%20-%20https://sketch.com%20--%3e%3ctitle%3e3ECBBF50-FD08-412A-943D-2CFC5E025A16%3c/title%3e%3cdesc%3eCreated%20with%20sketchtool.%3c/desc%3e%3cg%20id='Stylesheet'%20stroke='none'%20stroke-width='1'%20fill='none'%20fill-rule='evenodd'%3e%3cg%20transform='translate(-926.000000,%20-668.000000)'%20id='BUTTONS'%3e%3cg%20transform='translate(100.000000,%20597.000000)'%3e%3cg%20id='BTN-/-RIGHTS'%20transform='translate(816.000000,%2065.000000)'%3e%3cg%20id='ICON-/-Rights'%20transform='translate(10.000000,%206.000000)'%3e%3cpath%20d='M13.5752841,0%20L1.34943182,0%20C0.59375,0%200,0.586666667%200,1.33333333%20L0,22.6666667%20C0,23.4133333%200.620738636,24%201.34943182,24%20L17.6505682,24%20C18.40625,24%2019,23.3866667%2019,22.6666667%20L19,5.36%20L13.5752841,0%20Z%20M13.5752841,1.89333333%20L17.0838068,5.36%20L13.5752841,5.36%20L13.5752841,1.89333333%20Z%20M1.34943182,22.6666667%20L1.34943182,1.33333333%20L12.1988636,1.33333333%20L12.1988636,5.36%20C12.1988636,6.10666667%2012.7926136,6.69333333%2013.5482955,6.69333333%20L17.6235795,6.69333333%20L17.6505682,22.6666667%20L1.34943182,22.6666667%20Z'%20id='Shape'%20fill='%23595959'%20fill-rule='nonzero'%3e%3c/path%3e%3ccircle%20id='Oval'%20stroke='%23595959'%20stroke-width='1.5'%20fill='%23595959'%20cx='9.5'%20cy='13.5355339'%20r='5.5'%3e%3c/circle%3e%3cpolyline%20id='Path-2'%20stroke='%23FFFFFF'%20stroke-width='1.5'%20transform='translate(9.500000,%2013.035534)%20rotate(45.000000)%20translate(-9.500000,%20-13.035534)%20'%20points='8%2016.0355339%2011%2016.0355339%2011%2010.0355339'%3e%3c/polyline%3e%3c/g%3e%3c/g%3e%3c/g%3e%3c/g%3e%3c/g%3e%3c/svg%3e)
Social policy and applied social science
Social policy is conventionally understood as a form of applied social science. ‘Broadly defined’, Donnison wrote, ‘it is an attempt to apply the social sciences . . . to the analysis and solution of a changing range of social problems’ (cited in Birrell et al., Reference Birrell, Hillyard, Murie and Roche1971: 9). The claim that social science is being ‘applied’ to social policy might be taken to mean that social policy relies on the basis of knowledge drawn from social science research: that what is being applied are concepts, findings and facts, borrowed eclectically from a range of subjects. But the links between social policy and the social sciences run deeper and stronger than occasional scavenging: social policy draws no less on the rationales, methodologies and patterns of thinking of the social science disciplines. Nearly 40 years ago, Gans argued that ‘policy researchers . . . must have – and create – a policy oriented social science, independent of but related to and not estranged from the academic disciplines’ (Gans, Reference Gans and Horowitz1975: 3). That work has still to be done.
This article focuses on a strand in social science that is of particular importance for policy research: the principle of generalisation. Research in social policy is often concerned with questions like ‘what works?’ or ‘can this be done elsewhere?’ The circumstances being considered are taken to refer not just to the specific aspects of one policy, like an evaluation or a measure of outcomes, but on generalisable principles or regularities that can be taken from one set of circumstances and used in another. However, the process of generalisation is beset with problems. The argument that follows here reviews some of the issues, and outlines an alternative paradigm that may be able to avoid some of the difficulties.
Approaches to generalisation
The process of generalisation is understood in different ways. For some, it depends on forming and testing theory. Researchers begin with a theoretical model – perhaps an ideal type, perhaps an equation, perhaps (as in critical social policy) an interpretation of society – and work within the paradigm to confirm or deny the relationships the theory implies. This is sometimes referred to as a ‘deductive’ approach in research. Examples are discussions of whether frustration leads to aggression, unemployment leads to crime, liberal values lead to lower public spending or higher benefits lead people to choose not to work. David Gordon argues that all scientific research is theory-led. He thinks that for any theory to be scientific, it should be falsifiable, testable, have predictive value and the results must be reproducible; and he adds Lakatos's criteria that the theory should map out directions for further research and lead to the discovery of novel phenomena (Pantazis et al., Reference Pantazis, Gordon and Levitas2006: 36–7).
For others, generalisation depends on regular and consistent relationships between observed phenomena. Using statistical data, we begin with observation and attribute underlying relationships to the associations. An example is the work of Noble and his colleagues in their work for the Index of Multiple Deprivation:
It is hypothesised that an underlying factor exists . . . that makes these different states likely to exist together in a local area. This underlying factor cannot be measured directly but can be identified through its effect on individuals . . . These variables need to be combined at an ecological level to create an area score. Fundamentally this score should measure, as accurately as possible, the underlying factor . . . The premise . . . is that the underlying factor is imperfectly measured by each of the variables in the dataset but that the variables that are most highly correlated with the underlying factor will also be highly correlated with the other variables. (Social Disadvantage Research Centre, 2003: 53)
This is presented as if it was deductive in form – there is a hypothesis to be tested. But the ‘hypothesis’ is simply that there is some unspecified phenomenon that can be discovered from patterns in the observed data.
