Measuring voluntary social distancing behavior during the COVID-19 pandemic =========================================================================== * Youpei Yan * Amyn A. Malik * Jude Bayham * Eli P. Fenichel * Chandra Couzens * Saad B. Omer ## Abstract Staying home is an important part of the effort to contain COVID-19 and limit deaths. Every state in the United States has enacted policies to encourage distancing and staying home. An important question is how these policies interact with individuals’ voluntary responses to COVID-19 cases and deaths. We find evidence of a non-trivial voluntary response to local and nationally reported COVID-19 cases and deaths. ## Main Text Worldwide, people are staying home to reduce transmission of the SAR-CoV-2, the virus that causes COVID-19. Economic theory has long suggested that people may alter behavior to avoid becoming sick.(*1*, *2*) Yet, many people are under government orders to reduce time in public involuntarily. As governments begin to ease these restrictions, it is important to disentangle the extent that the observed shifts in behavior are voluntary. This information is critical for assessing the epidemiological risk and potential short-run economic benefits of relaxing stay-at-home orders. The United States provides a good case study because all 50 States and the District of Colombia in the United States have issued emergency orders, and 40 (39 States and the District of Colombia) have issued stay-at-home orders. However, stay-at-home orders were not all issued at the same time or in response to the same number of reported cases (Figure 1). ![Fig. 1.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2020/05/06/2020.05.01.20087874/F1.medium.gif) [Fig. 1.](http://medrxiv.org/content/early/2020/05/06/2020.05.01.20087874/F1) Fig. 1. Time distribution of states issuing orders and counties reporting a first COVID19 case. Here, we use the variation in policy responses along with cell phone data to measure the amount of time Americans stayed home and show a strong voluntary avoidance behavior. This behavior is largely crowded out and replaced by state-of-emergency orders, though stay-at-home orders lead to substantially more time spent at home. Americans also respond to local and national information differently. Those living in well-connected metro counties put greater weight on national information than those is remote and less connected non-metro counties. The voluntary response to COVID-19 cases and deaths locally and nationally suggests that relaxing restrictions and relying on greater voluntary action carries some risk, but we should not expect people to revert to pre-epidemic behavioral patterns. However, relying on voluntary behavior depends on clear and accurate information and consistent messaging. During the 2003 SARS epidemic, people in Beijing and Hong Kong took voluntary actions to avoid public places with tourism as well as local and international transport.(*3*, *4*) A survey in England, Scotland and Wales found 38% of the participants practicing recommended behavior with a subset practicing avoidance behavior during the 2009 H1N1 pandemic.(*5*) Empirical evidence from the 2009 H1N1 shows that Americans reduced the amount of time spent in public voluntarily enough to have affected peak prevalence and cumulative H1N1 cases.(*6, 7*) Early evidence from the current pandemic supports the theory that people take defensive action to avoid illness.(*8*, *9*) US states declared a series of orders to keep people home and encourage physical (social) distancing. We explore emergency and stay at home orders, which include shelter-in-place orders. The implementation of the state-of-emergency declaration and the stay-at-home, including shelter-in-place, order vary over time (Figure 1). Only 14 counties reported the first case before February 24. We exploit the variation in policy implementation to identify the impact of policy and infection risk influence the time Americans spent at home. We use SafeGraph smart device median home dwell time data at the census block group level averaged to the county level for all counties in the United States and Washington DC between January 01, 2020, and April 09, 2020. Of the 3,142 counties(*10*), 1,116 were classified as metropolitan counties, which generally have higher shares of urban and suburban communities then non-metropolitan counties. The median time spent at home on January 1 was 740.2 mins (IQR: 674.4, 791.6). There was