2020 05 28 COVID19 Report 23 Version2

28 May 2020 Imperial College COVID-19 Response Team
Report 23: State-level tracking of COVID-19 in the United States

Version 2 (28-05-2020)
H Juliette T Unwin∗, Swapnil Mishra2,∗, Valerie C Bradley∗, Axel Gandy∗, Michaela A C Vollmer, Thomas Mellan, Helen
Cou-pland, Kylie Ainslie, Charlie Whittaker, Jonathan Ish-Horowicz, Sarah Filippi, Xiaoyue Xi, Melodie Monod, Oliver
Ratmann, Michael Hutchinson, Fabian Valka, Harrison Zhu, Iwona Hawryluk, Philip Milton, Marc Baguelin, Adhiratha
Boonyasiri, Nick Brazeau, Lorenzo Cattarino, Giovanni Charles, Laura V Cooper, Zulma Cucunuba, Gina Cuomo-
Dannenburg, Bimandra Djaafara, Ilaria Dorigatti, Oliver J Eales, Jeff Eaton, Sabine van Elsland, Richard FitzJohn, Katy
Gaythorpe, William Green, Timothy Hallett, Wes Hinsley, Natsuko Imai, Ben Jeffrey, Edward Knock, Daniel Laydon, John
Lees, Gemma Nedjati-Gilani, Pierre Nouvellet, Lucy Okell, Alison Ower, Kris V Parag, Igor Siveroni, Hayley A Thompson,
Robert Verity, Patrick Walker, Caroline Walters, Yuanrong Wang, Oliver J Watson, Lilith Whittles, Azra Ghani, Neil M
Ferguson, Steven Riley, Christl A. Donnelly, Samir Bhatt 1,∗ and Seth Flaxman∗
Department of Infectious Disease Epidemiology, Imperial College London

Department of Mathematics, Imperial College London
WHO Collaborating Centre for Infectious Disease Modelling
MRC Centre for Global Infectious Disease Analytics
Abdul Latif Jameel Institute for Disease and Emergency Analytics, Imperial College London
Department of Statistics, University of Oxford
∗
Contributed equally.
1
Correspondence: [email protected].
2
Methodological correspondence: [email protected]
SUGGESTED CITATION
H Juliette T Unwin, Swapnil Mishra, Valerie C Bradley et al. Report 23: State-level tracking of COVID-19 in the United
States - VERSION 2 (28-05-2020), doi: https://doi.org/10.25561/79231.
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0

International License.
DOI: https://doi.org/10.25561/79231 Page 1

Summary
As of 20 May 2020, the US Centers for Disease Control and Prevention reported 91,664 confirmed or probable COVID-
19-related deaths, more than twice the number of deaths reported in the next most severely impacted country. In
order to control the spread of the epidemic and prevent health care systems from being overwhelmed, US states have
implemented a suite of non-pharmaceutical interventions (NPIs), including “stay-at-home” orders, bans on gatherings,
and business and school closures.
We model the epidemics in the US at the state-level, using publicly available death data within a Bayesian hierarchical
semi-mechanistic framework. For each state, we estimate the time-varying reproduction number (the average number of
secondary infections caused by an infected person), the number of individuals that have been infected and the number
of individuals that are currently infectious. We use changes in mobility as a proxy for the impact that NPIs and other
behaviour changes have on the rate of transmission of SARS-CoV-2. We project the impact of future increases in mobility,
assuming that the relationship between mobility and disease transmission remains constant. We do not address the
potential effect of additional behavioural changes or interventions, such as increased mask-wearing or testing and tracing
strategies.
Nationally, our estimates show that the percentage of individuals that have been infected is 4.1% [3.7%-4.5%], with wide
variation between states. For all states, even for the worst affected states, we estimate that less than a quarter of the
population has been infected; in New York, for example, we estimate that 16.6% [12.9%-21.4%] of individuals have been
infected to date. Our attack rates for New York are in line with those from recent serological studies [1] broadly supporting
our modelling choices.
There is variation in the initial reproduction number, which is likely due to a range of factors; we find a strong association
between the initial reproduction number with both population density (measured at the state level) and the chronological
date when 10 cumulative deaths occurred (a crude estimate of the date of locally sustained transmission).
Our estimates suggest that the epidemic is not under control in much of the US: as of 25 May 2020, the reproduction
number is above the critical threshold (1.0) in Above 1: 26 [95% CI: 18-34] states. Higher reproduction numbers are
geographically clustered in the South and Midwest, where epidemics are still developing, while we estimate lower repro-
duction numbers in states that have already suffered high COVID-19 mortality (such as the Northeast). These estimates
suggest that caution must be taken in loosening current restrictions if effective additional measures are not put in place.
We predict that increased mobility following relaxation of social distancing will lead to resurgence of transmission, keep-
ing all else constant. We predict that deaths over the next two-month period could exceed current cumulative deaths
by greater than two-fold, if the relationship between mobility and transmission remains unchanged. Our results suggest
that factors modulating transmission such as rapid testing, contact tracing and behavioural precautions are crucial to
offset the rise of transmission associated with loosening of social distancing.
Overall, we show that while all US states have substantially reduced their reproduction numbers, we find no evidence
that any state is approaching herd immunity or that its epidemic is close to over.
We invite scientific peer reviews here: https://openreview.net/group?id=-Agora/COVID-19

1 Introduction
The first death caused by COVID-19 in the United States is currently believed to have occurred in Santa Clara, California on
the 6th February [2]. In April 2020, the number of deaths attributed to COVID-19 in the United States (US) surpassed that
of Italy [3]. Throughout March 2020, US state governments implemented a variety of non-pharmaceutical interventions
(NPIs), such as school closures and stay-at-home orders, to limit the spread of SARS-CoV-2 and help maintain the capacity
of health systems to treat as many severe cases of COVID-19 as possible. Courtemanche et al. [4] use an event-study
model to determine that such NPIs were successful in reducing the growth rate of COVID-19 cases across US counties.
We similarly seek to estimate the impact of NPIs on COVID-19 transmission, but do so with a semi-mechanistic Bayesian
model that reflects the underlying process of disease transmission and relies on mobility data released by companies
such as Google [5]. Mobility measures reveal stark changes in behaviour following large-scale government interventions,
with individuals spending more time at home and correspondingly less time at work, at leisure centres, shopping, and
on public transit. Some state governments, like the Colorado Department of Public Health, have already begun to use
similar mobility data to adjust guidelines over social distancing [6]. As more and more states ease the stringency of
their NPIs, future policy decisions will rely on the interaction between mobility and NPIs and their subsequent impact on
transmission.
In a previous report [7], we introduced a new Bayesian statistical framework for estimating the rate of transmission and
attack rates for COVID-19. Our approach infers the time-varying reproduction number, Rt , which measures transmission
intensity. We calculate the number of new infections through combining previous infections with the generation interval
(the distribution of times between infections). The number of deaths is then a function of the number of infections
and the infection fatality rate (IFR). We estimate the posterior probability of our parameters given the observed data,
while incorporating prior uncertainty. This makes our approach empirically driven while incorporating as many sources
of uncertainty as possible. In this report, similar to [8, 9], we adapt our original framework to model transmission in the
US at the state level. In our formulation we parameterise Rt as a function of several mobility types. Our parameterisation
of Rt makes the explicit assumption that changes in transmission are reflected through mobility. While we do attempt to
account for residual variation, we note that transmission will also be influence by additional factors and some of these are
confounded causally with mobility. We utilise partial pooling of parameters, where information is shared across all states
to leverage as much signal as possible, but individual effects are also included for state- and region-specific idiosyncrasies.
Our partial pooling model requires only one state to provide a signal for the impact of mobility, and then this effect is
shared across all states. While this sharing can potentially lead to initial over or under estimation effect sizes, it also
means that a consistent signal for all states can be estimated before that signal is presented in an individual states with
little data.
We infer plausible upper and lower bounds (Bayesian credible interval summaries of our posterior distribution) of the
total population that have been infected by COVID-19 (also called the cumulative attack rate or attack rate). We also
estimate the effective number of individuals currently infectious given our generation distribution. We investigate how
the reproduction number has changed over time and study the heterogeneity in starting and ending rates by state, date,
and population density. We assess whether there is evidence that changes in mobility have so far been successful at
reducing Rt to less than 1. To assess the risk of resurgence when interventions are eased, we use simple scenarios of
increased mobility and simulate forwards in time. From these simulations we study how sensitive individual states are

to resurgence, and the plausible magnitude of this resurgence.
Details of the data sources and a technical description of our model and are found in Sections 4 and 5 respectively.
General limitations of our approach are presented below in the conclusions.
2 Results
2.1 Mobility trends, interventions and effect sizes
Mobility data provide a proxy for the behavioural changes that occur in response to non-pharmaceutical interventions.
Figure 1 shows trends in mobility for the 50 states and the District of Columbia (see Section 4 for a description of the
mobility dimensions). Regions are based on US Census Divisions, modified to account for coordination between groups
of state governments [10]. These trends are relative to a state-dependent baseline, which was calculated shortly before
the COVID-19 epidemic. For example, a value of −20% in the transit station trend means that individuals, on average,
are visiting and spending 20% less time in transit hubs than before the epidemic. In Figure 1, we overlay the timing of
two major state-wide NPIs (stay at home and emergency decree) (see [11] for details). We also note intuitive changes in
mobility such as the spike on 11th and 12th April for Easter. In our model, we use the time spent at one’s residence and
the average of time spent at grocery stores, pharmacies, recreation centres, and workplaces. For states in which the 2018
American Community Survey reports that more than 20% of the working population commutes on public transportation,
we also use the time spent at transit hubs (including gas stations etc.) [12].
To justify the use of mobility as a proxy for behaviour, we regress average mobility against the timings of major NPIs
(represented as step functions). The median correlation between the observed average mobility and the linear predic-
tions from NPIs was approximately 89% (see Appendix A). We observed reduced correlation when lagging (forward and
backwards) the timing of NPIs suggesting immediate impact on mobility. We make no explicit causal link between NPIs
and mobility, however, this relationship is plausibly causally linked but is confounded by other factors.
The mobility trends data suggests that the United States’ national focus on the New York epidemic may have led to
substantial changes in mobility in nearby states, like Connecticut, prior to any mandated interventions in those states.
This observation adds support to the hypothesis that mobility can act as a suitable proxy for the changes in behaviour
induced by the implementation of the major NPIs. In further corroboration, a poll conducted by Morning Consult/Politico
on 26th March 2020 found that 81% of respondents agreed that “Americans should continue to social distance for as long
as is needed to curb the spread of coronavirus, even if it means continued damage to the economy” [13]. While support
for strong social distancing has since eroded slightly (70% agree in the same poll conducted later on 10 May 2020), the
overall high support for social distancing suggests strong compliance with NPIs, and that the changes to mobility that
we observe over the same time period are driven by adherence to those policy recommendations. However, we note
that mobility alone cannot capture all the heterogeneity in transmission risk. In particular, it cannot capture the impact
of case-based interventions (such as testing and tracing). To account for this residual variation missed by mobility we
use a second-order, weekly, autoregressive process. This autoregressive process is an additional term in our parametric
equation for Rt and accounts for residual noise by capturing a correlation structure where current Rt is correlated with
previous weeks Rt (see Figures 13).