A third approach is ‘realism’, which is an attempt to uncover the underlying mechanisms that explain relationships in practice. I am using this word, rather than ‘positivism’, for two reasons. One is that the term ‘positivism’ has also been used to encompass the kind of empiricism I have referred to in the previous category. The other is that ‘realism’ has been distinguished from positivism in some important texts, notably by Pawson and Tilley in their seminal book, Realistic Evaluation (Pawson and Tilley, Reference Pawson and Tilley1997: ch. 3). Realism (or ‘critical realism’) supplements traditional empiricism partly by emphasising the structural aspects of social relationships, but primarily by its emphasis on looking for a ‘generative mechanism’ as a form of explanation – in other words, relationships of cause and effect (Sayer, Reference Sayer2000). This can be represented as theory-led or data-led, but it is distinct from both: it is led by the identification of a process or structural relationships.
There are important differences between these approaches, but some elements are true of all three. To be able to take principles or lessons from one case to others, there has to be something about the observation or analysis that is transferable. Whether researchers are looking for patterns in data, describing correspondences between theory and practice, or reviewing the implications of systems or structures, they have to build on some sense of the relationship between observations. The pattern of generalisation in each of these paradigms is to juxtapose elements of knowledge, and show a relationship between those elements that can be taken and used in a different context. The attempt to generalise in these terms is not true of every aspect of social science, but it is the cornerstone of a wide range of methods in policy research.
Induction
The rationale for generalisations about relationships is commonly attributed either to induction, or to cause and effect. Induction is supposed to depend on the proposition that when people observe that two things happen together, they come to believe that those two things are connected: that the presence of one indicates the presence of the other. There are some examples of this kind of reasoning in social science – there are long-established empirical associations, like the link between social class and health, which have usually been found to hold good even though they are subject to competing explanations. For the most part, though, it is an obvious fallacy to say that because things happen a lot together they will happen together the next time – so obvious, one has to say, that it should be immediately clear that there is something wrong with the basic concept. The label was used by philosophers like Russell to discredit the approaches they disapproved of (Russell, Reference Russell1912: 37). Doyal and Harris, similarly, identify induction with ‘crude empiricism’ (Doyal and Harris, Reference Doyal and Harris1986). Of course, no one wants to describe themselves as ‘crude’.
Induction in practice covers a wide range of approaches. When we are reviewing evidence for policy, describing and classifying a set of phenomena, or conducting a statistical analysis to examine the relationships between elements, we are not usually doing the same kind of thing. Reviewing evidence may be interpretative, but, for the most part, it depends on an interaction between evidence and explanation. The process of describing and classifying phenomena is partly done by theoretical selection, partly by the application of principles like taxonomy or comparison, and partly by adapting observation to circumstances. Inductive statistics are used as both a descriptive and a predictive tool, but the predictions are based in a theory of probability governing distributions rather than assumptions about the regularity of the relationships. The core of the activity is that data are sifted and sorted to establish patterns of the occurrence and recurrence of juxtapositions; that is a way of generating new observations for the purposes of reflecting on and interpreting the data. These methods build an understanding of relationships in different ways; none of them rests intrinsically on the kind of presumptive claim that induction is supposed to make by its critics.
The process of identifying associations, and generalising from them, is an important part of social science. Several writers have tried to rehabilitate the idea of induction, redefining it in an attempt to get a concept that at least makes some kind of sense. For Babbie, inductive research is any research that implies a movement from particular cases to generalisations (Babbie, Reference Babbie2001: 34–5). Others use the term ‘induction’ to refer to research that is led by data rather than theory. For Gilbert, induction is ‘a method of reasoning that derives generalisations by seeking the common aspects of a number of specific cases’ (Gilbert, Reference Gilbert2008: 508). Znaniecki's idea of ‘analytic induction’ recasts it in the mould of deductive reasoning (cited in Patton, Reference Patton2002: 94–5). People do translate associations into relationships. That process is rarely atheoretical; it is much more likely to be part of an attempt to make sense of patterns in empirical data. Often what induction is doing is to gather evidence that is consistent with a process, a mechanism or a causal relationship, that will make it possible to generalise the relationship in other contexts. This kind of relationship lies at the root of generalist methodologies.
Causal explanations
There is a spectrum of causal explanations. The most direct connection of cause and effect is that A causes B if its presence is necessary and sufficient for the presence of B. ‘Necessary’ means that B does not occur when A is not present. ‘Sufficient’ means that if A is present, B will also be. On the face of it, these criteria are strong enough to mean that they are almost never applicable to problems in social science; the only circumstances where they both apply are likely to verge on tautology. The existence of the structure of employment in a formal economy is necessary, and probably sufficient, for the presence of unemployment; the development of compulsory schooling is necessary, and probably sufficient, for there to be a problem of truancy. Neither statement tells us much. It is rather more typical of causal explanations in social science that they are partial. We can ‘explain’ aggression in terms of frustration, or increased demand in terms of incentives, not because all stimuli lead to a common response, but because at least some will. ‘The business of science’, Russell suggests, ‘is to find uniformities’ (Russell, Reference Russell1912: 35). In social terms, consistent explanation of the behaviour of a proportion of a population is enough for causation to be considered to apply. The phrase that Sayer uses to describe this sort of causation is a ‘regularity’ (Sayer, Reference Sayer2000): within this model of causation, A is a cause of B if it regularly and consistently leads to B.
