Purposive Action Under Conditions of Unpredictability: Lessons from Development Practice and Some Suggestions

I. The Issues

Most, likely all, of those involved in the generation of ideas relevant to development practice would not assert that their knowledge is predictive. That is, they do not say, or believe, that they reliably know and can therefore guarantee that X will lead to Y. Most, likely all, however, would assert that some theory of change, or causal relationship between actions and outcomes (that X will lead to Y), should underpin development work, at least formally, and their knowledge is suitable to this (though ‘nothing guaranteed’). This is, for some, including me, more than a little contradictory. Devices like theories of change or the log frame deploy assumptions about instrumentality that require knowledge of causal relationships. But this contradiction, if that is a fair description, seems to be robust and of long duration.

Arguably, this is reflected in the contrast between textbooks about knowledge areas that deploy accepted predictive knowledge, such as bridge design, and those used to teach students of development: The latter are in essence a history of competing beliefs about what will lead to what, whilst the former are not. ¹

This contradiction is generally resolved, I think, by asserting that reliance upon the causative logics in specific ideas relevant to development practice is to ‘suggest’ that X will lead to Y. This leads I think to three issues that are not generally appreciated:

First, these logics, if deployed into instrumental actions, are not robust guides (‘nothing guaranteed’). This inevitably follows from the fact that they are not predictive. ²

Second, the reality within which development practitioners work is, then, one where beliefs in causative logics (such as theories of change) as guides to the formal organization of instrumental action are problematic: Not being predictive, they are not robust guides to what will happen, so unexpected consequences are likely, if not unavoidable. Further, they must coexist with other beliefs in a situation where differences cannot be resolved easily by reference to agreed facts (which tends to be the case if knowledge is predictive), so they tend to result in familiar tensions, such as evaluations that cannot easily be based upon single truths and are therefore contested. Managing the politics of these differences well is arguably the essence of good development practice, or, in other words, working non-instrumentally. But formally, development work is meant to be organized instrumentally. Therefore, there are tensions between form and content.

Third, (in part confirming these tensions between the form and content of development practice) there is very little thought-through and formalized understanding of non-instrumental actions, as can be seen from any search of the relevant knowledge databases. ³ The view that interventions should be organized formally as instrumental sucks oxygen from discussion of alternatives.

I suggest that one way to start getting around this is by developing empirically based arguments that, in each context, advise for or against instrumental action using causal logic. This allows for greater discussion of the possibilities of organizing non-instrumentally. In other words, I think that better practice can be obtained by, first, improving our understanding of this situation, so, for example, we should be able to assert that certain development contexts are unsuited to reliance upon beliefs in a reliable causal logic, and second, based upon that understanding, focussing upon research into development practices that do not rely upon any idea that there is a known causal logic (that is, non-instrumental action), that may then be thought-through and formalized. This would allow focus upon the (non-instrumental) content rather than the (instrumental) form of practice.

My suggestions go well beyond by now rather old ideas that we need to support causal logics that are better as they come from a better place, such as the ‘poorest of the poor’ (Chambers, 1983), or to realize that suggested causal logics are socially constructed (I take Mosse (2005) as one example amongst many). This is because, consistent with the general and longstanding view that development interventions should be instrumental, these and other approaches still assert the existence of causal logics, that X will lead to Y, that are well enough known to be reliably deployed into instrumental action.

II. Understanding the Situation: Expanding the Argument

This commentary is, thus, trying to make suggestions. Some of the arguments are presented ‘in full’ elsewhere. ⁴ These come down to the ideas that, first, we need to take non-instrumental action far more seriously, and second, that the way to do so is to re-assess development practice, for all that it is formally instrumental, as essentially non-instrumental.

It would seem self-evident that development interventions are generally thought of as ‘instrumental’—do this, and that will happen (X will lead to Y). The log frame is a particular example. Generally, experts deploy ideas of causation to support practitioners’ use of resources to secure outputs and outcomes. This is an ‘instrumental logic’. Clearly, there are other approaches, as people can seek better situations without knowing beforehand exactly how to get there or agreeing afterwards on what caused what to happen. But I think it is self-evident that knowable instrumentality is fundamental to how development is construed, and how it is (generally) meant to be done (its formal organization). Yet, most development thinkers assert that their knowledge is not predictive. Clearly, this position is very reasonable, as any discussion with natural scientists about method will show—why they seek and test theories for predictive power, and how starkly this differs from what development thinkers do. In this sense, a search for predictive power can be usefully seen as a criterion for gauging theories that are present in some sciences, typically natural science, but not in others.

