medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020.

The copyright holder for this preprint (which

was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.

Full lockdown policies in Western Europe countries have no evident

impacts on the COVID-19 epidemic.
Thomas Meunier1,2∗
1
Woods Hole Oceanographic Institution, Falmouth, Massachusetts
2
Ensenada Center for Scientific Research and Higher Education, Ensenada, BC
April 24, 2020

Abstract taken in response to the epidemic, from no action at

all (Sweden) to full lockdown (Italy, France, Spain
This phenomenological study assesses the impacts and United Kingdom), including police-enforced
of full lockdown strategies applied in Italy, France, home containment. Other countries, such as the
Spain and United Kingdom, on the slowdown of the Netherlands and Germany, opted for a measured
2020 COVID-19 outbreak. Comparing the trajectory response, encouraging social distancing without
of the epidemic before and after the lockdown, we locking their population down.
find no evidence of any discontinuity in the growth While new medical treatments proposed to cure
rate, doubling time, and reproduction number trends. COVID-19 cases are required to be validated through
Extrapolating pre-lockdown growth rate trends, we controlled double blind studies, the benefits and risks
provide estimates of the death toll in the absence of of social distancing strategies are not subject to any
any lockdown policies, and show that these strategies comparative tests. However, full lockdown measures,
might not have saved any life in western Europe. We such as those decided in Italy, France, Spain and
also show that neighboring countries applying less United Kingdom have not been experienced in West-
restrictive social distancing measures (as opposed to ern Europe countries for centuries, and their effects
police-enforced home containment) experience a very in contemporary population’s mental and physical
similar time evolution of the epidemic. health is largely unknown. The COVID-19 epidemic
episode was shown to, by itself, affect mental health,
including anxiety syndromes and depression [21]
Introduction and the consequences of isolation could enhance
these conditions. In the absence of any control
The recent COVID-19 outbreak in Europe has group, the impacts on western Europe’s population
challenged the governments responsiveness in front will not be measurable until months. Nevertheless,
of an unpredictable and unprecedented situation. increased mortality due to difficulties of access
Since most countries were unprepared to face such to basic health care, increased mental conditions
an unexpected epidemic, lack of testing capacities linked to isolation, as well as social consequences of
yielded most policies to shift towards social distanc- economic recession, despite being unquantifiable so
ing measures rather than modern laboratory-based far, is to be expected. Such measures are thus only
quarantine [10]. A broad range of public actions were appropriate if their impacts on limiting the epidemic
∗ Correspondence
spreading save more lives than their inherent death
to T. Meunier, tmeunier@whoi.edu

1
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.

toll. Attempting a real-time assessment of full changes are noticed in the trend of the daily death
lockdown policies efficiency thus seems crucial to growth rate, doubling time, or reproduction number,
help public action decisions in the forthcoming weeks. weeks after lockdown policies should have impacts.

Recent modeling results suggestes that China’s

full lockdown policy was successful in containing the 1 Methods
epidemic [9]. In an attempt to predict the efficiency
Although Epidemic outbreaks are complex dynamical
of similar policies in Western Europe countries (Italy,
systems, the daily new cases number most generally
France, Spain and United Kingdom), Picchioti el
follows a similar time evolution: after an exponen-
al. (2020) [16] implemented a SEIR model, testing
tial growth, infections slow and eventually decay ex-
different lockdown parameterizations, and suggested
ponentially as, whether group immunity is reached,
that early public containment measures could be
or seasonal factors or public actions slow the virus
efficient. However, as acknowledged by the authors,
reproduction. This behavior has been observed for
real-time parameterization of a model for an un-
seasonal influenza [14], H1N1 [7] as well as for re-
known disease is a difficult and uncertain task, and
cent coronavirus epidemics such as SARS [1, 5, 8] or
the effects of lockdown may vary from one country
MERS [3]. It is usually well described by exponen-
to another. Although modeling studies offer valuable
tial functions such as the logistic distribution or the
insights and possible scenarii for forthcoming events,
Gauss function. The Gauss function is defined here
and might provide a deep understanding of the
for time evolution of new cases as:
epidemic’s dynamics a posteriori, they require vali-
dation, which can only be provided by thorough data (t−Tp )2

analysis. In that regard, the observational efforts of c(t) = e τ2 (1)

