[T]he modern (forced) separation of the discipline of economics from other social sciences must be recognized as quite misguided. Indeed, this separation merely makes it difficult for economics to advance in pace with other branches of social science... (Lawson, 2003, p. 162)

Introduction

Political economy in general has long been critical of mainstream economics. Yet, of late, the literature has tended to express this critique not so much with respect to specific theoretical or empirical claims per se, though these are clearly important, but in terms of the methodological apparatus of economics. Among many contributions, Hodgson (1988) argues that mainstream economics simply emphasises one method of analysis which expresses theoretical explanation in terms of the achievement of an agent’s objective under effectively full information. This is elaborated upon by highlighting positions of attained equilibrium. As such, ‘economics forgot history’ (Hodgson, 2001). Lawson (1997, 2003) from a critical realist perspective has criticised the deductivism of the mainstream; while Dow (1990, 1996) crticised its monist tradition, and Davidson (1996) its reliance on the ergodic axiom.

In many respects, each of the above critiques compares the mainstream view of eco-nomics with respect to an alternative, specific view of economics described as a school of thought such as Institutionalism or Post Keynesianism.1 In contrast, the work of Tony Lawson is rooted in a more abstract agenda. His book Economics and Reality (1997) provides a critique of economics as social theory, through its critical emphasis upon mainstream method. Lawson’s recent contribution, Reorienting Economics (2003), moreover, has moved this debate further by proposing, as implied by the epigraph above, that the important fault line lies between economics and social science and that the latter can embrace different traditions of heterodox thought in economics once matters of ontology are treated explicitly.

This paper is a contribution to this debate. Here too, rather than focusing upon the attributes of specific economic analysis, it focuses upon research method, and its methodological justification, to argue that exploring the ontological assumptions under-pinning specific enquiry may reveal how disciplinary boundaries could be broken down and thus act as a precursor to interdisciplinary social science, of which economics could be a part. Central to this endeavour, it is argued, is the use of retroduction, the logic of inference espoused by critical realists. Retroduction, discussed more fully later in the paper, can be contrasted to other research strategies such as deduction or induction, as not simply developing specific claims from general premises nor general claims from specific premises, respectively, but the ‘mode of inference in which events are explained by postulating (and identifying) mechanisms which are capable of producing them’ (Sayer, 1992, p. 107). This paper argues that retroduction requires the ‘triangulation’ of research methods. Such triangulation can, under certain assumptions, be argued to unite research contributions in such a way as to transcend the use of specific methods in a disciplinary sense. This follows from drawing a distinction between methods of analysis and research methodology and, in particular, the ontological justification offered for the use of different methods of analysis. A corollary of this argument is that a middle ground is found lying between naïve realism and pluralism of method.

Two important and connected caveats are worth noting at this point, which stem from sharing the more abstract level of discussion of Lawson (1997, 2003). The first is that such a focus does not deny, and indeed expects, that the epistemologies of particular schools of thought may well be consistent with the current discussion and, as such, reveal a presentation of retroduction from an alternative perspective, one grounded in more specific enquiry. This said, there is a potentially conflicting alternative perspective. For example, drawing essentially upon Marxian thought, Fine (2004) argues that critical realism has a trans-historical emphasis and, as such, does not locate itself in relation to key theoretical concepts or historical epochs but, rather, that ontological and methodological concepts are developed independently of context. Fine (2004) can make this case in large part because, echoing Bhaskar (1978), critical realists continually maintain the ontological boldness of the approach, coupled with its epistemological caution and that both ‘science’ and ‘social science’ share common principles in analysis, if not procedures, because of a particular structured ontological perspective (see also Lewis, 2004). Critical realism thus retains a ‘naturalism’ that is not accepted by some writers. However, one either accepts or rejects this argument based upon the extent to which one feels that it is possible to maintain an abstraction that refers to higher levels than those reflecting specific circumstances.

The second caveat is that this paper does not aim to produce a simple ‘recipe’ of research methods for researchers. The specific nature of research questions and programmes will govern the choice of specific methods. However, because the logical structure of explanation is always implicitly or explicitly distilled from an ontological perspective, this suggests constraints upon the use of methods for particular purposes. This issue is primarily discussed in this paper. Necessarily, thus, the discussion of research methods is conducted at a high level of abstraction, which inevitably involves some compromise with the conceptual and operational nuances of particular methods. Naturally, this paper maintains the reasonableness of this approach. This is not just because higher order categorisation is valid but because it is important to frame the broad structure of research projects as consistent with the nature of the material that is being investigated. This importance is, of course, now recognised not just in philosophical discourse, but by the major funding bodies such the Economic and Social Research Council (ESRC) in the UK as a practical issue. As a result, there is a growing list of their funded workshops for experienced researchers as well as postgraduates that address these concerns.¹

An important corollary of both of these caveats, therefore, concerns the methodology– method relationship. It is often maintained that specific methods of analysis presuppose particular methodological presumptions. It is on this basis that Lawson (1997, 2003) assesses economics in a philosophical, and specifically ontological, sense by criticising the ubiquity of formal mathematical and statistical methods in neoclassical economics and arguing for their lack of relevance to an open social system. Other writers specifically concerned with social research methods, and considered later in the paper, likewise assess which methods of analysis are consistent with specific philosophical approaches in social research.

This paper argues that this concern with the ‘internal’ methodological consistency of argument, by requiring methods to reflect a particular ontological structure, does put constraints upon the use of research methods and it is here, of course, that Lawson (1997, 2003) finds unease with the use of formal mathematical and statistical methods. However, to address this unease, this paper argues that precisely because critical realism maintains a different ontological structure from that implied in existing discussions of research methods, it can be shown that specific methods of analysis do not, of necessity, need to be linked to other ontological positions, but that their combination can be employed in analysis. Thus, combining methods in the process of inference does have an ontological— and not merely a practical—basis.

In this regard, this paper shares Olsen and Morgan’s (2005) distinction between methodology and method. They argue that methods are techniques of data collection and transformation, whereas methodologies comprise combinations of methods, the practices involved in implementing them and the interpretation placed on this act by the researcher. This reveals, as echoed by Grix (2002), that methodology is essentially concerned with the logic of enquiry and ‘inparticular with investigating the potentialities and limitations of particular techniques or procedures. The term pertains to the science and study of methods and the assumptions about the ways in which knowledge is produced’ (p. 179).Consequently, this paper focuses specifically upon the alternative nature, practice and interpretation placed upon the combination of different methods of analysis.

To explore the issue of retroduction as ‘mixed-methods triangulation’, which promotes interdisciplinary research, Section 2 introduces and examines alternative concepts of triangulation, and provides a critical commentary on the use of the triangulation metaphor. It argues that mixed-methods triangulation (MMT hereafter) is a more appropriate nomenclature. Furthermore, it is on MMT—as opposed to other types of triangulation—that this paper focuses. Section 3 explores the different uses of MMT in both economics and elsewhere in social science. Sections 4 and 5 then explore the potential philosophical underpinnings to MMT and conclude that, through drawing upon critical realism, MMT can be rendered logically intelligible. Moreover, by drawing upon the concept of contrastive explanation, it is argued that MMT is an operational statement of retroduction. Section 6 closes the discussion with some indications of how adopting this approach could reintegrate economics into (a redefined) social science.

Triangulation

Triangulation has its applied origins in navigation and surveying, whereupon taking measurements from two separate locations one can derive, or predict, a third measurement or location. In social research in its broadest sense, triangulation implies combining together more than one set of insights in an investigation, and there are many early implicit uses.² A useful taxonomy is provided by Denzin (1970), which is presented in Table 1, as forms that are now frequently referred to in the literature, though it should be recognised that this list is not exhaustive, neither are the types implied to be necessarily mutually exclusive.³

In economics generally, the use of triangulation, beyond the weakest form of the interaction of modeller and model, is limited. As Downward and Mearman (2002, p. 410) note, for example:
based on text such as: . . . ‘[e]stimation methods or estimators are a second important tool in our tool kit and . . . [are] . . . necessary but insufficient for solving the model discovery problem’ [Hendry (1995, pp. 16–17) might appear to advocate triangulation] . . . such appeals are made prior to, and in the aid of, purely econometric analysis’.
Econometric analysis remains primary and other methods are auxiliary to it.

