Since the start of the COVID-19 pandemic, all eyes have been on movements in the daily deaths figure. Analysed by age, sex, and geographical area, these figures provide information on how the virus spreads and on the extent of mortality and excess mortality, which can be used to assess the effectiveness of measures to combat it. These epidemiological statistics were developed in France during the flu epidemic of 1889–1890, known as the Russian flu. This article recounts the history of how the concept of excess mortality was used, revealing the ways commentaries and interpretations of the new figures were shaped by the demographic and political context in France at the time, marked by a growing fear of depopulation.
1In 1895, the French sociologist Gustave Le Bon cited the 1889–1890 influenza epidemic to illustrate the complex relationship between figures and how people and health authorities assess the gravity of an event:
Whatever strikes the imagination of crowds presents itself under the shape of a startling and very clear image, freed from all accessory explanation, or merely having as accompaniment a few marvellous or mysterious facts: examples in point are a great victory, a great miracle, a great crime, or a great hope. … The epidemic of influenza, which caused the death just a few years ago of five thousand people in Paris alone, made very little impression on the popular imagination. The reason was that this veritable massacre was not embodied in any visible image, but was only learnt from statistical information furnished weekly. An accident which should have caused the death of only five hundred instead of five thousand people, but on the same day and in public, as the outcome of an accident appealing strongly to the eye, by the fall, for instance, of the Eiffel Tower, would have produced, on the contrary, an immense impression on the imagination of the crowd.
3Figures, projected in models of pandemics,  presented one after the other in epidemiology bulletins as an epidemic develops or aggregated in retrospective mortality counts, contribute to the complex social construction of flu epidemics before, during, and after they arise. Figures produced today by statisticians and epidemiologists clarify the contours of collective events that are hard to apprehend at the individual scale, even when it comes to counting victims, since the clinical symptoms of a given disease vary and are often confused with winter colds and other passing ailments. Only laboratory diagnosis can unquestionably determine the presence of a viral disease, but that method is often considered unnecessary for disorders that are generally benign.
4No one today would disagree that statistics are a relevant means of capturing the scope of flu epidemics. However, statistics conceal controversies not only around how they are interpreted but also how they are produced, a reflection of the different uses statistics are meant to serve. During the 2009 H1N1 flu pandemic, the definition of the term ‘pandemic’ was itself an issue. Epidemiologists’ initial, catastrophist mortality projections, based on how the new virus was circulating through the world population, were consistent with the popular representation of pandemics as phenomena that kill high numbers of people around the globe. When it was finally observed that this flu episode had been fairly benign despite its spread across the planet, people’s reaction was to distrust the figures along with the readiness policies (Doshi, 2011).
5This article studies the path-breaking uses of excess mortality calculations in France during and after the 1889–1890 epidemic of what was known as Russian flu. It seeks to understand how a few statisticians proceeded in ‘figuring’ out the Russian flu; to describe the methods they used to make sense of the data collected; and to discover how the knowledge produced after the episode impacted on the understanding of this epidemic event. 
6To comprehend the origins of the scientific innovation of excess mortality, several contextual features need to be specified. The first is that the reappearance of influenza in the winter of 1889 came as a great surprise as the disease had been absent from the continent since the middle of the century. Advanced societies were seeing infectious diseases recede at the time (Condrau and Worboys, 2007). In many European countries, the pandemic gave rise to a major health crisis.  Journalists of the informative press dubbed this disease the ‘Russian flu’.  Thanks to the technological invention of the telegraph, the press was able to cover the event at an international scale, thereby amplifying the epidemic’s eruption in public space (Vagneron, 2014). The second, longer-term contextual feature concerns the ‘avalanche of printed numbers’ (Hacking, 1982) that had begun to invade European societies in the 19th century. This phenomenon, both scientific and popular, shed light on multiple realities of social life while upsetting people’s experiential frameworks. As historiography has shown, the area of health and medicine was fertile ground for producing and developing statistical knowledge because such knowledge was used both to understand phenomena and to administer modern states in an environment of rising international competition (Porter, 1995; Desrosières, 2000).
7In that specific context, what statistical practices were used in France to make the demographic, medical, and even political dimensions of the flu and the epidemic intelligible? What in the observation scales that the statisticians chose for their studies gave them access to specific knowledge of the epidemic in Paris, throughout France, and perhaps across the world (Lepetit, 1993)? How did their statistical practices fit with other types of knowledge about the disease, namely how it spread? To answer these questions, I analyse a corpus of statistical studies of the Russian flu epidemic in France conducted between 1890 and 1893, primarily by three renowned statisticians. I describe how that new knowledge contrasted with, confirmed, and supplemented other medical knowledge and contemporaneous experiences of the epidemic. Those studies extracted the epidemic from its circumstances,  situating it instead within other interpretive contexts that worked to transform the meaning of the event. We shall see that statistical knowledge of the epidemic and the disease led to a new way of apprehending the phenomenon. I also draw on recent research in the sociology of science (Jas and Jouzel, 2015) that makes it possible to analyse in greater detail the mechanisms by which new knowledge is either used to take into account the problem described or, on the contrary, makes the problem easier to avoid.
8Studies by Jacques Bertillon (director of the Paris Statistics Bureau), Victor Turquan (director of France’s General Statistics Office), and Paul Roux (Interior Ministry’s Department of Public Assistance and Hygiene) conferred new importance on the epidemic due to the unprecedented overall mortality counts they found during that event.  The figures were clarified with the help of the notion of excess mortality (surmortalité), which was calculated for the entire country for a period of approximately 12 weeks. The studies were also implicated in debates on how the flu was transmitted. However, while the use of statistics to decipher the flu was hailed as a decisive innovation, it did not lead the French national health and scientific authorities to include influenza on their long-term agenda. For in the flu year of 1890, France experienced negative natural increase, and the issue of depopulation relegated the epidemic to a secondary concern—an ‘accident’ in the history of national mortality that did not weigh much alongside the preoccupying general decline in national births.
9The first section of this article describes how the epidemic in Paris, as revealed by the excess mortality caused by the flu, proved a particularly opportune and fertile research topic. The second analyses how study of the flu epidemic became caught up in contemporaneous debates on French population trends. The third and last section retraces how an Interior Ministry study provided key support for what was at the time the minority hypothesis that the disease was contagious, by showing that it had spread through France by rail.
