Understanding the Impact of COVID-19 Policy Change
in the Greater Seattle Area using Mobility Data
Roy Burstein​1​, Hao Hu​2​, Niket Thakkar​1​, Andrew Schroeder​3​, Mike Famulare​1​, and Daniel Klein​1
1​
Institute for Disease Modeling, ​2​Bill & Melinda Gates Foundation, ​3​Direct Relief

Results as of March 29, 2020 8 p.m.; ​covid@idmod.org

This report focuses on mobility only. We have prepared a ​follow-up study​ exploring the relationship
between mobility and COVID-19 transmission in King County, WA. Results of this study should be
interpreted with respect to the stated assumptions and limitations.

Executive Summary 
To stem the spread of COVID-19, Washington State instituted increasing levels of separation policies
between March 11 and 24, 2020, including closing schools and prohibiting large groups. These policy
changes were supported by ​computational modeling​, which explored social distancing at 25%, 50%, and
75% reduction in person-to-person contact. Due to the long period between first exposure to COVID-19
and the onset of symptoms (~5 days) , delays to seeking a test (1+ days), and time required to process
tests (1+ days), ​it takes about a week for any policy changes to have an impact on case counts,​ and
even longer to see an impact on COVID-19 mortality.
Real-time data give us insight into trends before they appear in epidemiological data. Aggregated data
from ​Facebook Data for Good​ allow us to measure coarse changes in where people spend their time and
how much they move around the greater Seattle area; data are aggregated in space and time, and
individual users cannot be identified. ​Since March 2, we’ve observed a persistent decline in mobility​,
with movement from other regions into Seattle and Eastside dropping by about 90% as of March 28
relative to baseline movement as observed in January and February. Also, we see strong evidence that
people have been staying at home, as daytime population occupancy counts in residentially zoned areas
of Seattle have increased by 27% on average compared to a baseline (pre-social distancing) period, while
the daytime population occupancy in mixed-use areas (primarily comprising downtown Seattle)
decreased 43% on average.
While Facebook mobility data gives us insight into changes in mobility patterns, it is not a direct measure
of social distancing, and it cannot tell us about COVID-19 transmission on its own. That being said, the
data is encouraging, particularly since similar interruptions in mobility have been shown to reduce
disease transmission in our area, as we saw during ​'snowpocalypse’​ last year. In a ​follow-up to this
study​, we use a transmission model-based approach to connect the mobility data to daily COVID-19 test
results provided by the Washington Department of Health to more definitively address this issue. For
the purposes of this report we have not yet linked mobility data to epidemiological trends, ​though while
the data are encouraging, we also see some evidence of fluctuating adherence to social distancing
policies, emphasizing that we remain in a precarious position​. For social distancing to be effective in
restricting COVID-19 transmission, ​widespread adherence over the coming weeks is critical​.

1

 Background
The novel coronavirus SARS-CoV-2 emerged in Wuhan, China, in ​late November or early December 2019​.
As of March 26, 2020, it is responsible for 462,684 confirmed cases and 20,834 deaths of the disease
COVID-19 (​WHO​). After initial emergence in China, travel associated cases started to appear in other
parts of the world with strong travel connections to Wuhan (​http://rocs.hu-berlin.de/corona/​). The first
confirmed case in the US was a travel-associated case in Snohomish County, WA, screened on January
19, 2020. In the six weeks following late February, a ​second presumptive case​ was identified roughly 10
miles away from where the first case was treated. As of the​ ​evening of March 19​, ​Washington State
reports 1,376 confirmed cases and 74 confirmed deaths associated with COVID-19​ with the majority
from King and Snohomish counties. 
Washington State has instituted several policies to slow the spread of the virus. On March 4, Public
Health Seattle & King County (PHSKC) ​announced​ recommendations that workplaces should enact
measures to allow people who can work from home to do so. Employers such as Microsoft, Amazon,
Google, Facebook, and others subsequently ​announced​ work from home policy starting March 5. Then,
on March 11, Governor Jay Inslee ​banned​ gatherings of more than 250 people in King, Pierce and
Snohomish counties, including weddings. The following day, all schools in King, Pierce and Snohomish
counties were ​closed​ through April 24, and on March 13, school closures were ​extended​ to all
Washington schools. Two days later, Governor Inslee ​announced​ closures of gyms, entertainment and
recreational facilities, as well as bars and restaurants (take-out/delivery remain available), and
gatherings of more than 50 people were prohibited. Finally, on March 23rd, Governor Inslee ​announced
the “Stay home, stay healthy” policy - a two week shelter-in-place order for all but essential services.
These policy measures were supported by COVID-19-specific ​computational modeling​ of the potential
impact of social distancing policies. The computer simulations compared “business as usual” to three
levels of social distancing--25%, 50%, and 75%. A key finding was that only the 75% level resulted in
reductions in the number of active cases over time.