Further, Downward and Mearman (2005) note that, in the process of providing advice on monetary policy, the Bank of England’s economists use various forms of evidence. Models are triangulated with people; different economists’ estimates are triangulated with each other (investigator triangulation); different types of data are used (data triangulation); different trials of the same technique and different types of technique (within-method) are combined. This evidence could seem to suggest extensive triangulation, but in fact, again, the analysis is driven by the needs of the large forecasting model, and other techniques are subservient to that. For example, the Bank of England (2004, p. 188) states that: ‘the new

Table 1. A taxonomy of triangulation

1. Data triangulation

Involves gathering data at different times and situations, from different subjects. Surveying relevant stakeholders about the impact of a policy intervention would be an example. An alternative would be address concerns about the inadequacy of available data. Economic forecasters who rely on national accounts for their modelling exercises find that there is a lag between that data and prevailing economic conditions. They often make use of different data sources (and types) to fill this gap. In this case, different types of data might be used; for example, survey data might be used alongside time series data. The Bank of England is one body which employs both this procedure and this rationale (see Britton et al., 1999).

2. Investigator triangulation

Involves using more than one field researcher to collect and analyse the data relevant to a specific research object. Asking scientific experimenters to attempt to replicate each other’s work is another example.

3. Theoretical triangulation

Involves making explicit references to more than one theoretical tradition to analyse data. This is intrinsically a method that allows for different disciplinary perspectives upon an issue. This could also be called pluralist or multi-disciplinary triangulation.a

4. Methodological triangulation

Involves the combination of different research methods. For Denzin, there are two forms of methodological triangulation. Within method triangulation involves making use of different varieties of the same method. Thus, in economics, making use of alternative econometric estimators would be an example. Between method triangulation involves making use of different methods, such as ‘quantitative’ and ‘qualitative’ methods in combination. It is here that the most interesting issues arise as discussed below in detail. MMT here is the a priori commitment to inference from between-method triangulation.

a: As discussed below, a key argument of this paper is that such pluralism, and that implied by other forms of triangulation, can be underpinned by a coherent ontological or epistemological position. Source: Downward and Mearman (2004A).

Bank of England quarterly model is . . . the main tool in the suite of models used by its staff and the [Monetary Policy Committee] in its deliberations’. Moreover, excluding the practical concerns of policy economists, there is little evidence of triangulation. It also seems to be the case that academic economists use triangulation least of all. Examples tend to exist outside mainstream, neoclassical research. Thus, Downward (1999) is an exception that explores pricing from a Post Keynesian perspective. Olsen (2003) and Olsen et al. (2003) make use of triangulation (specifically MMT) to explore gender issues in the context of development economics, while Olsen (2004) indicates its use in feminist economics more generally.

Yet, in the social sciences, and implied by Denzin’s taxonomy, triangulation is much more widespread. For example, Danermark et al. (2002, p. 152) claim that within the sociological community the view is widely supported that there is no universal method and that there is a need for multi-methodological approaches. Thus, in the applied social sciences, MMT is common in nursing, health and education, and tourism (see, for example, Shih, 1998; Hirst, 1993, Downward and Mearman, 2004B).

As Downward and Mearman (2004B) note, there are two main arguments put forward to justify triangulation. The first, put in an early explicit form by Webb et al. (1966), is that triangulation increases the ‘persuasiveness’ of evidence. Many early studies, however, tended to emphasise concerns to enhance the empirical reliability of quantitative measures through triangulation (Campbell and Fiske, 1959). Currently, however, the emphasis is more general and shows concerns for enhancing the ‘validity’ of insights, or for adding ‘completeness’ to accounts (Shih, 1998). Thus, triangulation elaborates understanding (Jick, 1979) or confirms the accuracy of data (Denzin, 1989). More explicitly, Danermark et al. (2002, p. 153) note the uses of quantitative analysis to ‘test’ the validity of qualitative insights, or to use qualitative work as preparation for quantitative work, and to elucidate a phenomenon in as much detail as possible.^4>

The second argument for triangulation, for example, put forward by Cresswell (1995) and Tashakkori and Teddlie (1998), is that one should combine methods on typically pragmatic grounds. It is in this context that triangulation occurs in economics, for example, in the forms of triangulation by economic policy forecasters, as discussed above (such as the Bank of England), who adopt multiple methods, perspectives (to some extent) and data as a response to poor past predictive performance (see Whitley, 1997; Pagan, 2003). Such pragmaticism extends to Bryman (2004), who argues that methods can be mixed, but that one method will always assume a primacy over the others.

These justifications raise three important and related issues that require further consideration. To begin with, the appropriateness of the label or metaphor of triangulation is questionable. Blaikie (1991) argues that triangulation as a navigational, surveying technique presupposes that measurement ‘is of the same kind and is based on a common ontology and epistemology’ (p. 118). Subject to a given degree of error, triangulation simply helps to define a different location. There is no conception of cross-checking observations to produce a more accurate picture, or balancing alternative points of view that may be motivated by the examination of apparently different phenomena.

Second, the pragmatic motive for ‘triangulation’ in social science research, or lack of concern for the appropriateness of the analogy, can thus be viewed as an instrumentalist (methodological) position.⁵ This suggests the necessity of exploring the legitimacy of triangulation on methodological grounds. Finally, one should note that the most controversial element of ‘triangulation’ is ‘methodological triangulation’, specifically ‘between-method triangulation’. It is here that the legitimacy of combining insights produced from different methods of analysis needing to be explored in methodological terms is most obvious. As implied earlier, it can be argued that research methods have a distinctive ontological and epistemological context. These issues are now discussed in more detail.

Methodology and triangulation

In the above views, or definitions of triangulation, is an implicit argument maintaining the complementarity of the data or investigations undertaken. This is, however, contentious, and this is made most clear in considering the fourth form of triangulation in which Denzin (1970) distinguishes between within-method and between-method triangulation. As implied above, the former involves using varieties of essentially the same method to investigate an issue, such as in the three cases above. The latter implies combining data generated by different methods and, in particular, quantitative and qualitative methods. In as much as these methods can be argued to presuppose different ontological assumptions, it is here that potential clashes arise. The literature has recognised methodological clashes, though the ontological status of the clash tends to be ignored. It is upon the context of triangulation as MMT that the rest of the paper focuses.

Olsen (2004) explores MMT in social science in some detail by reviewing social science research texts. In Olsen’s terms, the traditional social science perspective is that there is an ‘epistemological chasm’ between quantitative and qualitative research methods. This argument draws upon Walby (2001), who argues that these chasms have a disciplinary basis but are hard to justify philosophically. In this paper, we argue that the epistemological chasm can be viewed as explicable and consistent with specific ontological viewpoints; hence an examination of ontology is essential to explore the basis upon which methods can be mixed and some movement toward interdisciplinarity established.

For example, from one perspective, Silverman (1993) argues that a requirement of social research is that it employs qualitative methods. These methods reflect an ‘interactionist’ epistemology⁶ and, say, an interviewer creating the interview context and the interviewee engaging in a dialectic with the definition of the situation. In this respect, research reflects social relationships which are inherently subjective and not objective.

Interactionism, as an epistemology, has been much less influential in economics than in other social sciences. In the latter literature, it embraces a wide range of methods and methodological positions. Content analysis, discourse analysis, grounded theory, ethnography, postmodernism, post-structuralism, hermeneutics and phenomenology are examples of the epistemological variants.⁷ But, in general, interactionism recognises hermeneutic concerns that social phenomena are intrinsically meaningful; that meanings must be understood; and that the interpretation of an object or event is affected by its context. As Sayer (1992, p. 36) notes, ‘there is an interpenetration and engagement of the “frames of meaning” of the reader and the text [text here meaning anything which can be understood]. We cannot approach the text with an empty mind in the hope of understanding it in an unmediated fashion.’ Furthermore, such verstehen of objects is universal (Sayer, 1992, p. 37). Thus, theory- and value-neutral observations are impossible. Consequently, Sayer (1992, p. 35; 2000, p. 17) argues, for the above reasons, that meanings cannot be measured, counted or understood. Unsurprisingly, therefore, interactionist approaches tend to focus on the limitations of quantitative analysis in the social arena. On the basis of this dual, Silverman rejects quantitative methods as inappropriate to social research.

