key: cord-0779230-vskolc4g
authors: Shino, Enrijeta; Binder, Michael
title: Defying the Rally During COVID‐19 Pandemic: A Regression Discontinuity Approach
date: 2020-07-23
journal: Soc Sci Q
DOI: 10.1111/ssqu.12844
sha: 09f86f2f97cc4fad18e9e26ac2fac7d8a3518d05
doc_id: 779230
cord_uid: vskolc4g

OBJECTIVE: Do people set aside their partisan differences and rally around elected officials during a pandemic? President Trump's delegation of responsibility to the states during the COVID‐19 pandemic placed governors on the frontlines of the battle; some have shined and garnered positive national attention, others have wilted under the pressure of the national spotlight. METHODS: We use regression discontinuity design and exploit a discontinuity in the state's political events to assess the support of a governor's response to the pandemic. RESULTS: Using survey data from Florida's registered voters, we find that Governor DeSantis's approval dropped by 7 percentage points following his “Safer at Home” order press conference on April 1. CONCLUSION: Our results suggest that under certain circumstances partisanship can blunt a “rally around the flag” effect. This finding provides context to understanding when and under which circumstances elected officials can expect increases (or decreases) in public support.

New York announced their "stay-at-home" orders on March 19 and 20, while Florida's governor waited until Surgeon General Adams urged all states to follow suit (Daugherty, 2020) . On April 1, 2020 at 12:00 pm, DeSantis announced the issuing of Executive Order 20-19 "Safer at Home." We use this event in Florida politics to analyze how the public reacted to the preventative measures taken by the governor.

Using an original survey of Florida registered voters, the main question we address in this study is how the Florida governor's announcement of the "Safer at Home" order affected his public support. Did Floridians set aside their partisan views and rally around the governor during the pandemic? We employ regression discontinuity design to compare the public's support of how the Florida governor managed the pandemic before and after the press conference. We find that, even in times of unprecedented crisis, partisanship affects people's perceptions of whether the pandemic is more of an economic or public health threat to our society. Most importantly, we find that support for the governor dropped by 7 percentage points after his announcement. This effect is primarily driven by a decrease in Democratic support.

For this study, we conducted an email survey of registered Florida voters. The data collection began on March 31 and ended on Saturday, April 4, 2020. It was administered via Qualtrics and had a 1.1 percent response rate. The sample frame comprises 3,244 registered Florida voters, 18 years of age or older. The email addresses used for this survey were obtained from the Florida Division of Elections' February update and were selected using probability sampling among registered voters in the Florida voter file. The margin of sampling error for the total sample is ±1.7 percentage points. Variables such as partisan registration, sex, race, and age come from the voter file list. To ensure a representative sample of registered voters, the 10 Florida designated market areas were stratified. Quotas were placed on each of these stratified areas to ensure a proportionate number of completed surveys from across the state. Due to the unique population of Miami-Dade County, it was separated to create an 11th strata from the 10 designated media markets.

Since the beginning of his term in January 2019, Florida Governor Ron DeSantis has had a high approval rating among both Republicans and Democrats in the state; in March 2019, October 2019, and even February 2020, DeSantis's job approval was a doubledigit net positive among Democrats. Unlike other U.S. governors, whose approval ratings have increased during the COVID-19 pandemic, DeSantis's approval rating dropped by 7 points (Mehta, 2020) . In an effort to take both health preventative measures and continue to keep the economy open, Governor DeSantis's strategy of handling the pandemic has been criticized as slow and confusing (Rohrer, 2020) . Pictures of Florida's crowded beaches with Spring Breakers had deadly consequences and garnered national ridicule (Schorsch, 2020; Mazzei and Robles, 2020) , escalating Floridians' anxiety of a severe virus outbreak in the state. "Nearly 2,000 Florida healthcare professionals signed an open letter to DeSantis pleading with him to take more aggressive actions to hinder the spread of the virus, including enacting a statewide shelter-in-place order" (Allyn, 2020). DeSantis had also been criticized for lacking an original strategy for Florida while being labeled "a 'mini-Trump' governor who borrowed the president's playbook" (Luscombe, 2020) . This media coverage and elite discourse likely played into DeSantis's divergent approval ratings.

