Structural Injustice and the Limits of Blame

Introduction

Public responses to unrecorded alcohol in Kenya have long combined policing with exhortation. Crackdowns on production and sale, arrests, the closure of distilleries and awareness campaigns are the instruments described in the survey study on which this article draws (Cherpka, 2024). Behind them lies a recognisable moral picture. Producers and sellers are wrongdoers; consumers are agents who knowingly chose a prohibited and hazardous product; and responsibility for the resulting harm belongs to whoever can be shown to have done the choosing or the selling. Young (2011) calls this the liability model of responsibility. It is backward-looking, it isolates an agent whose conduct caused a harm, and it pairs that agent with blame and sanction.

Unrecorded alcohol is not a marginal phenomenon. It is defined by its position outside taxation and governmental control, and global modelling attributes about a quarter of adult per capita consumption to it, with shares of 43.1% in low-income and 54.4% in lower-middle-income countries (Probst, Manthey, et al., 2018; Probst, Fleischmann, et al., 2019). Its moral interest lies in what that position implies: products that escape quality assurance, licensing and tax reach consumers who cannot check what they are drinking.

The model is so familiar that its fit with a given domain is rarely examined. That examination is the purpose of this article. The question is not whether drinkers choose; they do. The question is whether the pattern of their choices, taken as a whole, is intelligible as an aggregate of separate wrongs or as the output of a structure that channels people towards a hazard. The distinction carries moral weight. Where the liability model is misapplied, it produces blame that is unfair to those who carry the heaviest burdens, shelters actors whose positions give them real power over the conditions of supply, and aims remedies at the wrong target.

The article defends a bounded thesis. In the case examined, quantitative survey evidence meets the conditions that Young’s account sets for structural injustice and defeats the empirical presuppositions on which the liability model depends. Responsibility is therefore better understood through her social connection model, as shared, forward-looking and differentiated by social position. The principal ethical element at issue is the attribution of responsibility: who may justly be blamed, and who is obliged to act. A secondary element follows from it, namely the injustice of unequal exposure to products whose contents consumers cannot verify.

Two literatures rarely meet on this ground. Philosophical work on structural injustice is theoretically refined but seldom tested against measured data from African urban settings, while public health studies of alcohol seldom engage normative theory. This article uses a single dataset to bring them into contact. Section 2 sets out the theory and reconstructs testable conditions from it. Section 3 describes the evidence and its limits. Section 4 develops the argument. Section 5 distributes responsibility among actors, Section 6 answers objections, and Section 7 draws normative implications.

Theoretical frame: Young’s social connection model

Young (2011) holds that structural injustice exists when social processes expose large categories of people to systematic threats of domination or to deprivation of the means to develop and exercise their capacities, while the same processes enable others to enjoy wide opportunities. Its distinctive feature is that no single act or policy need be wrong for the injustice to exist. Structures consist of the accumulated and mutually adjusting actions of many agents who follow accepted rules and pursue reasonable interests. McKeown (2018) distinguishes several readings of the ‘connection’ on which responsibility rests; this article assumes only that agents who contribute to the processes concerned are connected to them.

Against this background the liability model asks who caused the harm, whether they acted voluntarily and knowingly, whether background conditions were fair, and what sanction fits (Young, 2011). Its legitimacy rests on three presuppositions, labelled here L1 to L3. L1: the conduct to be assessed is traceable to a choice that expresses the agent’s own reasons, such as preference or appetite. L2: a culprit, or a identifiable set of culprits, can be found whose conduct is a sufficient explanation of the harm. L3: the background conditions against which choices are made are broadly fair. When these fail, applying the model does more than lose accuracy; it misallocates blame.

The social connection model replaces it. Responsibility on this model is not a matter of fault, is forward-looking and concerned with changing structures, is shared among all who contribute to the processes, and can be discharged only through collective action (Young, 2006, 2011; but see Beck, 2023, on whether responsibility and blame can be so sharply separated). The sharing is not equal. Young offers four parameters of reasoning to differentiate it: power to influence the processes, privilege in benefiting from them or being insulated from their costs, interest in changing them, and collective ability to organise effectively.