a gradual decline in time at home from January 1 to late February (minimum of 403.2 mins (IQR: 357.5, 435.6) on February 24). In late February time spent at home increased and reached a median of 735.5 mins (IQR: 659.4, 795.4) on April 9 (Figure 2). Time spent at home peeked on April 4, 829.3 mins (IQR: 743.1, 902.2). There was a dip in the time spent at home the weekend of March 21 and 22, which was the weekend that state and some local governments began asking people to stay at home. The SafeGraph retail visitation data suggest that people congregated in stores like Costco, Walmart, and Target to prepare for restrictive policies. This preparation event may have led to more transmission undermining the very goal of distancing policies.(*11*) ![Fig. 2.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2020/05/06/2020.05.01.20087874/F2.medium.gif) [Fig. 2.](http://medrxiv.org/content/early/2020/05/06/2020.05.01.20087874/F2) Fig. 2. Average of the median time spent at home in a census block group per day in the US (January 1, 2020 and April 9, 2020). People began to decrease their time spent at home prior to government orders (Figure 3). However, time spent at home was at its lowest level and began to rise prior to emergency declarations. Median time at home continued to increase following emergency declaration (Figure 3). ![Fig. 3.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2020/05/06/2020.05.01.20087874/F3.medium.gif) [Fig. 3.](http://medrxiv.org/content/early/2020/05/06/2020.05.01.20087874/F3) Fig. 3. Median time spent at home per day in the US by days from the emergency order by state governments. We regressed the log of time spent at home on the log of local and national case (or death) reports and the presence of distancing policies to disentangle voluntary avoidance behavior from the effect of policies. We merge county-level COVID-19 case and death data from the New York Times (NYT) Daily Reports with the SafeGraph time at home measure and a detailed dataset on policies to examine the role of policy and risk on avoidance behavior. We provide point estimates and lower bound estimates. The lower bound is less dependent on temporal variation of voluntary response. We focus on case reports, because are correlated with death reports (Table S1 and S2). We limit the data to case reports starting on February 25th, three days before the earliest emergency declaration. In the supplemental material we show the results are robust to this choosing an earlier start date. Americans increased time at home in direct response to local and national case reports, to emergency orders, and to stay at home orders. The voluntary and mandatory response have not been additive (Table 1). The magnitude of voluntary response is partially crowded out or offset by policies. Americans more strongly responded to local case reports than national case reports. We estimate that doubling county reported cases leads Americans on average to spend 2.3% (CI: 1.4%, 3.1%) more time at home. However, 72.1% of the increase in time at home was crowded out by the response to the declaration of the state of emergency (Column 2 in Table 1). A lower-bound estimate for the voluntary response to cases reported in one’s own county a 1.7% (CI: 0.8%, 2.6%) increase in voluntary time at home, and that 64.2% of this time is offset by the declaration of the state of emergency (Column 4 in Table 1). View this table: [Table 1.](http://medrxiv.org/content/early/2020/05/06/2020.05.01.20087874/T1) Table 1. Influences of COVID19 policies to case reports and policies (log-log model – February 25 to April 10, 2020). Americans also respond to national case reports with a similar magnitude to county case reports (Table 1). Doubling national cases leads Americans to increase time at home by 4.6% (CI: 4.3%, 5.0%), but 31.8% of the increase is crowed out by the declaration of state-of-emergency (Column 2 in Table 1). The response to national case reports, conditional on the lower bound of the response to county case reports, is 4.9% (CI: 4.6%, 5.1%), with 35.1% offset by the declaration of the state of emergency. Results are generally robust to a variety of model specifications (Table S3-S11). The emergency and stay home orders led Americans, on average, to increase time at home by 12.4% (CI: 9.9%, 14.9%) and 23.6% (CI: 18.4%, 28.8%), but similar to the declaration of state emergency, the stay-at-home order also has a crowding out effect to the response of county case