CASEART.DOC

Pre-publication version of paper submitted to the Second World-wide Research Symposium on Purchasing and Supply Chain Management, 1-3 April, 1998, London

Generalising from single samples - Evaluating the contribution of case-study analysis to empirical purchasing research

Introduction

[The ’one-shot’ case-study] design is an observation only of what exists at the time of the study, as such, it has no control over extrinsic and intrinsic factors. In addition, it does not allow for manipulation of the independent variable or for before-after or control-group-experimental-group comparison. Furthermore, since case studies analyze single unsampled systems, they are weak on generalization as well. Studies that employ the One-Shot Case-Study design have no checks on internal validity and thus are of little use in testing causal relations. Indeed this design has been denoted by some methodologists as pre-experimental.

Nachmias & Nachmias, (1976), p.42

Criticisms of this kind may have encouraged the abandonment of the case-study approach to empirical research in some academic fields. Mitchell (1983), p. 187 observes that it seems to have disappeared, for example, in Sociology in the early 1950s when it was replaced by the familiar jargon of variable analysis, sampling techniques et al. Much the same is true of Economics where the approach has been largely relegated to the status of a teaching aid. Nevertheless empirical research based on case-studies is alive and well in a variety of other fields such as Politics, Anthropology and Psychology. Moreover, a brief survey of the Purchasing literature indicates that the case-study approach has, in the absence of any critical debate, already become an extremely popular research strategy in this newly formed field of study.[i] In the light of the severity of some of the methodological attacks on the case-study, it would seem prudent for purchasing academics to pause and take a considered look at the debate before continuing with their current, apparently unquestioning embrace of this particular research strategy.

The Nachmias attack is based on an application of the ontology and epistemology of Humean positivism which regards knowledge of the world as proceeding from a search for, and uncovering of, patterns or regularities in samples of data relating to events in the empirical realm. The resulting regularities may then be formulated as ‘laws’ that can be applied as generalisations concerning the relationships between the relevant variables in the parent population of phenomena. The subsequent explanation of phenomena in social systems thus -consists of generalisable laws that permit the researcher to make predictions concerning future relationships between variables.

If the positivist approach to understanding social phenomena is accepted, then it is reasonable to ask if the case-study research strategy is capable of uncovering generalisable relationships between variables. Stated baldly, this line of criticism asks - if the research objective is the creation of a generally applicable understanding of purchasing phenomena, what is the point of looking closely at a sample of one? How can a general law be drawn from an appreciation of the way in which one company, department or individual behaves? It is, of course, possible to challenge the positivist ontological and epistemological foundations upon which such questions rest, and that task will be tackled below. However, before the validity of the positivist methodology is itself brought into question, the claim of Nachmias & Nachmias and others[ii] that the case-study approach is incapable of producing useful and valid generalisations, will be challenged on two grounds. One major dissenting line of thought deals with the nature of the process of causal explanation, the other with the nature of the phenomena under examination.

The process of explanation - analytical induction and logical inference

In the 1930s, before positivism had established its grip upon much empirical research and the plethora of statistical techniques and PCs capable of easily performing the necessary calculations had been invented, voices could be heard resisting the new epistemology. Thus Znaniecki (1934) can be found arguing strongly in favour of analytical induction in which:

...certain particular objects are determined by intensive study, and the problem is to define the logical classes which they represent. No definition of the class precedes in analytical induction the selection of data to be studied as representative of this class....

Znaniecki, (1934), p. 249

The procedure he championed involved establishing definitions or descriptions of concepts and processes based upon an identification of their ’essential’ characteristics:

Analytic induction abstracts from the given concrete case, characters that are essential to it and generalizes them, presuming that in so far as essential, they must be similar in many cases...... Thus, when a particular concrete case is being analyzed as typical or eidetic, we assume that those traits that are essential to it, which determine what it is, are common to and distinctive of all cases of a class.

ibid., p. 250

If the essential characteristics of a phenomenon can indeed be identified, then it will be logically possible to generalise from a single sample of a phenomenon by means of, for example, case study analysis.

Half a century later, in a development of this kind of inductive argument, Mitchell (1983) can be found discussing the difference between statistical and logical inference, stating:

Scientific or causal - or perhaps more appropriately - logical inference, is the process by which the analyst draws conclusions about the essential linkage between two or more characteristics in terms of some systematic explanatory schema - some set of theoretical propositions.

Mitchell, (1983), p. 199-200

and concluding that:

... the process of inference from case studies is only logical or causal and cannot be statistical and that extrapolability from any one case study to like situations in general is based only on logical inference. [Thus] We infer that the features present in the case study will be related in a wider population not because the case is representative but because our analysis is unassailable.

Mitchell, (1983), p. 200

Hence, if research into a single sample/case leads to the discovery of an ‘unassailable’ causal explanation of a phenomenon, then the researcher may be confident that the same explanation will apply to other members of the parent population from which the sample was taken. [iii]

The excavation of causation - skeletal specifications

Mitchell and Znaniecki provide arguments for the combination of very small or single samples with induction in the search for generalisable causal explanations. However, neither author explores the question of precisely how causation might be uncovered in the practice of empirical research, and the limitations that practice might impose upon the possibility of producing generalisable causal explanations. Some light is thrown on this subject by the literature surrounding the idea that it is possible to usefully distinguish between an entire phenomenon and a sub-section of that whole that can be called an underlying ‘process’.