Why Young and not a rival? Role-based alternatives such as Zheng’s (2018) refine who should do what, but they presuppose the structural diagnosis that Young’s model supplies. Relational accounts of autonomy (Mackenzie & Stoljar, 2000) explain why socially shaped choices remain the agent’s own, but they do not allocate duties. Young’s model takes background conditions seriously and answers the question of who must act. It is also open to empirical test, because its conditions can be stated in observable terms.

I reconstruct those conditions as follows. The operationalisation is mine, offered as a defensible reading and not as Young’s own formulation. C1 (patterning): harm-linked behaviour is distributed non-randomly across positions in the social order, so that disadvantaged position predicts exposure. C2 (diffuse production): the behaviour and its supply are sustained by many actors whose individual conduct is ordinary or only weakly wrongful, with no single sufficient culprit. C3 (constrained capacity): those affected report conditions that restrict their capacity to develop and exercise their capabilities and treat the behaviour as a response to those conditions and not as an end sought for its own sake. If C1 to C3 hold, L1 to L3 fail in corresponding places: C3 undermines L1, C2 undermines L2, and C1 together with C3 undermines L3.

Evidence base and its limits

The empirical base is a survey of current and former consumers of unrecorded alcohol in Embakasi East Sub-County, Nairobi, conducted for a master’s project at the University of Nairobi (Cherpka, 2024). Respondents were recruited through purposive and snowball sampling within seven clusters selected with probability proportional to size, a design suited to a stigmatised and hidden population. Tabulated analyses use 118 respondents, of whom 87 (74%) were active and 31 (26%) former consumers. A structured questionnaire recorded demographic characteristics, frequency of outlet visits, quantity consumed per visit, awareness of producers and outlets, and five-point ratings of reasons for drinking across three blocks: social norms, family circumstances and social environment. Table 1 profiles the sample.

Source: Cherpka (2024), Tables 4.1–4.4 and Figure 4.2.

Four limits govern what these data can bear. First, the sample is non-probabilistic, so the figures describe patterns among these respondents and do not estimate population prevalence. Second, the design is cross-sectional and self-reported, which permits association and forbids causal claims; Section 6 returns to this. Consumption of unrecorded alcohol is hard to measure even at national scale, where estimates rest on expert assessment and modelling (Probst, Manthey, et al., 2018), and a study of two Kenyan towns could offer no definitive evidence on consumption levels (Mutisya & Willis, 2009). Third, the quantity item is not fully coherent. The descriptive table reports a maximum of four litres per visit, while the cross-tabulations carry a band labelled ‘more than four litres’. This article therefore reads the bands as ordinal, as respondents’ relative placement as heavy or light consumers, and not as literal volumes. Fourth, several inferential statistics in the source, including parts of the repeated-measures and regression analyses, could not be reconciled with the tabulated frequencies, and they are not relied upon here. The evidence used consists of row percentages and mean ratings that any reader can recompute from the source tables. The claim is correspondingly modest: the article identifies patterns that a normative account must explain, and it does not estimate effects.

From data to argument

Condition 1: patterning

Heavy per-visit consumption is steeply graded by material position. Table 2 reports, for each income band and education level, the share of respondents placing themselves in the highest quantity band.

Source: Cherpka (2024), Tables 4.14 and 4.15 (row percentages).

Half of respondents earning less than KES 10,000 a month, and 47.1% of those earning KES 11,000–30,000, reported the highest band. The corresponding figures were 9.7% in the KES 31,000–40,000 band and 5.9% in the KES 41,000–50,000 band. The ratio between the lowest and highest income bands is about 8.5 to one. The shape is a step more than a slope: the sharp fall occurs at roughly KES 30,000, below which 45% of the sample falls (Table 1). Education shows the same ordering. Respondents with no formal education or with secondary schooling were roughly three times as likely as undergraduates to report the highest band, and none of the 35 respondents holding a diploma or post-secondary certificate did so. Those last groups are small, and the figures are indicative, not precise.

