The Mortality Impact of Cholera in Germany

Stylized fact 1: Cholera was disproportionately urban

Cholera was predominantly an urban phenomenon with city death rates consistently outpacing rural ones – at times by an order of magnitude. Figure 3 presents two complementary measures for this urban cholera penalty.

Fig. 3

Urban Excess Cholera Mortality over Time. Note: Panel a displays annual average cholera death rates as percentages in urban areas (gray bars) and rural areas (black pluses). Panel b plots exponentiated coefficient estimates from a panel regression of logged death rates on a year-interacted urban dummy variable, controlling for time and county fixed effects. For example, a value of 3.5 indicates that cholera death rates were 3.5 times higher in urban areas than in their immediate rural hinterlands, conditional on time-constant cross-location heterogeneity and overall time trends. Gray spikes in Panel b denote 95 percent confidence intervals, and coefficients with a significance level below 1 percent are omitted. Ratios are displayed on a logarithmic scale for clarity.

Panel a of Figure 3 displays the annual average cholera death rate (in percent), separately for urban and rural areas. Urban areas consistently exhibited much higher mortality rates than rural areas. This difference scaled with overall cholera severity and was most pronounced during major outbreaks (e.g. 1831, 1849, 1852, and 1866). Panel b offers a refined perspective by showing conditional urban-to-rural cholera death rate ratios. These are estimated in a regression model that compares cities specifically to their immediate rural hinterlands. This approach controls for shared local conditions – geography, climate, and some socioeconomic factors – and isolates the urban-specific cholera penalty (see Figure note for details). On average, urban areas experienced cholera death rates 3.5 times as high as those in their rural surroundings. This relative urban penalty remained mostly stable over time but reached a factor of 15 during the particularly devastating 1866 outbreak.

Stylized fact 2: Mid-sized cities were most affected

Not all cities were affected equally by cholera; population density played a critical role. Panel a of figure 4 presents binned scatter plots showing the variation in cholera death rates across city size, separately for each pandemic. Polynomial regression fits (thick lines) exhibit an inverse U-shape, particularly during the 2nd and 3rd cholera pandemics (1831 to 1838 and 1848 to 1864, respectively). Death rates peaked in mid-sized cities with approximately 3,000 inhabitants, while both smaller and larger cities experienced significantly lower mortality. In contrast, rural areas (dashed lines) displayed a more monotonic upward trend, with higher death rates in more populous rural regions. ^[26]

Fig. 4

Population Size and Cholera Mortality. Note: Panel a shows average cholera death rates by population size bins, separately for urban (circles and solid lines) and rural (squares and dashed lines) areas. Pandemic-specific death rates were calculated by pooling all cholera deaths and dividing them by mid-pandemic population. Panel b shows residualized cholera death rates, pooling all years (1831 to 1895), but controlling for location and region-specific time fixed effects. This reflects within-location differences over population size levels, abstracting from time-constant local factors and shocks common to all locations in a region. Fit lines stem from local polynomial regressions, and population size is shown on a logarithmic scale for clarity.

Panel b of Figure 4 shifts the focus from comparing differently sized locations during a given pandemic to comparing a single location to itself at different population-size levels. Specifically, the panel shows residualized death rates and population levels after controlling for all time-constant location characteristics and region-specific time trends in a two-way fixed effects model. The inverse U-shape persists: for an average city, death rates increased with population size up to a threshold (1,000 inhabitants), then declined as population size grew larger. A similar, though less pronounced pattern, is visible for rural areas.

Stylized fact 3: Cholera began in the North-East and shifted South-West over time

Figure 2 highlights a strong Eastern bias in cholera exposure, with high mortality clusters in Imperial Germany’s easternmost territories, particularly around what is Central Poland today. Figure 5 further explores this geographical bias and reveals a gradual South-West shift in cholera’s epicenter over time.