The second kind of explanation is generative: that phenomenon A is at the root of a mechanism that produces phenomenon B. Population growth causes shortages of resources (Meadows et al., Reference Meadows, Meadows and Randers1992); high-rise housing causes social malaise (Coleman, Reference Coleman1985); benefits cause long-term poverty (Murray, Reference Murray1984). These explanations have several things in common. They rely on a direct relationship between variables. Because they separate the causes from the context, they have to be selective. The principal criticisms, unsurprisingly, are based on qualifying the propositions, reviewing contrasting evidence and putting the arguments back in context (see, respectively, Todaro and Smith, Reference Todaro and Smith2006: ch. 6; Spicker, Reference Spicker1987; Walker, Reference Walker1994). The associations are difficult to falsify, however; even where they are transparently untrue, like the claims of Malthusianism (population growth over two hundred years has still not led to a global shortage of resources), they will be explained away by apparently special circumstances (they will be true eventually, or they would still be true, if only . . .). A related but weaker form of generative explanation is that condition X produces the circumstances in which condition Y can develop. Unemployment does not cause crime, but it creates the conditions in which crime can flourish (Box, Reference Box1987). This is essentially a variant of the same principle; the relationships are more remote, but the implication is that Y is additionally dependent on other, unspecified generative processes and that X provides a partial explanation. Regularities, Pawson argues, ‘occur because of the action of underlying mechanisms in particular contexts’ (Pawson, Reference Pawson1989: 324).
A third concept of causation is that factors are predictive – the presence of X makes it probable that Y will also occur. This is the root of the approach to statistical method which supposes that statistical variation in a dependent variable can be ‘explained’ by variation in an independent variable, and that the higher the association, the greater the contribution one factor could be taken to make to the other. This might be the result of a direct generative relationship, but it might also indicate a set of relationships: both X and Y might be produced by a common constellation of causes (such as poverty leading to ill health and lack of amenities, rather than lack of amenities leading to ill health directly). This is a much more adaptable and robust understanding of causation than either of the first two, but it has several pitfalls. One problem is that the conceptualisation of factors in isolation is often suspect – Pawson, indeed, argues that it is always suspect – because social factors can only have an influence in a broader social context (Pawson, Reference Pawson1989: ch. 2). Another is that prediction still requires a generative mechanism or process as the basis for explanation; associations without mechanisms are unexplained. Further, ‘probability’ is often understood in terms of a relative likelihood of association: minority ethnic groups are said to be ‘more likely’ than others to be poor, or people in lower social classes are more likely to have lower educational attainment. This type of association is not evidently predictive; a higher association carries no implication that most, or even many, will be affected by the associated issues.
Fourth, there is causation as narrative. For realists, Sayer argues,
causation is not understood on the model of regular successions of events, and hence explanation need not depend on finding them . . . The conventional impulse to prove causation by gathering data on . . . repeated occurrences is therefore misguided; at best these might suggest where to look for candidates for causal mechanisms. . . . Explanation depends instead on identifying causal mechanisms and how they work, and discovering if they have been activated and under what conditions. (Sayer, Reference Sayer2000: 14)
Identifying a cause depends on the construction of a sequential account. It is easy to gain the impression from some of the literature – both positivist and realist – that a causal relationship is a matter of empirical fact, and the answer is there, lurking under the surface, if only we analyse the information in the right way. What happens in the construction of a causal narrative is quite a different process. In any exercise where there are competing factors, and competing explanations, identifying causes is a matter of judgement. Leaving it to the computer does not avoid the problems of judgement; multivariate analysis depends on a series of judgements, about the selection of units for analysis, the choice of distributions, transformations, outliers and multicollinearity. When academics trawl the data in the hope of seeing a causal relationship, they will almost always find something. It is nearly always possible to see patterns in an amorphous mass of complex data, if you squint at the dots on the page in a certain way; but change the selection of dots, alter the basis on which the figures are derived, and the lines start to look different. The identification of general relationships in complex quantitative data is a characteristic element of comparative social policy (see Castles, Reference Castles2004, Reference Castles2005 for examples; Hantrais and Mangen, Reference Hantrais and Mangen1996; Clasen, Reference Clasen1999, for reservations). Causal explanation becomes, not the discovery of a single truth, but a form of interpretative analysis.