I – A particularly opportune and fertile research topic: the epidemic in Paris as revealed by flu-related excess mortality
10In an editorial for the Revue d’hygiène et de police sanitaire of January 1890, Dr Eugène Vallin, a leading figure in the field of public hygiene in the late 19th century, expressed his contempt for the now ‘fashionable’ subject of the flu:
Vallin’s scepticism was not shared by all. By late January 1890, the epidemic had become one of the most heavily discussed subjects in academies and scientific and medical journals. In all this research, statistical studies were of crucial importance. The data used were very different from the type referred to in medical dictionary entries on the flu published prior to 1889, most of which drew on more or less dated medical topographies derived from neo-Hippocratic theories that associated the emergence of epidemics not only with geographic but also climatic and social local conditions (Vagneron, 2015).For the last month, all the talk in the government councils, the academies, the newspapers, the salons—especially the salons—has been of the flu epidemic. The Revue in its turn can hardly refrain from putting in a word. And yet what do we know? Above all on the matters of pathogenesis, hygiene, or prophylaxis? Next to nothing. Surely it would be wiser to wait until some discovery is announced, a few genuinely scientific facts. Yet we have to yield to the current and behave like everyone else.
1 – Excess mortality in Paris according to Jacques Bertillon
11Paris was the first area examined by Bertillon and Turquan, two of the most prestigious representatives of ‘administrative statistics’ in the service of the state, a practice institutionalized in France in the late 19th century (Dupâquier, 1983; Brian, 1989). In contrast to the situation for the rest of the country, statistical health data on Paris existed and were accessible, making it the perfect site to investigate (Palsky, 2002).
12Bertillon began studying the flu immediately after it erupted in Paris. Parisians’ collective experience of the epidemic was shaped by information published in the city’s weekly statistical bulletins (Bulletins hebdomadaires de statistique de la ville de Paris) that was then duly taken up by medical journals and the daily press (Vagneron, 2014). The Bulletins hebdomadaires provided the public with near-instantaneous information on how the epidemic was developing, confirming what Parisians had seen and experienced (increasing numbers of burials, high levels of disease in the city), while revealing less perceptible aspects at the individual level (vulnerability by age, exacerbation due to other diseases, etc.). Drawing on the available data, Bertillon provided a detailed analysis of the epidemic in a series of lectures to learned societies, and in 1892 he published a report on the event in the Paris statistical yearbook (Annuaire statistique de la ville de Paris).
13Bertillon was quick to point out in his first lectures why the statistical approach was so useful in studying the flu. Speaking to the Society of Public Medicine and Occupational Hygiene on 22 January, he noted ‘the power of the statistical method’:
The flu was hidden behind many diseases. Statistics has enabled us to determine what those diseases are and the frequency with which they are complicated by flu. They have also enabled us to determine the age and sex of sick persons whom we may expect will develop those complications. No other method could have performed the same services for us.
15Bertillon’s work on the flu made three contributions to understanding it. First, the weekly death count enabled him to specify the chronology of the epidemic. He suggested bracketing a period marked by an exceptional rise in overall mortality in the city’s population compared to a period defined as normal. It turned out that mortality from December 1889 to January 1890 was quite different from measurements for the same period in the 3 preceding years (1886– 1888). Bertillon concluded there had been 6 consecutive weeks of excess mortality, from the 50th week of 1889 (2nd week in December) through the 3rd week of 1890, with a peak in the last week of 1889. This statistical period was much longer than the one during which the attention of health authorities and the public was riveted on the epidemic. That period had begun only in late December, when the Paris hospital system was overwhelmed by an influx of patients despite the reassuring discourse from certain eminent public hygiene experts. 
16Second, Bertillon studied the demographic impact of the epidemic; specifically, mortality by sex and age. On age, he observed that children were only slightly affected, whereas adult mortality had tripled and mortality among older persons had doubled (Bertillon 1890a).  On sex, Bertillon found nearly identical mortality levels for boys and girls, whereas adult men were hit twice as hard as women. Older women were more heavily impacted than older men, but that sex-related difference was slighter than the opposite one among younger adults.
17Third, Bertillon explored causes of death implicated in the excess mortality found over the 6-week period. He began by explaining that his information on the burden of flu-related mortality did not come directly from the cause-of-death reports signed by the city’s certifying physicians.  According to those official certificates, only 250 Parisians had died of the flu during the epidemic. And yet there had been overall excess mortality, as shown by comparing the deaths figure with those from the preceding years. Bertillon explained that mortality in the Paris statistics was dispersed among different causes:
The reported flu figures were very low compared to the overall mortality figure; influenza was mentioned very few times. Indeed, it is only through its complications that it can bring on death. The violence of a flu epidemic can only be revealed by statistics.
19‘Diseases of the respiratory organs’ had risen sharply. Bertillon found spikes in five main causes of death in this category: pneumonia, bronchopneumonia, acute bronchitis, chronic bronchitis, and phthisis [pulmonary tuberculosis]. Lung congestion was not on the list, but Bertillon noted 140 deaths from it in the 52nd week, as against ‘the average figure’ of 15 (Bertillon, 1890a, p. 153). He concluded that respiratory tract diseases were a decisive factor in the increase of mortality in Paris and attributed that increase to effects of the flu.
20Bertillon ruled out ‘epidemic diseases’ as causes of the excess mortality because they had remained stable —a logical finding, he explained, because several of those diseases struck children first, whereas children had been relatively spared by the Russian flu. He noted an increase in deaths from all chronic diseases that might involve pulmonary complications, stating that the flu ‘accelerates individuals’ ends’ (Bertillon, 1890a, p. 156). In 1892, he compared his statistics with statistical data from cities across Europe, putting forward an unprecedented map (Figure 1) representing the spread and intensity of the flu in major European cities from 3 November 1889 to 1 February 1890. The map also represented weekly fluctuations in flu-related excess mortality compared to the preceding years.