Tobias (2020) [20] represent an interesting approach.
where c(t) is the daily number of new cases, t is time,
The latter recently claimed that the full lockdown
Tp is the time of the peak (maximum infection), and τ
policies in Spain and Italy have had positive results
is a time scale defining the duration of the epidemic.
in slowing the epidemic. However, their methods,
In the assumption of a steady relationship between
based on fitting linear trends to the logarithm of
the number of cases and the number of fatalities (the
the daily new cases and daily death numbers, and
fatality rate µ is time-independent), the daily death
comparing them before, and after the lockdown
number d(t) is linked to the daily new cases number
policies, might not be appropriate. As will be shown
c(t) through the time-lagged proportionality relation:
below, to assess the trajectory of the epidemic, one
should look for trends in the time derivative of the d(t) = µc(t − Td ), (2)
logarithm of daily numbers, rather than trends in
the logarithm of the daily numbers itself. where Td is the average time between infection and
death. It thus follows a similar Gaussian-like law.
Here, we show that the available data exhibit no However, the latter never is a purely Gaussian func-
evidence for any effects of the full lockdown policies tion, and is often skewed or exhibits more complex
applied in Italy, Spain, France and United Kingdom patterns [8], so that the definition of Eq (1) is not
in the time evolution of the COVID-19 epidemic. Us- exactly correct and needs to be generalized.
ing a phenomenological approach, we compare the It is natural and convenient to express time evolution
evolution of the epidemic before and after the full of c(t) as a power function with a time varying expo-
lockdown measures are expected to produce visible nent. As for any strictly positive function, c(t) can
results. Our approach have similarities with Tobias be written in the form of a generalized exponential
(2020)’s [20]: it is focused on incident rather than function:
cumulative data, and it compares pre-lockdown and
post-lockdown trends. However, here, no positive c(t) = c(0)eγ(t)t (3)

2
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.

where the time varying function γ(t) will be referred and Γ(t) can be retrieved as in Eq (6):
to as the growth rate of c(t). Expressing time evolu-
ln(D(t + δt)) − ln(D(t))
tion of the daily new cases number in the form of Eq Γ(t) = (9)
(3) can be thought of as a generalization of Richards’s δt
phenomenological model [18]. The doubling time (T2 ) is related to Γ following:
If γ is a constant, c(t) is a pure exponential. If ln(2)
γ(t) is a linearly decaying function of the form γ(t) = T2 = (10)
Γ
β + αt, with α < 0, c(t) is a Gauss function, and Tp
and τ can be expressed in terms of the slope and so that we can obtain instantaneous estimates
intercept of γ(t): of T2 (t) without waiting for the number of total
r fatalities to actually double.
−1
τ= , (4)
α To assess the efficiency of lockdown policies, we
−β first compute the growth rate γ(t) from the daily
Tp = − (5) deaths observations and apply linear regression to
2α
estimate its trend before the lockdown should have
In that case, the short time asymptotic limit is close
any visible effects (t < Tld + Td , where Tld is the start
to a pure exponential growth, since t << Tp . It then
date of the lockdown measures). We then predict
slows down as t approaches Tp , and starts to decay
values for each variable of interest after the lockdown
as t passes Tp . Eq (3) however allows any form of
should have visible effects by extrapolating the linear
γ(t) and is not restricted to Gaussian or exponential
evolution of γ(t) after this date. This allows us to
behaviors. Values of γ(t) can be retrieved from any
compare observed values of growth rate, daily deaths,
time series of the daily new cases number, or equiva-
doubling time, and total fatalities number, with
lently the daily death number :
the values expected from the pre-lockdown trend
d(t + δt) ln(d(t + δt)) − ln(d(t)) (what would have happened if nothing had changed).
γ(t) = ln( )/δt = (6)
d(t) δt
To assess the evolution of the epidemic with a more
γ(t) can thus be defined as the time derivative of the
classical approach, we also compute an instantaneous
natural logarithm of the daily death number. This
reproduction number as well as an estimate of the re-
method is commonly used in the study of transient
production number, based on the daily deaths data
perturbations growth in fluid mechanic’s generalized
and Eq (2), which links the daily deaths number, the
stability theory [6, 12]. In this work, we will be
fatality rate, and the daily new cases number. The
primarily studying the time variations of γ, and
reproduction number is the number of secondary in-
search for visible trends in the latter.
fections provoked by a typical case [1]. In practice,
the reproduction number shows large variability de-
Since the time necessary for the number of fatalities
pending on a number of factors such as the age or
to double (hereafter doubling time) is a commonly
the region [15]. However, a mean estimate is useful
used diagnostic of an epidemic evolution, it is also
to assess the epidemic stage. Here, we approximate
computed in this work. The total number of deaths at
the instantaneous reproduction number Ri as the ra-
time t is the time integral of the daily death number:
tio of the number of new cases and the total number
Z t
of contagious cases at time t:
D(t) = d(s)ds (7)
0 c(t)
Ri (t) = R t , (11)
Since D(t) is also a strictly positive function, it can t−T c
(c(s) − d(s))ds
be expressed in a similar form as Eq (3):
where Tc is the time during which an infected person
D(t) = D(0)eΓ(t)t (8) remains contagious, and s is an integration variable.