The problems of causal analysis
The ideas of ‘cause and effect’ court controversy (which is why realists like Pawson and Tilley try not to use them and talk about ‘generative mechanisms’ instead). David Hume, during the Scottish Enlightenment, challenged the basic premise that sequences of events can be meaningful, or that juxtaposition can be seen as explanatory in any sense (Hume, Reference Hume and Selby-Bigge1739, Book 1, III, chs. 14–15). The reasons why he was doing that were rooted in his time: Hume was challenging the idea of miracles and the supernatural, and arguing for a rationalist approach to scientific methodology, in place of the narrative accounts that dominated discussion. Hume's scepticism can be seen as the father both of logical positivism (which, unlike the ‘positivism’ of sociology, sought to confine scientific and philosophical interpretation to what could be established) and of phenomenology (which, although it referred primarily to the study of social meaning, was used to justify radical scepticism about structural relationships). In recent times, both deconstruction and post-modern approaches have come to occupy the sceptical position.
What we have learned from this scepticism is that claims about cause and effect cannot be trusted. Sequence is not consequence; associations are not proof of generative mechanisms. Few issues can be identified in terms of a direct relationship between cause and effect. Causal analysis, or a focus on a ‘generative mechanism’, calls both for induction – movement from the specific to the general – and the interpretative leap which makes it possible to say that there is a process that explains the associations. Often the analysis rests on the imposition of a preconceived model on the data. Interpretative analysis can usually be re-interpreted.
The most obvious problem with causal analysis in social policy is that it is difficult. That is not an insuperable objection, but it should make commentators more hesitant than they seem to be. Social phenomena are typically complex; many issues are multifaceted, and even relatively simple social phenomena are likely to be influenced by a range of different factors. The history of social policy is festooned with examples of questionable claims made about causes: for example, that genetic inheritance is at the root of social problems (see Pick, Reference Pick1989; Carlson, Reference Carlson2001); that maternal neglect causes juvenile delinquency (Bowlby, Reference Bowlby1953); or that welfare provision causes poverty and idleness (Spencer, Reference Spencer1853; Murray, Reference Murray1984). Looking for an underlying explanation for social phenomena almost always calls for simplification – stripping away extraneous data to see what lies at the core. Unfortunately, any selection of a limited number of factors from a mass of detail is vulnerable to misinterpretation. It follows that people who think they know the causes of social phenomena are usually wrong.
Identifying how people respond under certain conditions should in principle make it possible to determine how they will respond when the specific conditions are altered. In principle, a process like a randomised control trial makes it possible to identify what effects result from the policy intervention, distinguished from the confounding effects of the social environment. That, Pawson and Tilley argue, is the trouble with the idea: social phenomena cannot be separated from the social environment. ‘Our argument’, they write, ‘is that precisely what needs to be understood is what it is about given communities that will facilitate the effectiveness of a program! And that is precisely what is written out’ (Pawson and Tilley, Reference Pawson and Tilley1997: 52). They suggest that the way to deal with the problems of attributing causes is to put a greater emphasis on the underlying generative mechanism. But a generative mechanism will only produce similar effects when other things are equal – and, as Pawson and Tilley themselves argue, they never are. There is an implicit contradiction in trying to hold to a generalised explanatory mechanism while at the same time emphasising that it can be understood only under localised conditions. They are trying to have it both ways.
It would be more understandable that social scientists want to hold on to the idea of causation, if explaining the cause pointed to useful prescriptions for policy. It rarely does. The best social science is often so specific, and so finely qualified by context and circumstances, that it is difficult to generalise from it. Titmuss's argument that the Second World War created the conditions in which the British welfare state could be developed (Titmuss, Reference Titmuss1950) is clearly a causal account, and it is hard to think of good arguments against it. This kind of narrative is valuable for understanding particular phenomena in retrospect, but it does not necessarily provide a basis for the kind of generalisations that people want to draw from policy research – it does not explain how the conditions can be replicated or what governments should do. (Titmuss did in fact go on to draw a more contentious general lesson: that as war makes greater demands of a society, it has direct implications for social policy, including a demand for improved health in the population – Titmuss, Reference Titmuss1955 – but it is questionable whether that is actually true or, if it is, whether it has any implications for future policy at all.) For a causal explanation to be generalisable, we have to be able to identify and isolate the factors at play; without that, the explanation cannot be applied to new situations.
Even if there is an apparently compelling causal explanation, it does not follow that we consequently know what to do. Understanding the development of industrial society does not help much when it comes to dealing with structural unemployment. Being told that the poverty of large cities is the result of irreversible capitalist decline – the answer submitted by the Community Development Projects (CDPs) (Loney, Reference Loney1983) – is useless by their own account. Worse, it was a betrayal of the people the CDPs were supposed to help: they refused to use the resources and licence they were given to engage with activities that might actually do anything about poverty. If you fall down a well, knowing about gravity is not going to do much to get you out of it, and an analytical knowledge of alternative methods of water supply is not going to help. Knowing how to climb is.
