, and for the notion of common descent

by gradual transformation. As to the
mechanism of natural selection itself, of
course, no experimental demonstration of the
formation of new species by its agency is
available, but the subsequent development of
such adjacent sciences as genetics and
ecology have both sustained and modified
Darwin’s conception, whilst elements of the
process are relatively isolable, and have
been examined by means of adaptations of
experimental technique. For example,
numerous investigators have exposed different colour-varieties of insect larvae
against various backgrounds in the vicinity
of the nests of insectivorous birds to discover differential rates of predation on
them. These ‘experiments’ can be combined
with statistical ‘thought-experiments’ to
determine the effects on the gene-pool of a
population through successive generations of
such differential predations.

Finally, on the ‘human’ side of Roy
Bhaskar’s natural/human opposition, there is
an unwillingness to conceive of forms of
historical causality as really distinct from
individual human agency, despite the prominence of the argument for the sui generis
reality of social structures. It 1S th1S
remnant of what has been called ‘the problematic of the subject’ which further sustains the ontological and epistemological
dualism of PN.

At the beginning of this paper, I quoted
Roy Bhaskar as advocating an anti-positivist
naturalism, according to which ‘it is possible to give an account of science under
which the proper and more-or-Iess specific
methods of both the natural and social
sciences can fall’ .27 It seem5 to me that
the anti-naturalist conclusions of PN are
part of a demonstration that RTS faIled in
this respect, and that the model of science
produced in that work requires revision to
take into account, in particular, epistemologically significant characteristics of
historical, developmental, and non-experimental natural and social sciences. This
would involve a-SYstematic attempt to adequately characterise and analyse the conditions of possibility of the non-experimental
empirical controls which I have above
sketched in relation to evolutionary biology
and Durkheim’s work on suicide. I remain
convinced that the outcome of such investigations would be a confirmation of the broad
outlines of Roy Bhaskar’s realist model, if
not of some of its more detailed articulation.

In case I should be misunderstood as advocating the kind of conception of a monolithic unified science which for so long
characterised logical empiricist orthodoxy,
it may be necessary to point out that my
arguments against Roy Bhaskar’s antinaturalism are designed less to show that
the social sciences are (or could be) more
like the natural ones that he supposes, than
to show that the natural sciences, or, at
least, some of them, are more like the sociaJ
than he supposes. More importantly, though,
I remain committed, as he does, to the view
that there are significant differences in
the methods of the different sciences, which
are grounded in real differences in the subject matters of those sciences and the relationships of those sciences to their subjectmatters. 28 Where I differ from Roy Bhaskar

and other anti-naturalists is that I think
these differences to be almost always of a
methodological rather than epistemological
kind, and that I do not, whereas Roy Bhaskar
does, align the whole range of methodological diversity along a single fault-plane,

dividing the natural and the social.

Methodologically, if not epistemologically,
the sciences display a ‘family resemblance’,
of cross-cutting and overlapping differences
and similarities of method.



This includes: Roy Bhaskar, A Realist Theory of Science, Leeds,
1975, and Hassocks and New Jersey, 1978, ‘Feyerabend and
Bachelard: Two Philosophies of Science’, New Left Review 94
(1975); ‘On the Possibility of Social Scientlflc Knowledge and
the Limits of Naturalism’, in Mepham and Ruben (eds.), Issues
in Marxist Philosophy, Vol. Ill, Hassocks, 1979; and The
Possihilllity of Naturalism, Brighton, 1979; RussellN:”” Keat,
‘Positivism, Naturalism and Anti-Naturalism in the Social
. Sciences’, Journal for the Theory of Social Behaviour, I,
pp.3-l7; Russell Keat and John Urry, Social Theory as Science,
London, 1975; T. B~nton, Philosophical Foundations of the
Three Sociologies, London, 1977, and ‘Natural SClence and
Cultural Struggle’ in Mepham and Ruben (eds.), op.cit., Vol.II;
Andrew Collier, ‘In Defence of Epistemology’, in Radlcal
Philosoph¥ 20, Summer 1978; and David Thomas, Naturalism and
Soclal SClence, Cambridge, 1979.

2 Roy Bhaskar, The Possibility of Naturalism, op.cit., p.3.

See, for example, RTS, pp.30-36, where ‘presuppositions’ become
‘conditions of posSIDility’ on p.36. Connectedly, we get
shifts from the ‘intelligibility’ to the ‘rationality’ and
‘existence’ of practices such as scientific perception, as if
these were equivalent.

4 RTS, p.17.

5 RTS, p.20.

6 RTS, p.244
7 RTS, p.245.

8 See, for example, my Philosophical Foundations, op.cit., p.38ff
9 PN, p.14.

10 PR, p.60.

11 PN, P .18.

12 PN, p.43.

13 PN, p.46 and p.65.

14 See PN, p.64ff.

15 See PN, p.64ff.

16 PN, p. 48.

17 See. for example, Steven Lukes, Power: A Radical View, London,
1976. My paper, ‘Objective Interest and the SocloIogy of
Power’ (unpublished) presents this argument at greater depth.

18 For example PN, p.50.

19 I am now somewhat sceptical as to the power of this argument,
but I retain it because it seems to me to have intrinsic
interest. I suspect that to see where it goes wrong, assuming
it does, would be illuminating.

20 Most well known of these is, perhaps, Peter Winch’s Idea of a
Social Science, London, 1959.

21 PN, p. 49.

22 PN, p. 16 5 •
23 See my ‘Natural Science and Cultural Struggle’, op.cit. (n.l).

24 E. Durkheim, Suicide, London, 1952.

25 p.15 above.

26 PN, p.6lff.

27 PN, p. 3.

28 PN, p. 3.

29 r-should like to thank the organiser of the Sociology Graduate
Seminar at the University of Sussex for providing the stimulus
to write this paper, and the participants in that seminar for
helping me to clarify and correct my ideas. I should also
like to thank Andrew Sayer, whose written comments were most






Download the PDFBuy the latest issue