If developmental ideas self-identify as not predictive, yet development practice is instrumental, then this is—I think—a contradiction. In my experience, this view is heterodox. The tension can be managed by treating development ideas as good guesses or reasonable expectations, but this, arguably, is an attempt to accommodate the formal instrumental focus of development practice, that which is required by donors and by communities of practice. In effect, even if developmental ideas take the form, through their causative logics, of suggesting instrumentality, it may not be good enough to argue that the instrumentality is known with uncertainty, so ‘nothing is guaranteed’. Indeed, it is, I think, worth pointing out that predictive knowledge also does not guarantee anything because it tends to make assumptions that end up with its predictions being probabilistic. Thus, a degree of error comes with the package and can probably be reduced (at a cost) by more detailed calculations and more spending on measurement. It is clear that such language is not deployed when discussing development theories. It is surely rare that people pose the question ‘to what extent can we guarantee that X will follow from Y’ or suggest that with a larger sample or more detailed thought (which costs more), greater reliability may be obtained.

Strikingly, there is also little thought-through discussion of non-instrumental action. The point can be made that much development practice, whilst taking the form of instru- mental action, given that its associated beliefs are not predictive, is surely in content non-instrumental. If ‘nothing is guaranteed’, even probabilistically, then practitioners live in a world of unforeseen outcomes, and the problem of which guarantor to listen to. One way of putting this is that whilst the form may be instrumental, the content is not.

Tensions accompany the fact that develop- ment thinkers assert that their knowledge is not predictive. For example, they often privilege their own causal logics, as is required by standard principles for intervention design, and the required instrumentality.

Consider the article by Mosse (2005), by a well-known thinker who would not assert that their views were predictive. He critiques donor knowledge as socially constructed, and not predictive. For instance, he assets that

Meetings with top government officials, aid agencies, NGOs and academics establish the prevailing rural development discourse, its key players and those who need to be enrolled as supporters. (p. 27)

And:

[P]rojects have to be conceived of as predictive models in which the elements are systematically and causally related, and where outcomes of actions are certain. (p. 37)

He sees this as deeply flawed, for

The development policy models through which resources and political support are so successfully mobilised are rarely those best suited to understanding the social and historical context of development action. (p. 47)

This suggests that alternative models are better suited to informing development action.

In this critical reflection on a development project in which he was an adviser, he sees knowledge generated by participatory methods and anthropological knowledge as offering a better understanding of the context for development action. For instance, he states that ‘Bhil identity was and is the product of relationships with outsiders’ (p. 47). So ‘When given the opportunity to increase agricultural earning, those who could afford to would divert their energy from migration to their land’ (p. 72, emphasis added). Similarly, he asserts that causal logics arrived at through participatory methods are a good basis for developmental action. So, despite critiquing policymakers’ models as predicative, he puts forward the causal inferences of anthropological and participatory knowledge as being better for informing instrumental action.

There is nothing exceptional about the story Mosse tells: it is familiar. For me, there is a tension here between the assertion that developmental knowledge is not predictive and the deployment of its causal logics into instrumental action. Clearly, it is correct to say that such knowledge is not predictive, but it is also correct to say that they make claims about the validity of their causal logics that are vital to the foundations of the instrumental rationality of standard project designs.

It may clarify what I am trying to argue by re-considering how some scientific methods generate what is generally understood to be predictive knowledge.

Scientific theories, whether natural science or not, are generally constructed based on what is known empirically by the theorist, considering existing theories and their relative empirical success. This can be described as an ‘inductive’ stage. Theorists can be expected to try hard to match their theories to relevant data, though at this stage this is not done by seeking predictive power by empirical testing of deductions from theory. Rather, as such theories compete with others, perhaps with equally confident theorists, predictive method, if deployed, adds, to requirements that theories be logically coherent and respectful of relevant facts, another criterion as to how theories are to be compared. This is done by seeing whether deductions from them are predictive, which requires judgements about measurement and assumptions to ground theories in what they purport to be about. ⁵ In what is generally understood to be the natural science method, the requirement that they be tested predictively can be thought of as a requirement that they be tested against each other by their ability to manage data. Clearly, this requirement is part of some scientific methodologies, typically ‘natural science’, but not all.

This, equally clearly, is not what happens when people choose between developmental theories containing causal logics, each of which states, in effect, that there is a reasonable expectation if we do X, Y will follow. To repeat, with development interventions construed as instrumental, and required formally to be designed as such, beliefs in causal logics must underpin instrumentality, even if accompanied by assertions that ‘nothing is guaranteed’.