3
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.

It is important to note that we make the assump- To compute the growth rate of the daily deaths
tions that new cases start to be contagious right af- number γ(t) and of the total fatalities number Γ(t)
ter infection (zero generation time), and that all cases as well as the doubling time T2 (t), the daily deaths
within the contagion period Td are equally likely to data are first low-pass filtered using a 3-days running
produce secondary infections. While our simplifying mean on the logarithm ln(d(t)).
hypothesis of zero generation time might yield to an The values used here for the time between infec-
underestimate of the reproduction number, it does tion and death Td ranges between 14 and 20 days,
not affect its general trend, which is what this study with a median reference value of 17 days. It cor-
is focused on. The reproduction number R, which is responds to the averaged value of hospitalization to
the number of persons that will be infected by each death reported by Russell et al. (2020) [19] (13 days)
contagious person during the time Tc is approximated plus a period of 1 to 7 days between infection and
as: hospitalization. Two different values of the time dur-
ing which an infected person remains contagious (Tc )
Z t+Tc
R(t) = Ri (s)ds (12) were tested to infer the reproduction number: 14, and
t 21 days. The former is the duration of the quarantine
applied to any confirmed cases in most countries, and
The epidemic is in a growing phase if R > 1 and the latter is a longer estimate used for comparison
decays otherwise. since the 14 days value is uncertain [19]. For com-
puting estimates of the daily case number, we used a
fatality rate of 1.7 %, which is a median value between
Russell et al. (2020)’s [19] estimates of the Infection
2 Data Fatality Ratio and Case Fatality Ratio onboard the
Diamond Princess passenger ship. The latter also
Because of the important proportion of asymp- closely matches South Korea’s fatality rate (1.6%)
tomatic cases of COVID-19 [2, 17, 13] and of [4], which is one of the most reliable national esti-
the testing policies of most countries, which are mate so far, given the wide-range testing policy and
restrained to severe and potentially deadly cases, the the advanced stage of the epidemic in this country.
daily number of new confirmed cases is not a reliable
variable to assess the evolution of the epidemic. We
thus only used the daily deaths number to estimate 3 Results
the growth rates and their trends. The daily number
of new cases is inferred from the daily death number Time evolution of the reproduction numbers Ri
and a fixed fatality rate using Eq (2). and R is shown in figure 1 for France, Italy, Spain,
The data used in this study are produced by the and United Kingdom. 4 different estimates are
European Center for Disease Prevention and Control proposed in each figure: the estimates defined in
(ECDC). Because of a lack of daily deaths reports Eq (12), computed for values of contagion duration
in nursing homes in France until 02 April 2020, of Tc = 14 and 21 days, as well as values of the
that were suddenly corrected in the dataset, we instantaneous reproduction number multiplied by
only use hospital deaths data for this country, for the contagion duration (Ri Tc ). For all four variables,
consistency of the time series. The daily and total time evolution exhibits a similar pattern: a steady
deaths numbers in France are thus greatly underes- decreasing trend from 3.5 to 6 secondary infections
timated in our study, but one should keep in mind per case in the beginning of the epidemic to less
that we are focused on time evolution and trends than unity 20 to 40 days before the reference date
rather than absolute values, so that time-consistency (April 24). In all four countries, no discontinuity in
is the single most important requirement for the data. the general decaying trend is observed around the
full lockdown’s start date. Even though this date