Fig. 5

The Geographic Epicenter of Cholera. Note: Panel a shows the geographic centers of annual outbreaks, computed as mean latitude and longitude coordinates, weighted by the cholera death rate. Circle sizes indicate the overall severity of each outbreak, with larger circles representing more deadly years. For major outbreaks, the year is also labelled within the circles (e.g. “50” for 1850). Black crosses mark the locations of eight large cities for reference. Panel b projects the annual geographic centers onto a diagonal axis from Straßburg in the South-West to Königsberg in the North-East. The gray spikes depict the average bidirectional squared deviations from this projected center, capturing the spatial dispersion of each year’s outbreak. Horizontal dashed black lines show pandemic-specific average positions, weighted by each epidemic’s overall death rate. Letters on the vertical axis indicate city locations: Königsberg (Kaliningrad today), Posen (Poznań today), Breslau (Wrocław today), Berlin, Hamburg, Munich, Cologne, and Straßburg.

Panel a of Figure 5 shows the geographic center of each outbreak, calculated as the mean latitude-longitude coordinate of all affected locations, weighted by cholera death rates. Most outbreaks were concentrated in Germany’s North-Eastern provinces near Posen, including the most severe ones (e.g. 1831, 1848/1849, and 1866). However, secondary waves following major outbreaks (e.g. 1832, 1850, and 1867) shifted toward Central Germany, while smaller outbreaks in 1853, 1859, and 1892 focused on Northern Germany. The 1854 outbreak, centered in the South-West (Bavaria and Alsace-Lorraine), stands out as a notable exception.

Focusing on the most apparent spatial trend, Panel b projects the geographic centers onto a diagonal axis running from South-Western Straßburg to North-Eastern Königsberg (see figure 5 note for details). Although year-to-year variation (gray points) and within-year dispersion (gray spikes) are substantial, the pandemic-specific averages (dashed black lines) reveal a clear South-West drift over time. Each pandemic saw distinct outbreaks in the North-Eastern territories, but cholera’s overall geographic focus moved West over time.

Stylized fact 4: Cholera’s diffusion speed increased across and within locations

Figure 6 examines the temporal dynamics of cholera’s diffusion, showing how its spread accelerated both across regions and within locations over the 19th century. While cholera death rates declined over time, diffusion speed increased, indicating growing efficiency in outbreak propagation. Panel a measures cross-location diffusion speed as the average kilometers travelled per day by cholera in each year (see figure 6 note for details). Dashed black lines, indicating averages per pandemic, suggest acceleration over time. In the 1830s, cholera travelled at an average speed of 18 km/day. In 1892, this had increased to 36 km/day. Circles are proportional to the severity of each outbreak, indicating that deadlier outbreaks also diffused faster.

Fig. 6

The Acceleration of Cholera Diffusion across and within Locations. Note: Panel a shows cross-location diffusion speed, i.e. average location-pairwise distance (in kilometers) divided by the time (in days) between the start of their outbreaks. Panel b depicts within-location diffusion speed, measured as the average number of deaths per day during an outbreak spell. Circle sizes in both panels are proportional to the overall death rate; dashed black lines indicate pandemic-specific averages, weighted by each outbreak’s overall death rate. Uncertain dates have been approximated, with early, mid, late, and unknown days within a known month coded as days 5, 15, and 25, and 15. Results remain similar when these observations are dropped instead. Observations with unknown outbreak months have been dropped.

Panel b of Figure 6 focuses on within-location diffusion speed, measured as the average number of deaths per day within each affected location during an outbreak spell. Similar to cross-location trends, within-location diffusion also accelerated over time, though less dramatically. In earlier outbreaks, daily deaths averaged around 1.2, but by 1892, this number had almost tripled, indicating more intense and faster local outbreaks despite the overall reduction in mortality. Circle sizes again correspond to outbreak severity, reinforcing the observation that larger and deadlier outbreaks were associated with faster diffusion.

Stylized fact 5: Outbreaks grew more uniform across locations but clustered in space

Figure 7 compares to what extent local cholera outbreaks were similar to each other across locations. In Panel a, the coefficient of variation (CV) measures the relative dispersion of cholera death rates across locations within a given year. The decline in the CV over time suggests that locations became more similar in terms of local outbreak intensities. This trend suggests an overall convergence in the severity of cholera’s impact across locations over the 19th century, except for the last epidemic, which was highly dispersed with its disproportionate impact on Hamburg.