Even earlier, Chambers (1983, 1997) argued that through rapid or participatory rural appraisal (RRA or PRA), better knowledge would be acquired ‘and the last put first’. Chambers did not assert that through what he proposed predictive knowledge would be achieved. Yet, then as now, these preferred Knowledges had causal logics and these, he said, not those of the experts he was attacking, should be deployed instrumentally. They were better. Even if ‘nothing was guaranteed’, activities X would (generally, even if not always…) lead to outcomes Y. As ever, the instrumental design of interventions requires the deployment of a causal logic.

Contemporary scholars such as Eyben (2019) grapple with broadly the same problem some 35 years later. The problem, in my view, is the tension between assertions that knowledge is not predictive and the deployment of them into instrumental action. If there is to be a reliance upon instrumentality, then I think we would expect that the focus of those analysing processes of knowledge construction would be to improve the links between instruments and outcomes. Yet the focus, in some recent work, is elsewhere—upon the sociology of practice. This continues the point I made above, which is that, unlike practices based upon knowledge that self-identify as predictive (say, ‘bridge-building’), relevant textbooks on development in effect teach histories of development practices and their relevant ideas (‘Structural Adjustment’, ‘Rights Based Approaches’, ‘Participation’, etc.).

Recent work supports this interpretation. For example, Dar (2014) argues, in a critical vein, that aid recipients’ accounting practices are under pressure from donors to comply with how donors understand the world, and so he understands them by treating them as contestations over discourses. Kenny (2012) also explores how ‘people engage with powerful discourses’ (p. 1175). One does not find such foci in studies of engineering (though one does in studies of engineers).

To manage this situation better, I suggest that we stress the deep and pervasive influence of instrumental logic in the norms of most formal development practice, for it maps, as is obvious, into organizational requirements—how aid ‘should be done’. Those who understand that their theories and knowledge are not predictive see their causal logics deployed into practices that use them to argue that, though ‘nothing is guaranteed’, X will lead to Y. The contradiction is evident in Mosse (2006), where he reports the donor’s unhappiness when X did not lead to Y: In my reading, it was felt that the project failed to meet its stated targets, even though it also seemed that the project had (more or less) got process right.

It is worth considering the formal aspects of development practice in more detail.

One important expression of the formal instrumentality of mainstream development practice is the ‘log frame’, part of the ‘logical framework approach’ (LFA) central to the normative position of the OECD’s DAC. ⁶ I do not think I need to explain to readers in detail what this approach is and how it is meant to work. There is much empirical work still to be done. But the ‘log frame’ is, as is well known, an organizational tool that is inherently instrumental. It requires interventions to link a set of inputs, through a matrix, to a set of outputs. The matrix chosen is meant to reflect evidentially based research that supports two core assertions: first, that the inputs can reasonably be expected to lead to the outputs; and second, that these outputs can reasonably be expected to lead to certain developmental outcomes. Even if ‘nothing is guaranteed’, the role of causal logic in development thinking is crucial and essential. So, to return to the example of Mosse (2005), the causal logic of ideas he supported—such that the Bhil would invest in working their land rather than migrating once they were able to generate better incomes from farming identity—became, in effect, part of the justification for the ‘X will lead to Y’ formal instrumentality of the project. It seems clear to me from Mosse (2005 and 2006) that whilst in form it was instrumental, the actual content of the project was not.

It has long seemed to me, and this is a heterodox view, that this means that known predictability is thus essential to the LFA, both in the ‘internal’ relationships between inputs and outputs and the ‘external’ relationships between outputs and outcomes. It is then striking to me that the associated knowledge self-identifies as non-predictive, yet its causal logics are deployed instrumentally. The equivalent, in natural science, would be to use empirically untested theory as a basis for design, for example, a bridge. The issue is that of the requirement that development practice be instrumental. For me, perhaps reflecting a STEM background and long experience in development practice, ⁷ it makes more sense to take the bull by the horns and say that knowledge that is not predictive should not be deployed into instrumental actions.

But—so what?

I fear that, whilst a predictive criterion for choosing between theories is part of scientific methodology, there is no equivalent criterion in managing contestation between development ideas, and that this probably helps explain the patterns of contestation—the fads, interplay of agency interests and tendencies, as the quotes from Mosse above argue, for ‘those who pay the piper to call the tune’. This all comes out in the content of relevant textbooks (such as Willis, 2011).