4
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.

coincides approximately with the 1-crossing of R lockdown with the same slope as before lockdown).
in France, Italy, and Spain, the latter is only the Comparing these estimates (dashed green line) with
follow-up of a longer term decay. Note that, despite the observed values (blue squares) shows that the
the simplifying assumptions used here, our estimates initial pre-lockdown trend yields a steeper rise in the
are of the same order of magnitude as Liu et al. doubling time than what is observed after lockdown
(2020)[11]’s interval [1.4-6.49] for the outbreak in policies should have visible impacts. The visual
Wuhan. impression of an accelerating T2 growth in the data
is thus not to be attributed to the lockdown effects,
Analysis of the evolution of the growth rate γ(t) but rather to the inherent growth of the T2 function
confirms this long-term trend in the trajectory of when Γ(t) reaches small values. Figure 3 thus also
the epidemic before any full lockdown policies were underlines the lack of evidence of any effects of the
effective (figure 2). A general decaying trend of γ(t) full lockdown.
is evident from the beginning of the epidemic in all
4 countries, although some variability exists around Similarly to the doubling time, we estimated
the linear trend, with a nearly periodic oscillation the evolution of the daily deaths number in the
of 5 to 8 days. Linear regression satisfyingly models hypothesis of a continuation of the pre-lockdown
the time evolution of γ, with coefficients of determi- trend in γ(t) after the lockdown policies should have
nation r2 (fraction of the variance explained by the visible impacts. Our results show that, even though
model) of 0.67, 0.72, 0.73, and 0.61 for France, Italy, the dates of the daily deaths peaks in France, Italy,
Spain, and United Kingdom, respectively. Linearity and Spain roughly correspond to the dates where
of γ(t) suggests that time evolution of the epidemic lockdown effects should be visible, the peak dates
is consistent with a Gauss function. Comparing expected from the pre-lockdown trends are actually
the linear decaying trend before, and after the time the same. Moreover, daily deaths observations after
by which full lockdown policies should have visible this date show a slower decay than what would be
impacts, we find that the slope of γ(t) decreases in expected from the pre-lockdown tendencies. Forecast
France, Italy, and Spain after the full lockdown, and of the future evolution of the daily deaths number
remains constant in United Kingdom. The decay of using the same method with the linear trend of the
the epidemic has thus slowed since the lockdown is full time series and the latest observations is also
effective. Comparison of the general decay trends shown as the blue line (for indicative purpose only).
in France, Italy, Spain and United Kingdom with
that of the Netherlands provides a further insight Time evolution of the total death toll, both
on the effects of full lockdown: Netherlands decay observed, and predicted from pre-lockdown trends
trend is slightly slower than France and Spain is shown in figure 5. One would expect total dead
before lockdown, and is nearly similar to Italy and numbers to rapidly saturate at a value close from
United Kingdom’s. In all four countries, the decay that corresponding to the crossing of the curve
trend after the effective lockdown is slower than and the date of expected visible lockdown effects.
Netherlands trend. However, the total dead number kept on growing
after this date, closely following the values expected
Since a raw, visual analysis of the effects of full from pre-lockdown trends, and even reaching values
lockdown on the doubling time could mislead to the beyond the death toll expected from the latter.
impression that its increase is accelerated after the Again, the forecast dead number obtained from
lockdown is effective, we should carefully inspect the extrapolating γ(t)’s linear trend in the future is
results of figure 3. We computed an estimate of the presented for indicative purpose. It is however
doubling time, assuming the pre-lockdown trend in interesting to notice the consistency in the order of
γ(t) remains constant after the lockdown is effective magnitude of the final total deaths forecast at a 15
(we assume that γ(t) keeps on linearly decaying after to 30 days interval (at the time of the lockdowns and