Given that there is variation across different subgroups of the public on how they react to elected officials, often dependent on their level of economic, political, and personal considerations (Sniderman, Brody, and Tetlock, 1991; Krause, 1997; Baum and Kernell, 2001; Baum, 2002) , we analyze the heterogeneity of the rally around the flag effect across different subgroups of the constituency. In Table 1 , we show whether Democrats and Republicans hold similar perceptions of COVID-19. Referring to Table 1 , we find that 83 percent of Democrats consider COVID-19 to be a greater public health impact than economic threat compared to 52.2 percent of the Republicans. Apparently, partisanship is affecting the public's perceptions of COVID-19 as these numbers show that respondents are aligning their perceptions with their respective party's rhetoric. Does partisanship affect their support for the governor's job in handling the virus? Table 2 shows support among a majority of Republicans regardless of whether they consider COVID-19 to be a greater economic or public health threat. For example, 82.8 percent of Republicans who consider the virus to be an economic threat support the governor's handling of the COVID-19 pandemic compared to 73.5 percent of the those who think it is more of a public health risk. Conversely, 49.5 percent of Democrats who think that the virus is more of an economic threat support the governor's handling of it, compared to only 26.4 percent of those who think it is a greater public health risk. Clearly, partisanship has impacted the assessment of the governor's job in this circumstance. Did the press conference and DeSantis's "Safer at Home" order affect the governor's support? To answer this question, our identification strategy utilizes a sharp regression discontinuity design, as explained below.

We use a quasi-experimental regression discontinuity design to test how the approval of the governor's response to COVID-19 discretely changed following the "Safer at Home" executive order. Let each voter i in our random sample be characterized by a vector (y i , w i , d i ), where the scalar y i denotes the governor's support, the vector w i captures individual-level characteristics, d i = 1[x i > 0], our treatment variable, is an indicator function equal to 1 if the bracketed logical condition holds, and 0 otherwise, and x i is our forcing variable. The forcing variable runs from −25 to 65 hours, where 0 (cutoff point) denotes the time when the governor held the "Safer at Home" order press conference, April 1, 2020 at 12:00 pm. The regression discontinuity approach assigns observations to the treatment and control groups based on a discrete threshold of our continuous forcing variable x i . That is, we assign observations to the treatment group if x i > 0 and to the control group if x i < 0.

The main identification assumption required for the regression discontinuity design is that the conditional expectation functions of the potential outcomes are continuous on the support of the running variable x i . 1 Hence we can test the approval of the governor's response to COVID-19 by the value of the discontinuity of the expected value of the response at x i = 0 (Angrist and Pischke, 2008) . We specify the parametric model at the individual level, estimating the local average treatment effect as follows:

where is our parameter of interest that captures the approval of the governor's response to COVID-19. We also estimate a more flexible local regression model:

where f is a function of x i , and we model f using a second-and third-order degree polynomials.

In Figure 1 , we plot Floridians' support for the governor before and after the press conference. The cutoff is the time of the press conference, April 1, 2020 at 12:00 pm. As shown in Figure 1 , Floridians who responded to our survey before the governor's press conference to address the "Safer at Home" measure showed greater support for the governor compared to their counterparts who completed the survey after the state address. To be more concrete, the support for the governor after the press conference dropped by 7 percentage points among all respondents. Figure 1 provides a general overview of the governor's support among all respondents but it does not highlight how partisanship is affecting this relationship.

To understand which subgroup of Floridians is responsible for this shift in the governor's support, we replicate Figure 1 using split-sample analysis by party affiliation. Referring to Figure 2 (a), we find that Republicans, as expected, had a high support rate for the governor before the press conference, about 80 percent, and after the press conference it marginally dropped to 74 percent. A similar pattern is observed with no party affiliates in Figure 2 (b) . As shown in Figure 2 (b), no party affiliates' support for the governor before the conference was about 54 percent and after the conference it dropped to about 47 percent. It should be highlighted that the decrease in the governor's support after the press conference is General Support for Governor Before and After the Announcement NOTE: The dashed line is the cutoff point at the time when the governor held the press conference, April 1, 2020, at 12:00 pm. Hours before and after Governor's speech Probability of Governor approval not statistically significant for either Republicans or no party affiliates compared to their respective preconference support levels. The slight decline in the governor's support, in Figures 2(a) and 2(b), bounced back up to its preannouncement levels several hours after the event.

