Pre-publication version of :
„Problems with empiricism and the philosophy of science: Implications for purchasing research‟, European Journal of Purchasing and Supply, 1998, 4, 2/3,
Problems with empiricism and the philosophy of science:
implications for Purchasing Research
Following the establishment of IPSERA; the publication of the first book of readings taken from papers presented at an academic conference in the purchasing field (Cox 1996 a.); and the creation of an MPhil programme in Purchasing and Supply Management at Bath University, it is safe to assume that we are about to witness a rapid increase in the volume of empirical research undertaken in this field. There could be no better time to consider the philosophical foundations and problems of empirical research in Purchasing. The rationale for undertaking such an investigation stems not only from the obvious need for purchasing researchers to adopt the most professional and rigorous approach to their work, but also from the fact that many philosophers have drawn attention to the peculiarly intractable difficulties of conducting meaningful empirical research in the social sciences. This paper airs some of the more troublesome philosophical problems in the area and may serve as a useful, cautionary counterbalance to the wave of otherwise unbridled empirical enthusiasm that is likely to break over the profession in the next few years.
With the exception of Andrew Cox‟s (1996 b.) paper which is discussed below, no methodological discussion has appeared in the Purchasing Literature to date. The last major book by a recognised purchasing academic based on empirical research in the field (Lamming 1993), made no mention of the concept, and although the term does appear a few times in Cox‟s (1996 a.) text, it is used
1(inaccurately) to refer to specific research techniques or business practices.
Methodologies are concerned with the analysis of how research should be undertaken or how it can proceed, in other words the study of the means of attaining knowledge of the world, rather than the techniques of research or practice themselves. The Literature displays not only an absence of the philosophical usage of the concept, but also no discussion of either ontological issues referring to the kind of things that there are in the world or epistemological matters dealing with how those things can be made known to the researcher. Research methodologies necessarily embody both of these topics. Nevertheless, their absence from the Purchasing literature does not mean that methodologies are not in use. On the contrary, it is impossible to describe the world or undertake empirical research without reference, whether implicit or explicit, conscious or unconscious, to a philosophy, and thence, a methodology. As Hughes puts it:
1 Hines et al (Cox, (1996 a.), p. 71), for example, have a section on „Accounting and Cost Benefit Methodologies‟, but merely describe business techniques.
The relevance of the philosophical issues ....arises from the fact that every
research tool or procedure is inextricably embedded in commitments to
particular versions of the world and to knowing that world. To use a
questionnaire, to use an attitude scale, to take the role of a participant
observer, to select a random sample, to measure rates of population
growth, and so on, is to be involved in conceptions of the world which
allow these instruments to be used for the purposes conceived. No
technique or method of investigation is self-validating: its effectiveness, i.e.
its very status as a research instrument making the world tractable to
investigation, is, from a philosophical point of view, ultimately dependent
on epistemological justifications.
Hughes, (1980), p. 11
The dominant methodologies
Two methodologies dominate the empirical study of the social sciences - the phenomenological or interpretivist approach and the positivist approach. Empiricism itself refers to a set of philosophical beliefs formed around the idea
2that experience rather than reason is the source of our knowledge of the world.
In ordinary, rather than technical, philosophical usage, the term has also come to mean the practice of investigating the nature of the world using methods based on practical experience rather than theories or assumed principles. Positivism, meanwhile includes a focus on the approach to understanding the world employed in the natural sciences with its emphasis on facts as distinct from values or meanings, and the use of the „scientific method‟ in which theory is
3deduced as a result of formulating and testing hypotheses. This approach
identifies cause and effect through „the constant conjunction‟ of events, resulting
4in what has been called the „covering law‟ or „law-explanation‟ orthodoxy:
The basic theme will, I think, be familiar. It is that all science,
including history and the other social sciences, is devoted to the
pursuit of explanations, which take the form of general laws,
sometimes called covering laws. To explain an event is to relate
it to a general law, analysed as a universal generalisation. In a
rather hackneyed example, the freezing of my car radiator is
explained by the general laws governing the behaviour of water
plus the low temperature last night (initial conditions). The
roots of this conception of explanation lie in Hume‟s theory of
causation, according to which all we can ever observe is the
„constant conjunction‟ of events, such as freezing temperatures
2 For an account of the historical development of the philosophy of Empiricism from Hume through the phenomenalists and the subsequent private language debate, see the introduction to Morick‟s (1980) text. 3 See Hughes (1980), chapter 2 and Bhaskar (1986), chapter 3 for a critical, detailed analysis of the precise meaning and nature of Positivism 4 ‟To give a causal explanation of an event means to deduce a statement which describes it, using as premises of the deduction one or more universal laws, together with certain
singular statements, the initial conditions.„ Popper (1959), p.59
and burst radiators. This is all we can know, and all we need to
know for empirical science to be possible.