Findings from Facebook Disaster Maps Data
We used data from the ​Facebook Data for Good Project - Disaster Maps​ to track changes in population
and mobility between regions over time. ​These data are collected from mobile users with location
services enabled and are aggregated to coarse geographic levels as anonymous counts of users;
individual users cannot be identified.​ On a typical day, data are captured from around 230,000 people
across the Puget Sound region. Using these data, we make comparisons over a time period between
February 26 through March 28, and a baseline period 45 to 90 days before that (i.e., for movement
measures: mid-January to end-February, and for population occupancy measures: most of December
through February). Unless otherwise noted, all results are presented as a percent change in either
mobility or number of Facebook users at a location, relative to the same time, day of week, and location
during the entire baseline period. ​We emphasize we are not measuring reductions in social contact,
but rather changes in mobility and the places where people are spending their days.​ Though as
populations spend more time in residential areas and away from shared public and work spaces, it is

2

 likely they are coming in contact with fewer people. These datasets are described in detail ​here​, and a
broader discussion of the utility of these data for COVID-19 research ​here​.
Data confirms that mobility patterns have shifted tremendously in the greater Seattle area over the past
four weeks. We observe reduced overall movement across the city and region coupled with increasing
population concentrating in residential areas and away from commercial areas. These trends began
around March 2, the Monday after the first COVID-19 death was reported in King County, and have
steadily increased over time, during a period of increased policy intervention as well as COVID-19
awareness.
Movement during daytime (9 a.m. to 5 p.m.) from other regions to Seattle and to the Eastside had
declined from about 40% below baseline on March 5 after Public Health Seattle and King County
recommended a work from home policy, to over 60% on March 12 when Gov. Inslee banned gatherings
of over 250 people and closed K-12 schools through April 27th, to about 90% on March 28, the last day
for which we have data.
Decreases in movement into Seattle at nighttime (1 a.m. to 9 a.m.) have been less pronounced and had
taken longer to achieve until March 16, when Gov. Inslee announced the closure of all entertainment
and recreational facilities, including gyms, bars, and restaurants. As of March 28, the decrease is close to
90%.

Figure 1.​ Percent change in mobility from other regions into Seattle (left) and the Eastside (right). Top and bottom
panels of each figure represent daytime and nighttime trends. The dashed horizontal line is baseline level, and
negative percent changes indicate reductions in mobility.

3

 Table 1​. Cumulative changes in mobility relative to baseline. Positive numbers indicate reductions.
Date

Event

Day into
Seattle

Day into
Eastside

Night
into
Seattle

Night
into
Eastside

29-Feb

First announced death

-0.4%

6%

8%

-0.6%

05-Mar

On 04-Mar PHSKC recommended workplaces should
enact measures to allow people who can work from
home to do so. Multiple employers announced work
from home policy.

38%

41%

13%

28%

12-Mar

Gov Inslee banned gatherings of more than 250 people
(11-Mar) and ordered closure of all schools in King,
Pierce and Snohomish counties (12-Mar)

62%

63%

-3%

34%

16-Mar

Gov. Inslee announced on the evening of 15-Mar that
all entertainment and recreational facilities, including
gyms as well as bars and restaurants will close;
however, restaurants will be allowed to do take out
and delivery.

66%

68%

-8%

31%

28-Mar

Stay home, stay healthy in place since 23-Mar. Last day
of available data.

91%

89%

82%

90%

In addition to declines in mobility around the region, Figure 2 (below) shows that many more people are
staying away from commercial areas and remaining in residential areas of Seattle. By March 24, we
observed an average 28% and 26% increase in daytime population occupancy in multi-family and
single-family zoned areas of Seattle over the baseline period. The same day, daytime occupancy at the
University of Washington Seattle campus had declined 55%, mixed-use zoned areas (primarily
comprising downtown) declined 42% percent, and commercially zoned areas (primarily SoDo, Harbor
Island, and Fishermen's Terminal) declined by 31%. During weekends, the trends revert to some extent
toward parity with baseline, as, for example, more people typically stay home during the weekend.

4

 Figure 2.​ Lines show the average change in daytime population occupancy relative to the baseline period (January
and February) across zoned areas of Seattle. Reduction in commercial, mixed-use, and UW areas are in stark
contrast to increases in single- and multi-family zones. Grey vertical bars indicate weekends. Facebook population
tile layers were overlaid with a ​city parcel map​ containing zoning information on over 178 thousand city parcels.
Only tiles for which all contained parcels were of the same zone were kept, these remaining tiles are mapped in the
legend of the plot.