reports (44.2% from Column 1 in Table 1) and national case reports (69.0% from Column 2 in Table 1), which is also robust to alternative model specifications (Table S3). Local context matters, and we examined metropolitan counties and non-metropolitan counties separately. Doubling county reported cases in non-metropolitan counties increases the reported time at home in those counties by 15.5% (CI: 5.5%, 25.5%), which is 2.3 times greater than the 6.6% (CI: 3.3%, 9.9%) increase reported in metropolitan counties (Table S4). These non-metropolitan counties are less responsive to national case counts. Rural counties may be more disconnected from national trends, so people rationally put more weight on local information. Conversely, well connected metropolitan areas are more responsive to national trends. Table 2 shows an accounting of effects for metropolitan and non-metropolitan counties with the corresponding lower bound estimates. View this table: [Table 2.](http://medrxiv.org/content/early/2020/05/06/2020.05.01.20087874/T2) Table 2. Accounting for the effect with an example of representative counties’ responses 20 days after emergency declarations, with 146,153 case reported nationally. Results base on Table S4. Putting our results in context, a representative metropolitan county, 20 days after an emergency order experienced an average of 131 local cases and spent an additional 291.4 (CI: 133.7, 476.9) minutes at home in response. In contrast, a representative non-metropolitan county experienced 7 cases 20 days after an emergency order and increased time at home by 252.9 (CI: 80.0, 465.9) minutes with only 7 cases. It would take 10 additional local county cases reports to exceed the corresponding response in a metropolitan county with 299.36 (CI: 93.11, 561.71) mins. The response to national reports in a non-metro county was half of the corresponding response in a metropolitan county. We find that this voluntary response was largely crowded out by a declaration of emergency. The final net effects are positive for both types of representative counties, but we estimate that 71.5% (for the nonmetropolitan county) and 84.7% (for the metropolitan county) of the time reallocation is induced by the response to county and national cases. Americans may still have adapted and increased time at home in response to the increasing county and national reports of COVID19 cases, even if no orders were issued. We use the regression results to predict the counterfactual time at home in the absence of both orders (Figure 4). Americans in both types of counties exhibited voluntary avoidance behavior in response to the COVID19 pandemic. Voluntary behavior would have led to about a 200-minute increase in time spent at home conditional on the case trajectory, with the order adding about another 150200 minutes. However, if Americans had spent less time at home, then there may have been more cases and more death, which could have led to a stronger avoidance response than the current prediction. Yet, given testing constraints, it is unlikely more cases would have been reported, and there is mounting evidence that the number of cases is many times the reported amount.(*12, 13*) ![Fig. 4.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2020/05/06/2020.05.01.20087874/F4.medium.gif) [Fig. 4.](http://medrxiv.org/content/early/2020/05/06/2020.05.01.20087874/F4) Fig. 4. Predicted and lower bound voluntary and order-enforced staying home behavior. The counterfactual lines are setting both orders and the interaction terms to zero. Our results show that Americans have increased time at home during the COVID-19 epidemic and that a non-trivial share of this response was voluntary. Governments have imposed restrictive policies to encourage distancing during this pandemic. Now some governments are considering relaxing some of these restrictions. It is important to understand the individual actions that people take to protect themselves from risk in order to understand the public health benefit and tradeoffs of such policies. Evidence from a recent poll (e.g., [https://poll.qu.edu/images/polling/us/us04082020_uksb19.pdf](https://poll.qu.edu/images/polling/us/us04082020_uksb19.pdf)) appears to corroborate our results. The voluntary nature of a substantial share of the response is robust to model specification, but ultimately challenging to measure precisely. Emergency orders, school closures, and stay at home orders may increase the salience of case and death data. This increase in voluntary social distancing behavior can be grounded in availability heuristic.