Thus, there appears to be limited legitimacy for MMT under this perspective. Admittedly, different observations could be compared, but commensurability is necessarily incomplete, and it is not clear how the insights would be combined. But what is the logical basis of this invalidity? The key here is to recognise that interactionism as defined above embraces a specific ontological position. This is one in which the object of analysis is not meaningfully distinct from the subject of analysis but is inevitably context specific. Consequently, where, say, case studies are used, the uniqueness of the case and of the context of the investigation render comparisons difficult. Likewise, any attempts to generalise (let alone universalise) from a particular case would be considered fraught with conceptual problems. Thus, in such cases where attempts have been made to combine qualitative with quantitative insights by exploring the frequency of some qualitative phenomenon, under a strict interactionist view such a procedure is clearly problematic. Moreover, the scope for MMT is almost nil: qualitative analysis could not be combined with quantitative analysis if the latter was considered invalid.

In contrast, positivism also remains influential in social science, despite having a long and tangled history and being difficult to define (see Section 4). Positivism stands in contrast to interactionism in maintaining that valid objects are always observable and measurable (Sayer, 2000) and essentially exist in a value-free sense, and in which as the subject of research they are not completely defined by the context. This ontological distinction implies, consequently, that an epistemology emphasising quantitative methods is recommended, and qualitative data are viewed as questionable because they can allow values to enter the analysis as investigators bring their own theoretical concepts or standards (of various kinds) into observation. Unlike interactionism, however, and with the exception of between method triangulation, there is scope for triangulation under this positivist perspective, particularly where different quantitative methods are to be used (triangulation of method) and for different quantitative measures to be combined (data triangulation).

Further, while authors such as Frankfort-Nachmias and Nachmias (2000) embrace such a positivist perspective, they acknowledge that, where quantitative analysis, which remains the preferred method of analysis, is impossible, qualitative analysis can be used. This is suggestive of methodological triangulation being advocated on practical grounds. However, Frankfort-Nachmias and Nachmias (2000) argue that, even with qualitative work, value-free hypotheses should be tested. Therefore, it is argued that triangulation is possible to allow ‘value-free’ concepts to be explored by different methods—in the manner of data triangulation discussed above. Moreover, where qualitative data is used, the instinct for the positivist researcher is to quantify it, for example through coding.

Frankfort-Nachmias and Nachmias also discuss the notion of intersubjectivity (p. 15). Intersubjectivity is clearly linked to investigator triangulation defined above. However, again, intersubjectivity is treated sceptically: all investigators must have the same value systems and standards for interpretation. If investigators have different perspectives when observing, this reduces the ability to triangulate their insights. In this case, thus, the desire for triangulation is driven by concerns for the validity of data, based on replication. In the same way Frankfort-Nachmias and Nachmias also argue for triangulation in the validation of sets of conclusions by confirmation based on the conclusions of other investigators. In this regard, positivist approaches acknowledge that attempts to make triangulation between methods legitimate requires reference to a positivist ontology and epistemology.

This discussion makes clear the main methodological issues pertaining to triangulation. First, those advocating either a singly positivist or interactionist method implicitly reject triangulation, and the legitimacy of this case essentially rests upon ontological grounds. There is, in essence, no issue to debate. Consistency of argument rules out the possibilities. Second, if the same sort of data are triangulated, then this could be legitimate from a positivist perspective. Once again, there is no particular methodological issue to debate.⁸ Third, then, it is in the context of combining methods that methodological contention arises. If one accepts that methods are tied to a distinct ontology, as discussed above, then combining a ‘positivist’ and ‘interactionist’ approach is simply not logically tenable, given their different ontological presuppositions, and this, of course, undermines the pragmatic use of MMT discussed in the previous section. To provide legitimacy for MMT, therefore, clearly requires a different set of ontological presuppositions.

It is argued in the remainder of this paper that critical realism can provide a basis for rethinking MMT. More importantly, the logic of retroduction can provide this legitimacy. In so doing this provides an opportunity to express how economics can be reoriented into social science, now defined in critical realist terms.

A critical realist critique

It is clear from the above discussion that a methodological justification for combining methods, that is, to explore the complementarity of multiple research findings, requires an explicit analysis of the ontological bases of various logics of inference. This section explores this issue by reconsidering both interactionism and positivism. This proceeds by exploring, first, the essential links between inductivism, positivism and deductivism and, second, the deductivist/positivist–interactionist dual.

The previous section argued that the basic tenet of positivism is to envisage explanation in which value-free observation of objective reality is the key. Induction, as a research strategy, is central to this view. There is debate about the origins and main tenets of induction, but Harré(1986) argues that three main principles describe induction. These are that knowledge grows through the accumulation of ‘well-attested’ facts; laws can be inferred from the accumulation of these facts; and that our degree of belief in such laws grows proportionally with the number of confirmed instances of phenomena. As Blaikie (1993) argues, thus, ‘[t]he inductive strategy embodies the realist ontology which assumes that there is a reality “out there” with regularities that can be described and explained, and it adopts the epistemological principle that the task of observing this reality is essentially unproblematic . . . that there is a correspondence between sensory experiences and the objects of those experiences . . . ’ (p. 137).

As a logic of inference, induction is central to accounts of positivism, as implied in the previous section. However, positivism also has a series of variants and categorisations. For example, Halfpenny (1982) discusses up to 12 versions in sociology. Walters and Young (2001) also discuss the fluidity of the term and its content in economics. In general, Blaikie (1993) summarises these accounts as embracing the perspective that experience is the only reliable source of knowledge and should form the basis of abstract conceptual development. This is as opposed to drawing upon values and normative propositions which cannot contribute to knowledge. Consequently, experience is of independent atomic events which can be developed as law-like statements that subsume specific cases. Outside the literature on triangulation, the induction problem, coupled with the inability to purge value from observations, has been the centre of widespread philosophical criticism of theapproach. The induction problem implies that for successive observations upon phenom¬ena to form a reliable basis of causal explanations between them requires assuming ‘the uniformity of nature’ or that the patterns will, of necessity, repeat themselves. In other words, the explanation is assumed rather than demonstrated (Blaug, 1980). The concept of value-free observation relies on the assumption that judgements of value have no empirical content and thus are inconsistent with tests of them through experience (Giddens, 1974), yet it ignores the fact that decision criteria themselves are judgements (Rudner, 1953), let alone that concepts themselves are theoretically laden (Schutz, 1963).

It is often argued that the development of Popper’s falsifiability criterion bypassed these problems (Popper, 1959) because the logical demonstration of falsehood of value-driven hypotheses with reference to a particular set of observations circumvents the problem of induction (Bunge, 1996).⁹ Thus, Popper’s work, among other contributions, has been seen to be one of the underpinning planks of the hypothetico-deductive approach that now populates research methods texts (Ryan, 1975; Blaug, 1980).¹⁰ Crucially, it is here that positivism and deductive logic can become enmeshed in much of the practice of social scientific research. Either informally or formally, through statements of initial conditions and assumptions, deduced consequences or predictions are assessed empirically. It should be noted that Popper did draw distinctions between the logic of falsification and the underlying ‘psychology’ with which the theoretical propositions to be tested emerged. Thus ‘[t]he initial stage, the act of conceiving or inventing a theory, seems to me neither to call for logical analysis nor be susceptible to it’ (Popper, 1959, p. 31). However, Popper (1972) argues clearly that ‘[t]he role of logical argument, of deductive logical reasoning, remains all important for the critical approach; not because it allows us to prove our theories, or to infer them from observation statements, but because only by pure deductive reasoning is it possible for us to discover what our theories imply, and thus to criticise them effectively’ (p. 51). Importantly, in this regard, Popper retains an emphasis upon nature comprising uniformities, and the process of theorising as the imposition of regularities that are then critically compared with nature (Blaikie, 1991).¹¹ Yet, and as clearly identified by Popper, deduction as a form of argument does not require references to empirical categories. Deduction is simply the process of establishing the logically correct conclusions from the components of an argument. Therefore, there is a difference between pure deduction, the hypothetico-deductive approach and the inductive reasoning associated with positivism as typically defined. This is because, in the latter cases, quantitative evidence acts as an arbiter.

However, there is a factor common to the variants of explanations just discussed and which characterise their essential logic: explanations, derived from their respective methods of investigation, are presented in the form of ‘covering laws’, that is relationships between variables that transcend space or time. Lawson (1997, 2003) and Sayer, (2000) describe the approach as ‘Humean’, because causality is associated with the succession of events, as ‘correlations of a causal-sequence sort’ (Lawson, 2003, p.25).¹² Ontologically speaking, a closed system is assumed such that causes act in a consistent manner (the ‘Intrinsic Condition of Closure’ (ICC)) isolated from other causes (the Extrinsic Condition of Closure (ECC)). In such circumstances, events, our empirical description of them, and the causes of the events are conflated. Critical realists would describe such approaches as naïve, simple or empirical realism.