Fig. 7

The Cross-location Similarity of Outbreaks. Note: Panel a shows outbreak-specific coefficients of variation (CV) for death rates, i.e. their standard deviation over their mean. A higher CV suggests a more dispersed distribution. Panel a uses a logarithmic scale for clarity. Panel b shows outbreak-specific Moran’s I estimates for death rates, computed using inverse straight-line distance weighting. A higher Moran’s I suggests stronger spatial clustering of similarly affected locations. Circle sizes in both panels are proportional to the overall severity of each outbreak; dashed black lines indicate pandemic-specific averages, weighted by each outbreak’s overall death rate.

Focusing on the spatial dimension of dispersion, Panel b of Figure 7 plots annual Moran’s I values for cholera death rates. These measure the spatial autocorrelation of outbreaks, with higher values indicating stronger spatial clustering of similarly impacted locations. While the variation over time is high, pandemic-wise averages (black dashed lines) suggest that highly impacted locations increasingly clustered in specific regions rather than being more evenly distributed across space. ^[27] The opposing patterns in Panels a and b suggest a complex trend with respect to the concentration of outbreaks. While locations generally converged in terms of outbreak severity (via the CV), outbreaks also became more spatially clustered over time (via Moran’s I).

The origins of public health and sanitary innovation

Pandemic shocks are often associated with significant innovations, investments, and institutional reforms. However, the conditions under which such changes occur and the mechanisms driving them remain contested. Quantitative-causal frameworks are well-suited to investigate and generalize these dynamics. For example, recent studies suggest that the 1918/1919 Influenza pandemic spurred formal healthcare expansion, reinforced religious and scientific convictions, and undermined social trust. ^[29] Cholera historiography contains similarly intriguing claims that, however, have not yet been explored by quantitative social scientists.

Historians highlight 19th-century cholera epidemics as pivotal to the rise of public health and major sanitary investments. ^[30] Yet, key aspects of this narrative remain unclear. First, cholera’s perceived importance often exceeded its actual mortality impact relative to endemic diseases like typhoid. ^[31] Second, while rich anecdotal evidence links cholera to public health advocacy and reform, these processes spanned decades, often extending beyond cholera’s major mortality episodes. ^[32] Finally, the failure to replicate similar sanitary revolutions outside the Western world suggests that factors such as ideology, political institutions, economic incentives, and medical knowledge mediate the cholera-sanitation nexus. ^[33]

In contrast to the now decade-spanning literature on Influenza and the Black Death, researchers have only very recently begun to explore cholera’s role in public health in settings that abstract from single case studies. For example, recent studies identify capital-skill production complementarities as a main driver of elites’ willingness to finance sanitary infrastructure in face of cholera epidemics, and, show that 19th century British expert’s opinions on the contagiousness of cholera were shaped by their economic stakes in oversea trade relations. ^[34]

Epidemic shocks, political crisis, and social change

Epidemic shocks can ignite far-reaching political and social dynamics, though their directions and magnitude are difficult to predict as demonstrated by the COVID-19 pandemic. ^[35] Cholera, due to its salience, wide geographic spread, and focus on modern mass societies, provides a promising historical lens to examine how epidemics shape society and politics. Recent studies emphasize cholera’s role as one of the most conflict-prone epidemics in history, unleashing social clashes, scapegoating, and the infamous cholera riots across Europe and the world. ^[36] This perspective builds on an earlier tradition that has linked 19th-century European cholera epidemics to revolutionary and disruptive events, identifying them as both catalysts for and amplifiers of political change. ^[37]

Cholera also played a role in major international turning points, such as the 1830/1831 Polish November Uprising, the 1848 Revolutions and civil wars, and the Crimean War (1853 to 1856). ^[38] On a brighter note, Europe’s shared exposure to cholera spurred early international sanitary cooperation, laying the groundwork for the eventual creation of the World Health Organization. ^[39]

Despite this rich historical backdrop, scholars have barely begun to explore cholera’s sociopolitical impacts within explicitly quantitative-causal frameworks. A comparison to the well-studied 1918/1919 Influenza pandemic demonstrates the dormant potential. For example, recent studies show that exposure to this pandemic caused voters in Weimar Germany to reward parties seen as competent in public health while punishing governments that demonstratively neglected health infrastructure. ^[40]