(*14*) With increased media coverage of morbidity and mortality associated with COVID-19 along with emergency and stay at home orders, individuals may easily recall the adverse effects of spending too much time in public leading to voluntary social distancing. On the hand, public policies may have encouraged employers to allow greater flexibility to work from home and diminish cultural presentism in the workplace. We did not test this hypothesis explicitly, and it would be difficult to separate such a signaling or salience effect from a voluntary effect. The national-wide trends and preparation behavior suggests there may be spillovers from orders in one state to voluntary behavior in another. Our results cast doubt on the dichotomy between public health and the economy. If people voluntarily spend less time in public, they are choosing to forgo economic activity. While businesses may not have closed as suddenly as they did when forced by policy, it is unlikely that they would have experienced the same level of patronization necessary to sustain their workforce. Future research may quantify these impacts by integrating economic data into the current analysis. Our results are consistent with previous analyses of voluntary distancing during pandemics. Bayham and colleagues found that Americans spent 2.4 additional minutes at home per 1000 cases confirmed by the CDC during the H1N1 pandemic in 2009.(*6*) The current pandemic has appeared to generate significantly more concern as headlines are dominated by COVID-19 news. Our results are also consistent with the 3.4% reduction in city mobility per day controlling for implementation of social distancing policies across 41 cities found by Malik and colleagues in their analysis of Citymapper data during this pandemic.(*8*) These result how that humans respond to risk and adapt their behavior voluntarily during disease outbreaks. Public policies led to greater levels of time spent at home, but have not been completely additive to voluntary efforts. This additional distance may be necessary, since private behavior may not fully internalize the costs of infecting others or congesting hospitals. Still, the public measures appear to be crowding out a substantial amount of the voluntary effort. Our results suggest that relaxing policies will not result in people returning to pre-epidemic behavior as long as people perceive a risk of infection. However, behavioral change as cases decline may be different than as they rose, given the diminishing response to cases and deaths. Nevertheless, testing must produce clear, accurate, and locally relevant information for people to make decisions about behavior. The importance of local information is highlighted by the information that residents of metro and non-metro counties are using to make decisions. Non-metro residents are placing relatively greater weight on local information, whereas urban residents are putting greater weight on national information. Recent efforts have attempted to paint this as a political divide, which may correlate with the metro and non-metro divide. Yet, the weighting of information differently may be a rational response to what is the most relevant information as function of general connectivity. The implication, however, is that in order to rely on a voluntary response it may be important to ensure testing, and perhaps oversample, rural counties in order to provide locally relevant information. ## Data Availability Data used for this analysis are publicly available from Safe Graph, New York Times, and COVID-19 US state policy database maintained by colleagues at Boston University. ## Author contributions YY and AAM conducted the analysis; JB and EPF designed research in consultation with SOB, YY, AAM; YY, AAM and EPF lead writing; all contributed to writing. ## Competing interests Authors declare no competing interests. ## Data and materials availability All data are publicly available with links in the main text or the supplementary materials. ## Materials and Methods Data on emergency declarations and stay at home orders, which includes shelter in place orders, came from Julia Raifman and colleagues at Boston University.(*15*) Reported case and death data came from the New York Times.