Alternatives to generalisation
The approaches to social science considered so far claim more for social policy than they can reasonably hope to deliver. That has led some critics to abandon them altogether. The main alternative to generalisation has been sceptical pragmatism. For the sceptic, there are no general rules, just precedents. Pragmatism – an incremental focus, proceeding without depending on theoretical models – focuses on things that actually produce good effects, rather than the things that are supposed to. We can test, cautiously; we can imitate, in the hope that something that has worked will work under different conditions; we can diversify, choosing robust alternatives that allow for policy to be altered when it fails. No single policy will ever be good enough, because there is always the possibility of failure, or of confounding factors. It is not important whether or not a policy makes sense intellectually. The case for ‘muddling through’ is often attributed in the policy literature to Lindblom (Reference Lindblom1965), but, long before him, Edmund Burke made the case in rather more elegant terms:
By a slow but well-sustained progress, the effect of each step is watched; the good or ill success of the first, gives light to us in the second; and so, from light to light, we are conducted with safety through the whole series. We see, that the parts of the system do not clash. The evils latent in the most promising contrivances are provided for as they arise. One advantage is as little possible sacrificed to another. We compensate, we reconcile, we balance. We are enabled to unite into a consistent whole the various anomalies and contending principles that are found in the minds and affairs of men. From hence arises, not an excellence in simplicity, but one far superior, an excellence in composition. (Burke, Reference Burke1790: 209)
Pragmatism has a lot to commend it. It has the advantages of building on success, flexibility and robustness. It has the disadvantages, however, of requiring constant monitoring and frequent evaluation. It is slow. It may lead to failure to respond to critical problems in good time, while all around us things are getting worse. And it is often distorted by prejudgements (Spicker, Reference Spicker2008). In practice, then, Etzioni suggests, what policy-makers do becomes ‘mixed scanning’, hopping back and forth between pragmatic and theoretical views (Etzioni, Reference Etzioni, Gilbert and Specht1977).
In recent years, another set of approaches has developed into ‘post-positivism’. This is an ungainly and ambiguous term for what is, essentially, a form of critical scepticism, which treats scientific empiricism with distrust. Guba explains the approach in these terms:
[Post-positivism] asserts a relativist ontology on the assumption that all reality is mentally constructed and that there are as many realities as there are persons to contemplate them; that there are no general or universal laws that can be counted on in every situation but that the action or behavior noted in any context is uniquely determined therein; and that all elements of a context are continuously involved in ‘mutual simultaneous shaping’ in ways that render the concept of cause–effect meaningless. Further, the emerging paradigm assumes a subjective . . . epistemology, so that inquirer and respondents mutually share their construction in a hermeneutic circle throughout the inquiry and thus create the ‘reality’ which the inquiry may finally mirror. (cited in Durning, Reference Durning and Jabes2004: 690)
Reality, the post-positivists argue, is socially constructed, empiricism is ill-founded and unequivocal falsification is impossible (Kelly and Maynard-Moody, Reference Kelly and Maynard-Moody1993). ‘The social sciences’, Fischer writes, ‘have largely failed’ (Fischer, Reference Fischer1998: 129). If there is no clear, unambiguous ‘answer’ to be found, the duty of the social scientist is to facilitate deliberation about the issues. Post-positivism consequently defines the role of the researcher as promoting dialogue and engaging with diverse perspectives, often through the use of multiple methodologies (Kelly and Maynard Moody, Reference Kelly and Maynard-Moody1993; Fischer, Reference Fischer1998). I have some sympathy for the methods, which imply greater flexibility, and less presumption, in the interpretation of data than traditional social science can offer. The rationale for post-positivism, however, is difficult to take. The critique seems to be based in the idea that social science is all based in empiricism, which seems curiously dated, and that the social construction of phenomena justifies a subjective epistemology, which does not follow. Denying the validity of any evidence leaves us with no obvious criteria to choose between competing statements (Groff, Reference Groff2004). Its extreme relativism seems to deny the possibility of any empirical social science. We have to find a better way.
Towards a different approach
The challenge for Social Policy as a field of study is to ask how it is possible to develop a set of approaches – an applied social science – that can cope with the demands of policy in practice. The problem is not generalisation in itself. The major weakness of the kind of social science I have been considering is not that it tries to draw lessons, or even that it tries to find regularities, it is that it generalises about the wrong sort of thing. The focus on generative mechanisms or cause and effect has often proved to be a blind alley; and the problem of walking down blind alleys in social policy is not just that we are wasting time, but that there are ideologues waiting with bludgeons who are determined to make sure that we stay there. If policy research needs to draw lessons, perhaps what we need to do is to reconsider the type of lessons we are trying to draw.
Phronesis
Flyvbjerg argues for a different kind of knowledge in social science: the application of phronesis, the Aristotelian term for practical wisdom (Flyvbjerg, Reference Flyvbjerg2001). Phronesis is usually translated in terms of wisdom, prudence or judgement. Aristotle distinguishes phronesis both from episteme – generalised, scientific knowledge – and techne – the knowledge involved in making things (Aristotle, Ethics, Book 6). The distinction between theory and practice – episteme and techne – is relatively straightforward. Episteme is theoretical knowledge, and techne is applied – it is the distinction between physics and engineering. The two forms of knowledge are interrelated, and they overlap in practice, but they are discrete. The distinction between techne and phronesis is more difficult. They also overlap in practice, but they refer to different approaches and ways of applying knowledge. Techne and phronesis are both forms of applied knowledge; both include aspects of ‘craft knowledge’, the specific applied expertise that is peculiar to a trade or profession. Both can take in elements of non-propositional knowledge (Rycroft-Malone et al., Reference Rycroft Malone, Seers, Titchen, Harvey, Kitson and McCormack2004), including information that is experiential, tacit or implicit in the exercise of certain skills. However, Aristotle explains that ‘doing and making are different in kind’ (Aristotle, Ethics, Book 6, ch. 5: 177). If techne is concerned with knowing how to make something, or what to do to produce a specified effect, phronesis is concerned with judgement – understanding the implications, and making the right choices. The key question in phronesis, for Noel, is: ‘What should I do in this situation?’ (Noel, Reference Noel1999: 274).
