or otherwise of statistics.3 The most obvious, which he labels ‘metrological realism’, rests on the assumption of some permanent reality, which is independent of any observation apparatus – so that quantitative social sciences could ultimately attain equivalent status to natural sciences.
In contrast, ‘accounting realism’ relates to a more limited space (internal to an enterprise or economic), but offers the illusory promise of rational, testable, provable numbers through double-entry bookkeeping. Illusory, because the numbers necessarily depend on a whole series of judgements – from those involved in the underlying business (or government of, say, a national economy), to the accountants involved directly in compiling the public record, and eventually to those behind the accounting frameworks in use.
The validity of this ‘reality’ depends in turn upon trust in those accountants and others – not in some objectively verifiable and unique set of data. If you’re tempted to think accounting realism is anything but illusory, just ask anyone who has ever looked at the annual report of a multinational company to try to work out whether they paid the right tax, at the right time, in the right place.
Desrosières labels the third attitude ‘proof in use’. Here, researchers using a given dataset prepared by some other body may evaluate its ‘reality’ on the basis of its internal consistency and/or of how well the results of their analysis conform with their priors. Good data with genuine inconsistencies may be undervalued, and institutions publishing statistics may have more incentive to ensure internal consistency of their data – even to the point of deliberately censoring genuine datapoints that do not conform with expectations. Finally, these three attitudes are contrasted with a fourth: one that recognizes explicitly ‘that the definition and coding of the measured variables are “constructed”, conventional, and arrived at through negotiation’.4 There is no objective truth, but the constructed data can be more or less legitimate as a reflection of the different concerns and interests.
This point is extended in an important contribution by Wendy Espeland and Mitchell Stevens, which makes the case that ‘quantification is fundamentally social – an artefact of human action, imagination, ambition, accomplishment, and failing’.5 Measures not only reflect a view of people or things, but also lead people to change their behaviour – including policymakers, thereby creating the possibility of circular feedback between the (in any case overlapping) processes of government and measurement. That circularity is an inevitable feature, and can be both vicious and virtuous. In the case of the uncounted, poor data can promote poor policy, which in turn undermines the scope to improve data; but data improvements can also be self-reinforcing, driving a positive loop of better policy.
Specific metrics, above all where they allow ranking, can impose powerful (Foucaultian) discipline on the people and groups measured; that is, forms of accountability through transparency.6 And so influence over the choice of metric, even for a fixed dataset, is an important form of power. Sakiko Fukuda-Parr and others examined a wide range of metrics selected in the context of the UN Sustainable Development Goals (SDGs), illustrating exactly how competing actors have sought (and achieved) political influence over supposedly technical processes and decisions.7
The choice of who and what go uncounted, excluded either from the gathered statistics or from the chosen metrics, is equally a question of power. And the roles of power and social construction in counting are not optional. There is no ‘neutral’ option in which counting decisions are taken in a vacuum, free from political concerns. And there are no meaningful counting decisions that do not have political implications.
We cannot design a system that inoculates societies from these core characteristics of counting. But we can inoculate ourselves to a degree, from the ‘seduction of quantification’, by opening our eyes to it: by understanding the central dynamics, and the possible nature and extent of the biases that result. We can count better. And if we do, the world can be better.
In this book, I look at a range of important counting choices as they are actually made. I consider the implications for inequality, for governance and for human progress. I use the term ‘uncounted’ to describe a politically motivated failure to count. This takes two main forms, and each has direct implications for inequalities.
First, there may be people and groups at the bottom of distributions (e.g., income) whose ‘uncounting’ adds another level to their marginalization – for example, where they are absent from statistics that underpin political representation (‘who decides’) and also inform policy prioritization (‘what people get’). Second, there may be people and groups at the top of distributions who are further empowered by being able to go uncounted – not least by hiding income and wealth from taxation and regulation (‘what people are required to do’). The uncounted at the bottom are excluded; the uncounted at the top are escaping.
A further distinction in each case lies between what we can call ‘relative’ and ‘absolute’ uncounting. In the former, people or groups are included within the data sample, but are not differentiated. For example, household surveys may include (some) transgender people but without differentiation will fail to generate comparative data on how this group fares. Or consolidated company accounts may reveal information about a multinational’s global profits and tax, but without revealing the relative positions of its operations in Luxembourg, say, or Kenya. Absolute uncounting, meanwhile, reflects the complete failure to include a group – whether that be the absence of high net-worth tax evaders from wealth distribution data, or of some rural, indigenous populations from census data.
Questions of power being complex, there are also cases when marginalized groups may seek to be uncounted precisely in order to exert some power. Any desire to be counted in order to provide a basis for curtailing inequalities will be remote, when the purpose of a state’s counting is to impose greater inequalities.8 Think of oppressed populations fighting to avoid being singled out – whether against the use of the Star of David to isolate Jews in Nazi Germany, for example, or against the use of ‘pass books’ as tools of racial discrimination in South Africa, from the eighteenth century up until the apartheid regime; or the resistance in certain cases to group identification in census surveys (the ‘I’m Spartacus’ response).9
Hidden identity through collective pseudonyms has a long history as a tool of resistance also. Marco Deseriis tracks the use of ‘improper names’ in groups from the Luddites of the nineteenth century, to the modern-day Luther Blissett Project and the Anonymous hacker collective, and argues that they share three features:10
1 Empowering a subaltern social group by providing a medium for identification and mutual recognition to their users.
2 Enabling those who do not have a voice of their own to acquire a symbolic power outside the boundaries of an institutional practice.
3 Expressing a process of subjectivation characterized by the proliferation of difference.
Depending on the external conditions – the power faced, and its legitimacy to count or identify – the case for being uncounted, and the space to do so, will vary. There is a clear difference, however, between the ‘guerrilla’ tactics of the relatively powerless seeking to go uncounted in the face of a quantifying bureaucracy, and the exertion of power by those at the top to escape or circumvent counting.
Inevitably, counting at the national level is imperfect. Survey and census data tend to have major flaws, as does the administrative data used for taxation and voting – and yet these are the basis for any number of crucial policy decisions about where and how to allocate resources.
If the missing data were more or less random, any overall distortions would be limited. If, on the other hand, there were systematic patterns to the distortions, then we should be less sanguine. And, of course, it turns out that what goes uncounted is not random after