Figure 2 shows the average global effect sizes for the mobility types used in our model. Estimates for the regional and
state-level effect sizes are included in Appendix B. We find that increased time spent in residences reduces transmis-
sion by 31.8% [-10.6% - 66.7%], and that decreases in overall average mobility reduced transmission by 69.3% [41.2% -
85.7%]. These two effects are likely related - as people spend less time in public spaces, captured by our average mobility
metric, they conversely spend more time at home. Overall, this decreases the number of people with whom the average
individual comes into contact, thus slowing transmission, even if more time at home may increase transmission within
a single residence. We find less time spent in transit hubs is weakly associated -14.8% [-40.8% - 11.6%] with increased
transmission (the reverse of what we might expect). The impact of transit mobility is in contrast to what we observed in
Italy [8], and could reflect higher reliance on cars and less use of public transit in the US than Europe [14].
The learnt random effects from the autoregressive process are shown in Appendix C. These results show that mobility
explains most of the changes in transmission in places without advanced epidemics, as evidenced by the flat residual
variation. However, for regions with advanced epidemics, such as New York or New Jersey, there is evidence of additional
decreases in transmission that cannot be explained by mobility alone. These may capture the impact of other control
measures, such as increased testing, as well as behavioural responses not captured by mobility, like increased mask-
wearing and hand-washing.
2.2 Impact of interventions on reproduction numbers
We estimate a national average initial reproduction number (Rt=0 ) of 2.8 [1.9 Nebraska - 4.5 New York] and find that,
similar to influenza transmission in cities (see Dalziel et al. [15]), Rt=0 is correlated with population density (Figure 3)1 .
Dalziel et al. hypothesize that more personal contact occurs in more densely populated areas, thus resulting in a larger
Rt=0 .
Rt=0 is also negatively correlated with when a state observed cumulative 10 deaths (Figure 3). This negative correlation
implies that states began locally sustained transmission later had a lower Rt=0 . A possible hypothesis for this effect is the
onset of behavioural changes in response to other epidemics in the US. An alternative explanation is that the estimates of
the early growth rates of the epidemics in the states affected earliest are biased upwards by the early national ramp-up
of surveillance and testing.
In both these relationships there may are many confounding variables and we discourage causal attribution, these results
just serve as qualitative assessments.
In subsequent analysis we use Rt=EmergencyDecree rather than Rt=0 . Our reason for this is to allow more focus on
current trends. Despite both Rt=0 and Rt=EmergencyDecree being highly variable the majority of states have generally
decreased their Rt since the first 10 deaths were observed (Figure 4). We estimate that 27 states have a posterior mean
Rt of less than one but only 0 have 95% credible intervals that are completely below one. A posterior mean Rt below one
and credible interval that includes one suggests that the epidemic is likely under control in that state, but the potential for
increasing transmission cannot be ruled out. Therefore, our results show that very few states have conclusively controlled
their epidemics. Of the ten states with the highest current Rt , half are in the Great Lakes region (Illinois, Ohio, Minnesota
1 We also considered the relationship of Rt with a population density weighted by proportion of the total population of the state in each census
tract. This was less strongly correlated to Rt=0 .

Great Lakes Great Plains

● ●
● ●
●
●
● Emergency decree ●
●
● Emergency decree
● ●
50%
● Stay at home order issued 50%
● Stay at home order issued
Change in mobility
Change in mobility
Stay at home order eased Stay at home order eased
0% ● Indiana 0% ● Iowa
● Illinois ● Kansas
● Ohio ● Missouri
−50%
● Michigan −50%
● North Dakota
● Wisconsin ● Nebraska
● Minnesota ● South Dakota
b
ar
ar
r
ay
ay
ar
ar
r
ay
ay
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
M
6
20
6
20
24
24
9
23
9
23
4
18
4
18
Mountain Northeast Corridor ● Emergency decree
●
● ● Stay at home order issued
●
●
Stay at home order issued ●
●
● ● Stay at home order eased
50%
●
●
Stay at home order eased 50%
●
●
Change in mobility
Change in mobility
● ●
●
● Pennsylvania
● Utah
● New York
0% ● Arizona 0%
● New Jersey
● Colorado
● Rhode Island
● New Mexico
● Connecticut
−50%
● Montana −50% ● Massachusetts
● Nevada
● New Hampshire
● Idaho
● Vermont
● Wyoming
● Maine
b
ar
ar
r
ay
ay
ar
ar
r
ay
ay
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
M
6
20
6
20
24
24
9
23
9
23
4
18
4
18
Pacific South Atlantic ● Emergency decree
● ●
● ● Stay at home order issued
●
●
●
● ● Stay at home order eased
50% Stay at home order issued 50%
●
●
Change in mobility
Change in mobility
●
Stay at home order eased ●
● Maryland
● Florida
0% ● Washington 0%
● North Carolina
● California
● District of Columbia
● Hawaii
● Virginia
−50%
● Oregon −50% ● Delaware
● Alaska
● South Carolina
● Georgia
● West Virginia
b
ar
ar
r
ay
ay
ar
ar
r
ay
ay
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
M
6
20
6
20
24
24
9
23
9
23
4
18
4
18
Southern Appalachia TOLA

● ●
● ●
●
●
●
50% Stay at home order issued 50% Stay at home order issued
Change in mobility
Change in mobility
Stay at home order eased Stay at home order eased
0% ● Kentucky 0% ● Arkansas
● Tennessee ● Louisiana
● Alabama ● Texas
−50%
● Mississippi −50%
● Oklahoma
b
ar
ar
r
ay
ay
ar
ar
r
ay
ay
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
M
6
20
6
20
24
24
9
23
9
23
4
18
4
18
Figure 1: Comparison of mobility data from Google with government interventions for the 50 states and the District of
Columbia. The solid lines show average mobility (across categories “retail & recreation”, “grocery & pharmacy”, “work-
places”), the dashed lines show “transit stations” and the dotted lines show “residential”. Intervention dates are indicated
by shapes as shown in the legend; see Section 4 for more information about the interventions implemented. There is a
strong correlation between the onset of interventions and reductions in mobility.

Average mobility ●
Mobility
Residential ●
Transit ●
0% 25% 50% 75% 100%

(no effect on transmissibility) (ends transmissibility)
Relative % reduction in Rt
Figure 2: Covariate effect sizes: Average mobility combines “retail & recreation”, “grocery & pharmacy”, “workplaces”.
Transit stations is only used as a covariate for states in which more than 20% of the working population commutes using
public transportation. We plot estimates of the posterior mean effect sizes and 95% credible intervals for each mobility
category. The relative % reduction in Rt metric is interpreted as follows: the larger the percentage, the more Rt de-
creases, meaning the disease spreads less; a 100% relative reduction ends disease transmissibility entirely. The smaller
the percentage, the less effect the covariate has on transmissibility. A 0% relative reduction has no effect on Rt and thus
no effect on the transmissibility of the disease, while a negative percent reduction implies an increase in transmissibility.
● ● ●
●
● ● ●
4
● ●
●
3 ● ● Great Lakes ●
● Great Lakes
● ●●
● ● ● ●
● ●
● ● ● ● Great Plains ● ● Great Plains
● ● ● 3 ● ●
● ● ● ●
● ● ● ● ● ●
● ● Mountain ● ● ● ● Mountain
● ● ● ●
●●● ●●●
● ● ● ●
Rt = 0
Rt = 0
● ● ● ●
● ● ● ●● ● ● ●
2 ●●
● Northeast Corridor ● Northeast Corridor
● ●●● ● ● ●
● ● ● ● ● ●
2 ●●
● Pacific ● ● Pacific
●
●
● South Atlantic ● South Atlantic
● ● ●
● ●
1 ● ● Southern Appalachia 1 ●
● ● Southern Appalachia
● ●
● TOLA ● TOLA
0
0 250 500 750 1000 1250 Mar Apr May
Population density Date of 10 cumulative deaths
(a) (b)
Figure 3: Comparison of initial Rt=0 with population density (a) and date of 10 cumulative deaths (b). R-squared values
are 0.558 and 0.377 respectively.

Hawaii ●
Montana ●
Kansas ●
●
●
District of Columbia ● ●
●
Massachusetts ●
●
●
Oregon ●
●
●
New Jersey ● ●
●
South Dakota ●
Indiana ●
●
●
Colorado ● ●
●
Oklahoma ● ●
●
Delaware ●
●
●
Connecticut ●
●
●
Maryland ● ●
●
Louisiana ●
●
●
West Virginia ●
●
●
Great Lakes
Missouri ● ●
●
Idaho ●
●
● ●
●
Great Plains
New York ●
●
●
Vermont ●
●
●
●
●
Mountain
Arkansas ● ●
●
Washington ●
●
●
●
●
Northeast Corridor
Alaska ●
North Carolina ● ●
● ●
●
Pacific
Pennsylvania ●
●
●
Illinois ●
●
● ●
●
South Atlantic
Kentucky ● ●
●
South Carolina ●
●
● ●
●
Southern Appalachia
California ●
●
●
Michigan ● ●
●
●
●
TOLA
Nebraska ●
●
●
Alabama ●
●
●
Mississippi ● ●
●
New Mexico ● ●
●
Rhode Island ●
●
●
Minnesota ●
●
●
● Emergency Decree
Georgia ● ●
●
Virginia ●
●
● Current
North Dakota ●
Texas ● ●
●
Maine ●
●
●
Iowa ●
●
●
Ohio ● ●
●
Wyoming ●
Nevada ●
●
●
Arizona ● ●
●
Florida ● ●
●
Utah ●
New Hampshire ●
●
●
Tennessee ● ●
●
Wisconsin ●
●
●
0 2 4 6
Rt
Figure 4: State-level estimates of Rt=EmergencyDecree and the average Rt over the week ending 25 May 2020. The
colours indicate regional grouping as shown in Figure 1. We do not include estimates of Rt=EmergencyDecree for Alaska,
Hawaii, Montana, North Dakota, South Dakota, Utah, West Virgina and Wyoming as Emergency Decree was declared in
these states before we start modelling these states.
Indiana, and Wisconsin). In Figure 5 we show the geographical variation in the posterior probability that Rt is less than
1; green states are those with probability that Rt is below 1 is high, and purple states are those with low probability. The
closer a value is to 100%, the more certain we are that the rate of transmission is below 1 and that new infections are not
increasing at present. This is in contrast to many European countries that have conclusively reduced their Rt less than
one at present [7].
Figure 5 shows that while we are confident that some states have controlled transmission, we are similarly confident that
many states have not. Specifically, we are more than 50% sure that Rt > 1 in 26 states. There is substantial geographical
clustering; most states in the Midwest and the South have rates of transmission that suggest the epidemic is not yet
under control. We do note here that many states with Rt < 1 are still in the early epidemic phase with few deaths so
far.

