Radical
Statistics
BOOK REVIEW
The Politics of Large Numbers - a History of Statistical Reasoning
Alain Desrosières, (translated by Camille Naish), Harvard University Press, 1998. £27.95, hardback. ISBN 0-674-68932-1.
Review by Ray Thomas
A great strength of this book is its focus on the relationship between developments in statistical reasoning and the growth of official statistics. That relationship is not close. Official statistics and statistical method have by tradition occupied separate compartments of knowledge.
Many ideas in statistical method remained unused for decades. The factors that determined the growth of official statistics have rarely depended upon the availability of relevant statistical methods, and, typically, there has been little theory to support constructive use of official statistics or of other statistics that purport to describe aspects of society. Official statistics and statistical method began to achieve integration only when the process was supported by social and economic theory in the second half of the 20th century.
The basic ideas of sampling, for example, were developed in the 17th and 18th centuries. But official statistics did not make substantial use of sampling until the need to measure unemployment in the United States in the 1930s.
The development of econometrics provides another example. Desrosières includes a very impressive diagram (p. 280) showing the genealogy of econometrics from the 18th century to the middle of the 20th century. The development took so long because for most of this period mathematics, probability, and descriptive statistics were regarded as incompatible traditions. Mary Morgan's history of econometric ideas (Morgan, 1990) provides support for Desrosières' account.
The generous treatment of Britain makes The Politics of Large Numbers the most comprehensive study of developments in Britain. But Desrosières' canvas also includes developments in Germany, the United States and France. The size of the canvas must make the book one of the most enlightening studies of statistics of the century.
Britain set the pattern in many ways. Desrosières identifies the strength of the General Register Office (GOR) in the 19th century as its association with the Poor Law Unions. The variations in local mortality rates shown by the GRO established the value of vital statistics in Britain earlier than in any other country.
Desrosières identifies a characteristic pattern of synergy Britain; '..debates and statistical surveys in the reformist milieu independent of the state are taken up by ad hoc parliamentary commissions and result in new legislation which itself gives rise to a specialist statistical bureau' (p. 171). The 1870 Education Act leading to the creation of a statistical office in 1876, the establishment of the Labour Department in the Board of Trade in 1886, and the report of the Royal Commission for the reform of the Poor Law, are cited as products of this synergy.
In the United States the Census of Population dominated statistical developments until the 1930s. The Census constitutionally determines both the numbers of state representatives in Congress and the basis of Federal taxation. Census statistics also underpinned debate in the 1920s on immigration and the determination of immigration quotas. But US official statistics changed dramatically in the 1930s with the growth of social surveys.
In 1935 Congress approved funding for a postal census from which 7.8 million were identified as unemployed. A parallel sample survey found that only 71% of the unemployed had answered the postal census. The estimate of the number of unemployed was revised to 11 million.
Even President Roosevelt had been sceptical of the practicality of measuring unemployment. But the statistics produced in the 1930s were based on the three criteria that nearly half a century later became enshrined in the International Labour Organisation (ILO) definition of unemployment. To be unemployed a person had to be out of employment, seeking work, and available for work. Margot Anderson's (1988) history of the American census provides solid background for Desrosieres' account of these developments.
Statistics played a major role in the unification of Germany. The Prussian Bureau of statistics was established in 1810. Desrosières describes the work of one of the Bureau's most well known Directors - Ernst Engel as being: 'a typical nineteenth century statistician, an activist and organiser, (but) not very sensitive to the refinements of mathematics' (p. 181). Ernst Engel the author of Engel's law (that as household income increased the proportion spent on food declined) should not be confused with Marx's friend Friedrich Engels.
The Prussian Bureau of Statistics gradually extended its influence over the German Empire. The Bureau produced statistics of employment that were used to support legislation on disability insurance, compensation for work-related accidents in the 1880s and, in 1891, retirement pay. Desrosières indicates how the traditions set by the Prussian Bureau of Statistics survive even in contemporary German statistics that still rely on a negotiated balance between the federation and the Lander in controlling the Federal Office of Statistics (p. 184).
By contrast, the Statistique Générale de France was established in Paris in 1833 and statistics in France have remained under the control of a centralised administration. Hollerith machines were introduced in the 1890s enabling the production of workplace tables from the census of population that supported industrial analysis.