Sex shows a smaller gradient. Thirty per cent of men and 23% of women reported the highest band, while 11% of men and 32% of women reported the lowest. With 74% of the sample male, this fits the normative pressures discussed in Section 4.3 and shows that income is not the only axis of patterning. Condition C1 is satisfied in the sense required: position in the social order predicts exposure. The pattern is consistent with, though it does not test, the alcohol-harm paradox: a systematic review reports that people of low socioeconomic status suffer disproportionately greater alcohol-attributable harm (Probst, Kilian, et al., 2020). Patterning alone does not establish injustice, since poorer people might simply prefer more. This is why C2 and C3 carry the argument.

Condition 2: diffuse production

If a few culprits sustained the market, L2 would be defensible. The data point the other way. Of respondents rating home-brew sellers, 53% described them as operating independently, 27% as acting in complicity with others and 19% as an organised group. Organised operation is more common among sellers of counterfeit spirits (41%) and repurposed industrial ethanol (34%), which is where a culprit-centred response has its best claim to fit. Yet even there, about three in five and two in three sellers respectively were described as independent or complicit and not organised.

The physical infrastructure is ordinary too. Respondents located home-brew sales mainly in rented houses (41%), with a further 15% in abandoned buildings and 18% in the flats or houses of the sellers. Counterfeit and industrial products were placed chiefly in local bars and clubs (30% and 41%). The rental housing market and licensed premises are therefore part of the supply chain, though not as wrongdoers in any ordinary sense.

On the demand side, the factor most often chosen as the main social-environmental driver was the price of unrecorded alcohol relative to licit alcohol (37.3%), followed by the distance travelled to find it relative to licit alcohol (22.0%). Together they account for 59.3% of respondents. On five-point agreement scales, respondents gave the highest mean to the statement that unrecorded alcohol is cheaper and affordable (4.79), followed by the presence of clubs, abandoned buildings, farms or houses acting as outlets (4.60), the normalisation of drinking at social functions (4.48) and lax enforcement of alcohol control (4.22). Seventy-seven per cent of respondents knew where producers operated, and 37% said they learned of sale locations through social media, ahead of word of mouth (23%). Reviews from high-income settings find that outlet density is associated with higher consumption and harm (Campbell et al., 2009), which suggests a plausible mechanism for the availability findings, though not one demonstrated for this setting.

No actor in this picture is a sufficient cause. A landlord letting a house, a platform carrying a post, a retailer pricing licit products, an official who does not patrol, and a household serving unrecorded drink at a social function each act within ordinary rules, and the market results from their interaction. This is the profile of a structural process in Young’s sense. L2 fails for the bulk of the market, though not for every seller: organised counterfeiters remain apt targets of liability, which the social connection model does not deny.

Condition 3: constrained capacity

Table 3 summarises respondents’ mean ratings (1 = least important, 5 = most important) of the reasons they drink or drank.

Source: Cherpka (2024), Tables 4.18, 4.28 and 4.29; means of item means computed by the author. Item standard deviations in Table 4.18 range from 1.0 to 1.3.

The contrast is large. Hedonic reasons cluster around 2.1, while the other blocks average between 4.05 and 4.3. A gap of about two scale points exceeds one and a half standard deviations at the item-level dispersion reported. The highest single items were the lower price of unrecorded alcohol (4.79), the worsening of family finances (4.77), school dropout (4.64) and the availability of outlets in abandoned buildings and farms (4.60). In the norms block, drinking in order to forget financial or social problems and to affirm cultural or traditional values each scored 4.5, with asserting masculinity or femininity at 4.2 and peer loyalty at 3.9. Three family items scored lower: child-headed households (3.50), domestic violence (3.42) and single parenthood (3.06). The family mean of 4.05 includes them, so the pattern is not uniform, and the highest-rated items concern money and schooling.

Two features of this evidence matter. First, respondents did not describe drinking as the pursuit of pleasure. They described it as a way of living with deprivation and of meeting expectations. Former consumers rated ‘forget financial or social problems’ at 4.85, higher than active consumers at 4.32. The reason is at least as salient to those who have stopped as to those who continue, which is consistent with its naming a condition that outlasts the habit. Second, 80.5% of respondents agreed that the social and family environment encourages tolerance of unrecorded alcohol.