Probability Rt < 1
x < 20%
20% ≤ x < 40%
40% ≤ x < 60%
60% ≤ x < 80%
x ≥ 80%
Figure 5: Our estimates of the probability that Rt is less than one (epidemic control) for each state. These values are an
average over the week ending 25 May 2020.
2.3 Trends in COVID-19 transmission
In this section we focus on five states: Washington, New York, Massachusetts, Florida, and California. These states
represent a variety of COVID-19 government responses and outbreaks that have dominated the national discussion of
COVID-19. Figure 6 shows the trends for these states (trends for all other states can be found in appendix D). Regressing
average mobility against the timing of NPIs yielded an average correlation of around ∼ 97%. Along with the strong visual
correspondence, these results suggest that that interventions have had a very strong effect on mobility, which given our
modelling assumptions, translates into effects on transmission intensity. We also note that there are clear day-of-the-
week fluctuations from the mobility data that affect transmission; these fluctuations are small compared to the overall
reductions in mobility.
On February 29th 2020, Washington state announced the nation’s first COVID-related death and became the first state to
declare a state of emergency. Despite observing its first COVID-19 death only a day after Washington state, New York did
not declare a state of emergency until 7 March 2020. We estimate that Rt began to decline in Washington state before
it did in New York, likely due to earlier intervention, but that stay-at-home orders in both states successfully reduced Rt
to less than one. Approximately one week after New York, Massachusetts issued a stay-at-home order and the mean Rt
is currently (0.8 [0.5-1.2]). In Florida, Rt reduced noticeably before the stay-at-home order, suggesting that behaviour
change started before the stay-at-home order. However, increasing mobility appears to have driven transmission up
recently (1.2 [0.7-1.6]). California implemented early interventions in San Francisco [16], and was the first state to issue
a stay-at-home order [17], but the mean Rt still is around 1 (1.0 [0.6-1.5]). For all the five states shown here there is
considerable uncertainty around the current value of Rt .

Washington Timing
● Started
●
6,000 4
● Eased
40
Daily number of infections

Daily number of deaths
Interventions
3 ● Emergency decree
30 4,000 Restrict public events
Rt
Business closure
2
Restaurant closure
20
School closure
2,000
1 Stay at home mandate
10
Credible intervals
0 50%
0 0
95%
b
b
23 ar
ar
20 r
4 r
18 ay
ay
ar
ar
r
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
M
M
M
6
6
20
20
10
24
9
23
23
4
18
18
New York Timing
● Started
250,000 ●
Daily number of infections 6 ● Eased
900 200,000
Interventions
4
150,000 Restrict public events
600
Rt
Business closure
Restaurant closure
100,000
2 School closure

50,000
Credible intervals
0 50%
0 0
95%
b
9 b
23 ar
6 r
20 pr
4 r
18 ay
ay
ar
ar
4 r
18 y
ay
ar
ar
ay
ay
a
Ap
Ap
Ap
Ap
Ap
a
Fe
Fe
M
M
M
M
M
M
6
20
20
10
24
9
23
23
18
Massachusetts Timing
● Started
50,000 ●
300 6 ● Eased
40,000
Interventions
200 4
Rt
Business closure
Restaurant closure
20,000
2 School closure
100
Stay at home mandate
10,000
Credible intervals
0 50%
0 0
95%
b
9 b
23 ar
6 r
20 pr
4 r
18 ay
ay
ar
ar
r
ay
ay
ar
ar
ay
ay
a
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
M
M
M
6
20
20
10
24
9
23
23
4
18
18
Florida Timing
● Started
100 ●
● Eased
15,000
4 Interventions
75
Restrict public events
10,000
Rt
Business closure
50
Restaurant closure
2
School closure
25 5,000 Stay at home mandate
Credible intervals
0 50%
0 0
95%
b
9 b
23 ar
6 r
20 pr
4 r
18 ay
ay
ar
ar
18 y
ay
ar
ar
ay
ay
a
Ap
Ap
Ap
Ap
Ap
a
Fe
Fe
M
M
M
M
M
M
6
20
20
10
24
9
23
23
4
18
California Timing
40,000 ● Started
150 ●
● Eased
4
30,000 Interventions
100 3 ● Emergency decree
Rt
20,000 Business closure

2
Restaurant closure
School closure
50
10,000
Credible intervals
0 50%
0 0
95%
b
9 b
23 ar
6 r
20 pr
4 r
18 ay
ay
ar
ar
20 r
r
18 y
ay
ar
ar
ay
ay
a
Ap
Ap
Ap
Ap
Ap
a
Fe
Fe
M
M
M
M
M
M
6
20
10
24
9
23
23
4
18
Figure 6: State-level estimates of infections, deaths, and Rt for Washington, New York, Massachusetts, Florida, and
California. Left: daily number of deaths, brown bars are reported deaths, blue bands are predicted deaths, dark blue
50% credible interval (CI), light blue 95% CI. Middle: daily number of infections, brown bars are reported confirmed
cases, blue bands are predicted infections, CIs are same as left. Afterwards, if the Rt is above 1, the number of infections
will start growing again. Right: time-varying reproduction number Rt dark green 50% CI, light green 95% CI. Icons are
interventions shown at the time they occurred.

2.4 Attack rates
We show the percentage of total population infected, or cumulative attack rate, in Table 1 for all 50 states and the
District of Columbia. In general, the attack rates across states remain low; we estimate that the average percentage
of people that have been infected by COVID-19 is 4.1% [3.7%-4.5%]. However, this low national average masks a stark
heterogeneity across states. New York and New Jersey have the highest estimated attack rates, of 16.6% [12.9%-21.4%]
and 15.5% [11.8%-20.3%] respectively, and Connecticut, Massachusetts, and Washington, D.C. all have attack rates over
10%. Conversely, other states that have drawn attention for early outbreaks, such as California, Washington, and Florida,
have attack rates of around 2%, and other states where the epidemic is still early, like Maine, having estimated attack
rates of less than 1%. We note here that there is the possibility of under reporting of deaths in these states. Under
reporting of COVID-19 attributable deaths will result in an underestimate of the attack rates. We note here that we have
found our estimates to be reasonably robust in settings where there is significant under reporting (e.g. Brazil [9]).
Figure 7 shows the effective number of infectious individuals and the number of newly infected individuals on any given
day for each of the 8 regions in our model. The effective number of infectious individuals is calculated using the generation
time distribution, where individuals are weighted by how infectious they are over time. The fully infectious average
includes asymptotic and symptomatic individuals. Currently, we estimate that there are 949000 [189000 - 2691000]
infectious individuals across the whole of the US, which corresponds to 0.30% [0.06% - 0.84%] of the population. Table
2 shows the number currently infected across different states is highly heterogeneous. Figure 7 shows that despite
new infections being in a steep decline, the number of people still infectious, and therefore able to sustain onward
transmission, can still be large. This discrepancy underscores the importance of testing and case based isolation as a
means to control transmission. We note that the expanding cone of uncertainty is in part due to uncertainties arising
from the lag between infections and deaths, but also from trends in mobility. State level estimates of the total number
of infectious individuals over time are given in Appendix E and the current number of infectious individuals are given in
Figure 2.
2.5 Scenarios
The relationship between mobility and transmission is the principal mechanism affecting values of Rt in our model.
Therefore, we illustrate the impact of likely near-term scenarios for Rt over the next 8 weeks, under assumptions of
relaxations of interventions leading to increased mobility. We note that mobility is acting here as a proxy for the number
of potentially infectious contacts. Our mobility scenarios [18] do not account for additional interventions that may be
implemented, such as mass testing and contact tracing. It is also likely that when interventions are lifted behaviour may
modify the effect sizes of mobility and reduce the impact of mobility on transmission. Factors such as increased use of
masks and increased adherence to social distancing are examples. Given these factors we caution the reader to look at
our scenarios as pessimistic, but illustrative of the potential risks.
We define scenarios based on percent return to baseline mobility, which is by definition 0. As an example, say that
currently mobility is 50% lower than baseline, or -50%, perhaps due to the introduction of social-distancing NPIs. Then, a
20% increase of mobility from its current level is −50% ∗ (1 − 20%) = −40%. Similarly, if mobility in residences increased
by 10% following a stay-at-home order, our 20% scenario reduces this to an 8% increase over baseline. This assumes that

Table 1: Posterior model estimates of percentage of total population infected as of 25 May 2020.
State % of total population State % of total population

infected (mean [95% infected (mean [95%
credible interval]) credible interval])
Alabama 2.0% [1.3%-3.1%] Montana 0.2% [0.0%-0.3%]
Alaska 0.1% [0.0%-0.5%] Nebraska 1.4% [0.8%-2.5%]
Arizona 2.0% [1.3%-3.3%] Nevada 2.1% [1.4%-3.1%]
Arkansas 0.6% [0.4%-1.1%] New Hampshire 3.1% [1.7%-5.8%]
California 1.6% [1.1%-2.4%] New Jersey 15.5% [11.8%-20.3%]
Colorado 3.7% [2.7%-5.4%] New Mexico 2.8% [1.7%-4.5%]
Connecticut 13.4% [10.0%-18.3%] New York 16.6% [12.9%-21.4%]
Delaware 5.1% [3.4%-8.0%] North Carolina 1.2% [0.8%-2.0%]
District of Columbia 11.6% [8.4%-16.2%] North Dakota 1.5% [0.8%-3.0%]
Florida 1.4% [0.9%-2.1%] Ohio 3.0% [1.9%-5.0%]
Georgia 3.2% [2.2%-4.6%] Oklahoma 1.1% [0.8%-1.6%]
Hawaii 0.1% [0.0%-0.3%] Oregon 0.4% [0.3%-0.7%]
Idaho 0.7% [0.4%-1.1%] Pennsylvania 5.4% [3.8%-7.9%]
Illinois 6.2% [4.3%-9.3%] Rhode Island 9.8% [6.4%-16.3%]
Indiana 4.4% [3.2%-6.5%] South Carolina 1.3% [0.9%-1.9%]
Iowa 3.1% [1.9%-5.3%] South Dakota 1.0% [0.5%-1.7%]
Kansas 1.0% [0.6%-1.4%] Tennessee 1.0% [0.6%-1.6%]
Kentucky 1.3% [0.9%-2.0%] Texas 1.1% [0.7%-1.7%]
Louisiana 7.8% [5.9%-10.4%] Utah 0.9% [0.5%-1.6%]
Maine 0.8% [0.4%-1.3%] Vermont 0.8% [0.5%-1.3%]
Maryland 5.9% [4.2%-8.5%] Virginia 2.5% [1.7%-3.8%]
Massachusetts 12.2% [9.0%-16.4%] Washington 2.0% [1.4%-2.7%]
Michigan 6.3% [4.7%-8.3%] West Virginia 0.6% [0.3%-1.0%]
Minnesota 2.9% [1.9%-4.7%] Wisconsin 1.4% [0.9%-2.2%]
Mississippi 4.3% [2.7%-6.9%] Wyoming 0.6% [0.1%-2.0%]
Missouri 1.7% [1.1%-2.5%] National 4.1% [3.7%-4.5%]