From the perspective of critical realism, it can be argued that all these approaches demonstrate an ‘epistemic fallacy’—that they conflate the subject and object of analysis through the invocation of covering laws. The conception or knowledge of phenomena manifest in the theorist’s ideas and arguments is treated as logically equivalent to the phenomena under review. There is a further dimension to this fallacy rooted in the inferential apparatus both implied in, and entailed by, a closed system. This is that premises fully entail conclusions. Lawson (1997, 2003) describes this as deductivism, thus generalising the concept of deductive reasoning to be the organising principle of any arguments that invoke covering laws, whether they are presented as part of a specifically deductive, inductive-positivist, or hypothetico-deductive view. It is because deductive reasoning is directly concerned with, and thus can only cope with, knowledge that already exists or has been acquired that it promotes the epistemic conflation.

Yet, as also discussed above, rival philosophical positions to these do exist that reject the independence of thought and reality said to characterise variants of realism, and emphasise, in contrast, a dialectic of subjectivities. Here too, the epistemic fallacy is evident. The conception or knowledge of phenomena manifest in the theorist’s ideas and arguments is treated as logically equivalent to the phenomena under review.

Finally, the same fallacy is present in instrumentalism. As Lawson writes,
The problem . . . is that it is effectively indistinguishable from the view that knowledge is merely a creation of the mind and nothing else exists. Indeed in his critical study of Locke this is precisely the conclusion that Berkeley draws. If there is no claimed necessary connection between our ‘ideas’ and external reality . . . what is there to support the view that any external reality exists? (Lawson, 1988, p. 54)

From a critical realist perspective, an effective research method thus needs to overcome this fallacy. It is subsequently argued below that this can be achieved by linking the critical realist ontological perspective and the logic of retroduction to MMT.

Critical realism and triangulation

Critical realists (Lawson, 1997, 2003; Sayer, 2000) reject the conception of society and the economy as closed systems, arguing instead that reality is a structured open system in which the real, the actual and the empirical domains are organically related. The real refers to the intransitive dimensions of knowledge, which exist independently of our

understanding of the world, and in which actual structures and causal powers reside. The actual domain refers to what actually happens if causal powers are activated. Causes act transfactually, but because society is open causes, though operating consistently, may not reveal themselves in empirical regularities because of countervailing influences.¹³ Thus, in the empirical realm, the real and actual are observed and experienced. Also, therefore, the empirical level is the access point to the transitive dimension of knowledge, albeit filtered through the hermeneutic process.¹⁴ Thus, knowledge is expressed and informed by subjectivity. In general, therefore, critical realism embraces the fallibility of knowledge and cautions against a ‘complacent’ link being made between reality and our knowledge of it. Yet, it is in directly facing up to this ontologically defined fallibility, that concepts of validity through triangulation can be both rooted, and understood, as discussed later. At this juncture, though, it is clear that, under these ontological circumstances, research methods that embrace the logic of inference described generally as deductivism will be faulty. Adequate explanation will require ontic depth, that is, moving beyond the immediately postulated level of events and/or texts. Retroduction is advocated in this regard.

As Danermark et al. (2002) argue, retroduction is not so much a formalised logic of inference as a thought operation that moves between knowledge of one thing to another, for example, from empirical phenomena expressed as events to their causes. The key is that the researcher moves beyond a specific ontic context to another, hence generating an explanation that embraces ontological depth. The process of abduction, whereby specific phenomena are recontextualised as more general phenomena, can be a part of this process. Downward and Mearman (2002) thus argue that Gardiner Means’s theory of administered prices could be understood as research in this sense. The (macro) statistical finding that prices adjusted differently in both magnitude and frequency to changes in demand in Bureau of Labor statistics for different industries was explained with reference to case-study work on the differences in the administration of prices in firms of different sizes.¹⁵

This ontological position has been linked by critical realists to MMT but with large degrees of scepticism about the role of quantitative methods. Sayer (1992, 2000) argues that critical realism is compatible with a wide range of methods, with the key issue being that analysis should be matched to the appropriate level of abstraction and the material under investigation. Table 3, reproduced with some abbreviation from Sayer (2000), thus distinguishes between intensive and extensive research designs. The former is what is typically thought of as social science, i.e., qualitative research. It begins with the unit of analysis and explores its contextual relations, as opposed to emphasising the ‘formal relations of similarity’ between them, that is, producing taxonomic descriptions of variables, as is the case in the latter, i.e., quantitative design. Sayer (2000) does argue that the approaches have complementary strengths and weaknesses, though the causal insights from extensive research will be less. Moreover, one is reminded that the validity of the (qualitative) analysis of cases does not rely upon broad quantitative evidence. In this sense, the traditional view put forward for MMT as validating qualitative insights is not applicable.

Likewise, Danermark et al. (2002) make a case for ‘critical methodological pluralism’ and argue that the methodological bases of triangulating methods is not explored. They share Sayer’s perspective ‘that quantitative analysis can be fruitful . . . and are undoubtedly valuable—provided that their field of application is confined to what is suitable . . . The limitations are revealed mainly when it comes to explanatory ambitions’ (Danermark et al., 2002, p. 174).

Table 4, constructed from their text, nonetheless summarises a position in which four possibilities of MMT are examined, two of which are rejected because of the epistemic fallacy, and two of which are argued to be consistent with critical realism. These include the cases in which ideas generated by quantitative analysis are explored for causal links by qualitative work and where both are combined in the context of theory development.

They, too, reject the typical idea in triangulation that qualitative insights are validated by quantitative analysis, and that statistical generalisation can be the driving inferential claim. As with Sayer, the key to these arguments is the view that quantitative analysis is essentially unable to yield causal narratives and does not of right add validity to analysis.¹⁶

It seems, however, that there are problems with this characterisation. In particular, it contains the implicit notion that the methods still ultimately focus on a strictly separate domain (i.e., reflect a distinct ontological basis), and this underpins the duality of research design used in the discussions above. In this respect, the ontological clash implied in pragmatic combinations of methods is not fully resolved.

In contrast, in a number of papers Downward and Mearman address these issues and argue that combining methods is central to retroductive activity.¹⁷ The following discussion briefly restates these arguments. To begin with, it can be argued that the specific research methods within intensive and extensive designs differ more in emphasis than in kind through invoking degrees of closure (Downward, 1999, 2000; Downward et al., 2002;

Downward and Mearman, 2003). One way of viewing this is in terms of the differences, say, in measurement scales used to record or describe phenomena. Based on some interview data, responses may be coded differently, with the differences between the choice of measurement scales essentially hinging upon assumptions being made about the ability to reflect the direction and magnitude of differences in responses, and the manner in which responses can be combined, and compared across subjects, calculating averages, say, from ratio data as opposed to counting frequencies of nominal data. Yet, logically, ratio data can always be redefined as nominal data. More generally, moreover, if, in contrast, ‘qualitative’ methods seek to collate insights and offer stylised interpretations and narratives as text, they too assume qualitative invariance of meaning—or intrinsic closure. Note here that, despite beginning with specific contexts, unless we assume that everything is fundamentally unique, some form of generalisation is offered, if not based in statistical induction. While critical realism maintains that such generalisation can refer to the essential qualities of phenomena which are actually rooted in an open system, it seems clear that the process of referring to these essential qualities implies the assumption of intrinsic closure.¹⁸ This is not to say that, as Winch (1971) discusses, one is simply viewing differences in degree in the sense of their being due to more complex measurement per se. There are clearly conceptual differences, too, linked to the purpose to which the research is put. But, even in this regard, both Olsen and Morgan (2005) and Byrne (2003) note that ‘quantitative’ research as with all measurement shares an essentially hermeneutic process. What is at stake, thus, is more a shift of emphasis and ability, say in an interpretive context, to explore causes through references to agency, change, emphasis and action than simply more mechanically combined observations that might be better captured as events or outcomes in more ‘quantitative’ form.¹⁹ Moreover, it seems clear that, within these broad guidelines, the

level of abstraction required for the analysis ultimately determines which methods are used as, say, retroduction proceeds. The point is that methods are merely redescriptive devices revealing different aspects of objects of analysis. In this respect, particular research methods are not necessarily wedded to particular, and different, ontological presumptions.²⁰