21The map presents the pandemic at the scale of major European cities. Though Bertillon’s comments are brief, the map clearly conveys his concern to construct a space that would permit visual comparison of manifestations of a specific phenomenon. It is not surprising that this map, like other studies of the Paris statistical bureau (Fijalkow, 2000), was drawn up under Dr Bertillon’s direction. It is also clear that he was using urban statistics here not so much to identify targets for the local authorities to act on—in contrast to the ‘health records’ on apartment buildings housing tubercular residents that would later be required, or the identification of insalubrious city blocks—as to produce a body of knowledge that would be independent of local circumstances and perspectives, and to do so by standardizing large-number statistics for different cities. Excess mortality, which is relatively easy to calculate, has since been widely favoured for studying flu epidemics precisely because it is not subject to the influence of local data-collection particularisms. 
22Looking for a historical precedent with which to compare his results, Bertillon logically turned to statistics for the City of London, which ‘to its credit was the first to have understood the need to know what its inhabitants die of’ (Bertillon, 1892, p. 127).
23Bertillon’s first table was a compilation of weekly London deaths during five epidemics (1733, 1743, 1833, 1837, and 1847) as compared to the ‘number of weekly deaths during an ordinary period’ (p. 126). He then compared the 1889–1890 epidemic in Paris with the 1847 epidemic in London.  The similarities were striking: both epidemics struck older persons and adults harder than children, and adult men harder than women; most of the deaths counted were caused by respiratory diseases; chronic diseases likely to cause difficulty breathing were the most lethal; and the mortality counts for equivalent populations over a similar length of time (6 weeks) were similar (approximately 5,000 deaths). In London, however, deaths attributed to the epidemic seem to have been spatially distributed by inhabitants’ wealth, a feature that Bertillon did not observe for Paris. Because British statistics enjoyed such a strong reputation, comparing the 1889–1890 epidemic in Paris with the 1847 London epidemic had the effect of legitimating Bertillon’s own statistical study.
24It was as a doctor and a statistician, then, that Bertillon undertook his first studies of the flu. These combined competencies explain his status as a founding father of the International Classification of Diseases, begun in 1891 (WHO, 1950). Through his descriptions of the evolution of the excess mortality figure during the epidemic, Bertillon showed how the flu had been concealed behind a range of medical causes of death and clinical diagnoses. He later broadened his study to include social inequalities in health. Bertillon’s consistent use of international comparison over time shows his commitment to descriptive statistics that allow for comparison and generalization of the calculation methodology. In this, his pioneering research was very different from Victor Turquan’s first analyses.
Frequency of flu-related deaths in the main cities of central Europe (1892)
Frequency of flu-related deaths in the main cities of central Europe (1892)
(cont’d). Frequency of flu-related deaths in the main cities of central Europe (1892)
(cont’d). Frequency of flu-related deaths in the main cities of central Europe (1892)
2 – From statistical calculations to government action: excess mortality according to Victor Turquan
25The first two articles by the head of France’s General Statistics Bureau (SGF), Victor Turquan, published respectively in the Revue scientifique (Turquan, 1890) and the Journal de la société statistique de Paris (Turquan, 1891), were based entirely on data published by the Paris Statistics Bureau. Like Bertillon, Turquan had been a member of the Paris Statistical Society since 1882. However, Turquan’s view of disease was not so much medical and scientific as demographic and policy-oriented.
26Turquan chose to study weekly death counts in Paris from late October 1889 rather than starting a month later, as Bertillon had. He too focused on deaths caused by diseases of the respiratory tract and consumption. Like Bertillon, he believed that all late 1889 excess mortality could be attributed to an increase in deaths from ‘diseases of the organs of respiration’. But as he saw it, overall mortality in Paris first exceeded the weekly average in the 46th week of 1889, though it was not until around 15 December that the epidemic took a sudden turn for the worse (Turquan, 1891). He also extended the epidemic period into February, the assumption being that it had lasted not 6 weeks but 10 to 12, from the middle of the 46th week of 1889 to the middle of the 6th week of 1890. This led to a higher mortality count: Turquan estimated excess mortality in Paris to be between 5,000 and 6,000 (Turquan, 1891), whereas Bertillon continued to cite 5,000.
27Turquan then confirmed all of Bertillon’s analyses of mortality by age and sex (it was only on this point that he cited Bertillon at all). He agreed with Bertillon’s interpretation of excess mortality among adult men: their occupational activities forced them to work outdoors and to return to work too soon after falling ill. He also agreed with Bertillon that the epidemic had exacerbated mortality among people with chronic diseases.
28Turquan’s third section contains more original material—though of the sort expected from a French statistician at the time. In it, he worked to identify the Paris neighbourhoods where influenza had caused the most deaths. As he did not know morbidity by neighbourhood, he hypothesized that the lethality of the disease was the same in all of them. Following the social statistics study practice of the day (Desrosières, 1988; Brian, 1991), he compared average district deaths by population and presented ‘the average proportion of deaths caused by respiratory diseases’ for each of the 11 weeks of the epidemic (Turquan, 1891, p. 65). He presented his findings in tables and cartograms  in accordance with the demonstration method that French administrative statisticians had adopted in the 1870s, proudly pointing out that ‘a single glance at this map’ was enough to identify the districts in which the epidemic’s intensity had been particularly strong or weak. 
29Despite the roughness of his averages, Turquan interpreted his cartogram in predictable fashion, associating greater neighbourhood population density  with greater damage from the disease, while showing that the flu had indeed spread to all strata of Parisian society:
While all classes of Parisian society seem to have been equally afflicted, the same cannot be said of city districts: the west and a part of the centre, from Porte Saint-Denis to the Madeleine, were much less badly affected than the northeast and the southwest.
31This observation led him to make a policy suggestion—thereby bolstering the role of expert statisticians as a source of support for their own institution and government action. For Turquan, statistics had once again shown
the necessity, if only from the perspective of hygiene, of thinning out the population in places where it is too dense, by means of wide thoroughfares…. In this regard, the planned metropolitan railroad would be of great service, as it would make it easier for the overly concentrated populations in the centre to gain access to the faubourgs [outlying districts] and suburbs.