Pacific Great Plains Northeast Corridor
200,000 2,000,000
90,000
150,000 1,500,000
60,000
100,000 1,000,000
30,000
50,000 500,000
0 0 0
24 eb
9 b
23 ar
6 r
20 pr
4 pr
18 ay
ay
24 eb
9 b
23 ar
6 r
20 pr
4 pr
18 ay
ay
24 eb
9 b
23 ar
6 r
20 r
4 pr
18 ay
ay
a
a
Ap
Fe
Fe
Fe
M
M
A
A
M
M
A
A
M
M
A
M
M
M
M
M
M
F
F
10
10
10
Mountain Great Lakes Southern Appalachia
600,000
150,000
Number of people
100,000
100,000 400,000
50,000
50,000 200,000
0 0 0
24 eb
9 b
23 ar
6 r
20 pr
4 pr
18 ay
ay
24 eb
9 b
23 ar
6 r
20 pr
4 pr
18 ay
ay
24 eb
9 b
23 ar
6 r
20 r
4 pr
18 ay
ay
a
a
Ap
Fe
Fe
Fe
M
M
A
A
M
M
A
A
M
M
A
M
M
M
M
M
M
F
F
10
10
10
TOLA South Atlantic
400,000
150,000
300,000
100,000
200,000
50,000
100,000
0 0
24 eb
9 b
23 ar
6 r
20 pr
4 pr
18 ay
ay
24 eb
9 b
23 ar
6 r
20 r
4 pr
18 ay
ay
a
a
Ap
Fe
Fe
M
M
A
A
M
M
A
M
M
M
M
F
F
10
10
# new infections [95% CI] # new infections [50% CI] # current infections [95% CI] # current infections [50% CI]
Figure 7: Estimates for the effective number of infectious individuals on a day in purple (light purple, 95% CI, dark purple
50% CI) and for newly infected people per day in blue (light blue, 95% CI, dark blue: 50% CI).

Table 2: Posterior model estimates of the number of currently infectious individuals as of 25 May 2020.
State Number of infectious State Number of infectious

individuals (mean [95% individuals (mean [95%
credible interval]) credible interval])
Alabama 13,000 [2,000-35,000] Montana 0 [0-1,000]
Alaska 0 [0-2,000] Nebraska 4,000 [0-13,000]
Arizona 22,000 [5,000-60,000] Nevada 6,000 [1,000-17,000]
Arkansas 2,000 [0-8,000] New Hampshire 8,000 [1,000-24,000]
California 65,000 [11,000-190,000] New Jersey 49,000 [9,000-145,000]
Colorado 14,000 [2,000-48,000] New Mexico 9,000 [2,000-24,000]
Connecticut 31,000 [6,000-86,000] New York 68,000 [10,000-210,000]
Delaware 5,000 [0-16,000] North Carolina 14,000 [2,000-42,000]
District of Columbia 6,000 [1,000-16,000] North Dakota 2,000 [0-8,000]
Florida 32,000 [7,000-87,000] Ohio 57,000 [11,000-151,000]
Georgia 38,000 [10,000-95,000] Oklahoma 3,000 [0-9,000]
Hawaii 0 [0-1,000] Oregon 1,000 [0-3,000]
Idaho 1,000 [0-4,000] Pennsylvania 59,000 [9,000-182,000]
Illinois 87,000 [17,000-243,000] Rhode Island 16,000 [3,000-46,000]
Indiana 24,000 [4,000-74,000] South Carolina 7,000 [1,000-19,000]
Iowa 20,000 [5,000-50,000] South Dakota 1,000 [0-4,000]
Kansas 1,000 [0-5,000] Tennessee 11,000 [2,000-30,000]
Kentucky 5,000 [0-17,000] Texas 40,000 [8,000-112,000]
Louisiana 17,000 [3,000-47,000] Utah 5,000 [0-15,000]
Maine 1,000 [0-4,000] Vermont 0 [0-1,000]
Maryland 31,000 [6,000-87,000] Virginia 29,000 [8,000-71,000]
Massachusetts 44,000 [7,000-129,000] Washington 7,000 [1,000-24,000]
Michigan 27,000 [4,000-79,000] West Virginia 1,000 [0-4,000]
Minnesota 24,000 [5,000-64,000] Wisconsin 11,000 [2,000-30,000]
Mississippi 21,000 [5,000-54,000] Wyoming 1,000 [0-5,000]
Missouri 10,000 [2,000-29,000] National 949000 [189000 - 2691000]

Constant mobility Increased mobility 20% Increased mobility 40%
1,000
Washington
500
2,000
New York
1,500
1,000
500
0
500
Massachusetts
400
300
200
100
0
5,000
4,000
Florida
3,000
2,000
1,000
0
8,000
6,000
California
4,000
2,000
0
9 b
23 ar
6 r
20 r
4 r
18 ay
ay
15 n
29 n
13 n
27 l
Fe l
9 b
23 ar
6 r
20 r
4 r
18 ay
ay
15 n
29 n
13 n
27 l
Fe l
9 b
23 ar
6 r
20 pr
4 r
18 ay
ay
15 n
29 n
13 n
27 l
l
Ju
24 Ju
Ju
24 Ju
Ju
Ju
a
Ap
Ap
a
Ap
Ap
Ap
Fe
Ju
Ju
Ju
Ju
Ju
Ju
Ju
Ju
Ju
M
M
M
M
M
M
A
M
M
M
M
M
M
1
1
24
Figure 8: State-level scenario estimates of deaths for Washington, New York, Massachusetts, Florida and California. The
ribbon shows the 95% credible intervals (CIs) for each scenario. The first column of plots show the results of scenario (a)
where mobility is kept constant, the middle column shows results for scenario (b) where there is a 20% return to baseline
mobility, and the right column shows scenario (c) where there is a 40% return to baseline mobility.
people have begun to resume pre-stay-at-home behaviour, but have not yet returned to baseline mobility. We hold this
20% return to baseline constant for the duration of the 8-week scenario.
We present three scenarios (a) constant mobility (mobility remains at current levels for 8 weeks), (b) 20% return to pre-
stay-at-home mobility from current levels and (c) 40% return to pre-stay-at-home mobility from current levels. We justify
the selection of these scenarios by examining how mobility has changed in states that have already begun to relax social
distancing guidelines. For example, Colorado’s stay-at-home order expired on the 26th of April, and activity level reported
by the Colorado Department of Public Health has recovered approximately 30% of the decrease observed following initial
implementation of NPIs [6]. Figure 8 shows the estimated number of deaths for each scenario in the five states discussed
above: Washington, New York, Massachusetts, Florida, and California. Results for all the states modelled are included in
Appendix F. These estimates are certainly not forecasts and are based on multiple assumptions, but they illustrate the
potential consequences of increasing mobility across the general population: in almost all cases, after 8 weeks, a 40%
return to baseline leads to an epidemic larger than the current wave.

3 Conclusions
In this report we use a Bayesian semi-mechanistic model to investigate the impact of these NPIs via changes in mobility.
Our model uses mobility to predict the rate of transmission, neglecting the potential effect of additional behavioural
changes or interventions such as testing and tracing. While mobility will explain a large amount of the variance in Rt ,
there is likely to be substantial residual variation which will be geographically heterogeneous. We attempt to account
for this residual variation through a second order, weekly, autoregressive process. This stochastic process is able to
pick up variation drive by the data but is unable to determine associations or causal mechanisms. Figure 13 shows the
residual variation captured by the autoregressive process, and given these lines are flat for the majority of states, we can
conclude that much of the variation we see in the observed death data can be attributed to mobility. However, there are
states, such as New York, where this residual effect is large which suggests that additional factors have contributed to the
reduction in Rt . We hypothesise these could be behavioural changes but testing this hypothesis will require additional
data.
We find that the starting reproduction number is associated with population density and the chronological date of epi-
demic onset. These two relationships suggest two dimensions which may influence the starting reproduction number
and underscore the variability between states. We are cautious to draw any causal relationships from these associations;
our results highlight that more additional studies of these factors are need at finer spatial scales.
We find that the posterior mean of the current reproduction is above 1 in 0 states, with 95% confidence, and above 1
in 26 states with 50% confidence. These current reproduction numbers suggest that in many states the US epidemic is
not under control and caution must be taken in loosening current interventions without additional measures in place.
The high reproduction numbers are geographically clustered in the southern US and Great Plains region, while lower
reproduction numbers are observed in areas that have suffered high COVID-19 mortality (such as the Northeast Corridor).
We simulate forwards in time a partial return of mobility back to pre-COVID levels, while keeping all else constant, and
find substantial resurgence is expected. In the majority of states, the deaths expected over a two-month period would
exceed current levels by more than two-fold. This increase in mobility is modest and held constant for 8 weeks. However,
these results must be heavily caveated: our results do not account for additional interventions that may be introduced
such as mass testing, contact tracing and changing work place/transit practices. Our results also do not account for
behavioural changes that may occur such as increased mask wearing or changes in age specific movement. Therefore,
our scenarios are pessimistic in nature and should be interpreted as such. Given these caveats, we conjecture at the
present time that, in the absence of additional interventions (such as mass testing), additional behavioural modifications
are unlikely to substantially reduce Rt in of their own.
We estimate the number of individuals that have been infected by SARS-CoV2 to date. Our attack rates are sensitive to
the assumed values of infection fatality rate (IFR). We account for each individual state’s age structure, and further adjust
for contact mixing patterns [19]. To ensure assumptions about IFR do not have undue influence on our conclusions, we
incorporate prior noise in the estimate, and perform a sensitivity analysis using different contact matrices. Also, our attack
rates for New York are in line with those from recent serological studies [1]. We show that while reductions in the daily
infections continue, the reservoir of infectious individuals remains large. This reservoir also implies that interventions
should remain in place longer than the daily case count implies, as trends in the number of infectious individuals lags
behind. The magnitude of difference between newly infected and currently infected individuals suggest that mass testing