Indeed, different methods can be seen to be necessary to reveal different aspects of the constituency of phenomena, that is their ontic character, as structural, that is cause and effect, relations more broadly. Thus, the concept of cause in critical realism is tied to emergence from the interaction of human agency and institutions or structures. In this regard, the motivational (or otherwise) dimension of agency needs to be elaborated, as well as the mechanisms that facilitate action, or behaviour, coupled with the relational context of that behaviour. Each of these components clearly requires different methods of analysis to reveal their nature and action.²¹

As Downward and Mearman (2004A) argue, the logical and indeed explanatory coherence of this approach can be understood in linguistic terms. This carries with it the (linguistic) notion that a ‘question and answer’ theoretical structure is preferable to the view that embraces a deductive/inductive emphasis. Given any object of analysis, a question can be asked about it. Any answer that is offered logically links the object to the answer (Thagard, 1992; Sintonen, 1989). In this respect, it follows that, say, data reveal some patterns of general sets of occurrences, interpretive research can be used to pursue the reasons why it is so in this case and not in that. As well as the obvious statistical calibration of the observations, it is clear that the processes associated with real behaviour that produce the patterns can be explored. Of course, this is something espoused, say, by Danermark et al. (2002). The point here is that there is no need to view the techniques of analysis as, somehow, separable and that seeking generalisation through empirical analysis, of necessity, is rooted in consideration of a different object either in terms of implicitly linking a positivist ontology to empirical insights, or arguing that empirical methods merely refer to formal relations of similarity. In contrast, a nexus of mutually supportive explained propositions can be constructed in which the whole stands distinct from its parts.²² Therefore, these mutually supported propositions are where MMT adds ‘validity’. This is not in the sense of reverting to the primacy of either method and, by implication, a subjectivist or objectivist orientation, but in the sense of deriving better or fuller explanations through abstraction. The notion of explanatory power is, of course, relative, and reflects the proportion of questions which are left unanswered in any particular enquiry.²³

As Risjord et al. (2001) argue, in this context, the choice of method is not paradigmatic or one of ontology, but reflects the specifics of the question being asked. It is true that their concern is for linguistic consistency. It remains, however, that if the questions probe different features of a phenomenon, different methods might be needed, while focusing on the same phenomenon and simply stressing different aspects of abstraction. It seems clear, thus, that the logic of retroduction makes some form of MMT not only possible but also necessary to reveal different features of the same layered reality without the presumption of being exhaustive. Figure 1 presents the conceptual relationships involved. Broadly speaking, on the left-hand side lies the intransitive domain of structured reality in which real causes trigger real events: the logic of inference required is retroduction of the causes. Likewise, on the right-hand side lies the transitive domain in which knowledge of the intransitive domain is obtained in an epistemologically relative context (see Bhaskar, 1979, ch. 2).

Knowledge is derived from various forms of enquiry, for example, that share the aspiration of exploring patterns of events and their causes through MMT. These forms of enquiry, significantly, correspond to aspects of our empirical understanding of the world as ‘empirical’ but here, defined more generally in keeping with Olsen and Morgan (2005) and Byrne (2003) as reflecting a ‘quantitative hermeneutic’ and a conjoining of insights about aspects of the object under investigation beyond simple empiricism that relies on reference to empirical regularities of one sort or another. The dotted vertical line in the diagram, however, indicates the imperfect and partial correspondence of our knowledge with reality under scrutiny. Further, the ‘empirical’ can be said to bridge the intransitive and transitive domains, because it constitutes the experiences of agents, which are in reality, but are also the point of access to the actual events and causal mechanisms of reality from the transitive domain. Thus, the ‘empirical’ is drawn as straddling the line between the intransitive and transitive domains.

It is true, however, that the degree of closure invoked in a move towards increased generality increases and, of course, becomes most profound as one moves towards empirical magnitudes and subsequently from descriptive statistical analysis to that which

draws upon statistical testing.²⁴ Yet, at this point, it should be remembered that, regardless of ontological presuppositions, the internal logic of any quantitative analysis rests upon the robustness of the qualitative invariance invoked in causal mechanisms. The same will apply to stylised interpretive analysis. In this respect, the discovery and robustness of such causal claims emanating from the latter methods will, along with their implications expressed perhaps in terms of quantitative analysis, of necessity, always be open to revision. In this respect, probabilistic inferences are also conditional upon, and should be assessed in connection with, analysis of the nature of the object under investigation.²⁵

The above perspective, it is argued, is consistent with Lawson’s (2003) conception of explanatory method. Moreover, it goes some way towards reconciling the necessary closures that are invoked in the transitive domain, as epistemological expressions, and the nature of reality as an open system. These issues are now discussed.

Lawson (2003) emphasises the contrastive nature of adequate explanation and, in particular, that the experimental method of the natural sciences is a special case as a closed system. Here causal factors are isolated, uniform and stable and are represented as the ‘treatment’ that brings about changes in events. Explanation then resides in linking the contrasted states of events both before and after the application of the ‘treatment’, to the treatment as a causal mechanism.²⁶ In an open system, as discussed earlier, the conditions of closure do not in general apply, but partial and potentially unstable closures can occur producing demi-regularities, that is, partial patterns of events, which become unstable or ‘contrastive’ as the system changes (see also Downward et al., 2002; Downward and Mearman 2002; Mearman, 2002, 2004).

In this context, contrast explanation can clearly be of two types. The first could explore the basis of the partial stability, asking, perhaps less interesting, questions such as ‘Why X?’ with only an implicit alternative ‘and not Y’?. Under such circumstances, one has to establish the demi-regularity, and then the causal mechanisms that produce it. Retroduction as described above will meet this objective with quantitative analysis of data patterns and qualitative investigations of the agencies and structures that produce the behaviour. As Sayer (2000) notes, for such routinised practices ‘ordinary’ questions in the form of revealing behaviour through verbs might apply. More generally, of course, the issue is that research methods need to explore the ‘ontic’ character of the agency and structures involved.

As Lawson (2003) argues, more interesting questions are asked in social science of the explicit form ‘Why X rather than Y?’ Here surprising and previously unexpected changes occur which require understanding. Partial closures break down illustrative of the need to explain how this happened and with the possibility that, through contrast, the intervening mechanism(s) is (are) revealed. How then to proceed? Clearly, an open system precludes experimentation. Why not then simply compare observations? This could occur, say, according to a variety of well-documented forms. Factor analysis, cluster analysis, qualitative comparative analysis, grounded theory or case studies could provide vehicles or methods that, to varying degrees, explore the variety of forms of combination of phenomena.

In the former case, the variability of a data matrix is reduced through combining variables. There is a sense, thus, in which one moves away from the traditional thrust of econometric analysis where a ‘single’ dependent variable is identified. In cluster analysis, the reduction takes place around the data ‘cases’ as opposed to the variables. One could argue that here there is a further move away from the traditional analysis of econometrics to an exploration of the subjects as cases as opposed to the outcomes of behaviour expressed as variables. Likewise, qualitative comparative analysis (see, for example, Ragin, 1987), involves either exploring the construction of phenomena, or seeking to demonstrate ‘cause’, through the use of Boolean logic applied to a redescription of phenomena in non-parametric terms.²⁷ The objective is to reveal the different combinations of categories that are associated with cases in the data.

Regardless of the details of these approaches, however, what matters for the current discussion is the question that, while ‘contrastive’ in different ways, can these methods explore why different events are observed in an open system? This is clearly the intention of Ragin (1987). Yet, the logic of the approaches is that they, when viewed as methods in isolation, only operate on one level. Trying to establish cause from the derived comparisons would thus be to fall back into an analysis of ‘correlations of a causal sequence sort’ rather than emergence. As such, one will always be left with the question, ‘but what explains these contrasts?’ and so on, to which clearly the same data cannot provide an adequate, albeit always partial, answer. To assume so, would be to embrace deductivism and the implicit assertion that the current data contain the full set of conditions upon which fully adequate conclusions can be drawn. Clearly, there is a need to augment this analysis.

