When local governments’ stay-at-home orders meet the white house’s “opening up america again”

In the current polarized political climate, citizens frequently face conflicting directives from their local and federal government officials. For instance, on April 16th, 2020, The White House launched the “Opening up America Again” (OuAA) campaign while many U.S. counties had stay-at-home orders. We created a panel data set of U.S. counties to study the impact of U.S. counties’ stay-at-home orders on community mobility before and after The White House’s campaign to reopen the country. Our results suggest that before the OuAA campaign, stay-at-home orders substantially decreased the time spent in retail and recreation businesses. However, after the launch of the OuAA campaign, the time spent at retail and recreational businesses in a typical conservative county increased significantly more than in liberal counties (23% increase in a typical conservative county vs. 9% increase in a typical liberal county). We also found that in conservative counties with stay-at-home orders, time spent at retail and recreational businesses increased less than in those without stay-at-home orders. These findings illuminate that when federal and local government policies are at odds, residents decide which policies to adhere to based on the alignment between their political ideology and the government body. Our findings highlight the substantial importance of each government body in forming citizens’ behaviors, offering practical implications for policy makers during natural disasters.


Introduction:
The COVID-19 pandemic has been raging in the world since December 2019.The United States reported its first COVID-19 patient in January 21 st , 2020.1 As of August 12, 2020 there are over 5 million confirmed cases in U.S. with a death tally of over 165,000. 2 The immense impact of COVID-19 pandemic on the lives of billions of people has forced authorities to devise response strategies to contain the damage.From enforcing stay-at-home orders to restrict public and private gatherings to wearing masks and social distancing, the authorities have come up with a variety of response strategies.These restrictive strategies are effective only if they are adhered to by citizens.
Given the diverse set of opinions held by citizens, their level of adherence to the restrictions could be different.Results from initial research studies provide some evidence of these differences based on residents' political beliefs.For instance, Painter and Qiu (2020) found that residents' political beliefs affects their compliance with social distancing orders that are imposed as a response to the spread of COVID-19.Their findings reveal that counties that voted for President Trump in 2016 presidential election are more likely to disobey the social distancing orders.Another study by Grossman et al. (2020) also revealed that the Democratic counties were more likely to obey stay-athome orders enforced by the state governor.They also found that Democratic counties with Republican governors are more likely to stay at home when compared to other counties.Overall, the current research signals the impact of stay-at-home orders on residents' stay-at-home behaviors.
Although both studies offer similar findings, they both use data from SafeGraph's shelter-in-place data. 3This data set uses a sample of users' cell phone locations to determine the time stayed at home.However, we believe that another data set published by Google would be more informative as it not only includes data about stay at home time, but also it includes data about the amount of time users spend in retail & recreation locations (such as bars, restaurants, and gyms), grocery stores, parks, places of work, and transit locations.We believe that such data would be more informative in measuring the impact of stay-at-home orders.
Another distinction between our research and prior studies is about the federal government's intervention: "[w]e are starting our life AGAIN!,"4 said the U.S. president during his Coronavirus Task Force press briefing on April 16 th .A day earlier, during another press briefing, the president claimed that the U.S. has "passed the pick on new cases."5Yet a couple of days before that, on April 13 th , he also claimed "total authority"6 over governors.The three Coronavirus Task Force press briefings on April 13 th , 15 th , and 16 th mark a major shift in U.S. president's response policy regarding the pandemic.A new policy that was also cascaded in president's tweets (see Figure 1).What followed this new policy was a set of guidelines7 to open up America.The White House launched "Opening up America Again" (OuAA) website on Arpil 16 th .Along with these guidelines, president Trump used his powerful Twitter account to encourage protestors (mainly composed of his supporters) to "liberate" Michigan and Minnesota, two states with Democratic governors who imposed strict social distancing restrictions (Shear and Mervosh 2020).4-Conservative counties that had a stay-at-home order in place spent more time at home and less at retail & recreation places even after OuAA campaign launch when compared to conservative counties that did not have stay-at-home order in place.
Our results are based on a quasi-experimental setting and we have controlled for the number of cases and deaths per 100k population, county fixed effects, time fixed effects and the interaction of state and time fixed effects.We also examined the robustness of our findings by running the models using a matched sample of counties.We used county level data about residents' education, population, deaths, births, domestic and international migration, percent below federal poverty line, unemployment rate, and median household income.In this brief summary, we report our study design, data analysis, and preliminary findings.