and isolation could be an effective intervention.
Our results suggest that while the US has substantially reduced its reproduction numbers in all states, there is little
evidence that the epidemic is under control in the majority of states. Without changes in behaviour that result in reduced
transmission, or interventions such as increased testing that limit transmission, new infections of COVID-19 are likely to
persist, and, in the majority of states, grow.
4 Data
Our model uses daily real-time state-level aggregated data published by New York Times (NYT) [20] for New York State
and John Hopkins University (JHU) [3] for the remaining states. There is no single source of consistent and reliable data
for all 50 states. We acknowledge that data issues such as under reporting and time lags can influence our results. In
previous reports [8, 9, 7] we have shown our modelling methodology is generally robust to these data issues due to
pooling. However, we do recognise no modelling methodology will be able to surmount all data issues; therefore these
results should be interpreted as the best estimates based on current data, and are subject to change with future data
consolidation. JHU and NYT provide information on confirmed cases and deaths attributable to COVID-19, however again,
the case data are highly unrepresentative of the incidence of infections due to under-reporting and systematic and state-
specific changes in testing. We, therefore, use only deaths attributable to COVID-19 in our model. While the observed
deaths still have some degree of unreliability, again due to changes in reporting and testing, we believe the data are of
sufficient fidelity to model. For age specific population counts we use data from the U.S. Census Bureau in 2018 [21]. The
timing of social distancing measures was collated by the University of Washington [11].
We use Google’s COVID-19 Community Mobility Report [5] 2 which provides data on movement in the USA by states and
highlights the percent change in visits to:
• Grocery & pharmacy: Mobility trends for places like grocery markets, food warehouses, farmers markets, speciality
food shops, drug stores, and pharmacies.
• Parks: Mobility trends for places like local parks, national parks, public beaches, marinas, dog parks, plazas, and
public gardens.
• Transit stations: Mobility trends for places like public transport hubs such as subway, bus, and train stations.
• Retail & recreation: Mobility trends for places like restaurants, cafes, shopping centres, theme parks, museums,
libraries, and movie theatres.
• Residential:Mobility trends for places of residence.
• Workplaces: Mobility trends for places of work.
The mobility data show length of stay at different places compared to a baseline. It is therefore relative, i.e. mobility of
-20% means that, compared to normal circumstances individuals are engaging in a given activity 20% less.
2 We use mobility data from the report released on 23rd May, which contains data until 16th May 2020. For dates after 16th May 2020, we impute
the Google mobility data using a supervised machine learning approach with random forests, trained with the visitdata.org Foursquare data [22] as
predictors and Google data as labels.

5 Methods
We introduced a new Bayesian framework for estimating the transmission intensity and attack rate (percentage of the
population that has been infected) of COVID-19 from the reported number of deaths in a previous report [7]. This frame-
work uses the time-varying reproduction number Rt to inform a latent function for infections, and then these infections,
together with probabilistic lags, are calibrated against observed deaths. Observed deaths, while still susceptible to under
reporting and delays, are more reliable than the reported number of confirmed cases, although the early focus of most
surveillance systems on cases with reported travel histories to China may have missed some early deaths. Changes in
testing strategies during the epidemic mean that the severity of confirmed cases as well as the reporting probabilities
changed in time and may thus have introduced bias in the data.
In this report, we adapt our original Bayesian semi-mechanistic model of the infection cycle to the states in the USA. We
infer plausible upper and lower bounds (Bayesian credible intervals) of the total populations infected (attack rates) and
the reproduction number over time (Rt ). In our framework we parametrize Rt as a function of Google mobility data.
We fit the model jointly to COVID-19 data from all regions to assess whether there is evidence that changes in mobility
have so far been successful at reducing Rt below 1. Our model is a partial pooling model, where the effect of mobility is
shared, but region- and state-specific modifiers can capture differences and idiosyncrasies among the regions.
We note that future directions should focus on embedding mobility in realistic contact mechanisms to establish a closer
relationship to transmission.
5.1 Model specifics
We observe daily deaths Dt,m for days t ∈ {1, . . . , n} and states m ∈ {1, . . . , M }. These daily deaths are modelled using
a positive real-valued function dt,m = E[Dt,m ] that represents the expected number of deaths attributed to COVID-19.
d2t,m
The daily deaths Dt,m are assumed to follow a negative binomial distribution with mean dt,m and variance dt,m + φ ,
where ψ follows a positive half normal distribution, i.e.

!
d2t,m
Dt,m ∼ Negative Binomial dt,m , dt,m + ,
ψ
ψ ∼ N + (0, 5).
Here, N (µ, σ) denotes a normal distribution with mean µ and standard deviation σ. We say that X follows a positive
half normal distribution N + (0, σ) if X ∼ |Y |, where Y ∼ N (0, σ).
To mechanistically link our function for deaths to our latent function for infected cases, we use a previously estimated
COVID-19 infection fatality ratio (IFR, probability of death given infection) together with a distribution of times from
infection to death π. Details of this calculation can be found in [23, 24]. From the above, every region has a specific
mean infection fatality ratio ifrm (see Appendix G). To incorporate the uncertainty inherent in this estimate we allow the
ifrm for every state to have additional noise around the mean. Specifically we assume
ifr∗m ∼ ifrm ·N (1, 0.1).

We believe a large-scale contact survey similar to polymod [19] has not been collated for the USA, so we assume the
contact patterns are similar to those in the UK. We conducted a sensitivity analysis, shown in Appendix G, and found that
the IFR calculated using the contact matrices of other European countries lay within the posterior of ifr∗m .
Using estimated epidemiological information from previous studies [23, 24], we assume the distribution of times from
infection to death π (infection-to-death) to be
π ∼ Gamma(5.1, 0.86) + Gamma(17.8, 0.45).
The expected number of deaths dt,m , on a given day t, for state m is given by the following discrete sum:
t−1
X
dt,m = ifr∗m cτ,m πt−τ ,
τ =0
R s+0.5
where cτ,m is the number of new infections on day τ in state m and where π is discretized via πs = s−0.5
π(τ )dτ for
R 1.5
s = 2, 3, ..., and π1 = 0 π(τ )dτ , where π(τ ) is the density of π.
The true number of infected individuals, c, is modelled using a discrete renewal process. We specify a generation distri-
bution g with density g(τ ) as:
g ∼ Gamma(6.5, 0.62).
Given the generation distribution, the number of infections ct,m on a given day t, and state m, is given by the following
discrete convolution function:
t−1
X
ct,m = St,m Rt,m cτ,m gt−τ , (1)
τ =0
Pt−1
i=0 ci,m
St,m = 1 −
Nm
R s+0.5
where, similar to the probability of death function, the generation distribution is discretized by gs = s−0.5 g(τ )dτ for
R 1.5
s = 2, 3, ..., and g1 = 0 g(τ )dτ . The population of state m is denoted by Nm . We include the adjustment factor
St,m to account for the number of susceptible individuals left in the population.
We parametrise Rt,m as a linear function of the relative change in time spent (from a baseline)
3
X region state
Rt,m = R0,m · f (−( Xt,m,k αk ) − Yt,m αr(m) − Zt,m αm − m,wm (t) ), (2)
k=1
where f (x) = 2 exp(x)/(1 + exp(x)) is twice the inverse logit function. Xt,m,k are covariates that have the same effect
for all states, Yt,m is a covariate that also has a region-specific effect, r(m) ∈ {1, . . . , R} is the region a state is in (see
Figure 1), Zt,m is a covariate that has a state-specific effect and m,wm (t) is a weekly AR(2) process, centred around 0,
that captures variation between states that is not explained by the covariates.
The prior distribution for R0,m [25] was chosen to be
R0,m ∼ N (3.28, κ) with κ ∼ N + (0, 0.5),
where κ is the same among all states.
average transit
In the analysis of this paper we chose the following covariates: Xt,m,1 = Mt,m , Xt,m,2 = Mt,m , Xt,m,3 =
residential average transit
Mt,m , Yt,m = Mt,m , and Zt,m = Mt,m , where the mobility variables are from [5] and defined as follows
(all are encoded so that 0 is the baseline and 1 is a full reduction of the mobility in this dimension):

average
• Mt,m is an average of retail and recreation, groceries and pharmacies, and workplaces. An average is taken as
these dimensions are strongly collinear.
transit
• Mt,m is encoding mobility for public transport hubs. For states where less than 20% of the working population
transit
aged 16 and over uses public transportation, we set Mt,m = 0, i.e. this mobility has no effect on transmission.
transit
For states in which more than 20% of the working population commutes using public transportation, Mt,m is the
mobility on transit.
residential
• Mt,m are the mobility trends for places of residences.
The weekly, state-specific effect is modelled as a weekly AR(2) process, centred around 0 with stationary standard devia-
tion σw that, in every state, starts on the first day of its seeding of infectons, i.e. 30 days before a total of 10 cumulative
∗
deaths have been observed in this state. The AR(2) process starts with 1,m ∼ N (0, σw ),
∗
w,m ∼ N (ρ1 w−1,m + ρ2 w−2,m , σw ) for m = 2, 3, 4, . . . (3)
with independent priors on ρ1 and ρ2 that are normal distributions conditioned to be in [0, 1]; the prior for ρ1 is a
N (0.8, .05) distribution conditioned to be in [0, 1] the prior for ρ2 is a N (0.1, .05) distribution, conditioned to be in
[0, 1]. The prior for σw , the standard deviation of the stationary distribution of w is chosen as σw ∼ N + (0, .2).
∗
The standard deviation of the weekly updates to achieve this standard deviation of the stationary distribution is σw =
p
σw 1 − ρ21 − ρ22 − 2ρ21 ρ2 /(1 − ρ2 ).
The conversion from days to weeks is encoded in wm (t). Every 7 days, wm is incremented, i.e. we setwm (t) = b(t −
tstart start
m )/7c+1, where tm is the first day of seeding. Due to the lag between infection and death, our estimates of Rt in the
last two weeks before the end of our observations are (almost) not informed by corresponding death data. Therefore, we
assume that the last two weeks have the same random weekly effect as the week 3 weeks before the end of observation.
The prior distribution for the shared coefficients were chosen to be
αk ∼ N (0, 0.5), k = 1, . . . , 3,
and the prior distribution for the pooled coefficients were chosen to be
αrregion ∼ N (0, γr ), r = 1, . . . , R, with γr ∼ N + (0, 0.5),

state
αm ∼ N (0, γs ), m = 1, . . . , M with γs ∼ N + (0, 0.5).
We assume that seeding of new infections begins 30 days before the day after a state has cumulatively observed 10
deaths. From this date, we seed our model with 6 sequential days of an equal number of infections: c1,m = · · · =
c6,m ∼ Exponential( τ1 ), where τ ∼ Exponential(0.03). These seed infections are inferred in our Bayesian posterior
distribution.
We estimated parameters jointly for all states in a single hierarchical model. Fitting was done in the probabilistic pro-
gramming language Stan[26] using an adaptive Hamiltonian Monte Carlo (HMC) sampler.