5
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.

on April 24.). 4 Summary and discussion

This observational study, using a generalized phe-
nomenological method based on official daily deaths
records only, shows that full lockdown policies of
While comparing pre and post lockdown time France, Italy, Spain and United Kingdom haven’t
evolutions of the epidemic brings a useful insight had the expected effects in the evolution of the
on the impacts of home containment, it is also COVID-19 epidemic. Our results show a general
of interest to compare this evolution with that of decay trend in the growth rates and reproduction
countries applying different policies. Figure 6 shows numbers two to three weeks before the full lockdown
the evolution of the daily deaths growth rates γ(t), policies would be expected to have visible effects.
their linear trends, the reproduction numbers, and Comparison of pre and post lockdown observations
the doubling times for 10 countries. We selected reveals a counter-intuitive slowdown in the decay
countries that had over 1000 fatalities by April of the epidemic after lockdown. Estimates of daily
15, 2020, and chose to exclude data from China, and total deaths numbers using pre-lockdown trends
given the growing doubts on their accuracy. The suggest that no lives were saved by this strategy, in
time reference was chosen to be the day by which comparison with pre-lockdown, less restrictive, social
the total death toll exceeded 100 in each country. distancing policies. Comparison of the epidemic’s
Evolution of γ(t) shows a similar general decay trend evolution between the fully locked down countries
in all countries. It is interesting to note that, while and neighboring countries applying social distancing
the linear trends of the growth rates have similar measures only, confirms the absence of any effects
slopes in nearly all countries, they show a wide of home containment. Evolution of the epidemic in
range of intercepts (value of γ at t = 0), showing Sweden however indicates that, in the absence of
that although the slowdown of the epidemic follows any social distancing measures, the epidemic’s decay
a similar trajectory, each country started at very may be subject to larger fluctuations. This work
different levels of growth rate. This general decay thus suggests that social distancing measures, such
trend is accompanied by a regular decay in the as those applied in the Netherlands and Germany, or
reproduction number in all countries, with similar in Italy, France, Spain, and United Kingdom before
slopes and, again, a wide range of initial reproduc- the full lockdown strategies, have approximately the
tion numbers. As expected from a decreasing growth same effects as police-enforced home containment
rate and reproduction number, the doubling time policies.
is increasing in all countries from the beginning
of the time series. Figure 6 thus shows that time So far, the reasons for the relatively regular
evolution of the epidemic is homogeneous in Western decay of the epidemic remain largely unknown.
Europe, and that the main differences reside in the While social distancing efforts may contribute to it,
initial conditions at the beginning of the epidemic. environmental conditions could as well have played
In particular, the figure shows that countries with a role (possible seasonality of the virus). The group
social distancing policies, but no home containment, immunity hypothesis, though being unlikely if the
such as the Netherlands and Germany experience a reference fatality rates are correct, deserves a short
very similar decay of the epidemic in terms of growth discussion: computing the number of daily new cases
rate, reproduction number, and doubling time, to from the number of daily deaths following Eq (2),
countries with police-enforced home containment. and using a fatality rate of 1.7%, we forecast a ratio
On the other hand, results for Sweden suggest that of infected population at the end of the epidemic of
taking no action at all may yield a more variable 1.4%, 3.0%, 3.2%, and 2.1% in France, Italy, Spain,
decay of the epidemic. and United Kingdom, respectively. The latter is
obviously far from being able to yield any group

6
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.