An example of the kind of practical judgement in social policy I have in mind might be shown through the relationship between unemployment and sickness benefits. We have known since the days of the Poor Law that unemployment and sickness are linked administratively. Chadwick's report on the sanitary conditions of towns found that a large number of paupers were in fact sick, and that was why the Poor Law took on the responsibility of public health, and subsequently of the hospitals. When National Insurance was introduced in 1911, there had to be unemployment and sickness insurance, because without it unemployed people would have had to present themselves as sick to get relief, or conversely sick people would have had to present themselves as unemployed. This is also the source of one of Beveridge's key ‘assumptions’: that for the National Insurance scheme to work a health service had to be available as well. The argument is hardly considered in his report, but Beveridge was only stating something that, at that time, was obvious to everyone who had been engaged with social security. And, in recent times, the apparent levels of incapacity benefit have increased as a reflection of unemployment in the economy. The same principle applies in provision for people with disabilities and those with chronic sickness; if provision is made for one and not the other, people who need to claim will claim the only benefit that is available. The phronetic position, then, is that any measure relating to one of these issues has to take the other into account. This principle has been well established in social policy for over a hundred and fifty years. It is not self-evident, and it is not trivial. If it is a narrative, it is not dependent on a generative mechanism. It is quite different from the run of causal explanations in social science.
What kind of narrative is it, though? It is certainly a generalisation, but it is a generalised observation – an induction – rather than a theoretical statement. It is not an explanation of what happens; it is a description – and it is only an approximate one. It could be reformulated as a technical statement, but it is very rough and ready – a rule of thumb, rather than a systematic proposition. Once we know that one kind of issue leads to changes in another, we can put mechanisms in place to deal with those changes. It is a precept – a working principle.
Here, briefly, are some other examples of phronetic precepts from social policy research:
• Selective social policies characteristically fail to reach a proportion of the people they are intended to reach (Titmuss, Reference Titmuss1968).
• The private sector exercises adverse selection in order to limit costs (Barr, Reference Barr2004).
• Claiming behaviour is affected by knowledge of services, negative attitudes to services and the costs of claiming (Craig, Reference Craig1991).
• There is an ‘inverse care law’ in health care which means that while people from lower social classes are in the greatest need, they are also least likely to receive services (Tudor Hart, Reference Tudor Hart, Smith, Dorling and Shaw1971).
• People whose priority is based on how long they have waited for service are better able to exercise choice than those priority is based in need (Clapham and Kintrea, Reference Clapham and Kintrea1986).
Any of these propositions could be expounded on at some length, but that is not my purpose here. Rather, I am referring to them only as examples of the kind of generalisation that is interpretative, derived from the experience of policy and used to evaluate services or generate prescriptions for policy. None of these statements is universally applicable; they are all dependent on circumstances. None of them is causal in form, even if commentators might be tempted to offer causal explanations for the phenomena. None of them is self-evident – they had to be discovered empirically. None of the statements is genuinely ‘explanatory’, even if any of them can be placed within a plausible narrative. They are, rather, expressions of phronesis.
Phronesis: competing interpretations
Phronesis is an ancient idea and, like many very old concepts, it has had the time and space to be interpreted in different ways. Noel identifies three principal understandings of the term: rationalist, moral and situational (Noel, Reference Noel1999). The rationalist approach depends on action in relation to a general principle or precept, which is translated into practice by a process of deliberation. An example might be the way that the rational-comprehensive model of planning has been developed into precepts for practice (Spicker, Reference Spicker2006). A more ambitious interpretation might be found in Habermas's case for ‘rationalisation of the life-world’. Dunne, who mainly emphasises the particular elements in phronesis, is critical of the attempt:
(Habermas) has tried mightily to demonstrate the possibility of a practice which . . . will have a rigorously rational and in the end quasi-theoretical basis. . . . Aristotle believed that if one's subject matter is the practical and communal life of persons then one must renounce the methodological purism and the possibilities of generalisation and precision that are legitimate aspirations in properly theoretical endeavours. Habermas, Aristotle might say, cannot have his theoretical cake and eat it. (Dunne, Reference Dunne1993: 18)
A second view of phronesis relates to the normative or moral character of phronetic decisions: the question of what should be done cannot be answered without considering normative or evaluative issues. The kinds of question phronesis attempts to answer are often normative: ‘Where are we going?’, ‘Is this desirable?’ and ‘What should be done?’ (Flyvbjerg, Reference Flyvbjerg2001: 60). Aristotle explains:
Practical wisdom is a rational faculty exercised for the attainment of truth in things that are humanly good and bad. (Aristotle, Ethics, Book 6, ch. 5: 177)
This is sometimes interpreted to mean that phronesis, unlike techne, is exclusively normative and moral (Waring, Reference Waring2000; Rooney and McKenna, Reference Rooney and McKenna2006). Politics, to Aristotle, is not something that can be done by abstracting general principles and running things predictably. It is not a form of social engineering. Political science, he argues, is a demonstration of practical wisdom, rather than craft: and he goes on to clarify that what he means to refer to is politics as public administration, ‘politics in fact’ (Aristotle, Ethics, Book 6, ch. 8).