6 Acknowledgements
We would like to thank Amazon AWS and Microsoft Azure for computational credits and we would like to thank the
Stan development team for their ongoing assistance. This work was supported by Centre funding from the UK Medical
Research Council under a concordat with the UK Department for International Development, the NIHR Health Protection
Research Unit in Modelling Methodology and Community Jameel.
References
[1] S M Kissler et al. “Reductions in commuting mobility predict geographic differences in SARS-CoV-2 prevalence in
New York City”. In: (2020). URL: http://nrs.harvard.edu/urn-3:HUL.InstRepos:42665370.
[2] Santa Clara County Public Health. County of Santa Clara Identifies Three Additional Early COVID-19 Deaths. 2020.
URL: https://www.sccgov.org/sites/covid19/Pages/press-release-04-21-20-early.aspx.
[3] E Dong, H Du, and L Gardner. “An interactive web-based dashboard to track COVID-19 in real time”. eng. In: The
Lancet. Infectious diseases (Feb. 2020), pp. 1473–3099. ISSN: 1474-4457. URL: https://pubmed.ncbi.nlm.nih.
gov/32087114https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7159018/.
[4] C Courtemanche et al. “Strong Social Distancing Measures In The United States Reduced The COVID-19 Growth
Rate”. In: Health Affairs 39.7 (2020).
[5] A Aktay et al. “Google COVID-19 Community Mobility Reports: Anonymization Process Description (version 1.0)”.
In: ArXiv abs/2004.0 (2020).
[6] J Bayham et al. Colorado Mobility Patterns During the COVID-19 Response. 2020. URL: http://www.ucdenver.
edu/academics/colleges/PublicHealth/coronavirus/Documents/Mobility%20Report_final.pdf.
[7] S Flaxman et al. Report 13: Estimating the number of infections and the impact of non-pharmaceutical interventions
on COVID-19 in 11 European countries. 2020.
[8] M Vollmer et al. Report 20: Using mobility to estimate the transmission intensity of COVID-19 in Italy: A subnational
analysis with future scenarios. 2020.
[9] T A Mellan et al. Report 21 - Estimating COVID-19 cases and reproduction number in Brazil. 2020.
[10] M. Reston, K. Sgueglia, and C. Mossburg. Governors on East and West coasts form pacts to decide when to reopen
economies. 2020. URL: https://edition.cnn.com/2020/04/13/politics/states-band-together-
reopening-plans/index.html.
[11] N Fullman et al. State-level social distancing policies in response to COVID-19 in the US. 2020. URL: http://www.
covid19statepolicy.org.
[12] United States Census Bureau. Explore Census Data. 2020. URL: https://data.census.gov/cedsci/.
[13] Morning Consult. How the Coronavirus Outbreak Is Impacting Public Opinion. Accessed on 15/05/2020. 2020. URL:
https://morningconsult.com/form/coronavirus-outbreak-tracker/.
[14] J Stromberg. The real reason American public transportation is such a disaster. 2015. URL: https://www.vox.
com/2015/8/10/9118199/public-transportation-subway-buses.

[15] B D Dalziel et al. “Urbanization and humidity shape the intensity of influenza epidemics in U.S. cities”. In: Science
362.6410 (2018), pp. 75–79. ISSN: 0036-8075. eprint: https://science.sciencemag.org/content/362/
6410/75.full.pdf. URL: https://science.sciencemag.org/content/362/6410/75.
[16] City and County of Department of Public Health San Francisco. ORDER OF THE HEALTH OFFICER No. C19-07c. 2020.
URL: https : / / sf . gov / sites / default / files / 2020 - 04 / 2020 . 04 . 29 % 20FINAL % 20 % 28signed % 29 %
20Health%20Officer%20Order%20C19-07c-%20Shelter%20in%20Place.pdf.
[17] L Gamio S Mervosh J Lee and N Popovich. See Which States Are Reopening and Which Are Still Shut Down. 2020.
URL: https://www.nytimes.com/interactive/2020/us/states-reopen-map-coronavirus.html.
[18] KEC Ainslie et al. “Evidence of initial success for China exiting COVID-19 social distancing policy after achieving
containment”. In: Wellcome Open Research 5.81 (2020).
[19] J Mossong et al. “Social Contacts and Mixing Patterns Relevant to the Spread of Infectious Diseases”. In: PLOS
Medicine 5.3 (Mar. 2008), pp. 1–1. URL: https://doi.org/10.1371/journal.pmed.0050074.
[20] M Smith et al. Coronavirus (Covid-19) Data in the United States. 2020. URL: https://github.com/nytimes/
covid-19-data.
[21] Census reporter. Census reporter. 2020. URL: https://censusreporter.org.
[22] VisitData. VisitData Trends. 2020. URL: https://visitdata.org/index.html.
[23] R Verity et al. “Estimates of the severity of COVID-19 disease”. In: Lancet Infect Dis (2020).
[24] P Walker et al. Report 12: The Global Impact of COVID-19 and Strategies for Mitigation and Suppression. 2020.
URL: https : / / www . imperial . ac . uk / mrc - global - infectious - disease - analysis / news -- wuhan -
coronavirus/.
[25] Y Liu et al. “The reproductive number of COVID-19 is higher compared to SARS coronavirus”. In: Journal of Travel
Medicine (2020). ISSN: 17088305.
[26] B Carpenter et al. “Stan: A Probabilistic Programming Language”. In: Journal of Statistical Software 76.1 (2017),
pp. 1–32. ISSN: 1548-7660. URL: http://www.jstatsoft.org/v76/i01/.
A Mobility regression analysis
In Figure 9 we regress NPIs against average mobility. We parameterise NPIs as piece-wise constant functions that are
zero when the intervention has not been implemented and one when it has. We evaluate the correlation between the
predictions from the linear model and the actual average mobility. We also lag the timing of interventions and investigate
its impact on predicted correlation.

0.8
20
0.7
Number of States 15
Correlation
0.6
10
0.5
5
0.4
0.00 0.25 0.50 0.75 1.00 −20 −10 0 10 20

Correlation Lag
Figure 9: Mobility regression analysis.
B Effect sizes
Great Lakes ●
Great Plains ●
Mountain ●
Average mobility
Northeast Corridor ●
Pacific ●
South Atlantic ●
Southern Appalachia ●
TOLA ●
−50% −25% 0% 25% 50%
Figure 10: Regional average mobility covariate effect size plots.

Great Lakes ●
Great Plains ●
Mountain ●
●
Intercept
Northeast Corridor
Pacific ●
South Atlantic ●
Southern Appalachia ●
TOLA ●
0% 25% 50%
Figure 11: Regional intercept covariate effect size plots.
California ●
District of Columbia ●
Illinos ●
Maryland ●
Transit
Massachusetts ●
New Jersey ●
New York ●
Washington ●
−25% 0% 25%
Figure 12: State-level covariate effect size plots.

C State-specific weekly effects
Our model includes a state-specific weekly effect w,m (see equations 2, 3) for each week w in the epidemic period for
a state. As described in Section 5, We assign an autoregressive process with mean 0 as prior to this effect. This weekly
effect is held constant for the 4 weeks up to the present week. Figure 13 shows the posterior of this effect on the same
scale as in Figure 2, that is, the percent reduction in Rt with mobility variables held constant3 . Values above 0 have the
interpretation that the state-specific weekly effect lowers the reproduction number Rt,m , i.e. transmission for week t
and state m is lower than what is explained by the mobility covariates.
Alaska Vermont New Hampshire Maine
50%
0%
−50%
New York Connecticut Massachusetts
50%
0%
−50%
Washington Montana North Dakota South Dakota Minnesota Wisconsin Michigan Pennsylvania New Jersey Rhode Island
50%
0%
−50%
Oregon Idaho Wyoming Nebraska Iowa Illinois Indiana Ohio Virginia District of Columbia Delaware
50%
0%
−50%
Effect
Nevada Utah Colorado Kansas Missouri Tennessee Kentucky West Virginia North Carolina Maryland
50%
0%
−50%
California Arizona New Mexico Oklahoma Arkansas Mississippi Alabama Georgia South Carolina
50%
0%
−50%
Texas Louisiana Florida
50%
0%
−50%
Hawaii
50%
0%
−50%
Figure 13: Percent reduction in Rt due to the weekly, state-level autoregressive effect after the emergency decree.
3 Draws from the posterior are transformed with 1 − f (−m,wm (t) ), where f (x) = 2 exp(x)/(1 + exp(x)) is twice the inverse logit function.

D Model predictions for all states
State-level estimates of infections, deaths and Rt . Left: daily number of deaths, brown bars are reported deaths, blue
bands are predicted deaths, dark blue 50% credible interval (CI), light blue 95% CI. Middle: daily number of infections,
brown bars are reported infections, blue bands are predicted infections, CIs are same as left. The number of daily infec-
tions estimated by our model drops immediately after an intervention, as we assume that all infected people become
immediately less infectious through the intervention. Afterwards, if the Rt is above 1, the number of infections will start
growing again. Right: time-varying reproduction number Rt dark green 50% CI, light green 95% CI. Icons are interventions
shown at the time they occurred.
Alaska Timing
Started
1.00
3 Eased
200 Interventions
0.75
Emergency decree
2 Restrict public events
Rt
Business closure
0.50 Restaurant closure
100 School closure
1
0.25
Credible intervals
0 50%
0.00 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Alabama Timing
Started
6,000
4 Eased
30
Interventions
3
Emergency decree
4,000
Rt
Business closure
2
Restaurant closure
School closure

1
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Arkansas Timing
Started
8.0 4 Eased
1,000 Interventions
6.0 3
Emergency decree
Rt
Business closure
4.0 2
Restaurant closure
500 School closure

1
2.0
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Arizona Timing
12,500 Started
4 Eased
60
10,000
Interventions
3
Emergency decree
40
Rt
Business closure
2
Restaurant closure
5,000 School closure
20 1
2,500
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18

California Timing
150 40,000
Started
4 Eased

3
100 Emergency decree
Rt
2
Restaurant closure
School closure
50
10,000 1
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Colorado Timing
Started
8,000 4 Eased
100
Interventions
6,000 3
Emergency decree
Rt
Business closure
4,000 2
Restaurant closure
50 School closure
1
2,000
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Connecticut Timing
25,000 Started
200 6
Eased
20,000
Interventions
4
Rt
Business closure
100 Restaurant closure
10,000
School closure
2
50
5,000
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
District of Columbia Timing
20.0 Started
3,000 Eased
15.0 4 Interventions
Emergency decree
Rt
Business closure
10.0
Restaurant closure
2 School closure
1,000 Stay at home mandate
5.0
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Delaware Timing
16.0 Started
4 Eased
12.0 2,000 Interventions

3
Emergency decree
Rt
Business closure
8.0 2
Restaurant closure

4.0 1
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Florida Timing
100 Started
Eased
15,000
75 4 Interventions
Emergency decree
10,000
Rt
Business closure
50
Restaurant closure
2 School closure
5,000
25
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18