33Turquan then proceeded to apply standard interpretive frameworks (social inequality in the face of death by neighbourhood population density), visual representations and models (cartograms), and a proposal for public action—fixing the problem by regulating the excessive density of some neighbourhoods—to his routine statistical analysis based on averages and a particular locale. It did not matter that the medical community was still debating the aetiology of the flu since the social and economic factor of population density cited by Turquan was indifferent to the contagion-versus-miasma question.
34Through Bertillon’s and Turquan’s work on the case of Paris, excess mortality became an index of the impact of epidemics on populations. Its very generality meant that the statistic offered three advantages: it obviated the need to join an uncertain medical debate on flu aetiology; it made interurban comparisons possible; and, as we shall see, it could be applied to different spatial scales. Working from a medical perspective, Bertillon designated populations that were relatively vulnerable to the epidemic, while Turquan put forward an overall mortality count for the entire capital and territorialized cases by arrondissement in the aim of guiding government policy and actions. These two contributions from eminent statisticians legitimated using the statistical method to study the flu.
II – How the ‘accident’ of the epidemic dovetailed with deep concern about declining population size (dépopulation)
35With the shift in observation scale that characterized the first studies of the epidemic’s effects on the size of the French population came a shift in the meaning of the epidemic. The main issue was no longer medical causes of death or urban density, but rather how the size of the national population was evolving and how this should be interpreted scientifically and at the policy level. SGF statisticians led by Turquan set out to study the national territory and the population as a whole.
1 – 1890, Year I of dépopulation
36Commenting in late 1891 on ‘Le mouvement de la population en France pendant l’année 1890’ [Population change in France in the year 1890],  Auguste Vannacque presented that year as exceptionally unfavourable to French population growth. As head of the SGF’s accounting and statistics division and a member of the Paris Statistical Society, Vannacque observed that the slackening natural increase the SGF had found for the preceding years had, in 1890, resulted in negative population growth (deaths exceeding births):
A similar phenomenon occurred in even greater proportions in 1870– 1871, after the [Franco-Prussian] war, and in 1854–1855, after the cholera outbreak and the Crimean War. … The effect of the drop in births and the simultaneous increase in deaths is that the latter have exceeded the former by 81,572. The year 1890 therefore looks to be one of the century’s worst in terms of various population developments: there were only 12,000 more births than in 1871 (826,000 births in 1871), and the deaths figure is the highest it has been in twenty years. … But the following observations show that population developments in 1890 were affected by the so-called influenza epidemic that ravaged all of France in the first four months of the year. However unfavourable this situation is, then, it may be thought of as due to a fundamentally accidental cause.
38The year 1890 therefore showed a manifest but anticipated change in the country’s natural increase; population growth had tipped into negative territory, an event that had not occurred since the 1870–1871 war and the cholera epidemic of 1854. However, the shift could be ascribed to ‘a fundamentally accidental cause’: the influenza epidemic.
39Two distinct mechanisms were operative in the composite phenomenon of a fall in births and a rise in deaths. According to Vannacque, the first was due to a fall in marriages, which, in a Catholic country, had mechanically led to a fall in natality. To this could be added the long-term repercussions of mortality from the 1854–1855 cholera outbreak: the 35–36 years since that crisis amounted to the ‘ordinary duration of a generation’ (Vannacque, 1891, p. 352), in this case, a generation amputated by cholera. To explain the year’s mortality, Vannacque noted the considerable increase due to the flu in the first 4 months of 1890: half of the year’s excess mortality compared to 1889.
40Negative natural increase in 1890 dovetailed with an alarmist interpretation of the national demographic situation, a perspective that was gradually gaining ground among the nation’s elites. Toussaint Loua, Turquan’s predecessor at the SGF and secretary-general of the Paris Statistical Society, wrote the following commentary in the margins of Vannacque’s article, spelling out the meaning contained in the published figures:
There can be no concealing the fact that the information contained in the preceding report is extremely distressing, as it indicates a first step in the direction of depopulation in France. We have seen that deaths increased from one year to the next by more than 81,000, but this rise was due to an accidental cause. The truly unfortunate result in our population trend is the gradual decline in births; and in this connection we have ample reason to fear that the drop is not over, as the number of marriages has constantly declined: Caveant consules! 
42In alarmist tones, Loua had linked 1890 population trends to the general theme of dépopulation, the overall pessimistic understanding that the national population had been declining both quantitatively and qualitatively since the late 1860s (De Luca Barrusse, 2008a). France’s defeat in the war of 1870, the mortality caused by that war and the Paris Commune that followed, had led to negative natural increase, a finding that traumatized the elites in a context of increasingly fierce competition between European countries (Bourdelais, 1993). Debates once strongly focused on controlling fertility now yielded to calls for more births. In the first years of the Third Republic, the feared national weakening was measured by population censuses, which had become ‘indicators of the population’s health and vitality’ (De Luca Barrusse, 2008b, p. 255). It was in the last decade of the 19th century that the depopulation issue became the top concern in both government and public opinion, as evidenced by the founding in 1896 of the National Alliance for the Growth of the French Population (Alliance nationale pour l’accroissement de la population française) by Jacques Bertillon. 
43Loua’s commentary sums up the approach to combating demographic decline that gradually won out in the 1890s: policy should focus on increasing births rather than reducing mortality through public hygiene measures. In this framework, the flu epidemic was thought of as an ‘accidental cause’ situated outside the field of government action, whereas falling birth numbers justified calling on the authorities to take action. The shift to negative natural population growth in 1890 was key to the development of the pro-birth movement in France because it was the first such result since the war of 1870. The 1891 and 1896 censuses confirmed that shift. In the absence of an exceptional national disaster such as war or cholera, the flu epidemic of 1890 and its effect on the country’s population got absorbed into the general preoccupation with depopulation, a French specificity that continued into the 1930s.