Outhwaite, (1987), p. 7
As this extract indicates, David Hume sowed the seeds of this approach to the
5investigation of the world back in the nineteenth century. Hume‟s identification
of the significance of the constant conjunction of cause and effect experienced as event-regularities forms the basis for positivism‟s interest in both the frequency and sequencing of phenomena and its reliance upon quantitative research techniques. Despite having been under sustained attack from a variety of quarters for at least thirty years, positivism remains a powerful methodological
6 is influence on empirical work in the social sciences. Cox‟s recent discussion
clearly deep in the thrall of this particular methodology:
„...it is important for academics in procurement to recognize
that robust techniques for operational applications, can only be
refined if they are first grounded in a scientific approach.....Thus
theory building must start not from a reliance on descriptive
systematic observations of discrete events in the real world, but
with the testing of a general law developed through inductive
reasoning. By this means, empirical cases can be utilized to test
the validity of a general law or theory and, in so doing, ensure
the development of robust, predictive and operationally useful
concepts, tools and techniques.
Cox, (1996 b.), p. 59
Positivism was developed in the „hard‟, natural sciences where the covering-law
technique has proved extremely successful. However, when the methodology was applied in disciplines such as Sociology and Anthropology, serious ontological problems were encountered. The „anti-naturalist‟ school of thought believes that
the subject-matter of the social sciences is so different from that of the natural sciences that an entirely different approach to empirical work is demanded, the other, „naturalistic‟ school, believes that the empiricist principles used in the
7study of nature are applicable in the social sciences.
5 „The nature of experience is this. We remember to have had frequent instances of the existence of one species of objects; and also remember, that the individual of another species of objects have always attended them, and have existed in regular order of contiguity and succession with regard to them. Thus we remember to have seen that species of object we call flame, and to have felt that species of sensation we call heat. We
likewise call to mind their constant conjunction in all past instances. Without any further ceremony, we call the one cause and the other effect and infer the existence of
one from that of the other....Thus in advancing we have insensibly discover‟d a new
relation betwixt cause and effect, when we least expected it, and were entirely employ‟d upon another subject. This relation is their CONSTANT CONJUNCTION.‟
Hume, (1888), p. 87 6 Cox, (1996 b.) 7 See Bhaskar, (1989), p. 66
One key difficulty in any attempt to apply a positivist research methodology in the social sciences stems from the nature of the principal actors in all such studies - the human animal. The positivist ontology assumes that the actors in events being studied are uniform, atomistic, passive agents who do nothing more
8 Social systems, from than observe and record the constant conjunction of events.
this viewpoint, are no more than the sum of the individual actors. These assumptions are, however, untenable. Unlike chemicals or machines, human beings are capable of learning and changing both consciously and unconsciously, not only their own behaviour, but also the form and structure of any systems