Georgia Timing
Started
100 4 Eased
15,000

Interventions
75 3
Emergency decree
10,000
Rt
Business closure
50 2
Restaurant closure
School closure
5,000 1
25
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Hawaii Timing
Started
2.00
Eased
2.0
300
Interventions
1.50
1.5 Emergency decree
200
Rt
Business closure
1.00 1.0 Restaurant closure
School closure
100 0.5
0.50
Credible intervals
0.0 50%
0.00 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Idaho Timing
Started
6.0 4 Eased
600
Interventions
3
Emergency decree
4.0
Rt
Business closure
2
Restaurant closure
School closure
2.0 200 Stay at home mandate
1
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Illinois Timing
200 Started
6
Eased
40,000
150 Interventions
30,000 Emergency decree

4
Rt
Business closure
100
20,000 Restaurant closure
School closure
2
50
10,000
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Indiana Timing
12,500 Started
Eased
150
10,000
4 Interventions
Emergency decree

100
Rt
Business closure
Restaurant closure
5,000 2 School closure
2,500
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Iowa Timing
Started
4 Eased
7,500
Interventions
20 3
Emergency decree
5,000
Rt
Business closure
2
Restaurant closure
10 School closure
2,500 1
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18

Kansas Timing
Started
4 Eased

1,000 Interventions
10.0 3
Emergency decree
Rt
Business closure
2
Restaurant closure
5.0 500 School closure

1
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Kentucky Timing
Started
20.0 3,000
4 Eased
Interventions
15.0 3
Emergency decree
2,000
Rt
Business closure
10.0 2
Restaurant closure
School closure
1
5.0
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Louisiana Timing
Started
Eased
15,000
4 Interventions
100
Emergency decree
10,000
Rt
Business closure
Restaurant closure
2 School closure
50
5,000
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Maine Timing
Started
5.0 800 Eased
4.0 4 Interventions
3.0
Rt
Business closure
400 Restaurant closure
200
1.0
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Maryland Timing
125 Started
Eased
100 12,000
4 Interventions
Emergency decree

Rt

Restaurant closure
50 2 School closure
4,000
25
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Massachusetts Timing
50,000 Started
300 6
Eased
40,000
Interventions
Emergency decree
200 4
Rt
Business closure
Restaurant closure
2
10,000
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18

Michigan Timing
250 Started
6
Eased
30,000

200
Interventions
Emergency decree
4
20,000
Rt
Business closure
Restaurant closure
100 School closure
2
10,000
50
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Minnesota Timing
12,500 Started
Eased
30 10,000
4 Interventions
Emergency decree
Rt
20 Business closure
Restaurant closure
10
2,500
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Mississippi Timing
Started
4 Eased
30
7,500
Interventions
3
Emergency decree

Rt

2
Restaurant closure
School closure

2,500 1
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Missouri Timing
5,000
Started
60
Eased
4,000
4 Interventions
Emergency decree
40 3,000 Restrict public events
Rt
Business closure
Restaurant closure
1,000
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Montana Timing
300
Started
4.0
Eased
2.0
Interventions
3.0
200 1.5 Emergency decree
Rt
Business closure
2.0 1.0 Restaurant closure
School closure
100
1.0 0.5
Credible intervals
0.0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Nebraska Timing
16.0 2,500 Started
4 Eased
2,000
12.0 Interventions
3
Emergency decree
Rt
Business closure
8.0 2
Restaurant closure
4.0 1
500
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18

Nevada Timing
Started
4 Eased
3,000

Interventions
20 3
Emergency decree
2,000
Rt
Business closure
2
Restaurant closure
10 School closure
1,000 1
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
New Hampshire Timing
Started
Eased
5,000
20.0
4 Interventions
4,000
Emergency decree
15.0
3,000
Rt
Business closure
Restaurant closure
2,000
5.0
1,000
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
New Jersey Timing
500
Started
6
Eased
400 60,000
Interventions
Emergency decree
4
40,000
Rt
Business closure
Restaurant closure
200
School closure
2
20,000
100
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
New Mexico Timing
16.0 Started
4,000 Eased
12.0 4 Interventions
Rt
Business closure
8.0
2 School closure
4.0 1,000
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
New York Timing

250,000
Started
6
Eased
900 200,000
Interventions
Emergency decree
4
600
Rt
Business closure
Restaurant closure
100,000
School closure
2
300
50,000
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
North Carolina Timing

50 Started
Eased
40
6,000
4 Interventions
Emergency decree
4,000
Rt
Business closure
Restaurant closure
20 2 School closure
2,000
10
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18

North Dakota Timing

1,500
Started
6.0
Eased
2.0

Interventions
1,000 1.5 Emergency decree
4.0
Rt
Business closure
School closure
500
2.0 Stay at home mandate
0.5
Credible intervals
0.0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Ohio Timing
150
Started
30,000 Eased
4 Interventions
Rt
Business closure
Restaurant closure
2 School closure
50
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Oklahoma Timing
Started
2,000 4 Eased
20.0
Interventions
1,500 3
15.0 Emergency decree
Rt
Business closure
1,000 2
School closure
1
5.0 500
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Oregon Timing
12.0 800
Started
4 Eased
9.0
600 Interventions
3
Emergency decree
Rt
6.0 400 Business closure

2
Restaurant closure
School closure
3.0 200 1
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Pennsylvania Timing
40,000 Started
6
Eased
300
Emergency decree
4
200
Rt
Business closure
20,000
Restaurant closure
School closure
2
100
10,000
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Rhode Island Timing

Started
6
30 Eased
7,500
Interventions
Emergency decree
4
5,000
Rt
Business closure
Restaurant closure
School closure
2
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18

South Carolina Timing

3,000
40 Started
4 Eased

Interventions
30 3
Rt
Business closure
20 2
Restaurant closure
School closure
1,000
10 1
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
South Dakota Timing
Started
5.0
3 Eased
500
4.0 Interventions
3.0
Rt
300 Business closure
Restaurant closure
2.0 School closure
200 1
1.0 100
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Tennessee Timing
16.0 Started
6,000
4 Eased
12.0 Interventions
3
Emergency decree
4,000
Rt
Business closure
8.0 2
Restaurant closure
School closure
4.0 1
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Texas Timing
Started
60 Eased
20,000
4 Interventions

Rt
Business closure

2 School closure
5,000
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Utah Timing
3,000 Started
8.0 4 Eased
Interventions
6.0 3
Emergency decree
2,000
Rt
Business closure
4.0 2
Restaurant closure
School closure
1
2.0
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Vermont Timing
Started
4.0 500
Eased
400 4 Interventions
3.0
Emergency decree
300
Rt
Business closure
200 2 School closure
1.0
100
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18

Virginia Timing
60 Started
Eased

10,000 4 Interventions
40 Emergency decree
Rt
Business closure
Restaurant closure
2
20
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Washington Timing
Started
6,000 4 Eased
40
Interventions
3
Emergency decree
30 4,000
Rt
Business closure
2
Restaurant closure
20
School closure
1
10
Credible intervals
0 50%
0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
West Virginia Timing
Started
6.0 4 Eased
600
Interventions
3
Emergency decree
4.0
400
Rt
Business closure
2
Restaurant closure
School closure
2.0 Stay at home mandate
200 1
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Wisconsin Timing
Started
20.0
6,000 Eased
4 Interventions
15.0
Emergency decree
Rt
Business closure
2 School closure
5.0
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
Wyoming Timing
Started
4.0
3 Eased
900
Interventions
3.0
Emergency decree
600
Rt
Business closure
School closure
1
300
1.0
Credible intervals
0 50%
0.0 0 95%
b
ar
ar
ay
ay
ar
ar
ay
ay
ar
ar
ay
ay
Ap
Ap
Ap
Ap
Ap
Ap
Fe
Fe
M
M
M
6
20
20
20
10
24
23
23
23
4
18
18
18
E Effective number of infectious individuals for all states
The effective number of infectious individuals, c∗ , on a given day is calculated by weighing how infectious a previously
infected individual is on a given day. The fully infectious average includes asymptotic and symptomatic individuals. Esti-
mates of the effective number of infectious individuals for all states can be found in Figure 14.

Alaska Vermont New Hampshire Maine

1,250 25,000 4,000
1,000 20,000
2,000 3,000
750 15,000
2,000
500 1,000 10,000
250 5,000 1,000
0 0 0 0
New York Connecticut Massachusetts
23 eb
20 ar
18 pr
ay
M
A
160,000
M
F
1,500,000
24
120,000
1,000,000 200,000
80,000
500,000 100,000
40,000
0 0 0
Washington Montana North Dakota South Dakota Minnesota Wisconsin Michigan Pennsylvania New Jersey Rhode Island
40,000 1,600 200,000 500,000
30,000 200,000
60,000 40,000
6,000 3,000 400,000
30,000 1,200 150,000 150,000
40,000 20,000 300,000 30,000
4,000 2,000 100,000
20,000 800 100,000
200,000 20,000
2,000 1,000 20,000 10,000
10,000 400 50,000 50,000 100,000 10,000
0 0 0 0 0 0 0 0 0 0
Oregon Idaho Wyoming Nebraska Iowa Illinois Indiana Ohio Virginia District of Columbia Delaware
5,000 5,000 80,000 160,000
50,000 250,000 20,000
4,000 15,000
4,000 4,000 40,000 200,000 60,000 120,000 60,000
10,000 15,000
3,000
Total number of infectious people
3,000 3,000 30,000 150,000 10,000

40,000 80,000 40,000
2,000 10,000
2,000 2,000 5,000 20,000 100,000
20,000 40,000 20,000 5,000
1,000 1,000 1,000 10,000 50,000 5,000
0 0 0 0 0 0 0 0 0 0 0
Nevada Utah Colorado Kansas Missouri Tennessee Kentucky West Virginia North Carolina Maryland
23 eb
20 ar
18 pr
ay
M
A
M
F
50,000 30,000 30,000
24
8,000 40,000
15,000 15,000
12,000 40,000 3,000 75,000
6,000 20,000 30,000
20,000
10,000 30,000 10,000 2,000
8,000 50,000
4,000 20,000
20,000
5,000 10,000 10,000 5,000
4,000 2,000 1,000 10,000 25,000
10,000
0 0 0 0 0 0 0 0 0 0
California Arizona New Mexico Oklahoma Arkansas Mississippi Alabama Georgia South Carolina
23 eb
20 ar
18 pr
ay
M
A
M
F
200,000 25,000 8,000 100,000
24
60,000
50,000
20,000 10,000 30,000 15,000
150,000 6,000 40,000 75,000
40,000 15,000
100,000 4,000 30,000 20,000 50,000 10,000
10,000 5,000 20,000
20,000 10,000 5,000
50,000 5,000 2,000 25,000
10,000
0 0 0 0 0 0 0 0 0
Texas Louisiana Florida
23 eb
20 ar
18 pr
ay
23 eb
20 ar
18 pr
ay
23 eb
20 ar
18 pr
ay
23 eb
20 ar
18 pr
ay
23 eb
20 ar
18 pr
ay
M
A
M
A
M
A
M
A
M
A
M
M
F
F
120,000
24
24
24
24
24
90,000 90,000 75,000
60,000 60,000 50,000
30,000 30,000 25,000
0 0 0
Hawaii
23 eb
20 ar
18 pr
ay
23 eb
20 ar
18 pr
ay
23 eb
20 ar
18 pr
ay
M
A
M
A
M
A
M
M
F
F
2,000
24
24
24
1,500
1,000
500
0
23 eb
20 ar
18 pr
ay
M
A
M
F
24
Figure 14: Estimates for the effective number of infectious individuals over time. The light purple region shows the 95%
credible intervals and the dark purple region shows the 50% credible intervals.
To be more precise, the effective number of infectious individuals of infectious individuals, c∗ , is calculated by first rescal-
gτ
ing the generation distribution by its maximum, i.e. gτ∗ = maxt gt . Based on (1), the number of infectious individuals is
then calculated from the number of previously infected individuals, c, using the following:
t−1
X
c∗t,m = ∗
cτ,m gt−τ ,
τ =0
where ct,m is the number of new infections on day t in state m. A plot of gτ∗ can be found in Figure 15.
1.0
● ●
●
●
0.8
●
infectiousness g*τ
0.6
● ●
●
0.4
●
0.2
●
● ●
●
●
●
● ● ●
0.0
0 5 10 15 20
nubmer of days τ since infection
Figure 15: Infectiousness gτ∗ of an infected individual over time.