The contradiction (or tension, if the reader prefers) between knowledge that is identified as not predictive and practices that are deemed instrumental can be explored by considering the relative importance attached to results compared with process. I think it is self-evident that many development practitioners, close to their local context and working with knowledge that is more based upon trial-and-error than generalized causative theory, would often prefer to be given the money and the desired outcomes and then left alone—that is, to be judged by results. Working in a world of multiple truths is not usually made easier by the single truth designed into a project by the log frame or a theory of change, so that the (I would say non-instrumental) content differs from the instrumental form. What is going on is reported as the deployment of resources to generate certain outputs and outcomes, but something else is happening.

So, whilst it is obvious that the LFA has on the surface affinities with results-based management (RBM), it is striking that it also requires, as strict RBM avoids, attention to securing valid processes. The log frame and theories of change are about process—how activities lead to outputs and then outcomes.

There are some striking elements to this when considering the content of development work.

First, RBM tends to assume ‘single sovereignty’: that is, whether applied to a corporation or to relations between a supplier and an agency that receives services from that supplier, it is assumed that a single authority exists to determine ‘the truth of the matter’. This assumption clearly does not generally apply to aid work, except in unusual circumstances. Nor does it apply to situations where some stakeholders do not accept the authority of the ‘single sovereign’ (such as often happens with First Nations people). It is for this and other reasons that experienced practitioners refrain from asserting ‘truths’, so far as is possible, but of course this is constrained by the methodologies of instrumental actions with their requirements for a single causative logic. Content differs from form. Strikingly, theory of change methods usually require a single theory at the level of the project, thus inhibiting the formal management of multiple truths; it is hard to avoid privileging the causal logic of the intervention design when confronted with alternatives.

Second, there is the issue of counterpart resources. As is well known, whilst adequate to formally define the intervention, organizationally, the intervention logic may not be able to secure the outcomes its causal logic envisages. Those with other views generally possess resources that intervention needs and cannot itself supply. Again, content differs from form, and what is formally designed instrumentally is, in practice, better seen as non-instrumental.

Development activities usually require others’ inputs—will villagers attend the participatory meetings? Logically, this is outside the LFA, which manages the project’s own identifiable resources, and these cannot include all those used. Negotiation of such issues is a central part of the content of aid practice. Interventions typically, as practitioners know well, contain a range of views of what is happening and why, and indeed, some evaluations present accounts of these (Fujita, 2010): In such framings, the logic is not instrumental. But standard evaluation methods require conclusions to be reached in ways that imply single answers—‘Is the intervention doing the right thing? Is it achieving its objectives?’. This simply reflects the formal assumption of instrumentality.

III. Possible Alternatives?

I think it is clear that formal instrumentality, in a situation where relevant knowledge self-identifies as non-predictive, remains central to development practice. Dealing with this is the ‘nettle to grasp’.

In a review of theory of change for DFID, Vogel (2012), in looking for a pointer to general agreement on its definition offers Davies’ (2012) formulation that it is: ‘The description of a sequence of events that is expected to lead to a particular desired outcome’ (2–3 citing Davies 2012…). More strikingly, she places great emphasis upon the theory of change, making clear its envisaged ‘causal links’ (pp. 33, 35, 46 and 51). She makes it clear that the donor agency requires that these be exposited and given some empirical foundation. But they are not, explicitly, meant to be rigidly predictive (p. 6) but supportive of a dialogue ‘about what is realistic and feasible’ (p. 19). In other words, the intervention should have one and only one, theory of change, but be aware that there are others. The theory of change can change over time but remains singular. In any search for a politics of the intervention that seeks to manage differences, this insistence on singularity (at any point in time) can only be expected to get in the way: The basic proximate reason for this, in my view, is the formal retention of instrumentality. If the theory of change can be readily changed, then it is not consistent to argue that the intervention is instrumental, founded on reasonable beliefs that X will lead to Y. It is something essentially different.

Thus, we find the OECD DAC defining theory of changes as:

[U]nderstandings about why particular inputs or activities are expected to achieve intended results (p. 80)

This is instrumental logic, and so a singular theory of change, and this is also shown by widely cited works such as Funnell and Rogers (2011) and Blamey and Mackenzie (2007). ⁸ For the latter, favourably quoting Connell and Kubisch (1998):

A theory of change approach would seek the agreement from all stakeholders that, for example, activities A1, A2, and A3, if properly implemented and with the ongoing) presence of contextual factors X1, X2 and X3) should lead to outcomes O1, O2 and O3. (p. 445)

Should one give up if this agreement cannot be reached? I think not. But it is clear that a theory of change is, for them, an agreed-upon single truth, or belief or hope about what the intervention instrumentality is to do:

To elicit the Theory of Change underlying a planned programme, the evaluator works with a wide range of stakeholders in a collaborative manner. (p. 43, emphasis added)

What does this lead us to?