Completely different support patterns are observed for Democrats in our sample. As shown in Figure 2 (c), Democrats' support for the governor was significantly lower compared to that of Republicans and no party affiliates. Democratic support for the governor seemed to have had an increasing pattern prior to the announcement and was followed by a drastic drop right after; Figure 2 (c). Referring to Figure 2 (c), Democratic support for the governor dropped immediately after the press conference and stayed stagnant for the next 65 hours postannouncement.

In Table 3 , we show the marginal effect for support for the governor by party affiliation. All estimates shown in Panel A are estimated controlling for the respondent's age, race, gender, income, education, virus perceptions, risk of contracting the virus, financial concerns, and county fixed effects, 2 while in Panel B, we replicate the same models without county fixed effects to see how much of the effect is driven by county differences. 3 As shown in Table 3 , Democrats' support declined by 9 percentage points after the press conference on April 1. To ensure that the effects we find are not due to nonlinearity in the Governor Support Before and After the Announcement, by Party NOTE: The dashed line is the cutoff point at the time when the governor held the press conference, April 1, 2020, at 12:00 pm. Each graph shows the split-sample description of the data patterns. 0 5 10 15 20 25 30 35 40 45 50 55 60 65 Hours before and after Governor's speech Probability of Governor approval for NPAs (2)-(4) in Appendices A and B.

As shown in Table 3 , the Republican decline in support for the governor is not statistically significant. In addition, the drop in support among no party affiliates appears to be statistically significant, but the effect goes away when we run sensitivity checks for nonlinearity. 4 Our results are consistent when we replicate our analysis without controlling for county fixed effects, as shown in Table 3 , Panel B. (1) in Tables B1-B3; * * * p < 0.01; * * p < 0.05; * p < 0.1.

To sum up, we find that people are still using their party identification lenses to view the world and form political expectations during the COVID-19 pandemic. Florida Democrats did not rally and their support for the governor dropped after the "Safer at Home" announcement. Apparently, the media and elite discourse (Zaller, 1992; Groeling and Baum, 2008) blunted DeSantis's potential job approval gains. The fact that Democratic governors such as Andrew Cuomo of New York and Gavin Newsom of California reacted faster, and the fact that the timing of the "Safer at Home" announcement coincided with the White House's marching orders, likely increased the partisan division between Democrats and Republicans in the state. Overall, 70 percent (2,264 respondents) of our respondents considered COVID-19 to be a greater public health issue than economic threat and 43 percent of respondents thought the state government's actions had not gone far enough to help fight the pandemic. Leading up to the issuance of the "Safer at Home" announcement, there was uncertainty about how best to handle this novel pandemic; it is likely that Floridians were holding out hope that a shelter-in-place order was not necessary. However, as soon as DeSantis finally issued the order, Democrats blamed the governor for delaying the response to put in place preventative measures to protect public health. The decision to prioritize the economy over health and delaying the "Safer at Home" order likely hurt Governor DeSantis's support most among Democrats.

Moreover, Baker and Oneal (2001:682) argue that "what appears to matter most in regard to the size of the rally effect is not the nature of the dispute itself but how effectively the White House manages the presentation of the dispute through presidential statements, prominent media coverage, and the garnering bipartisan support." Even though Florida was not hit as hard from the virus, Governor DeSantis's handling of the situation has been heavily criticized. "DeSantis' bizarre decision to don just one glove during a recent briefing, drawing ridicule from the Internet at large questioning whether he understood how to properly use protective gear" (Schorsch, 2020) . Hindering the public's perceptions of DeSantis's efforts also included banning reporters from the governor's briefings, inaccurate reporting of COVID-19 cases, and a reliance on a very partisan president's statements. This combination of actions, elite criticism, and media coverage has led to a stark partisan divide and blunted positive reviews that typically accompany statewide or national emergencies. (2)-(4) are sensitivity checks for any possible nonlinearity in the data; * * * p < 0.01 * * p < 0.05 * p < 0.1. (2)-(4) are sensitivity checks for any possible nonlinearity in the data; * * * p < 0.01 * * p < 0.05 * p < 0.1. (2)-(4) are sensitivity checks for any possible nonlinearity in the data; * * * p < 0.01 * * p < 0.05 * p < 0.1. (2)-(4) are sensitivity checks for any possible nonlinearity in the data; * * * p < 0.01 * * p < 0.05 * p < 0.1. (2)-(4) are sensitivity checks for any possible nonlinearity in the data; * * * p < 0.01 * * p < 0.05 * p < 0.1. (2)-(4) are sensitivity checks for any possible nonlinearity in the data.; * * * p < 0.01; * * p < 0.05; * p <0.1.