These trends extend beyond Seattle and the Eastside. The maps in Figure 3 below show the increasing
concentration of people into residential areas across the Puget Sound region. These patterns become
increasingly evident over time, both in terms of intensity and consistency across the entire region.

5

 Figure 3.​ Changes in population occupancy in the greater Seattle region from February 26 to March 18 relative to
the 90 day period before Feb 26. Blue indicates increases, whereas red indicates decreases. The figure shows
people are spending less time in urban areas, such as downtown Seattle and Tacoma, and the Microsoft campus.

Additionally, we can observe change in population movement during the course of a day. By comparing
population occupancy at night to the same location during the day, we captured the extent to which
people tend to stay put over the course of a day, without relying on a baseline period for comparison.
Figure 4 (below) again indicates a general reduction in mobility, in concordance with the other
measures. We saw a brief increase in mobility during the weekend of March 21 and 22, when there was
a ​noted​ increase in outdoor gatherings corresponding to nice weather. The following Monday, Governor
Jay Inslee announced the “Stay home, stay healthy” policy. The following weekend there was another
increase in mobility the following Saturday, March 28th. It is difficult to determine from these data if this
increased mobility activity corresponds to non-compliance with the “Stay home, stay healthy” policy. For
example, this could be capturing, in part, increased weekend grocery shopping or outdoor exercise. As
more epidemiological data are available, we will be able to better understand what impact these
changes have on transmission and what measures best capture transmission-relevant social distancing
behavior.

6

 Figure 4:​ Relative fluctuations in daytime to nighttime population over time in King County. We notice an inversion in weekend
mobility behavior. During the social distancing period (after March 2nd) there is more day-night fluctuation on weekends than
during the work week, compared to the ‘business as usual’ period before that, where more people tended to stay home on the
weekends. Now as more people are working from home, the more mobility occurs during weekends. Day-night fluctuations were
measured as the absolute value of the relative difference between day and night population in each tile. The black line indicates
the average relative day-night movement across populated tiles in King County. Tiles with Facebook population occupancy <10
are masked from researchers due to privacy concerns.

Data derived from social media users is not without its limitations. For example, there is some concern
that the population of Facebook users does not represent the general population. To explore this, we
compared Facebook estimates from Eastside to Seattle commutes with ​WS-DOT​ toll data from SR 520
measuring reduced volume relative to the same day of the week in February. There was very high
agreement between these two measures with R​2​ = 0.94, see Figure 5.

7

 Figure 5: V
​ erifying Facebook mobility data with other mobility signals. We compared a measure of mobility from Eastside to
Seattle with traffic volume data from the SR-520 toll. The reduction of mobility detected in Facebook data is highly correlated
with the traffic reduction detected through tolls, the reduction in Facebook mobility data (in number of people) is higher than
WS-DOT tolls (in number of cars).

Epidemiological Implications of Mobility Findings
Facebook data indicates mobility in the greater Seattle area has reduced significantly from March 2 to
today, showing dramatic change in response to COVID-19. Still, ​mobility increases over the past three
weekends highlight a potential difficulty in maintaining widespread adherence to social-distancing
policies​. In a ​follow-up report​, we model diagnosis data, allowing us to start to understand how mobility
decreases can be connected to decreases in COVID-19 transmission. However, the link between mobility
and disease transmission is not one-to-one; even with zero mobility, we know some transmission will
continue within homes and other tight personal circles.

Conclusions to Date 
Washington State has rapidly instituted policy measures to slow COVID-19. Modeling indicates generic
“social distancing” can overcome the spread of COVID-19, but only if high levels (i.e., 75% reduction in
person-to-person contact) are achieved. Delays in how COVID-19 spreads and further delays in testing
render us unable to measure the impact of policy on new cases for a few more days, and the impact on
deaths for another few weeks after policy implementation.
Based on the aggregated Facebook data, we can say with confidence ​mobility behavior in the greater
Seattle area has reduced dramatically since early March​, when it first became clear community
transmission of COVID-19 was occuring in the area. ​Significantly fewer people are commuting to work,
and many more people are spending their days at home. ​However, we see evidence of inconsistency,
particularly during weekends. Mobility is not a direct measure of social distancing; while these data
indicate people are spending much more time at home and away from shared spaces (which has been
shown to reduce disease transmission​ before), we cannot use it to determine the percent reduction in
social contact. See the ​follow-up report​ to learn how these changes in mobility relate to reductions in
COVID-19 transmission.

8