Globalization, growth, and epidemic burden

The relationship between trade, globalization, and the spread of epidemic diseases has long intrigued economists and gained renewed attention during the COVID-19 pandemic. ^[41] Cholera, in particular, is often described as the disease of the First Globalization, with its pandemics closely tied to the expansion of global trade networks during the 19th century. ^[42] Within countries, growing transport connectivity and labor mobility also facilitated the spread of diseases. ^[43]

Cholera’s broad geographic reach was undeniably a byproduct of globalization – and, by extension, economic progress. Yet, it has also been portrayed as the symbol of pre-sanitary poverty, disproportionately affecting dense and impoverished urban areas. ^[44] A useful framework to further explore this duality is provided by Werner Troesken, who distinguishes between “diseases of poverty”, such as typhoid, and “diseases of commerce”, such as smallpox. ^[45] Institutional features conducive to economic growth – such as trade openness and labor mobility – can simultaneously mitigate poverty-related diseases while exacerbating commerce-related ones. ^[46] Cholera occupies a dynamic position within this dichotomy: while trade and connectivity facilitated its spread, it also spurred sanitary investments in highly integrated economies like Germany, where cities grew rich from trade and labor inflow.

The relationship between epidemic shocks, economic activity, and trade restrictions has been studied extensively in the context of the 1918/1919 Influenza pandemic. For instance, recent studies argue that exposure to the Influenza pandemic led to increases in tariffs, while migration barriers failed to limit mortality during four historical Influenza pandemics, including 1918/1919. ^[47] Other contributions show that interaction-reducing non-pharmaceutical interventions (NPIs) in the US did not suppress economic activity beyond the direct costs of Influenza itself; moreover, local Influenza outbreaks were severely amplified by industrial atmospheric pollution. ^[48] Despite cholera’s similarly complex entanglement with trade and economic growth, similar empirical studies do not yet exist.

The macroeconomic effects of epidemic shocks

How do epidemic shocks impact macroeconomic variables such as real wages and output? The COVID-19 pandemic has revived interest in whether evidence from historical epidemics can inform predictions about the macroeconomic consequences of future outbreaks. ^[49] Much of this literature focuses on the 1918/1919 Influenza pandemic and the Black Death, with cholera receiving relatively little attention. ^[50]

Overlooking cholera in this context represents a missed opportunity. Unlike pre-modern plague epidemics, 19th-century cholera outbreaks occurred during a period of economic transition, affecting both Malthusian economies and those shifting toward modern economic institutions characterized by economies of scale, human capital accumulation, and sustained growth. ^[51] Moreover, in contrast to the 1918/1919 Influenza pandemic, which saw widespread use of non-pharmaceutical interventions (NPIs), cholera epidemics largely played out in a laissez-faire policy environment, with few interventions apart from the initial quarantines and border control instances noted earlier. ^[52] Cholera epidemics also differed in their temporal dynamics, unfolding over multiple waves of varying magnitude across the 19th century, rather than constituting a one-time shock.

Potential avenues of research extend beyond real wages and output. For instance, evidence suggests that Prussian cities quickly recovered their pre-epidemic population levels after major cholera shocks. ^[53] Recent work has also shown that cholera epidemics increased inequality in 19th-century France, induced long-term health costs from early-life exposure in Japan, and stimulated technological innovation in the French agricultural sector due to labor scarcity. ^[54] These findings, which plausibly echo expected macroeconomic effects of epidemic shocks in modern contexts (e.g. COVID-19), highlight the broader potential of studying cholera to understand how economies respond to epidemics.

Cholera, climate change and resource management

Cholera remains a significant global threat. Studies suggest that it may become more severe in the future as climate change expands cholera’s oceanic habitats, clean water resources grow scarcer, and urbanization accelerates. ^[55] However, many of the hydrological, climatic, and socioeconomic factors that influence cholera evolve slowly over time. Moreover, epidemiological risk assessment models require extensive data. For these reasons, historical quantitative evidence plays an increasingly important role in outbreak modeling and speculation about cholera’s future epidemic significance.