(*16*) All orders took place between January 1, 2020 and April 9, 2020. Counties were classified as metropolitan or non-metropolitan following the US Department of Agriculture Economic Research Service classification. Data on time spent at home is based on anonymized and aggregated mobile device location data from SafeGraph.(*17*) We used two distinct but related products to quantify behavior during the epidemic. First, we used median home dwell time reported at the Census block group on each day. Dwell time is the time that a device is present within a home. We aggregated to the county and normalize visitation by the devices present in the home summarized at the county level. The normalization was required to compare estimates over time as the panel of devices changes over time. Finally, we construct a set of county-level weather controls by aggregating 4km gridded estimates of maximum and minimum daily temperature, maximum and minimum relative humidity, precipitation amount, surface sun radiation, and wind speed. Observation are county *i* on day *t*. We regress time spent at home measured as county means of median Census Block Group home dwell time in minutes, *Yit*, on reported cases (and deaths) in one’s own county *Cit*, reported cases (and deaths) nationally *Nt*. We consider the influence of policy signals and include level effects for whether a county is under a state of emergency order *Eit* or a stay at home order, *Sit*. We condition the regression on county-level weather, *Xit*, day of week fixed effects *wt* and country-specific fixed effects *aj*. Finally, cases and deaths grow exponentially during an epidemic. Therefore, we work in log minutes and log cases and deaths. We cluster standard errors at the state level to account for state-level serial correlation and heteroscedasticity caused by the phase in orders. The general specification is ![Formula][1] We interact government orders with county and national case or death reports in our primary specification to determine whether the orders complement or crowd out private behavior. We also add days of orders issued, *Edit* and *Sdit*, to explore if order effects may change over time. ![Formula][2] In an epidemic, connectivity is important. We examined how well-connected metro counties and relatively remote non-metropolitan counties responded differently to different information by splitting the sample along by metropolitan and non-metropolitan classification. There are two potential concerns with our empirical approach. First, time spent out of the home may lead to cases leading to reverse causality. This has been a concern in research in response other infectious agents.(*18*) However, in daily resolution of the data separates how actions taken influence cases by over about weeks, which is sufficient to avoid the reverse causality problem. The second and more important concern with our empirical approach is that county cases, or deaths are correlated with when governments enact orders. To address this, we estimate a lower bound of the effect of county cases by including day fixed effects. National cases are perfectly correlated with the day fixed effects so we do not include national case reports at this stage. We leverage the variation across counties within a day to conservatively estimate the effect of county cases on time spent at home. We argue that this is a conservative estimate because the daily fixed effects likely capture some of the behavioral response to changing risk as more cases are reported. We use Manski’s partial-identification strategy to estimate the response to national case reports and policies conditional on the lower bound effect of county case (or death) reports.(*19*) Since the response to county case reports is the main voluntary behavioral effect, this gives us a conservative measure of the role voluntary behavior. We, then impose the lower bound estimate for the behavioral response to county case reports and re-estimate the main specification, conditional on the imposed response to county case reports. In the supplemental material we perform a series of robustness checks. We consider the response to reported deaths (Table S1), a level-log and interaction specifications (Table S2, S3, S5, S6, S7, S8), extending the data series back to January 1st (Table S1, S2, S8, S9), and lagging cases by one day (Table S9, S10, S11). Deaths may be a stronger and clear signal, but they are also delayed. The main results are robust to focusing on the signal from deaths. The second robustness check shows that the results are robust to functional form assumptions. The