44In his 1897 work, Le problème de la dépopulation, Bertillon championed the decision to focus national recovery efforts on increasing births rather than lowering deaths. In it, he claims that measures to lower French mortality were ‘illusory’ because the country’s level was no higher on average than mortality in neighbouring countries. The statistician who earlier had noted how the flu ‘harvested’  the lives of older persons and those with chronic diseases now went so far as to assert that adult and especially older adult mortality could only assist the effort to increase births. For the death of an adult ‘leaves a job open and makes the creation of a new household possible’ (Bertillon, 1897, p. 42), and the death of an old man made it easier for the children to establish themselves. To support his argument, Bertillon used a forestry metaphor: ‘doctors with excess confidence in their art’ who acted to thwart premature death were like bad loggers who ‘would preserve trees indefinitely’: their actions ‘are of absolutely no interest with regard to population figures’ (Bertillon, 1897, p. 44).
2 – A major epidemic crisis at the scale of the century as a whole
45Despite the concern to increase births, the fact that the 1890 flu epidemic appeared accidental next to the social scourge of depopulation did not mean it was dropped as an object of investigation.  Turquan’s later articles detail the effects of the flu not only on the nation’s deaths but also its births.
46His second study of the epidemic to be published in La Revue scientifique in 1892 drew on the 1890 findings on French population trends. Turquan immediately took the same alarmist tone as Loua: ‘never since the disastrous years of 1870–1871 have birth and marriage figures been so low, and never has the deaths figure been so high’ (Turquan, 1892b, p. 81). After using civil register data to explain what made 1890 stand out in the century’s population trends, he presented the distinct effects of the epidemic in the three different population types and spaces identified by the SGF: rural and urban, and the Seine département.
47To clarify the exceptional nature of what happened in 1890, Turquan presented annual mortality since 1800. Deaths had only overtaken births twice in the century; not even the cholera outbreak of 1832 had brought this about as births were very high that year. According to Turquan, the most striking development revealed by nearly a century’s worth of data was actually the fall in French mortality, from nearly 30 deaths per 1,000 inhabitants early in the century to 20 in 1889. Indeed, the mortality rate for 1889 had been the lowest ever recorded, despite child mortality, which continued to darken the picture. This historical preamble was followed by a study of nationwide deaths in 1890 compared to 1889.
48If both deaths caused directly by flu and deaths from the flu’s exacerbation of all other diseases were attributed to the flu, the epidemic ‘seems to have carried off 40,000 people, starting at the beginning of the year’ (p. 83); that is, in the first 4 months of 1890. The epidemic first appeared in Paris in December, then moved to provincial cities in January, and to the countryside between January and March. Mortality was proportionately higher in Paris than the other locations, even though the latter were affected for a longer time. Turquan estimated that mortality in France had risen as high as 60,000 for the entire period in which the disease was active (starting in the last month of 1889)—the equivalent of 1.6 deaths per 1,000 inhabitants.
49He also noted an unexpected effect: the epidemic had had a negative impact on births.  Whereas monthly births had moved along at an ‘ordinary pace’ until August 1890, there were 8,000 fewer births in September of that year than in September 1889; 12,000 fewer in October; and 6,500 fewer in November. Turquan attributed this development to the flu epidemic:
For how can we explain the sudden, temporary fall in births exactly nine months after the ravages of the epidemic if not by the morbid state in which the greater part of the population in a condition to procreate found itself?
51This meant that an additional consequence of the flu epidemic was 26,000 to 27,000 un-conceived babies (Turquan, 1893). Deaths due to influenza thus came to ‘a minimum of 67,000 individuals’ (p. 85).
52Clearly, the contours of the 1889–1890 epidemic had been redrawn using national civil registry data. This had two effects: the flu became the accidental factor in a national demographic event at the scale of a century; the mortality caused by the epidemic was assessed by attributing to it both direct and indirect deaths and the birth deficit of 9 months later (a deficit compared to the number of births in the preceding years). Because the flu had played a role in the negative natural increase of 1890, it statistically embodied the increasing concern about population decline among the country’s elites. The calculations of decline in 1890 worked to relegate the epidemic to the rank of accidental cause, in contradistinction to the falling birth trend, a phenomenon more worrying to the elites than the issue of mortality. And the elites were ready to voice that fear, defend their cause, and call on the authorities to address the problem (Rollet, 2001).
III – How the flu unified France geographically: an Interior Ministry study and its contribution to the epidemiological debate
53Paris statistics and statistics from civil registers across France retrospectively revealed that flu-caused mortality had extended over a 6-month period. Whereas the civil register data did not clearly capture how the flu had spread through France, health data collected on the country’s main cities for a study by the Interior Ministry’s Public Assistance and Hygiene department were much more informative.
1 – A path-breaking Interior Ministry study based on health statistics
54As Adrien Proust, inspector-general of health services, noted in his official 1892 report on the epidemic (Proust, 1892b), the new data had been collected from 1890 to 1891 in response to an Interior Ministry circular of 28 December 1889. That document called on the mayors of the country’s 40 largest cities (population over 40,000) ‘to furnish a daily report on the number of deaths reported to the [municipal] civil registry office as compared with number of deaths on the same day of the five preceding years’ (Proust, 1892b, p. 552). Paul Roux, second-in-command at the Ministry’s public hygiene office,  recorded the counts and analysed the data; in 1891, he published his findings in the form of statistical graphs and tables in the Statistique sanitaire des villes de France for the year 1890.  This was the first nationwide health study by the new Department of Public Assistance and Hygiene, which had been integrated into the Interior Ministry in January 1889.
55The study went forward despite the department’s priority on public assistance.  The new department director, Henri Monod, had waited some time before intervening in epidemic management, as it was assumed the health crisis was restricted to Paris. When he did intervene, his primary task as defined by the Ministry was to organize assistance for the most vulnerable elements of the population (Vagneron, 2014). In this new institutional context, a nationwide investigation into the 1889–1890 epidemic was quite a remarkable initiative—and a possible opportunity for Monod, well-versed in public hygiene,  to achieve more of a balance between the ‘hygiene’ and ‘assistance’ components of his department. The department study, entitled ‘Étude graphique sur l’accroissement de mortalité occasionné dans les quarante principales villes de France par l’épidémie de grippe’  [Study in graphs of the increase in mortality caused by the flu epidemic in the forty largest cities of France], won the Interior Ministry’s silver medal for service related to an epidemic  for its main architect, Paul Roux. Proust and Turquan included material from the study in their reports on the epidemic, and it was cited in the Statistiques sanitaires des villes de France (1891).