The centralisation of statistics in France may help to explain one of the fascinating themes of Desrosières' book - that of principles of statistical classification. The first question to ask is 'Are statistical classifications real or created?'
Desrosières describes the naturalist Linné's system of classification of the 18th century that assumed that the categories created (i.e. genera) were real. A genus was easily determined by a limited number of simple characteristics. This contrasts with Buffon's more ambitious approach to classification that could take into account any characteristics. Buffon's system could support more detailed classification and regroupings as new characteristics were identified. Linné believed that the genus determined the characteristics. Buffon believed that characteristics determined the genus.
This contrast between realist and nominalist positions underpins much of Desrosières' discussion of the classifications of official statistics. But the second set of questions Desrosières asks are about the principles underlying created classifications. A substantial chapter on classifying and encoding makes a stimulating exploration of this area.
Desrosières deals with the development of industrial classifications - from being based upon materials used in the early 19th century, then on the destination of products, and then, by the end of the century based on techniques of production. Desrosières traces how the idea of unemployment, defined as loss of wage earning employment, emerged towards the end of the 19th century in France and Britain. It is worth noting that this was half a century before unemployment developed as a category in the United States and a full century before it was recognised in Russia!
Desrosières justifiably gives attention to the development of the Social Class classification in Britain and the ways in which this classification reflected the GRO's defence against the eugenicists (citing Szreter, 1984 and 1991). Desrosières traces the development of social economic groups in France from being craft-based in the early 19th century towards recognition of the employer/employee distinction, and then at the end of the century the full socio-economic group classification recognising various other hierarchical facets.
There is discussion of the international development of the categorisation of the cause of death. GRO consistently supported an etiological classification, i.e. seeking the original cause, but the medics, especially in France, supported a topographical classification, i.e. emphasising symptoms and their location.
Desrosières recognises the social science position that official statistics are social or organizational constructions. In opposition to the realist position Desrosières identifies three nominalist positions in the creation of these classifications. The historical approach recognises that social practices create classifications, leaving it to science to describe and analyse how these classifications came into being.
The administrative approach emphasises that classifications are necessary for decision making. Governments need 'categories of action' such as poverty, unemployment, inflation, the trade balance, money supply, fertility, and cause of death, in order to govern. The relativist approach sees the production of statistics as the product of power relationships. It is necessary therefore to open up 'the black boxes' to show what the classifications of official statistics reveal about power relationships.
Desrosières inclines to a realist position in suggesting that these nominalist positions can lead to reality creation.
... conventions defining objects really do give rise to realities, in as much as these objects resist tests and other efforts to undo them. This principle of reality affords an exit from the dead-ended epistemological opposition between these two complementary and complicitous enemies, the realist and the relativist. It does not deny the reality of things once numerous persons refer to them to guide and coordinate their actions (p. 337).
Official statistics, in other words, are real because they are perceived as real. To quote what the American sociologist Cooley said early in the century (something Desrosières does not do) if people believe a thing to be true, then it is true for them in its consequences (Cooley, 1909).
In this way, Desrosières book encourages and opens up the way for both current and historical studies of the ways in which particular statistics become accepted as representing reality. These studies might aim to identify the factors that determined the production of the statistics, the categorisations used, and the production processes that are followed. Such studies could aim to identify the interests that are served by the statistical systems created, and the intended and unintended consequences associated with the operation of statistical systems.
REFERENCES
Anderson, M. J. (1988), The American Census: A Social History, New Haven: Yale UP.
Cooley, C. H. (1909), ial Organization, New York: Charles Scriber's Sons.
Morgan, M. S. (1990), The History of Econometric Ideas, Cambridge: CUP.
Szreter, S. (1984), 'The Genesis of the Registrar-General's Social Classification of Occupations', British Journal of Sociology, 35, 4: 529-546.
eter, S. (1991), 'The GRO and the Public Health Movement in Britain, 1837-1914', Social History of Medicine, 4, 3: 435-463.
Ray Thomas
Social Sciences
The Open University
35 Passmore
Milton Keynes MK6 3DY
Tel: 01908 679081; Fax: 01908 550401
E-mail: r.thomas@open.ac.uk
