Interactive companion

Pandemics and Cities

Evidence from the Black Death and the long run · Remi Jedwab, Noel D. Johnson, and Mark Koyama · Journal of Urban Economics, 2024 · Read the paper · Research

1Argument

The Black Death killed about 40 percent of Europe's population between 1347 and 1352. Some cities were spared and others lost more than half their people. Unlike a war or a natural disaster it destroyed no buildings, it did not target one group, and no government reconstructed anything afterwards. That makes it close to a pure population shock, and a test of what holds a city in place.

The paper uses city-level mortality for 165 cities that held about 60 percent of Western Europe's urban population in 1300. A city with 10 percentage points higher mortality was about 8.7 percent smaller in 1400. Two centuries later the average effect was zero, and the recovery came mostly through migration from places the plague had spared. The average hides a reshuffling. Cities with good land and access to trade, on a coast, a river, a road network, or in the Hanseatic League, recovered faster. Cities without them could stay smaller for centuries.

2The shock

Mortality rates come from Christakos et al. (2005), who compiled them from parish records, wills, tax rolls, chronicles, and other sources and checked conflicting estimates against each other. For most cities the source gives a percentage. For the rest the authors converted a literary description, a clergy death rate, or a desertion rate into a percentage using the rules in the paper. Populations come from Bairoch (1988), revised with later scholarship.

Figure 1. The 165 cities of the main sample. Circle area is population in 1300. In the first view color is the cumulative mortality rate in 1347–1352. In the second view color is the city's population in the chosen year as a share of its population in 1300, so a city that had regained its pre-plague size is shown in teal. Mortality from Christakos et al. (2005) as coded in the paper. Populations from Bairoch (1988) as revised by the authors.

3Where mortality fell

The design rests on mortality being unrelated to what made a city grow. The plague entered Europe at Messina in Sicily in October 1347, on Genoese ships from Kaffa in the Crimea. Messina was only the 45th largest city in Europe and it was not the only possible port of entry. The disease was most virulent in its early months, so cities reached early through trade with Messina suffered most. Paris, London, Cologne, and Lisbon were among the largest trading cities of Europe and were infected later, with lower mortality. The fleas that carried the disease were most active in warm weather, so a city whose outbreak peaked in summer lost more than one whose outbreak peaked in winter.

Figure 2. Mortality against the number of months between October 1347 and the plague's arrival in the city, for the 124 cities with an onset date. Hover for the city.
Figure 3. Mortality against log population in 1300 for the 165 cities, with the fitted line. The fit is Y = 42.5 − 1.01X with an R² of 0.00, as in Figure 3(a) of the paper.

Large cities were not hit harder than small ones. Across 27 city characteristics covering physical geography, access to markets and trade, and institutions, only proximity to rivers and latitude predict mortality on their own, and the river coefficient is negative. Once all controls enter together none is significant. The results also hold with country, state, and grid-cell fixed effects and with two instruments, distance to Messina and the month of peak infection.

4Recovery

The baseline regression relates each city's percentage population change over a period to its mortality rate, weighting each city by its population at the start of the period. Over 1300–1400 the coefficient is −0.87, close to the −1.00 the shock itself implies, so there was little recovery in the first half century. By 1300–1500 it is −0.28 and by 1300–1600 it is positive and insignificant. Before the plague, from 1200 to 1300, cities that would later suffer high mortality grew no differently from the rest.

coefficient on mortality, with robust standard error
weighted mean population change
cities
Figure 4. Percentage change in city population against Black Death mortality for the chosen period. Circle area is population at the start of the period, which is the regression weight. The line is the weighted least squares fit. The readouts are computed from the replication data and match Table 1 of the paper. Populations for 2015 are for the city itself rather than its agglomeration, as in the paper. Cities whose growth runs off the top of the chart are counted under it and still enter the readouts. Hover for any city.
Figure 5. The effect of mortality on population change since 1300, as beta coefficients, which express it in standard deviations so periods of different length can be compared. The 1200–1300 bar is the pre-plague placebo. Values from Table 1 of the paper.

The recovery has to have come from migration. Barcelona, Florence, Lübeck, and Venice regained their pre-plague populations within 5 to 30 years, which would have required rates of natural increase above 30 per 1,000, rates unheard of until the twentieth century. In England, villages far from cities were deserted more often where the plague had been mild, so people left the spared countryside for the stricken towns.

5Which cities came back

A zero average effect after two centuries does not mean every city recovered. The paper interacts mortality with eleven city characteristics in a single regression, so each interaction shows how much faster a city with that characteristic recovered from the same shock. A positive interaction means faster recovery. The outcome is standardized, so the interaction can be read against the direct effect of mortality in the same period.