It would be easy to argue that these methods of analysis, though not traditionally employed by economists are, nonetheless, essentially quantitative and thus unnecessary to social science. Yet the same logical problem applies to any method that seeks contrasts based on information that operates at the same (ontic) level. Thus, simply to contrast two interviews with individuals as a means of explaining differences in outcomes, by focusing upon a methodologically individualist comparison ignores the broader institutional context upon which the opinions of the interviewees, etc. were formed. In other words, key aspects of the ‘discourse’ are missing. This discourse may have both quantitative and qualitative characters, for example, the recurring financial losses of one of the interviewee’s organisations or the fact that one of the organisations has been recently taken over, leaving the interviewee with a new immediate superior. Clearly, too, the analysis needs augmentation.

Now, standing in contrast to these approaches, Grounded Theory (Glaser and Strauss, 1967) and case study methods (Yin, 1994) can actively embrace an epistemology of combining quantitative and qualitative methods in either an implicitly or explicitly con-trastive setting. Indeed, Lee (2002) argues specifically that Grounded Theory should form the epistemology of critical realism, as it is a process for theory development through data. While this is clearly a possibility, one might also argue that, because it is an ‘exhaustive’ search for stable conceptual categories, it can be a process of abduction rather than retroduction, though, clearly, if different methods of analysis underpin the different data sources, one can argue that it, too, captures the logic of retroduction. Yet, in this regard, there is no particular and explicit ontological constraint emphasised in the general literature that data of different levels should be combined. In this regard, Danermark et al. (2002) reject grounded theory as an inductive exercise. Yin (1994) shares this view, too, and contrasts grounded theory with case-study methods that ‘need’ prior theoretical content whose validity is established by analytic as opposed to statistical generalisation. There are echoes of the hypothetico-deductive model within this latter approach, however, though, unlike economics, analytic generalisation can proceed through outlining and comparing phenomena in quantitative and/or qualitative terms. It remains the case that there is no necessary perceived need for triangulation of methods. In themselves, thus, we would argue that these approaches require framing within more explicit ontological referents. To this end, we would argue that embracing contrastive explanation in critical realism requires embracing the process of retroduction as an organising epistemology. It is this that makes the ontological constraints upon research design clear and yet does not place an a priori constraint on the specific methods employed, nor their extent. This simply requires due care being paid according to the level of abstraction required.

Reordering economics into social science

The above discussion carries with it some implications for the way in which economics could be reoriented into the social sciences and, by implication, it reasserts or redefines social science. It is quite clear, for example, from examination of the classic works of Friedman (1953) or Blaug (1980) that scientific status is sought by economics, broadly defined to involve systematic explanations that are shaped by empirical evidence. The discussion in Section 4 indicated, of course, the specific conditions under which this can be perceived to take place. Nonetheless, there is a real sense in which economics is typically perceived to be closer to the ‘hard’ sciences than other social sciences because of the axiomatisation of the discipline (Hausman, 1998). In Lawson (1997, 2003), this is made most clear by detailed elaboration of the explicit and formal emphasis upon deductivism and, in particular, mathematical and statistical analysis motivated by models of atomistic rational choice. These features imply that, as a discipline, for example, characterised by Hirst (1993) or Toulmin (1972) as involving unique concepts, methods and aims, economics stands apart from other branches of social enquiry and, indeed, disciplines. The discussion above, moreover exemplifies this by outlining both the interactionist nature of social science, as well as arguments that social science should involve MMT.

This means that MMTas a research strategy raises interesting questions concerning the nature of social science. If one mixes methods of research and, in so doing, attempts to bring specific disciplinary tools to the analysis, such a ‘multi-disciplinary’ approach will entail the ontological clashes discussed earlier because, by construction, the different disciplines embrace different methods as a result of different ontologies as expressed by traditional philosophy of science. Put bluntly, there is no logical basis for attempting to merge a neoclassical optimising model with a hermeneutically constructed account of behaviour per se. While this might seem to be an obvious and even erroneous juxtaposition, the same logical principle would apply to attempts to unite insights that appear to refer to similar objects but which fundamentally differ in character. Thus, a transaction cost interpretation of the firm making or buying components that refers to more descriptively based evidence, yet maintains at its core a methodologically individualist and rational choice, that is, deductive approach, likewise cannot be linked consistently at the ontological level to a hermeneutic investigation of purchasing behaviour.

In contrast, to begin to ‘unite’ social science, one needs to attempt to transcend the separate disciplines to produce an ‘inter-disciplinary approach’. Social science, so defined, would naturally involve MMT, because the methods qua disciplinary boundaries are removed. It is argued in this paper that critical realism provides the methodological apparatus within which such a view of social science could be constructed. Aspects of the subject matter of the disciplines, if not the currently expressed nature of the disciplines, thus become branches or fields of the same domain of investigation brought together by MMT as retroduction. Lawson (2003), thus, indicates how substantive features of Post Keynesian, Institutionalist and Feminist economics could combine. This paper argues that MMT, as a manifestation of retroduction, would help to define the logical basis of research conducted under this approach.

Conclusion

This paper has addressed the issue of triangulation in social science research and argued that, viewed as MMT, it can be rendered a logically consistent approach through the lens of critical realism and as the manifestation of retroduction. As such, it can provide the basis upon which different insights upon the same phenomenon can be sensibly combined and thus has the potential to unite aspects of different traditions of economic and social thought. Indeed, it supports Lawson’s view that the exclusive insistence on mathematical and statistical modelling in economics is misguided. Rather than focusing upon the specific attributes of economic analysis, however, this paper justifies these claims by exploring the ontological assumptions underpinning social enquiry and thus reveals how disciplinary boundaries may be broken down and interdisciplinary social science, of which economics can be a part, established. Central to this endeavour, it is argued, is the use of MMT to unite contributions in such a way as to transcend the use of specific methods in a disciplinary sense.