F Scenario results for all states
We show here state level scenario plots of an increase of mobility 20% and 40% of current levels.
AK VT NH ME
40 300
100 30 200
200 150
20 100
50 100
10 50
0 0 0 0
1 Apr
M r
1 ay
1 un
l
NY CT MA
Ju
1 a
M
J
400 500
1
2,000
1,500 300 400
200 300
1,000 200
500 100 100
0 0 0
WA MT ND SD MN WI MI PA NJ RI
30 1,250 2,000 800
150 750 1,000 1,500 150
1,000 100 1,500 600
20 100 500 750 1,000 100
500 1,000 400
500 10 50 50 500
250 250 500 200 50
0 0 0 0 0 0 0 0 0 0
OR ID WY NE IA IL IN OH VA DC DE
250 160 400 800 1,500
300 200 600 1,500 2,000 60 150
120 300 600 1,000

150 400 1,500 40 100
200 80 200 1,000 400
100 1,000 500
100 50 40 100 200 500 200 500 20 50
0 0 0 0 0 0 0 0 0 0 0
1 Apr
M r
1 ay
1 un
l
NV UT CO KS MO TN KY WV NC MD
Ju
1 a
M
J
1,000 1,600 800
600
1
500 150 750
400 750 750 1,200 600 200 1,200
400 300 500 100 500 800 500
800 400
200 200 100
100 250 50 250 400 200 400 250
0 0 0 0 0 0 0 0 0 0
1 Apr
M r
1 ay
1 un
l
CA AZ NM OK AR MS AL GA SC
Ju
1 a
M
J
8,000 300 400 500 1,600 500
1
1,500 400 750
6,000 300 300 400 1,200 400
1,000 200 300 300
4,000 200 200 500 800
500 100 200 200
2,000 100 100 100 250 400 100
0 0 0 0 0 0 0 0 0
1 Apr
M r
1 ay
1 un
l
1 Apr
M r
1 ay
1 un
l
1 Apr
M r
1 ay
1 un
l
1 Apr
M r
1 ay
1 un
l
1 Apr
M r
1 ay
1 un
l
TX LA FL
Ju
Ju
Ju
Ju
Ju
1 a
1 a
1 a
1 a
1 a
M
M
J
J
5,000 600 5,000
1
1
4,000 4,000
3,000 400 3,000
2,000 200 2,000
1,000 1,000
0 0 0
1 Apr
M r
1 ay
1 un
l
1 Apr
M r
1 ay
1 un
l
1 Apr
M r
1 ay
1 un
l
HI
Ju
Ju
Ju
1 a
1 a
1 a
M
M
J
J
12.5
1
1
10.0
7.5
5.0
2.5
0.0
1 Apr
M r
1 ay
1 un
l
Ju
1 a
M
J
1
Scenarios Increased mobility 40% Increased mobility 20% Constant mobility
Figure 16: State-level scenario estimates for deaths. The blue ribbon shows the 95% credible intervals (CIs) for scenario
(a) where mobility is kept constant at current levels, the yellow ribbon shows the same CIs for scenario (b) where there
is a 20% return to baseline mobility and scenario (c) where there is a 40% return to baseline.
G Sensitivity analysis to infection fatality ratio
Geographic-specific contact surveys are important for calculating the weighted IFR values according to the methods in [23,
24]. There is no large-scale cross-generational contact survey, similar to the polymod survey [19], implemented in the
USA. Therefore, it was important to understand if the model was robust to changes in the underlying contact survey.
We calculated the IFRs using three different contact matrices: UK, France and Netherlands. We believe that the USA is
culturally closest to that UK out of those countries we had contact matrices for, but also considered France where we
saw the greatest mixing of the elderly and the Netherlands which showed the average behaviour of the European studies
used in [24]. We found that the IFR, calculated for each state using the three contact matrices, lay within the posterior
of IFR in our model (Figure 17). We also noted that our results remained approximately constant when using the IFR
calculated from the three different contact matrices as the mean of the prior IFR in out model, see Section 5.
Since we are using the same contact matrix across all the states, the differences in IFR are due to the population de-
mographics and not due to differential contacts. The low IFR in Texas and Utah reflects the younger population there
whereas the higher IFR in Florida and Maine is due to the older population. This is a limitation of our methods.

Alabama ● ●● ●
Alaska ● ●● ●
Arizona ● ●● ●
Arkansas ● ●● ●
California ● ●
● ●
Colorado ● ●● ●
Connecticut ● ●● ●
Delaware ● ●● ●
District of Columbia ● ●
● ●
Florida ● ●● ●
Georgia ● ●
● ●
Hawaii ● ●● ●
Idaho ● ●● ●
Illinois ● ● ●
Indiana ● ●
● ●
Iowa ● ● ●
Kansas ● ●● ●
Kentucky ● ●● ●
Louisiana ● ●● ●
Maine ● ●● ●
Maryland ● ● ●
Massachusetts ● ●● ●
Michigan ● ●
● ●
Minnesota ● ●
● ●
● France
Mississippi ● ●
● ●
State
Missouri ● ● ● ● Netherlands
Montana ● ●
● ●
● UK
Nebraska ● ● ●
Nevada ● ● ● ● Posterior
New Hampshire ● ●
● ●
New Jersey ● ●● ●
New Mexico ● ● ●
New York ● ●● ●
North Carolina ● ●
● ●
North Dakota ● ●● ●
Ohio ● ●
● ●
Oklahoma ● ●
● ●
Oregon ● ●● ●
Pennsylvania ● ●● ●
Rhode Island ● ●● ●
South Carolina ● ● ●
South Dakota ● ●● ●
Tennessee ● ● ● ●
Texas ● ●
● ●
Utah ● ●
● ●
Vermont ● ●● ●
Virginia ● ●● ●
Washington ● ●
● ●
West Virginia ● ●● ●
Wisconsin ● ●
● ●
Wyoming ● ●● ●
0.006 0.008 0.010 0.012

ifr
Figure 17: Sensitivity analysis for IFR. The red, green and blue dots show the IFR values calculated according to [23, 24]
using the French, Dutch and and UK contact matrices respectively. The purple dot shows the mean of our posterior
estimates for the IFR run using the UK contact matrix estimate and the purple error shows the 95% credible intervals of
the distribution.

2020 05 28 COVID19 Report 23 Version2

Uploaded by

Copyright:

Available Formats

2020 05 28 COVID19 Report 23 Version2

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2020 05 28 COVID19 Report 23 Version2

Uploaded by

Copyright:

Available Formats

28 May 2020 Imperial College COVID-19 Response Team

Report 23: State-level tracking of COVID-19 in the United States

Department of Infectious Disease Epidemiology, Imperial College London

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0

DOI: https://doi.org/10.25561/79231 Page 1

We invite scientific peer reviews here: https://openreview.net/group?id=-Agora/COVID-19

DOI: https://doi.org/10.25561/79231 Page 2

DOI: https://doi.org/10.25561/79231 Page 3

to resurgence, and the plausible magnitude of this resurgence.

2.1 Mobility trends, interventions and effect sizes

DOI: https://doi.org/10.25561/79231 Page 4

2.2 Impact of interventions on reproduction numbers

DOI: https://doi.org/10.25561/79231 Page 5

Great Lakes Great Plains

Southern Appalachia TOLA

Stay at home order eased Stay at home order eased

DOI: https://doi.org/10.25561/79231 Page 6

0% 25% 50% 75% 100%

DOI: https://doi.org/10.25561/79231 Page 7

DOI: https://doi.org/10.25561/79231 Page 8

2.3 Trends in COVID-19 transmission

DOI: https://doi.org/10.25561/79231 Page 9

Daily number of infections

300 Stay at home mandate

25 5,000 Stay at home mandate

20,000 Business closure

DOI: https://doi.org/10.25561/79231 Page 10

2.4 Attack rates

DOI: https://doi.org/10.25561/79231 Page 11

State % of total population State % of total population

DOI: https://doi.org/10.25561/79231 Page 12

Pacific Great Plains Northeast Corridor

TOLA South Atlantic

DOI: https://doi.org/10.25561/79231 Page 13

State Number of infectious State Number of infectious

DOI: https://doi.org/10.25561/79231 Page 14

Constant mobility Increased mobility 20% Increased mobility 40%

DOI: https://doi.org/10.25561/79231 Page 15

DOI: https://doi.org/10.25561/79231 Page 16

and isolation could be an effective intervention.

• Residential:Mobility trends for places of residence.

• Workplaces: Mobility trends for places of work.

DOI: https://doi.org/10.25561/79231 Page 17

5.1 Model specifics

where ψ follows a positive half normal distribution, i.e.

ifr∗m ∼ ifrm ·N (1, 0.1).

DOI: https://doi.org/10.25561/79231 Page 18

π ∼ Gamma(5.1, 0.86) + Gamma(17.8, 0.45).

The prior distribution for R0,m [25] was chosen to be

R0,m ∼ N (3.28, κ) with κ ∼ N + (0, 0.5),

where κ is the same among all states.

DOI: https://doi.org/10.25561/79231 Page 19

The prior distribution for the shared coefficients were chosen to be

αrregion ∼ N (0, γr ), r = 1, . . . , R, with γr ∼ N + (0, 0.5),

DOI: https://doi.org/10.25561/79231 Page 20

DOI: https://doi.org/10.25561/79231 Page 21

[21] Census reporter. Census reporter. 2020. URL: https://censusreporter.org.

[22] VisitData. VisitData Trends. 2020. URL: https://visitdata.org/index.html.