Writing in the context of educational research, Howard Becker (1990) cites the example of empirical research in mens’ and womens’ prisons. At first sight, the phenomena are quite different, Male prisons have, inter alia, well developed markets for drugs, clothes and so on, and the inmates devise prison-specific homosexual relationships that do not threaten their heterosexuality in the outside world. They also maintain a ‘convict code’ of behaviour based on loyalty towards other inmates and distrust of the guards. Female prisons, on the other hand, do not have developed markets, sexual relationships resemble pseudo-families and they have no convict code. Nevertheless, Becker argues that although the details of the phenomena are quite different, there is an underlying common ‘process’ in which the deprivations of prison life lead to the creation of a prison culture. Consequently he argues:

.....generalizations are not about how all prisons are just the same, but about a process, the same no matter where it occurs, in which variations in conditions create variations in results.

Becker, (1990), p. 240

Similarly, in a discussion of organisational research that echoes Mitchell’s earlier thoughts, Hartley (1994) argues that:

The detailed examination of organizations in a context can reveal processes which can be proposed as general or as peculiar to that organization. The detailed knowledge of the organization and especially the knowledge about the processes underlying the behaviour and its context can help to specify the conditions under which the behaviour can be expected to occur. In other words the generalization is about theoretical propositions not about populations.

Hartley, (1994), pp. 225-6:

This focus on underlying processes appears to be an attractive suggestion. In the field of purchasing phenomena for example, although it may well prove impossible to produce a general theory about, say, buyer-supplier negotiations that could be applicable to all types of company regardless of size, supplier base, product base or market structure, and every negotiator regardless of personal style, personality and the like, it might be possible to identify some underlying processes in negotiation practice that are so generalisable. This proposition appears all the more promising in view of the fact that it is reasonable to expect to find similarities in the underlying processes of individual cases in purchasing. Many of the buyers in companies have been through the same education process. In the UK there is only one professional Institute that represents professional buyers, consequently many buyers have studied the same academic syllabi and the same limited literature. Furthermore, with the increase in the popularity of practices such as bench-marking and the seeking out of ‘best practice’, many companies actively try to copy each other’s operations. In such circumstances it would be extremely surprising if empirical research did not uncover a number of similar, underlying processes.

We may usefully take a leaf out of the systems analysts’ book at this point. Significant proportions of their lives are concerned with the stripping away of the idiosyncratic surface details of processes within companies with a view to identifying and specifying what might be called underlying, ‘skeletal’ processes:

The physical model is contaminated with many physical and technology-related activities and memories simply because that is the way things are done in the current system. We want to separate the what from the how. Namely we want to separate the things that do the data transformations from the things that service the processes that do these transformations. Simply stated, we want to know what processes and data stores serve the users and which ones serve the system. In this way, we can distinguish between changes in the technology of how things are done from changes in the fundamental business practices implemented in the system.

Kowal, (1988), p. 155

Analysts seek to uncover what they term a logical model with which they can:

...briefly and concisely explain the system under study...... [and] describe the system in terms of its most important activities and task, its essential data, and the events to which the system responds.’

ibid., (1988), p. 271

There would seem to be little doubt then, that it is possible to strip away surface detail and reveal underlying processes. It is, moreover conceivable, as we shall see below, that those same processes may well be present in the broader population of phenomena.

In summary, support for the argument that it may be possible to use the case-study research strategy to produce valid, generally applicable explanations from a single sample of a phenomenon can be found in the ideas of:

·  analytical induction and the identification of the ‘essential’ characteristics of a phenomenon.

·  causal analysis and the construction of ‘unassailable’ analyses.

Moreover, the task of revealing underlying causal mechanisms may be facilitated by a search for:

·  skeletal specifications and the uncovering of common, underlying processes.

Unfortunately, both the ideas and the suggested empirical technique suffer from a variety of individual and shared defects.

Limitations on reliability - The problem of induction

Both analytical induction and causal analysis are necessarily subject to a potentially crippling logical difficulty memorably presented by Popper thus:

It is usual to call an inference ‘inductive’ if it passes from singular statements ..., such as accounts of the results of observations or experiments, to universal statements such as hypotheses or theories.

Now it is far from obvious, from a logical point of view, that we are justified in inferring universal statements from singular ones, no matter how numerous: for any conclusion drawn in this way may always turn out to be false; no matter how many instances of white swans we may have observed, this does not justify the conclusion that all swans are white.

Popper, (1972), p.27

This is the notorious ‘Problem of Induction’ originally formulated by Hume over two hundred years earlier.[iv] More recent treatments of the problem divide it into two parts:

...... the difficulty occasioned by the fact that all our putative knowledge of the natural world is in principle revisable (the little problem of induction)...[and] the difficulty occasioned by the fact that even if we had perfect knowledge of the natural world another world could come into being at any time in which that knowledge would be useless (the big problem of induction)

Harré and Madden, (1975), p.75

For positivists, by casting logical doubt on the reliability of any inductively generated theory, the problem of induction constitutes both a powerful argument for avoiding any inductively-based research technique and support for the hypothetico-deductive search for event-regularities adopted by positivism. However, the positivist research methodology is, of course, not without its own critics. The Realists have convincingly demonstrated that the reliable identification of cause and effect through the use of statistical analysis to uncover patterns in event is only possible in closed systems with stable environments and constant, unchanging processes and actors.[v] Such conditions rarely apply fully even under controlled conditions in chemistry laboratories. In social systems such as companies, where the external business environment is notoriously unstable and the presence of human beings in the internal company environment with their reflexive,[vi] unpredictable behaviour, unequivocally prevents system closure. The reliable identification of cause and effect from event-regularities as proposed by Hume[vii] is thus rendered impossible. Under the kind of conditions existing in purchasing environments, the casual significance of patterns in data is indeterminate.