→Asecondwaveofinfectionswoulddisproportionatelyharmtravel hubs
→Italiansarestillstayingathome,but travel in Germany has bounced back
Sources:Teralytics; Eurostat; US Census Bureau; national statistics; The Economist
*Not nationwide †A measure of how many different areas a given place exchanges visitors with regularly
‡Assuming countries’ transport patterns return to pre-pandemic behaviour, while holding current infection rates constant
Trips per day, seven-day moving average
February15th2020=100
Interconnectednessv covid-19cases,bygeographicarea
Germany,ItalyandtheUnitedStates,threeweeksafterlockdown
→Covid-19ismoreprevalentinbetter-networkedcities
Total covid-19 cases, actual v expected‡ May 11th 2020
Per 100,000 people,selectedcities,logscale
Italy UnitedStates Germany
Highercentrality
Lowercentrality
0
25
50
75
100
0
25
50
75
100
0
25
50
75
100
Feb Mar Apr May Feb Mar Apr Feb Mar Apr May
Milan
Berlin
NewYork
Chicago Munich
Venice
Nationwide Nationwide Nationwide
Lockdown *
10
100
1,000
-40 -20 Average 20 40 60
Network centrality† compared with country average, %
Casesper100,000
people,log scale
←Lessconnected Moreconnected→
Low High
Population density Trento is thinly populated,
but well connected. It has a
high infection rate
Milan
Rome
Ragusa
Berlin
Munich
Houston
New York
Bari
Berlin
Washington,DC
Houston
Milan
NewHaven
Philadelphia
Hanover
Rome
Frankfurt
Munich
Verona
50 100 200 500 1,000
↖Houston had fewer cases
than its high connectedness
and density suggest
Actual Expected
N
ow thatthe first wave of covid-19 in-
fections has crested, governments are
starting to relax their lockdowns. In Italy
shops will open their doors from May 18th.
Parts of Germany and America are also re-
opening on a state-by-state basis. Mobile-
phone data show that people are buzzing
around a bit more than they did in April.
Greater mobility raises the risk of a sec-
ond wave of cases. For countries where
policies are set locally, a big worry is that
outbreaks could begin in areas with lax
rules and spread elsewhere. In theory, this
risk should mirror “interconnectedness”—
the amount of travel to and from each re-
gion. One possible explanation for why
Lombardy was hit so hard by covid-19 is
that it is the best-networked part of Italy.
Teralytics, a Swiss technology firm, has
compiled data from Germany, Italy and
America that support this hypothesis. Each
time a mobile phone leaves one location
and arrives at a new one for an hour or
more—whether such travel is within a city
or for longer distances—Teralytics logs the
journey. In the week before lockdowns be-
gan, the firm recorded 5.7bn trips. Travel
fell by 40% once they were implemented.
To test how interconnectedness affects
vulnerability to covid-19, we built two sta-
tistical models to predict local infection
rates during the period just before lock-
downs. The first relied solely on each area’s
population density and income. The sec-
ond added on two measures of propensity
for travel: its number of journeys and its
“network centrality”, or how many other
places it tends to exchange visitors with.
The more elaborate model fared better,
with 30% more explanatory power than re-
lying on population density and income
alone. Interconnectedness matters a lot. In
all three countries, better-networked areas
had more infections than the simple model
predicted. Less-networked ones had fewer.
Governments should treat travel hubs
with caution. So far, many German cities
have seen surprisingly few infections—
perhaps because the country tests widely,
and began locking down earlier in its epi-
demic (as measured by the death toll) than
Italy did. Now that Germany is easing re-
strictions, its infection rate may rise again.
Well-networked Frankfurt is probably at
greater risk than, say, comparatively dis-
connected Hanover, and should reopen
relatively slowly. Milan in Italy, and Hous-
ton in America, should be cautious, too. 7
Phone data identify travel hubs at high
risk of a second wave of infections
The covid network
The EconomistMay 16th 2020 81
Graphic detailTravel and covid-19