References

1. For example, the ESRC has recently funded research seminars on ‘Making Realism Work’, ‘Focussing on the case in quantitative and qualitative research’ and training development workshops for postgraduate students on matching methods to social material.
2. In a sense, we all triangulate in making decisions, by combining arguments and evidence from a variety of sources. Likewise, the ‘Greats’ in Political Economy draw upon different evidential bases and arguments. So, too, peer review of academic research is a form of triangulation. What matters, then, for an analytical approach to triangulation, and which forms the basis for this paper, is an explicit examination of the logic within which such sources are drawn together in reaching a conclusion.
3. One could add to this list ‘hypothetical’ triangulation which in essence is what takes place in statistical induction. By engaging in statistical testing, the researcher is essentially making a claim with reference to hypothetical repeated sampling. In this respect, ontological assumptions act as the bridge between actual and more general claims being made.
4. Other motives which they consider, and which we agree are consistent with critical realism, are considered later in the paper. In contrast, however, we emphasise the combination of methods.
5. Instrumentalism focuses on using theories for practical purposes and arguably began in economics (see Friedman, 1953), with an emphasis upon prediction. There is an echo of positivism in the approach, in which data provide the arbiter in assessing the usefulness of theories. At the very least, the approach is inductive, yet this does not imply necessarily a quest for objective truth.
6. Here the term interactionism refers to symbolic interactionism, as associated with Weber and Mead (see Blumer, 1969). Interactionism should therefore not be confused with the view that mind and body are separate but interact. Interactionism is not identical, but similar, to ‘interpretivism’. Indeed, little would be lost by substituting one term for the other in our text.
7. See, for example, Blaikie (1993), who discusses interpretivism as emergent from the hermeneutic and phenomenological traditions of thought. Bryman (2001), moreover, reviews each of these approaches. Of course, whether each of these approaches simply reflects a specific method or is associated with a broader methodological position in which more than one method is employed is an area of debate and, as noted in the introduction, discussion of which is beyond the scope of this paper.
8. Data triangulation and investigator triangulation retain a strong inductive orientation. The same is true of theoretical triangulation if, say, this involved nested econometric models. Investigator and theoretical triangulation might involve different disciplinary traditions. This raises the same ontological questions discussed below and in Section 5.
9. Lakatos’s (1978) concept of scientific research programmes in which sophisticated falsification is required in the absence of crucial experiments is, in this regard, an extension of detail and aspiration than difference in logical position.
10. The deductive-nomological and inductive-statistical models of Carl Hempel (1965, 1966) can be viewed likewise.
11. Popper is often described as a critical rationalist in this regard in distinguishing his approach from positivist variants. On the one hand, the emphasis is upon criticism; that is refutation, rather than confirmation. On the other hand, one cannot approach ‘nature’ through observation in a value-free sense. Theory is required to guide falsification. Reason may be used to express a preference for a theory, but does not justify it scientifically. The process of falsification does this and thus separates science from non-science. The ability of researchers to falsify theories is known as Popper’s ‘demarcation criterion’. Only falsifiable theories are regarded as scientific. The logic of Popper’s position makes it easy to see how experimentation, as opposed to other forms of observation, is viewed as ‘the’ scientific method.
12. Whether or not Hume can be ascribed to this description is debated. See, for example, Dow (2002). In what follows, the adequacy of this label is not relevant to the arguments.
13. The concept of cause is thus not linked to the succession of events but rather an evolutionary concept of emergence, in which agency and institutions combine to bring about effects under the influence of their environment. Cause thus has ontological depth.
14. In social science, the researcher shares the hermeneutic moment of the objects of study (Bhaskar, 1978). Indeed, Sayer (2000) argues that the social researcher operates in a double hermeneutic of both the scientific and objects-of-study communities.
15. Generality here refers to essential constituents rather than, say, statistical generalisation.
16. In this respect, Danermark et al. (2002) argue that validity is explored through the strategic exploration of cases as opposed to seeking statistical significance through random sampling. In this respect, cases might explore extreme, varied, critical or normal circumstances to reveal different features of reality. Lawson (1997, 2003) for example emphasises a contrastive method to reveal causes which shares this aspiration, and Lawson (1997), in particular, is sceptical of statistical methods. Later in this section
17. Kemp and Holmwood (2003) make similar points that in the absence of experimental conditions realists can look to temporary closures, tied to a specific space-time context, examined empirically and statistically, to help to understand the structural, i.e., causal determinants of events. In this regard, they argue that the generic appeal to open systems is not helpful to social science. Moreover, they argue that, while structural accounts can easily be constructed, the key issue is to assess their validity. It is this issue that we are concerned with here.
18. This suggests that degrees of closure are central to epistemological claims if not ontological claims.
19. In this respect, while cause can still be understood in the comparative sense of the presence or absence of phenomena, as might be presented typically in terms of necessary and sufficient conditions, qualitative comparison can reveal more about the ways of acting of phenomena, and their complexity through their emergence into higher level regularities, rather than simply the presence or absence of attributes or regularities. From such a perspective, higher level emergent regularities or combinations of attributes can then be both understood in connection to their causes and also partially reveal them. This issue is discussed further below.
20. In terms of a distinction between them, the most that could be argued is that qualitative methods might involve less closure than quantitative, and that in many circumstances, they are more powerful (Mearman, 2004).
21. Contrast this with, say, the use of an equation to describe behaviour in economics. Here, motivation and action and subsequent consequences are assumed to be some form of optimisation exercise, which is described in static terms as the equation. Then the dependent variable, as effects, responds directly to the independent variables, as ‘causes’, with the relational nature of agency being reduced to the additive form in the equation directly, or indirectly as the economists moves from a theory of the individual to one of market, or more aggregate, behaviour.
22. This is central to the definition of interdisciplinary social science referred to in Section 5. Note here, that Downward’s (1999) account of Post Keynesian pricing specified the demi-regularities of pricing in the UK with reference to a survey of econometric studies, many of which were motivated by specific and often conflicting theoretical arguments. Drawing upon the descriptive behaviour of prices offered by each study, to provide a characterisation of the demi-regularities which could then be linked to insights from case-study research, implies that the newly constructed account of pricing is distinct from any of the specific studies reviewed.
23. Not every question need refer to causal relations or natural laws.
24. Downward and Mearman (2002, 2003) and the conclusion to Downward (2003) explore the changing emphasis of closure for varieties of statistical methods. An interesting point to note in the current context is that one can view statistical induction as a process of ‘hypothetical’ triangulation wherein validity is sought from hypothetical repeated sampling, with ontological assumptions about probabilities being required, that is, that they can act as indicators of summaries of complex covariation, if not their literal description, only if it is assumed that closure persists. While this might be useful as a vehicle for generating possible scenarios, clearly it implies a fragile basis for inferences outside such conditions.
25. This is consistent with Keynes’s view that probabilities are essentially ordinal and non-quantitative. Keynes emphasised the importance of rational belief rather than knowledge as a basis of argument (Keynes, 1973, VIII, p. 10). Rational belief resides in logical justification. There is a relative/absolute dimension to this. Probabilities, which are not necessarily numerically defined, are, on the one hand relative to given evidence. However, once given a body of evidence or initial proposition, probabilities concerning subsequent propositions are absolute or objective. Crucially for Keynes, relevant evidence is ascertained through a process of negative analogy. To avoid the problem of induction, Keynes argued that one should examine a particular phenomenon in different contexts. If a phenomenon appears to be common across various contexts, this indicates its relevance. In turn, this relevance adds weight to a particular account of that phenomenon. If the different contexts reveal non-common elements, the weight of an argument will decrease, revealing our ignorance.
26. The physical process or action of the treatment can, of course, be observed, but cause is still theoretical as the changed states are understood in connection to maintained hypotheses about the constitution of matter. At times, too, the processes are unobservable, in which case results are compared to hypothetical states that are predicted from theories. In experiments, thus, the concept of cause as emergence is essentially abstracted from as it becomes merged in a narrative that emphasises the presence or absence of attributes. This is because experiments engineer isolated states. While they help us to understand phenomena in isolation, therefore, this is no guarantee of predictability or stable application of that cause outside that isolation.
27. This is ‘demonstrated’ by the presence or absence of certain characteristics, albeit confined to the distinct set of cases.