C3 is satisfied: respondents explain their conduct chiefly by constrained circumstances and not by appetite. L1 therefore fails in the sense relevant to blame. The choices remain reason-responsive and remain the agents’ own (Mackenzie & Stoljar, 2000), but the model’s picture of a free preference for a prohibited good does not describe them. Pickard (2017), writing on addiction (a category that for her includes alcohol), acknowledges choice and agency while rejecting blame and stigma, and joins that stance to a commitment to social justice. The survey evidence suggests that pressure of this kind is, for much of the sample, the normal condition of life. Experimental and field evidence points the same way: financial concerns reduce cognitive performance among poorer participants (Mani et al., 2013).

The argument in standard form

The ‘proof’ here has a particular character, and it should be stated plainly. The empirical premises are supported by measured regularities, but the step to C5 is a judgement that those regularities satisfy conditions drawn from a normative theory, and the step to C7 is a conditional: if one accepts the theory’s principle that structural injustice grounds forward-looking responsibility, then the evidence engages it:

P1.  Heavy consumption of unrecorded alcohol is patterned by material position (Table 2).

P2.  Supply and access are sustained by many actors who act independently or within ordinary rules, and not by a few identifiable culprits (Section 4.2).

P3.  Respondents explain their drinking mainly by deprivation, coping and normative pressure and not by appetite (Table 3).

P4.  Just over half of respondents (51%) mainly consume counterfeit branded spirits or repurposed industrial ethanol, product types whose contents the consumer cannot verify, and the heaviest consumption is concentrated among the lowest-income respondents (P1).

C5.  From P1 to P3, conditions C1 to C3 hold, so the conditions of structural injustice are met and L1 to L3 fail for the bulk of the phenomenon.

C6.  From P4, the burden of an unverifiable hazard falls most heavily on the least advantaged, which is a distributive injustice independent of anyone’s blameworthiness.

C7.  Where L1 to L3 fail, the liability model misallocates blame, and a model of shared, forward-looking and differentiated responsibility is warranted.

The arguments above do not derive an ‘ought’ from an ‘is’. It shows that the descriptive presuppositions of one moral framework are false of this case and that those of another are true.

Distributing responsibility: the four parameters in Embakasi East

Young’s four parameters guide the division of moral labour. Table 4 applies them to the actors the evidence brings into view. The assignments are provisional and follow the survey’s own categories; they are not findings about particular institutions. Regional analyses lend some support to them: one review of sub-Saharan alcohol policy argues that action must address availability, including unrecorded production, outlet licensing, marketing and taxation (Ferreira-Borges et al., 2017), and another argues that the drinks industry has supplanted government in shaping alcohol policy in the region (Bakke & Endal, 2010).

Three points follow. First, the parameters differentiate. Those with power over premises, promotion and enforcement carry more of the duty to change conditions than do those at the bottom of the income gradient, whose interest in change is high but whose power is low. Second, liability survives where L2 holds: organised counterfeiters are wrongdoers in the ordinary sense, and nothing in the social connection model shields them. Third, consumers are not exempted. Their responsibility changes character, from deserving sanction to participating in remedy, and it is shared. Responsibility without blame is demanding and not lenient, because it cannot be discharged by punishing someone else (Young, 2011).

Objections and replies

Agency. Many poor people do not drink heavily, and respondents chose what they did. The reply is that the argument is probabilistic and not deterministic. The data themselves show that 5.9% to 9.7% of respondents in the higher income bands reported the highest band, and that 26% of the sample are former consumers who have stopped. Structure shapes the option set and the cost of each option; it does not abolish choice, and Young (2011) is explicit that the social connection model preserves agency by locating responsibility in what agents can do now. Nagel’s (1979) discussion of constitutive luck adds a further point. Who ends up in the heavy band is partly a matter of the circumstances into which one falls, and a theory of blame that is sensitive to such luck is suspect.