56The data collected covered the 40 cities indicated in the Ministry circular as well as some cities of over 30,000 inhabitants (included since 1886 in a new urban health statistics project that was completed only in 1906), for a total of 51 cities with over 30,000 inhabitants and a total population of 6,269,217 (according to the 1886 census). The cities were divided into three groups: Paris (population 2,260,943), cities with 95,000 to 400,000 inhabitants (11 cities with a total of 2,004,285 inhabitants), and cities with 30,000 to 80,000 inhabitants (39 cities with a total of 2,003,989 inhabitants). Deaths figures from November 1889 to May 1890 were compared to average mortality for the 3 preceding years. A ‘graphic table’, as it was called in the key (the term is Proust’s), presented monthly mortality per 1,000 inhabitants for all 51 cities and by city size categories from November to April as compared with the previously calculated averages (Proust, 1892a).
57Following Turquan’s method, Roux established a new national mortality estimate by extrapolating from the data on large French cities. For the 5,886,110 inhabitants of the 40 designated cities, he attributed 10,726 deaths to the flu epidemic, or 18.1 deaths per 10,000 inhabitants. Daily mortality in the city of Montpellier was found to have increased during the epidemic over the normal period by 111%; the figure for Versailles was 105%, and for Dijon, 100%, whereas in Bordeaux and Nantes the respective increases were lower: 39% and 29%. Average excess mortality for the 40 cities came to 65%. Roux found 64,000 deaths attributable to the epidemic for the country as a whole, a figure close to Turquan’s 60,000.
58His study likewise found that the epidemic had begun in Paris and only later spread to other cities. Roux then specified the daily advance of the epidemic in the 40 cities for which the relevant information was available. In 1892, Proust reproduced Roux’s graphs for the cities of Paris, Marseille, Saint-Étienne, and Le Havre (Proust, 1892a, pp. 96–97). In addition to the raw mortality data, the dynamic observed was graphed as a bell curve that smoothed daily fluctuations. For 31 of the cities, the period of maximum intensity was 5 to 18 January, whereas for four cities in the Paris Basin (Paris, Versailles, Saint-Denis, and Amiens) it was 29 December to 10 January; in four other major cities relatively distant from each other (Toulouse, Tours, Rennes, and Caen), the peak period began after 14 January.
59Roux’s study confirmed many statistical findings, but more importantly it had a considerable impact on epidemiological knowledge of the flu. For by mapping the spread of the disease to France’s main cities, Roux contributed what proved a decisive argument to the medical debate on how the flu was transmitted.
2 – A contagious disease spread by rail
60Similarly to the cholera debates that began in the 1830s (Ackerknecht, 1948), the arrival of Russian flu in France re-sparked lively controversies on flu aetiology and transmissibility. Prior to 1889, reputed medical dictionaries described the flu as one of the last diseases whose origin could be plausibly explained as some mysterious cause in the atmosphere. None of the main theories of disease transmission that had gained recognition before the development of bacteriology could account for the quick spread of the disease. Flu clearly was not accounted for by miasmatic theories, as it had been simultaneously described for sites with quite different geographical and social characteristics that were also too far away from each other to have been contaminated by a single first hub or cluster. Yet contagion seemed improbable too because cases seemed to rise faster than human bodies could move about. The only possible explanation for the presence of simultaneous cases seemed to be an unknown airborne cause (Vagneron, 2015).
61In the first weeks of the epidemic (December 1889), the controversy returned to the fore. Speaking on 17 December at the Academy of Medicine, the eminent pathologist Charles Bouchard once again expressed doubts— shared by most of his colleagues—that the flu was contagious, citing the argument of how quickly it spread.  However, in early January, several doctors openly criticized that opinion. One of them, Henri de Parville, writing in the Journal des débats politiques et littéraires, lamented the fact that the greatest scientists in Europe were claiming the flu was not contagious despite irrefutable evidence that it was (Parville, 1890). A few days later, a certain Dr Daremberg came out in favour of contagion, hypothesizing an as-yet unknown microbe (Daremberg, 1890). But as Louis Pasteur admitted on 13 January 1890 to a journalist at the daily L’écho de Paris, that microbe could not be found for the moment.  Pasteur also refused to give an opinion on whether the disease was contagious or miasmatic.
62Then Roux’s study arrived, with material for building a decisive argument for contagion. His map presented mortality in the 40 cities, dates for each city indicating when mortality began to worsen and when it began to fall, and the date on which daily mortality was highest (Figure 2). Over this information Roux had laid out the country’s main railroad routes. The dates for the appearance of the flu in the different cities showed that the epidemic had started in Paris, spread along the railroad routes to the main urban centres, and then moved on to secondary centres and the countryside. Roux’s study therefore offered evidence that the disease’s chronological spread lined up with the hierarchy of urban sites in France: from the capital to provincial cities. And the reason that some distant cities had been affected more quickly than cities closer to Paris was their position on the railroad network. By presenting and examining statistical data on 40 French cities, the study had found evidence of contagion through train travel.
Map of rising mortality caused by the 1889–1890 flu epidemic in France’s 40 main cities
Map of rising mortality caused by the 1889–1890 flu epidemic in France’s 40 main cities
63The statistics Roux contributed to the debate on whether the flu was contagious were not the only arguments supporting the contagion hypothesis, though that hypothesis was still energetically combated at the start of the epidemic. In France and elsewhere, other more standardly epidemiological ‘evidence’ and observations were put forward, including accounts of how the epidemic spread in isolated sites and institutions, such as asylums or lighthouses, and of the role of ocean liners in spreading the disease between ports of call (Parsons, 1891 and 1893; Proust, 1892b). Like Roux’s statistics, those arguments were not always received positively. The chain of flu outbreaks did not formally attest that cases had been imported because it was hard to determine the exact date of the first case in each community.