Of course, the limitations of historical comparisons must be acknowledged. Societal and economic structures have changed significantly, with important implications for cholera epidemiology. ^[56] The modern El Tor biotype differs from earlier strains in both longevity and fatality, and treatment methods have improved and become more accessible. ^[57] However, these limitations must be weighed against the opportunities historical research provides. Historical data offers rich variation across time and space, including fluctuations in climate and geography that contemporary data cannot deliver. A recent example for the creative use of long historical data series is a study of the relationship between changing climate conditions and cholera in Kolkata over 120 years. ^[58] Other examples are the use of historical records from London, Baltic port cities, and Denmark to estimate key parameters and transmission mechanisms within epidemiological models. ^[59]

6.1 Regional Excess Mortality Strongly Correlates with Cholera Epidemics

Excess all-cause mortality can be a useful proxy for cause-specific epidemic death counts. To assess excess mortality, I collect population and all-cause deaths data for each state in the dataset (see appendix B for details). Given their size, each Prussian district is treated as a separate “state”, resulting in 63 regional entities. All-cause mortality, derived from vital statistics, are available annually from 1816 to 1914, though some states have gaps in early years. ^[60]

I regress annual deaths D_r,t on state-specific base levels and time trends, adjusting for the population at risk P_r,t within a conditional exponential means model. Due to the observed mortality transition – a sudden, consistent downward trend in overall mortality – I allow the time trends to vary around a state-specific break year B_r, estimated through Supremum Wald tests. The regression model is as follows:

Using the fitted regression model, I obtain predicted deaths Dˆr,t for each state-year.

In Figure 8, I compare these predictions to observed deaths, expressed in per-capita terms to accommodate different state sizes. The tight, symmetric dispersion around the 45° line indicates that the observed mortality dynamics between 1816 and 1914 are well captured trough state-specific time trends, allowing for differential onset of the mortality transition. However, observations further away from the identity line indicate unexpected excess or deficit mortality (below or above the 45° line, respectively). Among these, I highlight state-years with cholera accounting for more than 10 percent (triangles) or 20 percent (crosses) of the observed deaths, according to my dataset. Almost all of these cholera years show high excess mortality, among them the most extreme years. ^[61]

Fig. 8

Excess Mortality and Cholera Years. Note: This graph plots observed per-capita deaths for 5,753 state-year observations against predicted deaths (also in per-capita terms) according to a time trend regression model (see text). Observations highlighted in red are cholera years, in which cholera accounts for more than 10 percent (triangles) or 20 percent (crosses) of all deaths. Out of the 49 (24) state-years in which cholera accounted for 10 to 20 percent (more than 20 percent) of all deaths, 45 (24) state-years also experienced excess mortality, i.e. more observed deaths than predicted deaths. Sources: Own calculations based on the dataset presented in this paper.

6.2 Climate and Infrastructure Risk Factors Predict Cholera Outbreaks

I test whether the observed mortality patterns align with established environmental and infrastructural risk factors. Studies demonstrate that cholera incidence correlates with climatic, hydrological and infrastructural conditions. Most evidence comes from coastal and tropical areas near cholera’s endemic regions, such as India and East Asia, but comparative studies from Africa, the Arabian Peninsula, and the Caribbean have also linked cholera risk to environmental factors. ^[62] Among the most established risk factors are high temperatures, heavy rainfall, and contaminated water supplies. ^[63]

The specific role of these three factors can vary substantially across locations, time and context. ^[64] However, at the most basic level, a large body of evidence links cholera risk to warm temperatures, heavy rainfall and poor sanitary conditions. ^[65] Antarpreet Jutla et al. present a simple integration of these risk factors in epidemic (i.e. non-endemic) settings. ^[66] Their framework posits that if all three conditions are met – high spring temperatures (inducing bacterial growth, low river levels and salination), followed by heavy summer rains (causing cross-contamination of water reservoirs), in regions with vulnerable water supplies – the likelihood of a cholera outbreak increases. If any condition is absent, cholera risk diminishes. This model’s simplicity, focusing on a limited set of factors, allows for the inclusion of interactions and temporal structure while remaining tractable. Moreover, each factor has been tested across diverse settings. ^[67]