9 They are in other words, active, self-aware, reflexive and that they are part of.
capable of perceiving and generating meaning. These alternative ontological
claims challenge the wisdom of trying to apply the positivist methodology in the social sciences. An extensive literature has developed around the suggestion that in order to „explain‟ what human actors are doing it is necessary to understand or
10reconstruct the social experiences of those actors. Awareness of this problem led
Schutz, for example, to argue that:
...the most serious question which the methodology of the
social sciences has to answer is: How is it possible to form
objective concepts and an objectively verifiable theory of
Schutz, (1962), p. 34
By way of illustration, imagine some research aimed at discovering „best practice‟ in the area of purchase order selection. This might, for example, involve recording data concerning the type of order selected by a variety of buyers in different sized companies, and a number of characteristics of the orders they were raising such as value, commodity type, the duration of the contract agreed and so on. Conventional statistical analysis might well uncover a data/event-regularity in the form of an association in large companies between large value purchases and, say, blanket orders. From an (extreme) positivist viewpoint, that would be the end of the affair. The event-regularity would enable the formation of a covering-law of the type - „In companies with more than (x) employees, when
the value of a purchase exceeds (?y), blanket orders are employed‟. To a positivist, that statement would constitute an „explanation‟ of the phenomenon.
There are many possible objections to this claim. Let us suppose that once uncovered, the „law‟ allowed us to predict (correctly) that in such companies, purchases above a value of (?y) do indeed tend to be treated with blanket orders. What would happen if we were to ask: „Why are they treated with blanket
orders?‟ Although several answers to this question may be possible, they will all necessarily be variations around the theme: „ Because in that kind of company, that is what happened in the past.‟ Thus, one may argue that the „explanation‟, explains nothing at all, it merely describes or reflects some regularities in the data; some „constant conjunction of events‟.
8 See Lawson,(1994) and Sayer,(1984). 9 Artificial Intelligence buffs may well argue at this point that neural networks, for example, are self-evidently capable of learning and changing. However, space limitations prohibit an exploration of the contentious issue of precisely what is meant by „learning„ in the current context, and its connection with the concept of consciousness. 10 This idea was given the name Verstehen by Weber and extensively discussed in his
Moreover, although the study might reveal a pattern in the data allowing the positivist researcher to generate a covering-law, the underlying reality of what was happening in these large departments might well be that the buyers had procedure manuals which stipulated the type of order to be used in different circumstances. Thus the choice of blanket orders might more usefully have been ascribed to the effect of buyers following their organisations‟ procedures. Now one might legitimately ask if the covering-law is actually explaining cause and effect. Is there a causal relation between the value of the purchase and the type of order, or is the choice of order actually determined or „caused‟ by the meaning the actors
11 From this have imbued and subsequently perceived in their procedure manual?point of view, the regularity uncovered by the research is only the external appearance of what the actors understand. The cause of the buyers‟ behaviour
lies not in the value of the order alone, but in the meaning the actors perceive in
It can be argued, furthermore, that humans actively create the systems that social scientists are trying to understand. This idea was extensively
12discussed by Weber in the later 1940s, and the notion that knowledge is thus
13socially constructed has been widely developed since that time producing a
14whole raft of different approaches many of which share an epistemological
emphasis on the importance of language:
......as far as social reality is concerned, it cannot be studied
independently of the theories, conceptions, subjectivities if you
will, of the members of that society; and, as an arguable
implication, there is no reality apart from the subjectivities for
our theories to correspond to. As far as social reality is
concerned, it is constituted subjectively. An alternative
formulation is to postulate social realities as being constructed
in and through meanings and to say that social realities cannot
be identified in abstraction from the language in which they are
Hughes, (1980), p 117
This line of reasoning attacks the epistemological foundations of empiricism and positivism, throwing further doubt on the relevance and validity of the search for event-regularities in the study of social phenomena or mechanisms. Without a consideration and understanding of the human or „social‟ aspects of purchasing systems, precisely how reliable and useful will Andrew Cox‟s suggested use of
case-studies to „...test the „validity of a general law or theory...‟ be?
The socially constructed nature of knowledge in human social systems causes particularly acute problems for researchers in their choice of data collection and manipulation techniques.
11 Arguments of this kind have been extensively explored in the field of Sociology. See, for example Ryan (1970), pp. 140-141 for a more detailed discussion, much of the Hermeneutic literature (see below) also deals with the task of uncovering and understanding meaning in human social relations. 12 Weber (1949), see pages 72-81 13 Schutz (1962) represents an outstanding contribution to the elaboration of Weber‟s analysis of these ideas. 14 See chapters 11 to 15 in Morick (1980).