Bibliography

• Bank of England 2004. The Bank of England Quarterly Model, London, Bank of England
• Bhaskar, R. 1978. A Realist Theory of Science, Sussex, Harvester Press
• Bhaskar, R. 1979. The Possibility ofNaturalism: A Philosophical Critique of the Contemporary Human Sciences, London, Verso
• Blaikie, N. 1991. A critique of the use of triangulation in social research, Quantity and Quality, vol. 25, 115–36
• Blaikie, N. 1993. Approaches to Social Enquiry, Cambridge, Polity Press
• Blaug, M. 1980. The Methodology of Economics, Cambridge, Cambridge University Press
• Blumer, H. 1969. Symbolic Interactionism: Perspective and Method, Englewood Cliffs, NJ, Prentice-Hall
• Britton, E., Cutler, J. and Wardlow, A. 1999. The bank’s use of survey data, Bank of England Quarterly Bulletin, May, 177–82
• Bryman, A. 2001. Social Research Methods, Oxford, Oxford University Press
• Bunge, M. 1996. The seven pillars of Popper’s social philosophy, Philosophy of the Social Sciences, vol. 26, no. 4, 528–56
• Byrne, D. 2003. Interpreting Quantitative Data, Sage, London
• Campbell, D. T. and Fiske, D. W. 1959. Convergent and discriminant validity by the multi-trait multi-method matrix, Psychological Bulletin, vol. 56, 81–105
• Cresswell, L. W. 1995. Research Design: Qualitative and Quantitative Approaches, Thousand Oaks, CA, Sage
• Danermark, B. Ekstrom, M. Jakobsen, L and Karlsson J. Ch. 2002. Explaining Society: Critical Realism in the Social Sciences, London, Routledge
• Davidson, P. 1996, Reality and economic theory, Journal of Post Keynesian Economics, vol. 18, no. 4, 479–508
• Denzin, N. 1970. Strategies of multiple triangulation, pp. 297–313 in Denzin, N. (ed.), The Research Act in Sociology: a Theoretical Introduction to Sociological Method, New York, McGraw-Hill
• Denzin, N. 1989. The Research Act, Englewood Cliffs, NJ, Prentice Hall
• Dow, S. C. 1990., Post Keynesianism as political economy: a methodological discussion, Review of Political Economy, vol. 2, no. 3, 345–58
• Dow, S. C. 1996. The Methodology of Macroeconomic Thought, Aldershot, Edward Elgar
• Dow, S. C. 2002. Historical reference: Hume and critical realism, Cambridge Journal of Economics, vol. 26, no. 6, 683–96
• Downward, P. M., 1999. Pricing Theory in Post Keynesian Economics, Aldershot, Edward Elgar
• Downward, P. M., 2000, A realist appraisal of Post Keynesian pricing theory, Cambridge Journal of Economics, vol. 24, no. 2, 211–24
• Downward, P. M. 2003. (ed.) Applied Economics and the Critical Realist Critique, London, Routledge
• Downward, P. M. and Mearman, A. 2002. Critical realism and econometrics: constructive dialogue with Post Keynesian economics, Metroeconomica, vol. 53, no. 4, 391–415
• Downward, P. M. and Mearman, A. 2003. Presenting demi-regularities: the case of Post Keynesian pricing, in Downward, P. (ed.), Applied Economics and the Critical Realist Critique, London, Routledge
• Downward, P. M. and Mearman, A. ( 2004A) On tourism and hospitality management research: a critical realist proposal, Tourism and Hospitality Planning and Development, vol. 1, no. 2, 107–22
• Downward, P. M. and Mearman, A. ( 2004B) Presenting ‘demi-regularities’ of pricing behavior: the need for triangulation, in Forstater, M. and Wray, L. R. (eds), Contemporary Post Keynesian Analysis: A Compendium of Contributions to the Seventh International Post Keynesian Workshop, New York, Elgar
• Downward, P. M. and Mearman, A. 2005. ‘Methodological Triangulation at the Bank of England: An Investigation’, Economics Working Paper 05/05, University of the West of England
• Downward, P. M., Finch, J. and Ramsay, J. 2002. Critical realism, empirical methods and inference: a critical discussion, Cambridge Journal of Economics, vol. 26, no. 4, 481–500
• Fine, B. 2004. Addressing the critical and the real in critical realism, in Lewis, P. (ed.), Transforming Economics: Perspectives on the Critical Realist, London, Routledge
• Frankfort-Nachmias, C. and Nachmias, D. 2000. Research Methods in the Social Sciences, 6th edn, New York, Wadsworth
• Friedman, M. 1953. The Methodology of Positive Economics, in Essays in Positive Economics, Chicago, Chicago University Press
• Giddens, A. 1974. Positivism and Sociology, London, Heinemann
• Glaser, B. and Strauss, A. 1967. The Discovery of Grounded Theory: Strategies for Qualitative Research, New York, Aldine
• Grix, J. 2002. The generic terminology of social research, Politics, vol. 22, no. 3, 175–86
• Halfpenny, P. 1982. Positivism and Sociology: Explaining Social Life, London, Allen & Unwin
• Harré, R. 1986. Philosophies of Science: An Introductory Survey, Oxford, Oxford University Press
• Hausman, D. 1998. Economics, Philosophy of, in Routledge Encyclopedia of Philosophy, London, Routledge
• Hempel, C. E. 1965. Aspects of Scientific Explanation and Other Essays in the Philosophy of Science, New York, Free Press
• Hempel, C. E. 1966. Philosophy of Natural Science, Englewood Cliffs, N.J., Prentice-Hall
• Hendry, D. F. 1995. Dynamic Econometrics, Oxford, Oxford University Press
• Hirst, P. 1993. Education, knowledge and practices, in Barrow, R. and White, P. (eds), Beyond Liberal Education, London, Routledge
• Hodgson, G. M. 1988. Economics and Institutions: A Manifesto for a Modern Institutional Economics, Cambridge, Polity Press
• Hodgson, G. M. 2001. How Economics Forgot History: The Problem of Historical Specificity in Social Science, London, Routledge
• Jick, T. D. 1979. Mixing qualitative and quantitative methods: triangulation in Action. In Van Manen, J. (ed.), Qualitative Methodology, London, Sage
• Kemp, S. and Holmwood, J. 2003. Realism, regularity and social explanation, Journal for the Theory of Social Behaviour, vol. 33, no. 2, 165–87
• Keynes, J. M. 1973. The Collected Writings ofJohn Maynard Keynes, Vol. VIII, London, Macmillan
• Lakatos, I. 1978. The Methodology of Scientific Research Programmes, Cambridge, Cambridge University Press
• Lawson, T. 1988. Probability and uncertainty in economic analysis, Journal of Post Keynesian Economics, 38–65
• Lawson, T. 1997. Economics and Reality, London, Routledge
• Lawson, T. 2003. Reorienting Economics, London, Routledge
• Lee, F. S. 2002., Theory creation and the methodological foundation of Post Keynesian economics Cambridge Journal of Economics, vol. 26, 789–804
• Lewis, P. 2004. Transforming Economics: Perspectives on the Critical Realist, London, Routledge
• Mair, D. and Miller, A. (eds) 1992. A Modern Guide to Economic Thought, Aldershot, Edward Elgar
• Mearman, A. 2002. To what extent is Veblen an open-systems theorist? Journal of Economic Issues, vol. 36, no.2, 573–80
• Mearman, A. 2004. ‘Open Systems and Economic Methodology’, Economics Working Paper 04/02, University of the West of England
• Olsen, W. K. 2003., Methodological triangulation and realist research: an Indian exemplar, in New, C. and Carter, B. (eds), Realism and Empirical Research, London, Routledge
• Olsen, W. K. 2004. Triangulation in social research: qualitative and quantitative methods can really be mixed, in Holborn, M. (ed.), Developments in Sociology: An Annual Review, Ormskirk, Lancs, UK, Causeway Press
• Olsen, W. K. and Morgan, J. 2005. A critical epistemology of analytical statistics: addressing the sceptical realist, Journal for the Theory of Social Behaviour, vol. 35, no. 3
• Olsen, W. K., Hara, T. and Sampson, E. 2003. Questionnaires in realist research: a case study of rural small-scale industry in Ghana, in Downward, P. M. 2003. (ed.), Applied Economics and the Critical Realist Critique, London, Routledge
• Pagan 2003. Report on Modelling and Forecasting at the Bank ofEngland, London, Bank of England
• Popper, K. R. 1959. The Logic of Scientific Discovery, London, Hutchison
• Popper, K. R. 1972. Conjectures and Refutations, London, Routledge
• Ragin, C. 1987. The Comparative method: Moving Beyond Qualitative and Quantitative Strategies, Berkley, University of California Press
• Risjord, M., Moloney, M. and Dunbar, S. 2001. Methodological triangulation in nursing research, Philosophy of the Social Sciences, vol. 31, no. 1, 40–9
• Rudner, R. S. 1953. The Scientist qua scientist makes value judgements, Philosophy of Science, vol. 20, 1–6
• Ryan, A. 1975. The Philosophy of the Social Sciences, London, Macmillan
• Sayer, A. 1992. Method in Social Science: A Realist Approach, London, Routledge
• Sayer, A. 2000. Realism and Social Sciences, London, Sage
• Schutz, A. 1963. Common sense and scientific interpretation of human action, in Natanson, M. A. (ed.), Philosophy of the Social Sciences, New York, Random House
• Shih, Fu-Jin 1998. Triangulation in nursing research: issues of conceptual clarity and purpose, Journal of Advanced Nursing, vol. 28, 631–41
• Silverman, D. 1993. Interpreting Qualitative Data. London, Sage
• Sintonen, M. 1989. Explanation: in search of the rationale, in Kitcher, P. and Salmon, W. (eds), Scientific Explanation, Minneapolis, University of Minnesota Press
• Tashakkori, A. and Teddlie, C. 1998. Mixed Methodology. Combining Qualitative and Quantitative Approaches, Sage, Thousand Oaks
• Thagard, P. 1992. Conceptual Revolutions, Princeton, N.J., Princeton University Press
• Toulmin, S. 1972. Human Understanding, Oxford, Clarendon Press
• Walby, S. 2001. Against epistemological chasms: the science question in feminism revisited, Signs, A Journal of Women in Culture and Society, vol. 26, no. 2, 485–509
• Walters, B. and Young, D. 2001. Critical realism as a basis for economic methodology: a critique, Review of Political Economy, vol. 13, no. 4, 483–501
• Webb, E. J., Campbell, D. T., Schwartz, R. D. and Sechrest, L. 1966. Unobtrusive Measure: Nonreactive Measures in the Social Sciences, Chicago, Rand McNally
• Whitley, J. 1997. Economic models and policy-making, Bank of England Quarterly Bulletin, vol. 37, no. 2, 163–73
• Winch, P. 1971. The Idea of a Social Science and Its Relation to Philosophy, London, Routledge & Kegan Paul
• Yin, R. K. 1994. Case Study Research: Design and Methods, 2nd edn, Thousand Oaks, CA, Sage