64However, the statistical evidence did support the ‘contagion’ side. Bacteriologists were having difficulty isolating a specific flu germ (Pfeiffer’s bacillus) that complied with the new science of bacteriology (Bresalier, 2012; Gradmann, 2014; Vagneron, 2015).  Proust, who drafted the official report on the epidemic in France, was careful to include a full-page reproduction of Roux’s map in the publication he delivered to the Academy of Medicine and the advisory committee on public hygiene in France (Proust, 1892a and 1892b). Reports on the epidemic were also being published in the early 1890s in other European countries—England, Germany, and Switzerland (Parsons, 1891 and 1893; Friedrich, 1894; Schmid, 1895)—so Proust had the opportunity to disseminate the new contagion theory beyond French borders:
When the flu was transported from one country to another, it jumped immediately from the first capital that was struck to the furthest point on the line of communication; that is, from Saint Petersburg to another capital, Paris or Berlin. Only later did it attack points in between. Those leaps backwards or in reverse can only be explained by person-to-person contagion.
66By identifying the role played by the period’s fastest means of travel and communication in unifying French territory and Europe as a whole during the pandemic, the Interior Ministry’s new office of public hygiene made a considerable contribution to the statistical description of a health event. We could even say that it constructed the national and international dimensions of that event. It is highly likely that the study also reinforced the institutional legitimacy of the ministry department that produced it and further illustrated the scientific utility of health statistics, a point that was still being contested in France (Lowenthal, 1902). The statistics-based argument also supported use of the term ‘pandemic’, which the press had only used until then in descriptions of historic pandemics and in reference to the rapid expansion of the telegraph across the European continent.
67But the path-breaking ad hoc study run by Roux did not lead to implementing routine flu surveillance at the Interior Ministry. Following the law of 15 February 1902 on public health protection, the flu was classified as a disease that did not have to be reported (order of 10 February 1903). Only in 1906, after Bertillon’s project for an international classification of diseases was adopted, did the flu join the causes-of-death list used in health statistics in France. As an independent cause of death, then, influenza appears late in the annals of French health statistics. Thanks to the new procedure of keeping health statistics, the seasonal flu epidemics of 1907 and 1911 stood out clearly (causing, respectively, 16,000 and 9,600 deaths). And yet, as we learn from remarks made in 1912 by Léon Mirman, the new director of the Interior Ministry’s Bureau of Public Health and Hygiene, the authorities had by then turned most of their attention to ‘social scourges’ (Mirman, 1912).
The statistical account of the epidemic was carried out with greater care than previous accounts, and constant progress in this science as applied to demographic movements suggests that if this long-standing pandemic reappears in Europe, we will know better how to study and count its developments, features, and the power it has to spread. At a time when clinical study, meteorological observation, and microbiology have acquired the means to conduct increasingly fine and accurate investigations, we have reason to hope that our current uncertainties about the flu will ultimately come to an end.
69As Proust stressed, in no flu epidemic prior to that of 1889–1890 had statistics been used so constructively in France to describe the course of an epidemic and its medical and demographic manifestations.
70In the space of 3 years, and in striking contrast to the medical corps’ uncertainty about the aetiology of the disease during and after the health crisis, the distinct statistical studies presented in this article led to new knowledge on the epidemic. How the collected data were processed, the way they were presented graphically, ensured unprecedented visibility for the medical, hygienist, and administrative statistical construction of the many dimensions of the pandemic in France.
71The statistics established the gravity of the epidemic at the scale of a century of national population trends. The 1890 shift into negative population growth caused by the 60,000 flu deaths and exacerbated by the falling trend in births was, for the 19th century, an exceptional statistical event comparable only to the negative growth situations observed during the cholera epidemic of 1832 and the war of 1870–1871. What made the rise in nationwide mortality during that 1889–1890 epidemic so striking was how sharply it contrasted with the effects of other epidemic diseases in the late 19th century (cholera, plague, and typhus): those diseases had been regressing since the 1870s and had almost disappeared in metropolitan France and other highly developed European countries, becoming much more sporadic and local (Hardy, 1993; Condrau and Worboys, 2007) and making the ‘ravages’ of other ‘social scourges’ that much more visible.
72The diversity of statistical approaches to the 1889–1890 flu attests that France was experiencing a golden age in administrative statistics. But the phenomenon of the epidemic itself and scientific statistical study of it reveal not only the scientific consensus on excess mortality during the winter of 1889–1890 but also the statisticians’ differing objectives. Bertillon’s, Turquan’s, and Roux’s distinct approaches demonstrate how statistics could be used differently according to whether the primary concerns were medical, demographic, or policy-related. Those divergences are reflected in how the spatial dimensions of the epidemic were constructed, whether or not the statistician was also implicated in local and/or national policy issues, and how targets for public action were designated. They are also expressed in the references used to assess the scope of the flu epidemic. Whereas Bertillon linked his study to the precedent of the 1847 flu epidemic in England, Turquan focused his on demographic trends over the 19th century within the French nation. But while Bertillon’s comparison between 1847 and 1889–1890 drew attention to flu epidemics, the link Turquan established between the 1890 shift to negative population growth and the longer-term ‘depopulation’ of France worked to subsume the flu under a ‘scourge’ deemed greater and more urgent for the country’s future. Paradoxically, the visibility given to the 1889–1890 flu epidemic by statistical study of it was ephemeral largely because the mortality found by that study was part of what led the country’s 1890 natural increase to topple into negative territory—a much more spectacular phenomenon.
73The success of measuring flu epidemics by excess mortality has held firm since the late 19th century. What is remarkable in this statistical method is that it allows for apprehending the flu and its complications indissociably during both pandemic and seasonal episodes—regardless of the disputes on aetiology that have continued to exercise the scientific community (Dehner, 2012). Mortality from the 1918 Spanish flu was constantly reassessed in the 20th century due in large part to the move to take excess mortality into account: estimated at 20 million victims in 1920, the count for that flu pandemic rose to between 50 million and 100 million worldwide in the early 21st century (Johnson and Mueller, 2002; Vagneron, 2018). But these disastrous and much circulated figures on the Spanish flu proved misleading in connection with the bird flu episode of 2009. The most alarmist projections in that case, based on the example of 1918 (Osterholm, 2005), ultimately proved unsound during the A/H1N1 flu episode (Wilson, 2012). Not only was the virulence of the 1918 virus exceptional, but excess mortality, due in 1918–1919 to complications, would be much lower today thanks to antibiotics, for example. While figures from the past have been used primarily to guide policy—specifically, to reinforce international mobilization against the next flu pandemic (Gilbert, 2015)—we also need to know how those figures were constructed, to prevent both their instrumentalization and the harm that would come from ignoring them.