The context-dependent nature of data
Whereas positivism is commonly associated with a quantitative approach to data generation and collection, in the interpretivist approach, qualitative techniques
15tend to dominate. There are a number of reasons for this difference in emphasis, not least the fact that although it is possible to quantify, record and subsequently use mathematics to manipulate measures of attitudes, opinions and so on, the methodological justifications for so doing are highly questionable. The unavoidable need to employ language in obtaining data in many social systems complicates the process of data analysis:
Practically adequate forms of quantifying using interval scales
can only be developed for objects and processes which are
qualitatively invariant, at least in their fundamentals. As such,
they can be split up and combined without changing their nature.
We can measure them at different times or places in different
conditions and know that we are not measuring different things.
Sayer, (1984), p. 177
Is it really valid, for example, to do something as apparently straightforward as take answers to questionnaires administered in a number of different company environments, reduce the responses to numbers and then sum them? The context-dependent nature of human attitudes and opinions (in particular) make
16the collection and analysis of such data extremely problematic. The
significance and relevance of these concerns, although recognised by researchers in disciplines such as Sociology, are frequently dismissed or overlooked in disciplines dealing with business topics. Information concerning many of the phenomena that Purchasing researchers are interested in for example, already
exists in quantifiable form. Companies routinely measure their own behaviour using money and thus record data in numerical form. However, this should not be a cue for researcher complacency. Much of the data generated by companies is not of the same order of „objectivity‟ or „reliability‟ as say, information on wind-
speed generated by a well calibrated anemometer. If, for example, Cox‟s
17argumentsin support of an emphasis on the search for „best practice‟ are taken up by the academic wing of the Purchasing profession, then there will be many future studies that include profit and purchase savings figures in attempts to identify „successful‟ companies or Purchasing Functions. Profits and savings are already conveniently recorded and expressed in numerical form. Nevertheless, the need for extreme caution should be apparent since, as any working accountant or buyer will tell you, these two, key variables are highly susceptible to human interpretation and manipulation, and should never be taken at their
face value. Reliable, „hard‟ data is difficult to obtain in the social sciences, and in
the absence of such data the process of identifying covering-laws capable of producing generalisable predictions or policy prescriptions becomes extremely problematic.
15 There are many texts available in this area, see for example, Gummesson (1991), Bogdan & Taylor (1975) or Van Maanen (1983). 16 Any researcher planning to try to measure attitudes or use questionnaires should read Cicourel (1964) and chapter 6 of Hughes (1980) with their alarming critique of these particular data collection techniques. 17 Cox (1996 a.), pp. 12-14
It would appear then, that the two main approaches to empirical research in the social sciences are quite distinct. On the one hand, positivism assumes passive human data recorders and employs deduction and the scientific method
18of the natural sciences, primarily implemented using quantitative research
techniques, to formulate and test hypotheses relating to cause and effect relations indicated by the existence of event-regularities with a view to inductively establishing generally applicable laws. On the other, the interpretivist methodology assumes that human beings are active rather than
19thus leading to socially constructed knowledge, tends to derive theory passive
20 from collected data rather than deductive hypotheses,and generates data using
qualitative research techniques with a view to identifying meaning and perhaps
21purpose rather than event regularities.
However, this clear, methodological dichotomy is not perfectly reproduced in empirical practice. In many social science disciplines empirical research seems to proceed using a mixture of the two methodologies and their associated techniques. In practice, as the examples shown below illustrate, it can be quite difficult to identify from the Literature precisely which approach is in use.
The mixing of methodologies
The following extracts were all taken from texts whose authors might reasonably be described as working in the „interpretivist‟ tradition. Thus, one mainstream qualitative research text offers a model of the „research cycle‟ which includes the stages of:
data analysis - description - generalization - explanation -
prediction - policy and practice
Marshall and Rossman, (1989), p. 23
Kirk & Miller speaking from an anthropological viewpoint argue that :
Qualitative research has gotten bad press for the wrong reasons
and good press for the wrong reasons. Complicating the
problem some nonqualitative enthusiasts brand qualitative
research as „descriptive‟, by which they mean nonquantitative.
