Parts (3) and (4): approaches to causation and their problems

Spicker identifies different approaches to causation in step (3). The standard required for causation to be a necessary connection (cf. Hume, Reference Hume1739, Reference Hume1748) fails even in modern ‘hard science’ grappling with, inter alia, the uncertainty principle in quantum mechanics, and chaos theory. Complete determinism is simply no longer credible in any kind of science. If it is required that causes are both necessary and sufficient for their effects, it would be impossible to claim that ‘smoking causes lung cancer’, for instance – because not all smokers contract this disease, and not all those with this disease were smokers. Yet most will agree that it is a causal claim with a great deal of support, tantamount to proof. Hence the adoption of the so-called Bradford-Hill (Reference Bradford Hill1965) criteria for determining a causal association, at least in epidemiology, but also with resonance in other subjects based on observation rather than experiment.Footnote ¹ These criteria include such notions as consistency and plausibility, which involve a degree of expert judgement and yet are clearly aiming at causal explanations.

Therefore perhaps the causal account that most will recognise in their empirical work is the probabilistic account of causation (Humphreys, Reference Humphreys1989). We do not attempt to predict outcomes at the level of individuals, but to judge probabilities and thereby to consider outcomes at the level of larger groups.

One of the key claims made is in part (4) of the argument – that causal accounts may not be trusted. Spicker draws on philosophical claims that ‘cause-and-effect’ may not be trusted by quoting Hume's Treatise of Human Nature (Reference Hume1739). Hume is, of course, arguing against causation in general, rather than in the social sciences. Causation within natural science is also based on shaky philosophical foundations, according to this influential account. Yet, in this appeal to the authority of David Hume, no other philosophical critiques are offered despite the fact that philosophical work on causation clearly did not stop with Hume, but continues with contributions looking at counterfactuals (Lewis, Reference Lewis1973) and at probabilistic causation (Humphreys, Reference Humphreys1989). The idea of entirely determined causation is no longer credible, but that does not mean that causation is itself undermined. Nor are the philosophical issues truly settled.

The main argument deployed to raise scepticism about causal accounts is that causal analysis is difficult because social phenomena are complex. Very little of the article is spent on this argument, but this is the pivot point of much of the whole logic of the piece. This ‘argument from complexity’ has been tackled in at least some of the literature on the philosophy of social science (McIntyre, Reference McIntyre1996: 17, 56). Spicker does not really present much of an argument for this, and merely states the idea of complexity – overall, it would be helpful to consider the evidence for this, and its implications. Certainly, McIntyre's in-depth analysis of this argument reaches a different conclusion to that of Spicker.

Spicker goes on to claim that it is no defence of causal analysis that it provides useful prescriptions for policy. It is certainly true that knowing the nature of a phenomenon provides no direct guidance on what policy should do to tackle social problems. Policy-makers need to know, arguably, what will help jobseekers to find work, and what will help in assisting smokers to stop, rather than knowing who is more likely to be unemployed, or what groups of people are the most likely to start to smoke. But that is to say that the wrong problem is often being studied by researchers, not that causal analysis itself is of no benefit. Spicker's person at the bottom of a well needs hypotheses on what approaches are feasible, and ideally evidence on whether trying to climb out has a greater chance of success than shouting for help.

Part (6): the alleged distinctiveness of selected phronetic generalisations in social policy

There have been a number of calls for a social science based more squarely on phronesis, and Spicker cites Flyvbjerg (Reference Flyvbjerg2001). But in a similar vein we may also list the work of Stephen Toulmin (Reference Toulmin2001). He traced the separation in Western thought into two different strands of ‘rationality’ (scientific, certain) and ‘reasonableness’ (narrative, uncertain), and described how the latter only came to be viewed as superior in modern times. Spicker would find some common ground in this analysis, I think. MacIntyre (Reference MacIntyre1985) also suggested a move towards Aristotelian approaches. These authors differ on key points, but all have made similar suggestions for tackling what they see as the difficulties and crises in social science research.

One of the key issues is how far Spicker's article is essentially applying the terms of the ‘Flyvbjerg debate’ within political science (Part 1 of Schram and Caterino, Reference Schram and Caterino2006) to social policy, or whether it is doing something different to, or beyond, that. The work of Flyvbjerg was influential in political science, and there are a number of responses from that discipline that may be used to critique the idea of a phronetic social science within social policy. For example, Laitin (Reference Laitin2003) conducted a robust defence of scientific method in political science, contra Flyvbjerg. Furthermore, even if a phronetic approach has its merits, why must it be advanced as a candidate for a new dominant methodology which replaces existing approaches, rather than simply adding to the set of approaches at our disposal? (cf. Jackson, Reference Jackson, Schram and Caterino2006).

Moreover, social policy research does not have the same kinds of problems that Flyvbjerg (Reference Flyvbjerg2001) was addressing when he made a comparable argument about more quantitative social sciences (addressing disciplines apparently afflicted either by ‘physics envy’ or ‘economics envy’). Most in social policy would agree that research should ‘address problems that matter to groups in the local, national and global communities in which we live’ and that ‘we must effectively . . . communicate the results of our research to our fellow citizens and carefully listen to their feedback’ (Flyvbjerg, Reference Flyvbjerg2005: 42). This may well contrast with some subjects where ‘quantitative analysis, formal modelling and hard science dominate the leading journals and research institutions’ (Kasza, Reference Kasza, Schram and Caterino2006: 222), but is hardly the case for social policy journals in the UK or an imminent risk (see Smith, Reference Smith2008: 327 for an analysis of Education, Sociology and Social Work). The reverse, a lack of quantitative and statistical capacity, is more likely to be the case (Mills et al., Reference Mills, Jepson, Coxon, Easterby-Smith, Hawkins and Spencer2006).