third test shows the results are robust to our truncated data. The reason for the truncated data is that the earlier data should not be helping identify the coefficients as behavior varied for reasons unrelated to policy. We are using the data from the New York Times, so it is likely that the people are receiving and acting on the information a day later. Our results are robust to this concern. ## Supplementary Tables View this table: [Table S1.](http://medrxiv.org/content/early/2020/05/06/2020.05.01.20087874/T3) Table S1. Impacts of COVID19 case and death reports and COVID19 orders using a log-log specification. View this table: [Table S2.](http://medrxiv.org/content/early/2020/05/06/2020.05.01.20087874/T4) Table S2. Impacts of COVID19 case and death reports and COVID19 orders using a level-log specification View this table: [Table S3.](http://medrxiv.org/content/early/2020/05/06/2020.05.01.20087874/T5) Table S3. Influences of COVID19 policies to the endogenous response behavior from the case reports using a level-log specification (February 25 to April 10, 2020). View this table: [Table S4.](http://medrxiv.org/content/early/2020/05/06/2020.05.01.20087874/T6) Table S4. Influences of COVID19 orders issued days to the heterogeneous voluntary response from the county case reports using a log-log specification (February 25 to April 10, 2020). View this table: [Table S5.](http://medrxiv.org/content/early/2020/05/06/2020.05.01.20087874/T7) Table S5. Influences of COVID19 orders issued days to the heterogeneous voluntary response from the county and national case reports using a log-log specification (February 25 to April 10, 2020). View this table: [Table S6.](http://medrxiv.org/content/early/2020/05/06/2020.05.01.20087874/T8) Table S6. Influences of COVID19 orders issued days to the heterogeneous voluntary response from the county case reports using a level-log specification (February 25 to April 10, 2020). View this table: [Table S7.](http://medrxiv.org/content/early/2020/05/06/2020.05.01.20087874/T9) Table S7. Influences of COVID19 orders issued days to the heterogeneous voluntary response from the county and national case reports using a level-log specification (February 25 to April 10, 2020). View this table: [Table S8.](http://medrxiv.org/content/early/2020/05/06/2020.05.01.20087874/T10) Table S8. Influences of COVID19 orders issued days to the heterogeneous voluntary response from the county case reports (January 1st to April 10, 2020). View this table: [Table S9.](http://medrxiv.org/content/early/2020/05/06/2020.05.01.20087874/T11) Table S9. Impacts of COVID19 lagged case and death reports and COVID19 orders using a log-log specification (January 1 to April 10, 2020). View this table: [Table S10.](http://medrxiv.org/content/early/2020/05/06/2020.05.01.20087874/T12) Table S10. Influences of COVID19 policies to the endogenous response behavior from the lagged case reports using a log-log specification (February 25 to April 10, 2020). View this table: [Table S11.](http://medrxiv.org/content/early/2020/05/06/2020.05.01.20087874/T13) Table S11. Influences of COVID19 orders issued days to the heterogeneous voluntary response from the lagged county case reports using a log-log specification (February 25 to April 10, 2020). ## Acknowledgement: Funding YY and EPF are supported by the Knobloch Family Foundation and AAM is supported by Yale Institute for Global Health. ## Footnotes * **One Sentence Summary:** People respond to COVID-19 risk and to state orders, though the response to orders crowds out some of the voluntary response. * Received May 1, 2020. * Revision received May 1, 2020. * Accepted May 6, 2020. * © 2020, Posted by Cold Spring Harbor Laboratory This pre-print is available under a Creative Commons License (Attribution-NonCommercial 4.0 International), CC BY-NC 4.0, as described at [http://creativecommons.org/licenses/by-nc/4.0/](http://creativecommons.org/licenses/by-nc/4.0/) ## References and Notes 1. 1. E. P. Fenichel, Economic considerations for social distancing and behavioral based policies during an epidemic. J Health Econ 32, 440–451 (2013). [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.jhealeco.2013.01.002&link_type=DOI) [PubMed](http://medrxiv.org/lookup/external-ref?access_num=23419635&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2020%2F05%2F06%2F2020.05.01.20087874.atom) 2. 