Data:
We created a panel data set of U.S. counties by collecting data about community mobility scores, COVID-19 new cases and deaths (adjusted by population), ideological orientation of counties, and state's COVID-19 response data (stay-at-home, shelter-in-place, and other types of restrictions and their timelines) for the period between the first week of March 2020 and the first week of June 2020.
From 3,141 U.S. counties, we removed any county that did not have a value for the variables in our models.For instance, many counties did not have a value for the community mobility indices (Retail and Residential described in table 1).After removing the counties with missing values, we ended up with panel data of 1,563 counties and 14 weeks.Below, we describe the sources of data and our data collection approach:

1-Community Mobility Data:
This data set was obtained from Google.8 Google's community mobility dataset "shows how visits to places, such as grocery stores and parks, are changing in each geographic region."9According to the documentations, this dataset shows how visits and length of stay at different places change compared to a baseline.The baseline is the median value, for the corresponding day of the week, during the 5-week period Jan 3-Feb 6, 2020.Google indicates that the data is included in the calculations based on user settings, connectivity, and whether there is any privacy concern (due to small sample size in some areas).If there are any concerns regarding the privacy of Google's users, the data fields will be left empty.Due to these omitted values, we did not include counties with missing values.
Google's community mobility data set includes six categories of places: grocery & pharmacy, parks, transit stations, retail & recreation, residential, and workplaces.From these six categories, we used retail & recreation and residential categories.The reason why we limited our study to these two categories is that these two categories portray a more accurate picture of community mobility during the pandemic.Retail & recreation category includes mobility trends for places like restaurants, cafes, shopping centers, theme parks, museums, libraries, and movie theatres.These businesses are nonessential businesses that could be avoided during the pandemic.If stay-at-home orders are effective, we would expect a decrease in retail & recreation trend.Residential category refers to the mobility trends of places of residence.If stay-at-home orders are effective, we would expect an upward trend in residential category.Other categories such as grocery & pharmacy stores, workplaces, and transit stations are either essential or determined by the employers rather than the residents.Therefore, trends in grocery & pharmacy stores, workplaces, and transit stations may not provide us with a reasonable pattern about residents' will in adhering to stay-at-home orders and other types of restrictions.We also excluded parks mobility trends because visiting such places would be possible with very limited risk of infection (people could stay six feet apart in the open areas).With that said, we created two dependent variables Retail & Residential based on Google's mobility trends for retail & recreations and residential places respectively.

2-COVID-19 Data:
This data set was obtained from NY Times' GitHub page.10This dataset includes the number of cases and the number of deaths per day per county per 100k residents.

3-Ideological Orientation Data:
This data set was obtained from American Ideology Project.11We used the 2016 release of "County-Level Preference Estimates".From this data file, we used variable mrp_mean, which is the estimate of the mean ideology of the county.This measure ranges from a negative number to a positive number.
The smallest value in mrp_mean represents the most liberal county, while the largest value in mrp_mean is associated with the most conservative county.Therefore, we can interpret mrp_mean as a metric for gauging the level of conservativeness of a county.Hence, in our study, we call this variable conservative.In the 2016 release of the data, conservative ranges from -1.098 (the least conservative county) to 0.842 (the most conservative county).The methodology for estimating ideological orientation scores is described in Tausanovitch and Warshaw (2013).