For me, the most basic conclusion is that formally organizing interventions based on instrumentality in the absence of reliable causative knowledge is a mistake. Attention needs to be paid to treating interventions as non-instrumental, not least to protect development theorists from over-reach, as donors deploy their causal arguments into organizational devices such as the LFA or theory of change, despite being told that they are not predictive.

I have two related suggestions:

First, it is useful to formally decide whether—or not—adequate knowledge is attainable (at a reasonable cost), and so whether instrumental action is possible, or impossible. Once posed, the options available, or that can be created, to manage such decisions are, I think, rather obvious. One is simply gathering a range of people with different interests who are familiar with the context and asking them whether or not they want to organize instrumentally. This is to ask them to make a judgement about whether they know enough about what causes what to proceed on that basis, or not? If not, then, in effect, they are to deploy non-instrumental action without asserting that they know what will cause what. This decision can obviously be managed in different ways, but it can also be articulated more formally—Fforde (2020) offers such a more formal approach, developing known statistical methods to create a decision process to assess the regularity of a context and the likely value of deploying researched causal logics into instrumental actions. The central issue is that if the context is not one where a reasonable basis exists for agreeing on there being a single acceptable causal logic, then interventions should be formally organized as non-instrumental. If it is not (yet) clear how formally to do this, then there is a job of work to be done to work out how. This is likely easier than it may appear. This leads to my second suggestion.

Second, because development knowledge are not predictive, as we all accept, development interventions, whilst formally instrumental, are generally in terms of their actual content non-instrumental, and should be accepted, designed, evaluated and researched as such. Unlike, say, bridge-building, they are not about the smooth deployment of a causal logic into securing, reliably, outputs from given resources. Rather, they are about managing worlds of multiple perspectives, truths and values. Whatever that world is, it is not one of instrumental action. For me, the obvious way forward is, having framed what they do in those terms, to research how practitioners and their advisers have been coping with non-instrumentality. For example, how do they manage a world of multiple truths, most of which are respectable, when the causal logic of the log frame or the theory of change, despite what the project documents say, coexists with other beliefs, hopefully in a positive manner (and much effort is often put into securing this rather than acrimonious dispute)? There is far more to be learned from this than the anthropologists’ common conclusions that there is a contrast between form and experienced meaning. Whenever the theory of change is not predictively valid, action—whatever is said about it—is actually non-instrumental action. Therefore, I suggest that the way to go is the formalization of (probably a large part of) existing development practice as non-instrumental action. Some linguistic changes are obvious: rather than talking of results or outcomes, which imply a knowable causality, we could instead find another term—perhaps ‘subsequent states’—that does not presume an interpretive singularity. Further though, such a formalization would require us to take a determined step away from the idea that there is a causal logic that is correct for everybody and to recognize that whilst the same event will likely be seen and experienced differently by different people, this does not necessarily stop them from enjoying shared ‘subsequent events’. The JICA review (Fujita, 2010) would seem squarely in this camp in its approach to both what can be agreed to be known, and the limits to this, and in its reported widespread liking of the ‘subsequent event’. Which is not to say that participants cannot agree, just that if they do not, good work can still happen.

IV. Postscript

Although space is limited, anonymous reviewers usefully raise two questions that I do not answer, but comment upon.

First, why does the situation I have addressed persist and exist? There is a vast literature on aspects of this. I would mention two points. Development practice is part of far wider practices, which include the very ‘20th century’ assertion of the positive value of expertise in policy development. Much of this literature, I think, tends to conclude, in various ways, that this valuation was wrong. We are living with the consequences.

Second, what is non-instrumental action in practice? Here it is clear that, as such, we have very little formal knowledge, as can be seen by a search for relevant research. However, given the apparent relative lack of knowledge suited to instrumentality, we must, as we do it so often, know much about non-instrumental action, but this is informal. Cash programming, for example, will, if done well, require processes for agreeing (or not) on whether different stakeholders agree (or not) that ‘subsequent events’ are worthwhile, and discussing different accounts (which may or may not agree) of how they happened. Any experienced practitioner can, if asked (and they should be), offer suggestions as to how such processes can be better organized.