Figure 6. The chosen interaction with mortality and the direct effect of mortality, by period since 1300, with 95 percent intervals from robust standard errors. All eleven interactions are estimated together. Computed from the paper's replication data with the specification of Table 8, and every coefficient and standard error matches the published table. The authors caution that these interactions are associations conditional on mortality and the other characteristics.

Coastal cities recovered faster from the start, and by 1500 a coastal city with high mortality had on average fully recovered relative to low-mortality cities. The Hanseatic interaction in 1300–1400 is −0.062 against −0.083 for mortality, so a Hansa town had recovered about three-quarters of its loss by 1400. Road and trade-route intersections mattered most in 1500. Rivers mattered from the seventeenth century, when investment in river transport and canals grew. Potato suitability helped from the seventeenth century, and land suited to pasture slowed recovery in the sixteenth century, when landlords turned to sheep because they needed less labor. The size of a city after the plague and its political institutions show no consistent recovery effect.

The authors use these estimates to ask which cities would be much smaller today without their favorable locations. Their list of large cities from 1300 is Venice, Florence, Cordoba, Naples, Cologne, Pisa, Toulouse, Rouen, and Marseille. Each was hit hard and each was coastal, on a river, at a road intersection, or in the Hanseatic League. They are outlined in black on the map in Figure 1.

6Permutations and the urban reset

Large cities in 1300 tended to be large in 1600, but the rank correlation leaves room for change. Regressing each city's rank in 1600 on its rank in 1300 gives an R² of 0.56, and many cities sit far from the diagonal.

Figure 7. Each city's population rank among the 165 in 1300 and in 1600, with 1 the largest. Cities above the diagonal fell in rank and cities below it rose. Color is mortality. Hover for the city.
Figure 8. Share of the population of all 1,801 cities living within 10 km of a coast, a river, or a Roman road or medieval trade-route intersection, or in a Hanseatic city, the definition in the paper's replication code. The dashed line continues the 1200–1300 trend. Computed from the replication data, which give about 89 percent in 1100, 83 percent in 1300, and 78 percent in 1700, with the trend reaching about 62 percent by 1700.

In the replication data 89 percent of the urban population lived near one of these trade factors in 1100 and 83 percent in 1300. Had the decline continued at its pre-plague pace the share would have fallen to about 62 percent by 1700. The actual share in 1700 was 78 percent. The paper's text gives 90, 83, 60, and 77 percent for the same four numbers. The trend broke immediately after the Black Death, which suggests the plague moved Europe's urban population toward better-connected places. The authors attribute this reset to the Black Death with caution, since other shocks could have produced a similar pattern.

7Pandemics and urban policy

The paper's findings bear on how cities respond to a large loss of population. Most cities recovered, and the reason was the fixed advantages of their locations and the investments already sunk in them, which kept wages high enough to draw migrants back. Mortality alone did not determine a city's future. A city with no locational advantage could stay permanently smaller relative to other cities after the same shock, because for those cities the effect of mortality in 1300–2015 is still more than two-fifths of its effect in 1300–1400. Urban systems as a whole were resilient, and the main mechanism of recovery was people moving to where labor had become scarce.

The setting differs from modern epidemics in ways the paper names. The 1918 influenza and COVID-19 killed particular subgroups of the population, while the Black Death did not target any group, destroyed no buildings, and drew no government help for recovery. The paper places itself in the literature on pandemics and cities surveyed by Baum-Snow, Glaeser, and Rosenthal (2022).

8Limits

Mortality is measured with error. About a third of the 165 rates are converted from literary descriptions, clergy deaths, or desertion rates. The paper shows the results hold when mortality is coarsely categorized and when only the most and least affected cities are compared.

Populations before 1500 are estimates, and the sample covers the 165 cities with both a 1300 population and a mortality rate. The paper shows the results hold with imputed mortality for a larger sample and when each major country is dropped in turn.

The recovery interactions in Section 5 are not instrumented. The paper presents them as associations conditional on the shock, since 165 cities cannot support instruments for eleven interactions.

Every figure except Figure 5 is computed from the paper's replication files. Table 1, including the 1300–2015 column, Table 2, Table 8, and Figure 3(a) reproduce exactly. Two numbers in the published text do not come out of the replication code. The rank regression behind Figure 7 gives the published R² of 0.56 and a slope of 0.76 against the 0.86 in the text, so the page does not report the slope. The trade-factor shares behind Figure 8 differ from the text by about one point in 1100 and 1700 and by two points in the 1700 trend, and the page reports the computed values. Figure 5 uses the published Table 1.