As an example of phronesis, Spicker quotes the link between unemployment and sickness benefits. He states that this is different from other kinds of causal explanations, but does not provide an argument for this being so. The hypothesis that ‘an increase in unemployment leads to an increase in claims for sickness (or disability) related benefits’ seems as causal an hypothesis as any other, and subject to empirical testing (and may be made falsifiable) – Beatty et al. (Reference Beatty, Fothergill and MacMillan2000) may be considered an example of such an approach. One might imagine there is then scope for academic judgement about how adequately to conceptualise measurements, time lags and the like when researching such a theory. Curiously, he states ‘It is not an explanation of what happens; it is a description’ – which begs the question of how a useful description cannot also invoke some kind of explanation.

Spicker, in fact, provides a set of examples, which are claimed to be non-causal, and all dependent on circumstances. In particular, a lot must hinge on how these studies differ, if at all, from the probabilistic model of causation outlined earlier. Two of the examples included are research into non-take-up of benefits (Craig, Reference Craig1991) and the ‘inverse care law’ (Tudor Hart, Reference Tudor Hart, Smith, Dorling and Shaw1971, my emphasis). Spicker argues that these are expressions of phronesis, and so cannot be said to be causal accounts or explanatory. In the next couple of paragraphs, I say a bit more about these studies, and critically evaluate whether or not they may be said to be causally based rather than examples of phronesis.

Research on take-up has been through a number of generations, with rigorous testing of different theoretical models and predictions. Kerr (Reference Kerr1983) constructed an empirical test of a model based around a set of thresholds rather than trade-offs. More recent research has sought to model the take-up process as related to the ‘costs of claiming’ (Pudney et al., Reference Pudney, Hancock and Sutherland2004). In each case, there were well-formed hypotheses which were then subject to empirical testing. The analysis of incomplete take-up of social security benefits is now a mature sub-field of investigation. For instance, it is well-established that the higher the level of entitlement, the greater the rate of take-up Oorschot (Reference Oorschot1995).

The latter example of an inverse link between the need for health care and its provision is presented as a ‘law’ rather than as a finding based on particular circumstances (Tudor Hart, Reference Tudor Hart, Smith, Dorling and Shaw1971). This law states that:

Tudor Hart marshalled a range of historical and statistical resources to substantiate his ‘law’. He did note that definitive statistical proof was not possible, because relevant data were not always available, but it seems part of his aim (and past studies provided considerable support).

It is argued that these kinds of statements are not universally applicable, but instead depend on the circumstances. But I would argue that no-one presenting a causal view of the world is unaware of the importance of context and circumstances. In arriving at conclusions, it is actually helpful (rather than undermining) to be able to look at different contexts, and incorporate the variation across different circumstances into any model. This may even be more formally described within the ‘most different’ nation approach to comparative study (Przeworski and Teune, Reference Przeworski and Teune1970). Aristotle himself was interested in classifying political systems based on their differences.

Part (7): distinguishing good from bad generalisations

In some respects the last point is among the most interesting and revealing of the approach. Clearly, Spicker is somewhat troubled by the implications of where his reasonings have led him. Without a ‘hard’ datum of evidence within a framework of cause-and-effect against which to test hypotheses, how can we reject (as Spicker wishes to) or refute the view of the likes of Charles Murray (Reference Murray1984) concerning the ‘underclass’? Once we move away from the standard deductive approach of formulating an hypothesis subject to empirical testing (‘falsifiable’), who is to distinguish which kinds of experiential phronetic generalisations may be relied upon from those we may wish to reject?

At the close of the article, Spicker is clearly concerned about how to separate what he sees as good or valid generalisations (for example, selective policies have incomplete take-up) from bad generalisations (such as the cycle of deprivation). Here his language betrays a return to a reliance on universal principles. Having argued that generalisations cannot take the form of laws, but are inevitably limited to context and the particular, and to judgement, he declares that some generalisations are nevertheless untrue. This is a strong universal claim, rather than one designed to apply in particular circumstances or only in certain contexts. But if generalisation is always subject to the kind of context and particular circumstances argued for elsewhere in the article, how is it possible to claim a generalisation of this kind is untrue (assuming it is not internally contradictory)? Of course, the answer is that it isn't possible.

The advice of experienced practitioners is not always to be trusted. In Baby and Child Care (Reference Spock1946), the renowned child care expert Dr Benjamin Spock said ‘it is preferable to accustom a baby to sleeping on his stomach’. Chalmers (Reference Chalmers2001: 322) described this as ‘apparently rational and authoritative advice’, but which ‘led to thousands, if not tens of thousands, of avoidable cot deaths’ (see Gilbert et al., Reference Gilbert, Salanti, Harden and See2005) Changes in the advice given to parents roughly halved the rate of cot deaths. When there is a conflict between the advice of practical wisdom and reliable empirical evidence, as in this case, it is clear which must give way.