4-County-level Restrictions Data:
This data set was obtained from a GitHub repository. 12The data summaries and the methods used for assembling the data sets are detailed in Killeen et al. (2020).There are multiple data files in the data repository, from which we used "interventions.csv"data file.This data file contains the dates that counties (or states governing them) enforced policies (such stay-at-home orders) to mitigate the spread of COVID-19 by restricting community mobility or gatherings.In addition to the dates of policy enforcement initiations, this data set includes the dates the polices were rolled back.From this data set, we used five types of restrictions: 1-Type 1: Stay-at-home orders 2-Type 2: Prohibiting gatherings of 50 or more people 3-Type 3: Prohibiting gatherings of 500 or more people 4-Type 4: Prohibiting dine-in restaurants and bars 5-Type 5: Closing entertainment businesses and gyms Type 1 is the main variable of interest in our study.We use Type 2 through Type 5 as control variables in our models.

6-County-level socio-economic data:
This data set was obtained from The U.S. Department of Agriculture (USDA). 13and includes information about the socio-economic indicators at the county level.In particular, this data set includes information about education level, population estimates including national and international migration, poverty, and unemployment.

Variables:
Table 1 reports the list of variables, their descriptions, and their summary statistics.Figure 2 reports the correlation coefficients.Our data set includes data about 1,563 counties over a 14-week period (from the 10 th week of 2020 through the 23 rd week of 2020).This is a period that covers the first peak in the number of cases in the U.S. and includes the time before and after the roll out of the OuAA campaign by The White House.Furthermore, many counties and states enforced stay-athome orders during this time period.Some of those restrictions were lifted again in the same time frame of our study.
We excluded observations with any missing value from the data set.This resulted in 18,769 observations with complete data.As mentioned before, the community mobility indices (Retail and

Econometric Model:
We use the Difference-in-Difference (DiD) study design to understand the impact of stay-at-home orders on community mobility.The enforcement of stay-at-home orders by counties over time creates a natural experiment setting that allows the comparison of difference in community mobility before and after enforcing the stay-at-home orders across the counties.We further address the endogeneity of stay-at-home order decisions using a matched sample of counties (a match between counties that enforced an order and counties that never did).To assess the effect of stay-at-home orders on community mobility indices, we employ the following model: Where i represents the county, s represents the state, and t represents the week. !"# is the community mobility index (i.e.Retail or Residential).We are interested in  $ through  ( . $ is the DiD coefficient  % through  ( capture the interaction effects between stay and conservative, post_reopen and conservative, and stay, post_reopen, and conservative respectively. ) through  %% capture the effects of the control variables. # captures time-fixed effects,  !" captures county-fixed effects, and  # ×  " capture the interaction between time-and state-fixed effects.

Results:
Table 3 reports the result of our DiD analysis.For stay-at-home orders to be effective, we expect a drop in Retail and a jump in Residential.In models 1 through 3, the coefficient for stay is negative and significant.That is, retail & recreation activities decreased in counties that had a stay-at-home order after controlling for number of cases and deaths per 100k, unemployment rate, other types of restrictions, county-fixed effects, time-fixed effects, and the interaction of time-and state-fixed effects.In models 4 through 6, the coefficient for stay is positive and significant.This indicates that stay-at-home orders were effective in keeping residents at home.In model 2, the interaction between stay and conservative is positive and significant.This means that more conservative counties had more retail & recreation activities than liberal counties.In model 5 the coefficient for this interaction is negative and significant.This indicates that the conservative counties spent less time at home during the stay-at-home enforcement.In model 3, the interaction between post_reopen and conservative is positive and significant, which indicates that the conservative counties had more retail & recreation activities than liberal counties after OuAA campaign launch.This coefficient is negative and significant in model 6, suggesting more conservative counties stayed less at home after OuAA campaign launch compared to less conservative counties.In model 3, the interaction between stay, post_reopen, and conservative is negative and significant.This means that conservative counties that had a stay-at-home order enforced had less retail & recreation activities after OuAA launch compared to conservative counties that did not.The three-way interaction coefficient in model 6 also suggest that the conservative counties with stay-at-home order in place spent more time at home after the launch of OuAA campaign compared to conservative counties that did not.