Phronesis, and policy research, are certainly imbued with value judgements. All the examples I have just given of phronesis in social policy have normative elements; they have an evaluative purpose, and it is difficult to separate them from associated moral judgements about what should be done in consequence. The kinds of questions that phronesis addresses call for judgements to be made. The key practical questions of social policy do not stop with technicalities, like ‘how is this effect produced?’; they are just as much concerned with the question ‘what should we do?’ Values are part of the framework in which decisions are made. There is no clear basis for distinguishing the role of empirical claims (for example, that employment prospects improve with personalised support) from normative ones (that personalised support for employment is consequently desirable), because both will be used in the same way. If phronesis is about forming guides to action, the norms that guide action are part of the process. Equally, if researchers are not bracketing off core elements they think are more relevant, norms are part of the package. Phronesis is a long way removed from the tradition of a value-free social science, which once dominated the field.
A third view of phronesis emphasises the situational context of decisions, and the importance of understanding and relating decisions to the context where they are taken (Noel, Reference Noel1999). Phronesis is all about flexible, practical judgement – a way to cope with messy, uncertain practicalities (Schwandt, Reference Schwandt2003). Both Schwandt (Reference Schwandt2003) and Dunne (Reference Dunne1993) borrow an analogy from Wittgenstein: if we try to walk on a flawless, frictionless surface like ice, we are unable to walk at all. We need to get ‘back to the rough ground’.
An important element of this interpretation is the link between phronesis and experience. Aristotle writes:
Intelligence apprehends the truth of definitions which cannot be proved by argument, while phronesis involves knowledge of the ultimate particular thing, which cannot be attained by science but only by ‘perception’. (Aristotle, Ethics, Book 6, ch. 8: 182)
For Dunne, phronesis is experiential, immediate, concrete and personal (Dunne, Reference Dunne1993: 228). Experiential knowledge has been identified, like phronesis, with professional craft knowledge learned ‘on the job’ (for example, Crook, Reference Crook2001; Estabrooks et al., Reference Estabrooks, Rutakumwa, O'Leary, McGrath, Milner, Levers and Findlay2005); but it can also refer to knowledge based directly in the personal experience of the observer (Heron and Reason, Reference Heron and Reason1997); and it has been used particularly in relation to the experience of witnesses, validating personal, non-expert knowledge (Borkman, Reference Borkman1976; Ruzek and Hill, Reference Ruzek and Hill1986). Phronesis and the idea of ‘experiential knowledge’ are not directly equivalent, however. Phronesis draws on experience, but it does not have to depend exclusively on the personal experience or character of the person who is using it. It might be shared, for example, in a community of practice; it might be communicated – and it follows that it can be generalised.
These three interpretations are not exclusive in practice, and discussions of phronesis tend to cut across them in different permutations. In each of its different guises, however, phronesis is concerned with action. That is what makes the concept so appropriate for social policy.
Phronetic generalisations
For Aristotle,
Practical wisdom being concerned with action, we need both kinds of knowledge; nay, we need the knowledge of particular facts more than the general principles. (Ethics, Book 6, ch. 7: 180)
Although phronesis does depend on what Flyvbjerg calls ‘little questions’ (Flyvbjerg, Reference Flyvbjerg2001: 133), it is possible to build those little questions into generalisations. The generalisations are about experience – about what happens – rather than about theoretical relationships. That means that they have to be understood in the circumstances where they are found.
Phronetic generalisations, like the generalisations on social policy considered earlier, have certain characteristics. First, whereas causes are specific, phronesis is approximate. Even if a phronetic statement can be taken to apply broadly in a range of circumstances, it might always be wrong. Phronesis is concerned with guiding action, not simply with statements that are true or false. Because it is instrumental in its nature, a precept might be applicable even if it is not universally true – and, because phronesis is based in observation of particular circumstances, it probably will not be.
Second, where causal analysis is universal, phronesis is particular. Causes describe underlying mechanisms; phronetic generalisations are necessarily contingent on their context. Every insight, every precept, is based on experience. In economics, one begins with the core principle and tests it by changing the parameters. In phronesis, by contrast, we generalise the experience without making assumptions about underlying relationships. When variations in conditions lead to different outcomes, we qualify the generalisation.
Third, causal generalisations depend heavily on selection – cutting away information to look at the essential core. Phronetic generalisations, by contrast, generalise by cross-referring (or triangulating) experiences from different sources, without eliminating inconvenient data. This is exemplified by grounded theory, a process where theory is generated interactively by interrogating and sorting data, continuing until all the data are classified that can be (Glaser and Strauss, Reference Glaser and Strauss1967).
It is worth adding some words of caution. The problem here is a matter of quality: some of the most influential generalisations in social policy do not bear scrutiny. Most of the examples of precepts I have given so far seem to me to be reasonably convincing, but many others are not. Some other examples currently influential in social policy might be these:
• the greater the punishment, the stronger the deterrent (Bentham, Reference Bentham and Harrison1789),
• poor people are trapped in a cycle of deprivation (UK National Action Plan, 2001: 15),
• if goods are free, demand will always exceed supply (Adam Smith Institute, n.d.), or that
• ‘complexity in the benefits system acts as a disincentive to entering work’ (Freud, Reference Freud2007: 9).