I use the model offered by Antarpreet Jutla et al. to indirectly test the plausibility of the German cholera dataset: If the model captures basic epidemic probabilities, and the data is accurate, cholera epidemics should be most severe in locations and at times subject to the presence of all three risk factors. To test this implication across locations i (in regions r) and years t, I define dummy variables for the absence of risk factors: Temperature risk is absent (T_i,t = 1) when spring temperatures are below their 15-years rolling average. Rainfall risk is absent (R_i,t = 1) when summer precipitation is below its 15-years rolling average. Finally, water quality risk is absent (W_i,t = 1) once a city builds a safe piped water supply. ^[68]

The estimation equation reads

where j = 1, … ,7 represents the seven possible combinations of risk-lowering factors – low spring temperature (T), low summer rainfall (R), and safe water infrastructure (W) – with the absence of all three as the omitted base category. Here, C is either an outbreak dummy indicator or the cholera death rate. Location fixed effects (α_i ) and region-year effects (λ_{r (i)×t}) control for unobserved, time-invariant risk factors and regional shocks.

The regression results in Table 2 indicate that locations meeting none of the three risk-lowering factors had an outbreak probability of 6 to 9 percent (cols. 1-3) and death rates of 2 to 5 percent (cols. 4-6), with cities showing a higher cholera incidence (cols. 2 and 5), especially those whose center is within one kilometer of a river (cols. 3 and 6). Below the baseline rates, the table shows marginal deviations (percentage point changes) from these baseline outcomes when one or more of the three risk factors is not met. The model’s prediction is that any absence of risk factors lowers cholera risk. The findings support this prediction as most coefficients are negative and many are significant. ^[69] This, in turn, indirectly validates the quality of the dataset employed for the estimation.

6.3 Cholera Events in the Dataset are well Represented in the German City Encyclopedia

While the dataset presented in this paper is the first to use systematic sources on cholera events in German cities, the multi-volume German city encyclopedia (Städtebuch) contains sections on historically significant epidemic events for a large sample of cities throughout history. ^[70] I rely on a recent transcription of these books and compare whether epidemic events mentioned in this source are also found in my dataset. ^[71] This is a meaningful validation, because the mention of a significant cholera epidemic in the city encyclopedia indicates that the respective authors referenced this event across multiple sources and deemed the information reliable. In Figure 9, I present several comparisons between both datasets.

Fig. 9

Cholera Event Mentions and Death Counts in the Städtebuch and Dataset. Note: This figure presents a comparison between cholera epidemics data mentioned in the German city encyclopedia and my dataset. Panel a is a crosstable, counting city-year instances according to whether cholera outbreaks are mentioned in my dataset (column A) or not (column B), and whether they are mentioned in the city encyclopedia (row X) or not (row Y). For example, 545 outbreaks are mentioned in both sources, representing 0.36 percent of all city-year instances. Panel b plots death counts reported in both datasets, based on 297 cholera epidemics for which such data is provided in the encyclopedia. Epidemics that are located on the dashed black 45° line have the same magnitude reported in both datasets. Sources: City encyclopedia data from Keyser et al. (Eds.), Städtebuch; Bogucka/Cantoni/Weigand, Princes and Townspeople.

In Panel a of Figure 9, I compare mentions of cholera events in 2,351 cities across both datasets. I register all “Cholera” and “Brechruhr” events in the city encyclopedia for the time span 1831 to 1895 (65 years), yielding 152,815 city-years. ^[72]

The cross-tabulation shows that, out of the 684 outbreaks mentioned in the city encyclopedia, 545 are also featured in my dataset, suggesting that these sources generally align. Moreover, a large share of the 139 events not listed in my dataset is explained by one-year deviations, e.g. when the same outbreak is coded under the years 1831 or 1832 in the two datasets, respectively. Beyond this positive check, the cross-tabulation also implies that 2,858 outbreaks filed in my dataset are not mentioned in the encyclopedia. This was to be expected, given that reporting in the Städtebuch depended on the subjective assessment of the respective authors, who often focused on significant epidemic events and did not necessarily aim at comprehensiveness. Given these limitations, the high odds ratio of agreement between both sources,

is a strong validation for my dataset.

In Panel b, I compare cholera death counts across both datasets for 297 city-years for which the city encyclopedia reports these. The high correlation (with an R-squared of 0.9) indicates agreement, although the city encyclopedia reports slightly higher death counts on average.