2. P. J. Francis, Optimal tax/subsidy combinations for the flu season. Journal of Economic Dynamics and Control 28, 2037–2054 (2004). [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/J.Jedc.2003.08.001&link_type=DOI) 3. 3. P. Beutels et al., The economic impact of SARS in Beijing, China. Trop Med Int Health 14 Suppl 1, 85–91 (2009). [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1111/j.1365-3156.2008.02210.x&link_type=DOI) [PubMed](http://medrxiv.org/lookup/external-ref?access_num=19508435&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2020%2F05%2F06%2F2020.05.01.20087874.atom) 4. 4. D. P. Durham, E. A. Casman, Incorporating individual health-protective decisions into disease transmission models: a mathematical framework. J R Soc Interface 9, 562–570 (2012). [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1098/rsif.2011.0325&link_type=DOI) [PubMed](http://medrxiv.org/lookup/external-ref?access_num=21775324&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2020%2F05%2F06%2F2020.05.01.20087874.atom) 5. 5. G. J. Rubin, R. Amlot, L. Page, S. Wessely, Public perceptions, anxiety, and behaviour change in relation to the swine flu outbreak: cross sectional telephone survey. BMJ 339, b2651 (2009). [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6MzoiYm1qIjtzOjU6InJlc2lkIjtzOjE3OiIzMzkvanVsMDJfMy9iMjY1MSI7czo0OiJhdG9tIjtzOjUwOiIvbWVkcnhpdi9lYXJseS8yMDIwLzA1LzA2LzIwMjAuMDUuMDEuMjAwODc4NzQuYXRvbSI7fXM6ODoiZnJhZ21lbnQiO3M6MDoiIjt9) 6. 6. J. Bayham, N. V. Kuminoff, Q. Gunn, E. P. Fenichel, Measured voluntary avoidance behaviour during the 2009 A/H1N1 epidemic. Proc Biol Sci 282, 20150814 (2015). [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1098/rspb.2015.0814&link_type=DOI) [PubMed](http://medrxiv.org/lookup/external-ref?access_num=26511046&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2020%2F05%2F06%2F2020.05.01.20087874.atom) 7. 7. E. P. Fenichel, N. V. Kuminoff, G. Chowell, Skip the trip: air travelers’ behavioral responses to pandemic influenza. PLoS One 8, e58249 (2013). [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1371/journal.pone.0058249&link_type=DOI) [PubMed](http://medrxiv.org/lookup/external-ref?access_num=23526970&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2020%2F05%2F06%2F2020.05.01.20087874.atom) 8. 8. A. A. Malik, C. Couzens, S. B. Omer, COVID-19 related social distancing measures and reduction in city mobility. *medRxiv*, (2020). 9. 9. S. B. Villas-Boas, J. Sears, M. Villas-Boas, V. Villas-Boas, “Are We #StayingHome to Flatten the Curve?” UC Berkeley: Department of Agricultural and Resource Economics, (2020). 10. 10. B. E. Sawe, “How Many Counties Are in the United States?” WorldAtlas, Ed. (2018). Available from: [www.worldatlas.com/articles/how-many-counties-are-in-the-united-states.html](http://www.worldatlas.com/articles/how-many-counties-are-in-the-united-states.html). 11. 11.SafeGraph, “Free Business Location & POI Data In ArcGIS” (SafeGraph Inc, San Francisco, CA, 2020). Available from: [https://www.safegraph.com/geospatial](https://www.safegraph.com/geospatial). 12. 12. E. Bendavid et al., COVID-19 Antibody Seroprevalence in Santa Clara County, California. *medRxiv*, (2020). 13. 13. C. F. Manski, F. Molinari, “Estimating the COVID-19 Infection Rate: Anatomy of an Inference Problem,” (National Bureau of Economic Research, arXiv:2004.06178, 2020). 14. 14. A. Esgate, D. Groome, An introduction to applied cognitive psychology. (Psychology Press, 2005). 15. 15.“COVID-19 US state policy database” (2020). Available from: [https://docs.google.com/spreadsheets/d/1zu9qEWI8PsOI\_i8nI_S29HDGHlIp2lfVMsGxpQ5tvAQ/edit#gid=0](https://docs.google.com/spreadsheets/d/1zu9qEWI8PsOI_i8nI_S29HDGHlIp2lfVMsGxpQ5tvAQ/edit#gid=0) 16. 16.The New York Times, “We’re Sharing Coronavirus Case Data for Every U.S. County” (2020). Available from: [https://raw.githubusercontent.com/nytimes/covid-19-data/master/us-counties.csv](https://raw.githubusercontent.com/nytimes/covid-19-data/master/us-counties.csv). 17. 17.SafeGraph, “Social distancing measures” (SafeGraph Inc, San Francisco, CA, 2020). Available from: [https://docs.safegraph.com/docs/social-distancing-metrics](https://docs.safegraph.com/docs/social-distancing-metrics). 18. 18. K. Berry, J. Bayham, S. R. Meyer, E. P. Fenichel, The allocation of time and risk of Lyme: A case of ecosystem service income and substitution effects. Environ Resour Econ (Dordr) 70, 631–650 (2018). 19. 19. C. F. Manski, Identification for prediction and decision. (Harvard University Press, 2009). [1]: /embed/graphic-7.gif [2]: /embed/graphic-8.gif