Robustness Checks:
Given that the decision to enforce stay-at-home orders could be endogenous, we used propensity score matching to find the best match for each county that had no stay-at-home enforcement with a similar county that enforced state-at-home order in place.We used county level data about residents' education, population, deaths, births, domestic and international migration, percent below federal poverty line, unemployment rate, and median household income to find counties (matches for counties that did not have stay-at-home orders).This resulted in 660 counties (330 counties without any stay-at-home order and 330 similar counties with some level of stay-at-home enforcement).We reran our analysis using the matched samples.The results are reported in Table 4.
According to the results reported in this table, our findings are robust.

Figure 1 .
Figure 1.President Trump's Tweets about his "Opening Up America Again" Campaign Residential) were obtained from Google's Community Mobility Report.According to Google, these data are based on "data from users who have opted-in to Location History for their Google Account". 14Retail and Residential indices reflect the change in users' locations based on a baseline.Per Google's documentation, the baseline is the median value, for the corresponding day of the week, during the 5-week period January 3 rd through February 6 th 2020.A negative value for Retail means that users spent less time in retail stores compared to the baseline timeframe.A positive score for Residential means that the users spent more time at a residential location (i.e.home) compared to the baseline timeframe.

Figure 3 .
Figure 3. Average County-level Stay-at-home Restrictions Color-coded Based on Longevity of Orders (counties with darker colors enforced stay-at-home orders for a longer period)

Figure 4 .Figure 5 .
Figure 4. Duration of stay-at-home Restrictions in U.S. States

Table 2
Retail changed from -29.250 to -15.637 (increase in retail & recreation activity) and the value for Residential decreased from 11.988 to 11.750 after the launch of OuAA campaign.For conservative counties without stay-at-home order the value for Retail changed from -24.656 to -4.126 and the value for Residential changed from 10.763 to 9.349.In liberal counties that with stay-at-home order in place, the value of Retail changed from -43.025 to -29.852 (increase in retail & recreation activity) and the value for Residential decreased from 18.028 to 15.183 after the launch of OuAA campaign after the launch of OuAA campaign.For conservative counties without stay-at-home order the value for Retail changed from -37.070 to -17.063 and the value for Residential changed from 15.898 to 12.318 after the launch of OuAA campaign.
, in liberal counties (conservative county = 0) that did not have stay-at-home order in place (stay-at-home order = 0), the value of

Table 3 . The Impact of "stay-at-home" orders and "Reopen America" on Community Mobility Indices in U.S. Counties
To better interpret the results for the three-way interaction terms, consider the following counties: .285after the launch of OuAA campaign.If this county did not have a stay-at-home order, the value for Retail would increase further to (13.695 × 1 × 0.495) = 6.799.According to Table1, this translate to 0.177 -0.318 = -0.141.If this county had a stay-at-home order in place, per results reported for model 3 in Table3, the launch of OuAA campaign.If this county did not have a stay-at-home order, the value for Retail would increase to (13.695 × 1 × -0.141) = -1.931.If this county had a stay-at-home order in place, per results reported for model 6 in Table3, the expected value of Residential would decrease from (1.813 × 1) + (-2.534 × 1 × -0.141) = 2.170 to (1.813 × 1) + (-2.534 × 1 × -0.141) + (-2.825 × 1 × -0.141) + (-0.862 × 1 × 1) + (2.606 × 1 × 1 × -0.141) = 1.339 after the launch of OuAA campaign.If this county did not have a stay-at-home order, the value for Residential would further decrease to (-2.825 × 1 × -0.141) = 0.122.