This pejorative use of the term is wrong-headed. Descriptive
work can be either qualitative or quantitative (e.g. descriptive
statistics). More important is whether or not research of any
category - whether qualitative or not - is in some way
hypothesis testing. When it is, such work has a potential to
modify a scientific paradigm directly. When not, the assembly of
18 A suggestion that was of course famously challenged by Kuhn, (1970). 19 See Fay (1987), chapter 3, section 3.2 20 Glaser and Strauss (1967). 21 „Qualitative researchers are characteristically concerned in their research with
attempting to accurately describe, decode and interpret the precise meanings to persons of phenomena occurring in their normal social contexts and are typically pre-occupied with complexity, authenticity, contextualisation, shared subjectivity of researcher and researched and minimization of illusion.‟
Fryer (1991), p. 3
„baseline‟ information makes a different and indirect
contribution to the evolution of science.
Kirk & Miller, (1986), p.71
Glaser and Strauss, whose idea of „grounded theory‟ lies at the heart of the qualitative approach discuss the problem of „verifying‟, suggesting that:
When the analyst turns to theoretical concerns, evidence is
invariably used as a test of his hypotheses - and thereby of the
relevance of his categories; comparative data give the best test.
Both implicitly and explicitly, the analyst continually checks out
his theory as the data pours in. Explicit verification beyond
testing his hypotheses may lead to establishing major
uniformities and universals, to strategic variations of theory
under different conditions, and to grounded modifications of
Glaser and Strauss, (1967), P. 26-7
One might be forgiven for mistaking this language as that of a full-blooded positivist. It would appear that the differences between the practices of the
positivist and interpretivist approaches to empirical research are not as pronounced as an understanding of the respective methodologies would suggest. Talk of hypotheses, testing, probabilistic causes, generalisations, uniformities and regularities abound throughout the qualitative research literature. The
attraction of the positivist approach for social scientists may spring partly from the fact that that Hume‟s insight concerning the relationship between event-
regularities and causation matches so closely our everyday experience of the
22world that it appears to be „self-evidently‟ „correct‟. Moreover, Sayer suggests
that one of the reasons why researchers try to form generalisations is that :
Many social scientists have believed that with further research
generalizations (of the type that lead to predictions of other
situations might be like) might be „firmed up‟ into laws of
human behaviour, whether deterministic or probabilistic,
although there is scarcely a scrap of evidence to suggest they
are succeeding. In other words, generalizations are seen by
some as an end in themselves, and as central to a conceptions of
social science as the search for order and regularity.
Sayer, (1984), p. 100
It is all the more unfortunate therefore, that perhaps the most damning of all recent criticisms of empiricism concerns the logical impossibility in the study of social systems of identifying meaningful event-regularities that can subsequently be used to formulate predictive generalisations.
The Realist Challenge - cause and effect in open systems
22 This may be little more than a linguistic phenomenon. See Whorf, (1941) p. 152 for an intriguing version of just such an argument.
Positivist methodology suggests that we can/should examine the world with a view to identifying event regularities, and from those, causal relations. However, even if we assume that we have circumvented the multifarious difficulties of obtaining valid, reliable data concerning human constructs and behaviour, this only remains a sensible strategy if the researcher can be sure that the event regularities uncovered in that data actually indicate the existence of causal relations. In the natural sciences this certainty is sought after by experimentally trying to create „closed‟ systems; that is systems in which all influences other than chance and the causal relation that is being examined, sought or tested
23 this is only have been eliminated. According to the Critical Realist approach
possible if the following two conditions are in place :
1. There must be no change or qualitative variation (e.g.
impurities) in the object possessing the causal powers if
mechanisms are to operate consistently. This is termed by
Bhaskar the „intrinsic condition for closure‟......
2. The relationship between the causal mechanism and those of
its external conditions which make some difference to its
operation and effects must be constant if the outcome is to be
regular (the extrinsic condition for closure).