Frequency of flu-related deaths in the main cities of central EuropeInterpretation (Bertillon’s specifications): ‘ The number of “deaths attributable to the flu” was assessed by subtracting the number of deaths observed in November 1888, December 1888, and January 1889 from those observed for the same three winter months in 1889 and 1890. Each horizontal line under a city’s name indicates 50 “deaths attributable to the flu” per 100,000 inhabitants in that city. Above the name of each city are 13 small circles, each representing a week (4 circles for November, 4 for December, and 5 for January). For weeks when the number of deaths observed exceeded twice the average number, the circle is black. For weeks when that number was greater than 1⅓ times the average, the circle has a vertical line through it. And for weeks when mortality was only slightly higher than average, the circle is empty.’
The 1889–1890 flu epidemic in France’s 40 main citiesInterpretation (Proust’s specifications): ‘The black line under city names indicates mortality per 10,000 inhabitants, at the scale of three deaths per millimetre. The dates below the city mark the start and finish of exceptional mortality. The letter M followed by a date indicates the day mortality was highest, taking the middle of the week with the highest mortality as indicated in the general overview table. Railroads.’
A pandemic is an epidemic that has spread beyond a local hub or centre.
My analytic framework here is similar to the one used by Laroche and Boudès (2006) to study official reports on France’s 2003 heat wave.
On how the epidemic spread across the rest of the globe, see Parsons (1891, 1893).
A few years later, the British physician Clemow (1894) was able to demonstrate that the first cases had indeed appeared in central Russia.
On the importance of sociological analysis of post-crisis formalizations, see, for example, Laroche and Boudès (2006).
The mortality indicator most often used in these studies was total number of deaths. The age structure of mortality was available only for Paris at the time. The mortality calculations Roux presented in his last study were based on data from the countries’ main cities.
See, for example, Jules Rochard, 1890, ‘Chronique de l’hygiène’, Le Temps, 9 January.
Six age groups were used in Paris: 0–1, 1–4, 5–19, 20–39, 40–59, and 60 and over.
The Paris system for certifying deaths was an exception in France at the time. In 1889, certification was the job of 88 physicians covering the districts of each arrondissement (Archives de Paris, D1K346 ).
The diseases in question were typhoid fever, smallpox, measles, scarlet fever, whooping cough, and diphtheria.
A few years later, in 1893, Bertillon promoted a new international classification of causes of deaths that was adopted in 1900 by the International Commission for the Revision of Nosological Nomenclature. One of Bertillon’s greatest ambitions was to produce a tool for standardizing health statistics (Vallin and Meslé, 1988).
To do so, Bertillon drew on a Register General study signed by George Graham that was most likely the work of the statistician William Farr. Though Bertillon does not cite Farr, he was surely familiar with Farr’s work on causes-of-death nomenclature (Dupâquier, 1983).
Cartograms are maps on which statistical differences take the form of visual differences. Bertillon’s father, Louis-Adolphe Bertillon, argued ardently in favour of this presentation method in the introduction to his La démographie figurée de la France, ou Étude statistique de la population française avec tableaux graphiques (1874).
Turquan calculated the number of deaths due to ‘diseases of the organs of respiration’ per neighbourhood compared to deaths from ‘all causes’. He then calculated that proportion per 1,000 inhabitants by the population of the given district. The results were as follows: Salpêtrière: 8.4 per 1,000; Santé, Plaisance: 4.3; Père-Lachaise: 3.9; Bercy: 1.5; Palais-Royal: 1.6; Invalides, Chaussée d’Antin, Porte Saint-Denis: 1.8. Turquan acknowledged that these proportions could not be extrapolated to people infected with the flu, on whom he had no data.
In 1887, Turquan had developed a map of the population density in France that drew considerable attention (Turquan, 1887).
An excerpt from a report to the Minister of Commerce, Industry, and the Colonies published in the Journal officiel of 21 October 1891.
‘Let the Consuls beware’: the opening words of a Roman Senate decree used in connection with extending the powers of the Consuls in response to crisis situations. The rest of the expression reads ne quid detrimenti respublica capiat (‘that no harm befalls the Republic’).
Initially called National Alliance Against Depopulation (De Luca Barrusse, 2008a).
The idea was that the flu had ‘accelerated the fatal dénouement’ among individuals whose death ‘would not have been long in coming’ (Bertillon, 1892, p. 109). Bertillon himself did not use the term ‘harvesting effect’ (effet de moisson), but the phenomenon does correspond to what is observed during heat waves, for example (Toulemon and Barbieri, 2008).
‘Social scourges’ included tuberculosis, alcoholism, syphilis, and (later) cancer (Faure, 1994).
Bertillon also mentioned this effect in 1892 in relation to the situation in Paris and other European cities. He had not previously observed any effect on the number of abortions or still births during the epidemic (Bertillon, 1892).
Roux became director of the public hygiene office in July 1901.
Statistique sanitaire des villes de France, Année 1890 et période quinquennale 1886-1890, Imprimerie administrative, Melun, 1891.
The ‘assistance’ component of this Interior Ministry department had three sections, whereas the ‘public hygiene’ component had only one, which had been recently transferred from the Commerce Ministry (Renard, 2000; Paquy, 2001).
Monod’s 1892 monograph on the 1885–1886 cholera epidemic in the Finistère [département of Brittany] received a considerable amount of attention (Monod, 1892).
No copy of the actual report could be found.
See the Bulletin de l’Académie nationale de médecine, 26, 1891, p. 728; session of 15 December 1891. The distinction was created in 1885, following the 1884 cholera epidemic.
Bulletin de l’Académie nationale de médecine, 1889, pp. 664–666; session of 17 December 1889.
L’écho de Paris, interview with Louis Pasteur conducted by Jules Huret, 13 January 1890.
The first flu viruses were identified only in the early 1930s, at which time Pfeiffer’s bacillus, renamed haemophilus influenza, was relegated to the status of a microbe responsible for certain flu complications.