Self-serving attribution. Consumers may externalise, citing price and hardship to excuse themselves, and the observed pattern of high structural ratings and low hedonic ratings is exactly what such a bias would produce. Three considerations limit the damage. The income and education gradients in Table 2 are reports of behaviour and do not depend on attribution. The supply-side facts, namely seller structure, outlet locations and relative price, are reports about the environment that respondents are well placed to observe. And C1 and C2 do not rest on attribution at all; only C3 does. The honest conclusion is that C3, and with it the failure of L1, is the least secure link in the argument. The distributive claim in C6 and the failure of L2 hold without it, though the case against blaming the individual consumer is then weaker.

Reverse causation. Heavy drinking may itself cause job loss, school dropout and family strain. That is likely in part, and cross-sectional data cannot separate the directions. The structural reading requires only reciprocity together with the independence of supply-side conditions. Price, distance and outlet location are not plausibly effects of an individual respondent’s drinking. A feedback loop in which hardship breeds dependence and dependence deepens hardship strengthens the structural reading, since self-reinforcing processes of this kind are what Young’s model is built to describe.

Dilution. If everyone is responsible, perhaps no one is, a familiar worry about diffuse responsibility. Table 4 is the reply in outline. The parameters differentiate responsibility, and because it can be discharged only collectively, it generates tasks and not merely attributions. Zheng (2018) argues that social roles supply further specificity, and landlord, platform operator and regulator are roles with identifiable duties.

Paternalism. A structural account might seem to license intrusive intervention in the lives of the poor. The Nuffield Council on Bioethics (2007) proposes an intervention ladder on which the more intrusive a policy, the stronger its justification must be, and the least intrusive effective measure is preferred. Relational autonomy points the same way: it respects agents by changing the conditions of choice, such as price, verification and location, and not by overriding the choices themselves (Mackenzie & Stoljar, 2000).

Normative implications

Four constraints follow. Each is conditional on the evidence and theory above, and each is better read as an ethical test of policy than as a policy programme.

Proportionality of blame. Sanctions aimed at consumers at the bottom of the gradient cannot be justified on grounds of desert, because the presuppositions of the liability model fail there. A state that penalises these consumers penalises position as much as conduct.

Differentiated and upstream duties. Coercive resources are best directed at actors with power, privilege or an organised interest in the supply chain (Table 4). Supportive duties fall on those able to change norms and prices.

Verification as justice. Since 51% of respondents mainly consume product types they cannot verify, and the heaviest consumers are the poorest, assurance of safety is a distributive matter. A harm-reduction approach is defensible to the extent that it addresses the context of drinking, such as poverty and unequal access, and not only its immediate hazards (Pauly, 2008).

Participation. Because responsibility is discharged through collective action, those most affected should co-author the measures that concern them. A regime that treats them only as objects of enforcement repeats, at the level of governance, the domination that the structural account identifies.

Conclusion

This article has argued that, in Embakasi East, quantitative evidence on unrecorded alcohol consumption satisfies the conditions of structural injustice and defeats the empirical presuppositions of the liability model. Heavy consumption is graded by material position, with a roughly 8.5-fold difference between the lowest and highest income bands; supply is diffuse, with most sellers described as independent or complicit; and respondents rate hedonic reasons at about 2.1 on a five-point scale against about 4.1 for family and environmental pressures. The ethical element at issue, the attribution of responsibility, is therefore better handled by Young’s social connection model than by a liability model: responsibility is shared, forward-looking and differentiated by power, privilege, interest and collective ability. Liability remains apt for organised wrongdoers, and consumers are not excused from the collective work of remedy.

The argument is limited by its evidence. The sample is small, non-probabilistic and self-reported, the quantity measure is ordinal at best, and the link from attribution to blame is the weakest in the chain. Future work should test the same conditions with probability samples, longitudinal designs and independent measures of price, outlet density and enforcement. It should also test whether the pattern holds in other jurisdictions, since nothing here licenses generalisation beyond the setting studied. The wider point is methodological: normative theories that state their conditions in observable terms can be confronted with data, and doing so sharpens both the theory and the policy it is asked to guide.