immunity. Under the rough assumption that 50 spread and fatality rates of covid-19. Proceedings
to 70% of the population needs to be infected to of the National Academy of Sciences, 2020.
ensure group immunity, it is possible to compute
an hypothetical fatality rate using Eq (2). We find [5] C. Dye and N. Gay. Modeling the sars epidemic.
that, if group immunity was responsible for the Science, 300(5627):1884–1885, 2003.
decay of the epidemic, the fatality rates would be [6] B. F Farrell and P. J. Ioannou. General-
of about 0.05%, 0.10%, 0.11%, and 0.07% for the ized stability theory. part i: Autonomous op-
50% hypothesis in France, Italy, Spain, and United erators. Journal of the atmospheric sciences,
Kingdom, respectively, and of 0.03%, 0.07%, 0.08%, 53(14):2025–2040, 1996.
and 0.05% for the 70% hypothesis. Obviously,
this is only a gedanken experiment, which is far [7] Y. H. Hsieh, D. N. Fisman, and J. Wu. On epi-
beyond the scope of this paper, and only serological demic modeling in real time: An application to
tests and further data analysis, once the epidemic the 2009 novel a (h1n1) influenza outbreak in
is completely instinct, will allow to discriminate canada. BMC research notes, 3(1):283, 2010.
between the possible reasons for its decay.
[8] L. S. Hung. The sars epidemic in hong kong:
As a concluding remark, it should be pointed out what lessons have we learned? Journal of the
that, since the full lockdown strategies are shown Royal Society of Medicine, 96(8):374–378, 2003.
to have no impact on the epidemic’s slowdown, one [9] C. C. Ku, T. C. Ng, and H. H. Lin. Epidemiologi-
should consider their potentially high inherent death cal benchmarks of the covid-19 outbreak control
toll as a net loss of human lives. in china after wuhan’s lockdown: a modelling
study with an empirical approach. Available at
SSRN 3544127, 2020.
References
[10] J. C. Lagier, P. Colson, H. T. Dupont, J. Sa-
[1] R. M. Anderson, C. Fraser, A. C. Ghani, C. A. lomon, B. Doudier, C. Aubry, F. Gouriet,
Donnelly, S. Riley, N. M. Ferguson, G. M. Le- S. Baron, P. Dudouet, R. Flores, and others.
ung, T. H. Lam, and A. J Hedley. Epidemiol- Testing the repatriated for sars-cov2: Should
ogy, transmission dynamics and control of sars: laboratory-based quarantine replace traditional
the 2002–2003 epidemic. Philosophical Transac- quarantine? Travel Medicine and Infectious Dis-
tions of the Royal Society of London. Series B: ease, page 101624, 2020.
Biological Sciences, 359(1447):1091–1105, 2004.
[11] Y Liu, A. A. Gayle, A. Wilder-Smith, and
J. Rocklöv. The reproductive number of covid-19
[2] Y. Bai, L. Yao, T. Wei, F. Tian, D. Y. Jin,
is higher compared to sars coronavirus. Journal
L. Chen, and M. Wang. Presumed asymptomatic
of travel medicine, 2020.
carrier transmission of covid-19. Jama, 2020.
[12] T. Meunier, C. Ménesguen, X. Carton, S. Le
[3] X. Chen, A. A. Chughtai, A. Dyda, and C. R. Gentil, and R. Schopp. Optimal perturbations
MacIntyre. Comparative epidemiology of mid- of an oceanic vortex lens. Fluids, 3(3):63, 2018.
dle east respiratory syndrome coronavirus (mers-
cov) in saudi arabia and south korea. Emerging [13] K. Mizumoto, K. Kagaya, A. Zarebski, and
microbes & infections, 6(1):1–6, 2017. G. Chowell. Estimating the asymptomatic pro-
portion of coronavirus disease 2019 (covid-19)
[4] J. B. Dowd, L. Andriano, D. M. Brazel, V. Ro- cases on board the diamond princess cruise
tondi, P. Block, X. Ding, Y. Liu, and M. C. Mills. ship, yokohama, japan, 2020. Eurosurveillance,
Demographic science aids in understanding the 25(10):2000180, 2020.

7
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.

[14] D. J. Muscatello, A. T. Newall, D. E. Dwyer,

and C. R. MacIntyre. Mortality attributable to
seasonal and pandemic influenza, australia, 2003
to 2009, using a novel time series smoothing ap-
proach. PloS one, 8(6), 2013.