-Coronavirus perceptions is coded 1 if the respondent thinks it will have a higher impact on the public health, and 0 if it will have a higher impact on the economy. -Risk of contracting the virus is coded 1 if the respondent and someone in his or her household is at higher risk of contracting the virus, and 0 if nobody in the household is at higher risk. -Financial concerns is coded 1 if the respondent is concerned about personal finances, and 0 otherwise. -Education is coded 0 if the respondent has a high school or lower degree, 1 some college degree, 2 college degree, and 3 a graduate degree. -Household income is coded 0 if the respondent's household income is up to $50,000, 1 if the income varies from $50,000 to $75,000, 2 if the income varies from $75,000 to $100,000, and 3 if it is higher than $100,000. -Party of registration is coded 0 for Democrats, 1 for Republicans, and 3 for no party affiliates. -Race is coded 1 for whites, 2 for blacks, 3 for Hispanics, and 4 other race.

-Age is coded 1 for 18-24 years, 2 for 25-34 years, 3 for 35-44 years, 4 for 45-54 years, 5 for 55-64 years, and 6 for 65 years or older. -Gender is coded 0 for male and 1 for female.

There has been a lot of talk lately about the new coronavirus (COVID-19) and the national emergency. I'd like you to tell me whether you approve or disapprove of the job Governor

How concerned are you personally about contracting the coronavirus (COVID-19)? • Very concerned • Somewhat concerned • Not Very concerned • Not at all concerned

COVID-19)? • Yes, I am higher risk • Yes, someone in my household is higher risk • Yes, both myself and someone in my household are higher risk • No, nobody in my household is higher risk 4. How concerned are you about the impact of coronavirus (COVID-19) on your personal finances? • Very concerned • Somewhat concerned • Not too concerned • Not at all concerned 5. Which is a bigger concern for you, the public health effects or the economic impact of the coronavirus (COVID-19)? • Public health effects • Economic effects 6

What is your annual household income? • Less than

Variable Information

Coded 1 if respondent approved of the job Governor Ron De-Santis is doing to address the pandemic, and 0 if disapproved. • Independent variables: REFERENCES Allyn, Bobby. 2020. Fact Check: Trump Praises Florida Governor Despite Criticism of Coronavirus Response

Mostly Harmless Econometrics: An Empiricist's Companion

Patriotism or Opinion Leadership? The Nature and Origins of the 'Rally'Round the Flag Effect'

The Constituent Foundations of the Rally-Round-the-Flag Phenomenon

Economic Class and Popular Support for Franklin Roosevelt in War and Peace

Trump Blames Hospitals for Masks and Ventilator Shortages

Intergroup Relations

Assessing the President: The Media, Elite Opinion, and Public Support

As DeSantis Defers to White House on Stay-Home Order, a Top U.S. Official Urges One

Crossing the Water's Edge: Elite Rhetoric, Media Coverage, and the Rally-Round-the-Flag Phenomenon

Joining and Leaving the Rally: Understanding the Surge and Decline in Presidential Approval Following 9/11

Explaining Presidential Popularity: How Ad Hoc Theorizing, Misplaced Emphasis, and Insufficient Care in Measuring One's Variables Refuted Common Sense and Led Conventional Wisdom Down the Path of Anomalies

Voters, Information Heterogeneity, and the Dynamics of Aggregate Economic Expectations

Rallying Around the Flag: Foreign Policy Events and Presidential Popularity

Florida's Slow Response: A 'Mini-Trump' Governor Who Borrowed the President's Playbook

Political Drama, Economic Conditions, and the Dynamics of Presidential Popularity

The Costly Toll of Not Shutting Down Spring Break Earlier

Manipulation of the Running Variable in the Regression Discontinuity Design: A Density Test

Most Americans Like How Their Governor is Handling the Coronavirus Outbreak

Presidential Popularity from Truman to Johnson 1

War, Presidents, and Public Opinion

Crowded Beaches, Confusion Amid Coronavirus Orders as DeSantis Won't Mandate Statewide Lockdown

Coronavirus vs. Governors: Ranking the Best and Worst State Leaders

How Ron DeSantis Is Winning the Pandemic Battle, but Losing the Perception War

Reasoning and Choice: Explorations in Political Psychology

The Nature and Origins of Mass Opinion