These are generalised statements that people will claim to be supported by experience. They all form part of political discourse. They offer, like the other precepts I have mentioned, a narrative. However, as far as I can tell, they are all untrue. Punishment is only one factor in a series of choices and considerations; poor parents do not produce poor grandchildren (Brown and Madge, Reference Brown and Madge1982; Coffield and Sarsby, Reference Coffield and Sarsby1980; Kolvin et al., Reference Kolvin, Miller, Scott, Gatzanis and Fleeting1990); the demand for free vasectomies is not infinite; and the benefits system, whether complex or not, has only weak or marginal effects on incentives to work (Atkinson and Mogensen, Reference Atkinson and Mogensen1993; Alcock et al., Reference Alcock, Beatty, Fothergill, Macmillan and Yeandle2003). But they seem proof against evidence, and neither argument nor research can seem to stop them, like the monsters in a horror film, from coming back to life whenever some teenagers start playing in the cemetery.
Any use of evidence in social policy is vulnerable to views that are ideologically committed. Typically, people tend to select methods and approaches that fit their world view to a cause. There are some outstanding examples in the study of poverty, where the advocates of different approaches are usually convinced that their particular approach is the one that will be most beneficial to the poor (for example, the debate between Sen and Townsend, in Townsend, Reference Townsend1993). The outcomes of social policy research really matter. They raise strong passions, and objectivity tends to fly out of the window. There are certainly arguments in social policy that are crassly polemical; in extreme cases, the results may have been fabricated. Examples are the work of Cyril Burt (Human Intelligence, n.d.) or the analysis of degeneracy in the US (Christianson, Reference Christianson2003). Ultimately, the problems of balancing empirical evidence may not be resolvable by an appeal to reason.
Phronetic generalisations can in some circumstances be falsified by counter-example – Flyvbjerg sees this as a particular strength of case-studies (Flyvbjerg, Reference Flyvbjerg2001: ch. 6). More typically, however, because phronesis is dependent on particular situations, it can be difficult to show that a falsification applies more generally. What will happen is that the area in which the generalisations apply starts to shrink when different scenarios are considered, reducing what seemed to be a generality to the status of a special case.
This, then, leads to one of the core problems about this kind of generalisation: how to distinguish good generalisations from bad ones. I do not think that I have an authoritative answer – I am not sure that there is one – but it should be possible at least to draw up some guidelines. The first is that phronesis has to begin with evidence rather than theoretical inference. All the precepts I used as examples at the outset were formed by people reviewing evidence and generalising from it. However, at least three of the four later statements are based on assumptions about what people believe ought to be true theoretically, rather than what has been found to be true from the evidence. The exception may be the assertion that poor people are in a cycle of deprivation: many practitioners think it is true even if the evidence from social science says it is not. The theoretical statements, which argue that deprivation is transmitted, have been in circulation for so long that it is difficult to say which came first.
The second guide is that generalised statements need to be cross-confirmed by example. Selectivity has problems related to means-testing and to need-based housing. The ‘inverse care law’ is paralleled by inferior provision in educational provision and social housing. By contrast, the later precepts can be set against some direct and immediate counter-examples from practice.
Third, none of the best precepts is deracinated. Conventional social science calls, typically, for the isolation of factors or variables from their social context. Phronesis does not work that way. It is not deductive, and it is not based in generative mechanisms. Where there are generalisations, they are made about processes in their social context, and there is no need to make further assumptions or to depend on explanations about the underlying factors.
Phronesis and method
The idea of phronesis does not hold all the answers. Phronesis on its own, without the scientific insight of episteme, would be, like pragmatism, vulnerable to conservatism – wedded to established practice and received wisdom. Its generalisations are difficult to validate, and to falsify. There are too many uncertainties and differences in interpretation to be certain what kind of prescription is being offered – but that is part of the point. Phronesis is not a substitute for episteme and techne, but it is an essential counterbalance. It begins with an acceptance of the ambiguity and confusion of policy, but it draws on the cumulative evidence that comes from interaction, observation and practice. Different interpretations of phronesis argue for an emphasis on a deliberative process of translating general principles into practice, acceptance of the normative foundation of policy, and practical and personal experience. It offers, then, both a rationale guiding action, and a focus for the development of alternative methods and approaches.
‘Phronetic social science’, Schram writes, ‘already exists; it is just not organized or recognized as such’ (Schram, Reference Schram2003: 849). In political science, the ‘perestroika’ movement has sought to shift the paradigms towards flexible methods, awareness of competing interests and the deliberative nature of policy analysis (Laitin, Reference Laitin2003; Schram, Reference Schram2004). There have been similar trends in social policy, which has come to be increasingly tolerant of uncertainty, ambiguity and the political character of much of the material it deals with. Researchers have sought to validate a diverse range of non-expert evidence, through a stress on experiential knowledge, testimony and voice (for example, World Bank Poverty Group, 1999; Rycroft-Malone et al., Reference Rycroft Malone, Seers, Titchen, Harvey, Kitson and McCormack2004). Many of our methods are based in action and experience; it is our methodology that needs to catch up. This article is intended to make a contribution to that project.