Sayer, (1984), p.122
The problem for the social sciences it that the systems they study are never
closed as so defined, but are instead, always and everywhere „open‟ in nature. The objects, processes or actors in social system are frequently subject to change or qualitative variation, as are the systems‟ external conditions. The constant conjunction of events, viz. regularities or patterns in empirical data suggesting an „if event A occurs, then event B occurs‟ type relationship, consequently can
occur without any underlying causal connection. Moreover, because casual factors in open systems may, for example, act to cancel each other‟s influence out, it is also possible that no regularities will appear despite the existence of causal relations between the factors being studied.
For an illustration of the former result, consider an imaginary empirical study designed to test the deductive hypothesis that there is likely to be a positive correlation between the level of formal qualifications held by professional buyers, and their ability to negotiate favourable contract terms in the purchase of a commodity (for the sake of this example assume that the word „favourable‟ is non-ambiguous !). Further suppose that the research set out to measure the performance of six well, and six poorly qualified buyers all negotiating the purchase of an identical commodity from the same salesman. The critical realist‟s intrinsic condition for closure states that the performance of the subject under study must be constant. The buyers being observed should not, for example, become less effective because, say, their marriages all collapse during the course of the study. The extrinsic condition states that the relationship between the
object of study and external influences must also remain constant. There must not, for example be a sudden global physical shortage of the commodity under negotiation, since this would cause a price increase beyond the agents‟ control. Now suppose that during the course of the study, of the six well qualified buyers, two suffered from a particularly debilitating strain of flu; one accepted an attractive job offer from one of her employer‟s competitors and lost interest in
23 See Bhaskar, (1978) and (1989), Sayer,(1984) and Collier, (1995)
negotiating a favourable deal and one was in a long-term corrupt trading relationship with the supplier and consistently offered them unusually high profit margins. Meanwhile assume that the poorly qualified buyers were all in
rude good health and trying their level best for their employers. It is possible that the research would „uncover‟ a negative correlation between the level of academic qualification and observed negotiating ability. However, due to the open nature of the system, the regularity would be spurious and the study would have failed to reveal a genuine causal relation. Hence the unavoidable conclusion that in open systems, where regularities appear at all, there is logically no way of knowing if they are indications of causal effects, and furthermore, that an absence of regularities is not a reliable indicator of an absence of causal relations. Without the ability to experimentally isolate the variables being studied, and thus create closed, or at least quasi-closed systems after the fashion of experiments in the natural sciences, the social sciences can never be sure of having correctly identified cause and effect. This conclusion may go some way towards explaining not only the general failure of the social sciences to identify many significant event regularities to date, but the also the prediction that the field of Purchasing, whose phenomena are always and everywhere associated
24with the effects of human behaviour, will prove equally barren.
What are we searching for?
The Critical Realist arguments pose something of a dilemma for researchers - if they can never be sure of identifying meaningful event-regularities and thus causal relationships, what then are they supposed to look for? How is empirical work in Purchasing to proceed?
Only predict ?
This discussion is confusing and unsettling. Meaningful or useful empirical research begins to look impossible or pointless. Precisely the same doubts and concerns have, of course, been confronted in other disciplines, and purchasing researchers might do worse than to consider the solutions developed elsewhere.
Economics, with its focus, inter alia, on the trading behaviour of
companies, buyers and sellers, is perhaps the most closely related, „reputable‟ discipline to purchasing. Faced with the extraordinary difficulty of isolating causal relations in economic systems, some economists have abandoned empirical work entirely, focusing instead on „high-theory‟ concerned with purely logical
issues, others have turned their backs a few decades ago on the research objective of trying to understand reality and settled instead for the much more modest goal of being able to make accurate predictions. Thus Milton Friedman famously argued that :
the only relevant test of the validity of a hypothesis is the
comparison of its predictions with experience. The hypothesis is
rejected if its predictions are contradicted (“frequently” or more
often than predictions from an alternative hypothesis); it is
accepted if its predictions are not contradicted; great confidence
is attached to it if it has survived many opportunities for
24 See Sayer,(1984), p.100, Lawson,(1994) p. 517