[15] H. Nishiura, G. Chowell, M. Safan, and

C. Castillo-Chavez. Pros and cons of estimating
the reproduction number from early epidemic
growth rate of influenza a (h1n1) 2009. The-
oretical Biology and Medical Modelling, 7(1):1,
2010.
[16] N. Picchiotti, M. Salvioli, E. Zanardini, and
F. Missale. Covid-19 italian and europe epi-
demic evolution: A seir model with lockdown-
dependent transmission rate based on chinese
data. Available at SSRN 3562452, 2020.
[17] D. Raoult, A. Zumla, F. Locatelli, G. Ippolito,
and G. Kroemer. Coronavirus infections: Epi-
demiological, clinical and immunological fea-
tures and hypotheses. Cell Stress, 4(4):66, 2020.
[18] F. J. Richards. A flexible growth function for
empirical use. Journal of experimental Botany,
10(2):290–301, 1959.
[19] T. W. Russell, J. Hellewell, C. I. Jarvis, K. Van-
Zandvoort, S. Abbott, R. Ratnayake, S. Flasche,
R. M. Eggo, A. J. Kucharski, and CMMID nCov
working group and others. Estimating the infec-
tion and case fatality ratio for covid-19 using
age-adjusted data from the outbreak on the di-
amond princess cruise ship. medRxiv, 2020.

[20] A. Tobı́as. Evaluation of the lockdowns for the

sars-cov-2 epidemic in italy and spain after one
month follow up. Science of The Total Environ-
ment, page 138539, 2020.
[21] X. Yin, J. Wang, J. Feng, Z. Chen, N. Jiang,
J. Wu, S. Yan, H. Li, C. Lv, Z. Lu, and others.
The impact of the corona virus disease 2019 out-
break on chinese residents’ mental health. Avail-
able at SSRN 3556680, 2020.

8
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.

Figure 1: Time evolution of the reproduction number for France, Italy, Spain, and Great Britain. The blue square
represent the daily reproduction number multiplied by an estimate of the time during which a case is contagious
(Ri Tc ). The green dots represent the reproduction number (R) computed as a time integral of the daily reproduction
number (Ri ) and thus takes into account time variations of the latter. The dashed vertical orange line is the start
date of the lockdown policies. Two values of Tc were used for each variable: 14 and 21 days. Time is referenced to
April 24, 2020.

9
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.

Figure 2: Time evolution of the growth rate of the daily deaths number (γ(t)) for France, Italy, Spain, and Great
Britain. The vertical orange dashed line shows the start date of the full lockdown policies. The orange shaded area
represents the time at which the lockdown should show some effects in the epidemic spreading (14 to 20 days), and
the thick line is the reference date (17 days). The blue squares represent the observations, and the thick gray line
represents the linear trend of the observations. The dashed green and red lines represent the linear trends before and
after the lockdown should affect the observations. Time is referenced to April 24, 2020.

10
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.

Figure 3: Same as figure 2 for the doubling time. The thick dotted green line represents the expected evolution if
the pre-lockdown linear trend in γ(t) is extrapolated beyond the day lockdown policies are expected to be effective
(What would happen without any lockdown, assuming that the growth rate’s evolution remains the same). Time is
referenced to April 24, 2020.

11
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.

Figure 4: Same as figure 3 for the daily deaths number. The green dotted lines represents the expected evolution
based on the pre-lockdown trend of γ(t), while the plain blue line represents the expected future evolution based on
the full time series trend of γ(t). The light blue shading represents an error margin computed by doubling or dividing
by two the slope of the linear fit to γ(t). Time is referenced to April 24, 2020.

12
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.

Figure 5: Same as figure 4 for the total fatalities number.

13
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.

Figure 6: Comparison of the time evolution of the epidemic for 10 countries. The time reference is chosen to be
the day each country reaches a total dead number of 100. The top left hand side panel shows the time evolution
of γ(t); the top right hand side panel shows the linear fits to γ(t); the bottom left hand side panel shows values of
the reproduction number R(t); the bottom right-hand side panel shows the instantaneous doubling time. Time is
referenced to April 24, 2020.

14
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.
medRxiv preprint doi: https://doi.org/10.1101/2020.04.24.20078717.this version posted May 1, 2020. The copyright holder for this preprint (which
was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
All rights reserved. No reuse allowed without permission.