Politics,Covid,and In-Person Instruction During the First Year of the Pandemic

Abstract

In spring 2020, nearly every U.S. public school closed at the onset of the Covid-19 pandemic. Existing evidence suggests that local political partisanship was a better predictor of in-person instruction than Covid case and death rates in fall 2020. We replicate and extend these analyses using data collected over the entirety of the 2020–21 academic year. We affirm that local political partisanship was an important initial predictor of county-level in-person instruction rates. We also demonstrate that, under certain conditions, Covid case and death rates were meaningfully associated with initial rates of in-person instruction. We reveal that partisanship became less predictive—and prior average student achievement became more predictive—of in-person instruction as the school year continued. We then leverage data from two nationally representative surveys of Americans’ attitudes toward education and identify an as-yet-undiscussed factor that predicts in-person instruction: public support for increasing teachers’ salaries.

Keywords

educational policy leadership policy politics politics of education

In response to the onset of the Covid-19 pandemic in spring 2020, nearly every K-12 school district in the U.S. closed school buildings and rapidly pivoted to online learning. As the next academic year began in fall 2020, educational leaders in school districts around the country reached dramatically different conclusions on whether to keep school buildings closed or to re-open for in-person instruction (Henderson, Peterson, & West, 2020; Lake & Dusseault, 2020). While some districts ushered students and educators back into their classrooms on the first day of school, others continued to embrace online learning. By the end of the 2020–21 academic year (AY2020-21), most K-12 students were back in school buildings (Gross et al., 2021; Henderson et al., 2021). This return to business-as-usual, however, was tenuous as the emergence of new and more infectious variants and strict quarantine rules frequently disrupted in-person instruction well into AY2021-22 (Cain Miller & Sanger-Katz, 2022).

With the benefit of hindsight, we may ultimately look favorably or unfavorably on educational leaders’ decisions during this turbulent time. Student performance across a wide range of subjects and grade levels dropped precipitously in the wake of the pandemic (NAEP, 2022b, 2022c; National Assessment of Educational Progress [NAEP], 2022a). There is an emerging consensus, grounded in empirical research, that extended school closures were deeply detrimental to many students’ academic progress and mental health and that the negative consequences were disproportionately experienced by low-income students and students of color (American Academy of Pediatrics et al., 2021; Goldhaber et al., 2022; Halloran et al., 2021; Lewis et al., 2021; Yard et al., 2021; but see Bacher-Hicks et al., 2022, for evidence that remote learning may have reduced incidences of bullying—including cyber-bullying—among school-age children). However, these adverse outcomes were not exclusively the result of extended school closures. Test scores declined across the country, both in districts that re-opened for in-person instruction swiftly and in districts that offered only remote instruction for almost the entire 2020–21 school year (Fahle et al., 2023). It is also worth remembering how challenging the decision-making context was at the time, with pervasive uncertainty and disagreement about the optimal course of action (Menasce Horowitz, 2020), inconsistent messaging from federal public health and education agencies (Mansoor, 2020), the public’s increasingly politicized responses to government guidance (Milosh et al., 2021), sharp divergences in parental preferences for online versus in-person instruction (Haderlein et al., 2021), and the basic dilemma posed by a pathogen that presented differential health risks to young students and older school employees (Centers for Disease Control and Prevention, 2021). Deciding whether or not to re-open schools for in-person instruction was, in many communities, a profoundly difficult choice with multiple and competing pressures. It was also a profoundly high-stakes choice with lasting implications for students, families, and educators. Understanding why some schools re-opened and others remained closed is an essential part of our collective effort to learn from this extraordinary event and to carry that knowledge forward to inform education policy responses to our next great challenge.

It is therefore an ideal time to revisit the early research on the factors that predicted in-person instruction in AY2020-21. With expanded data on weekly county-level rates of in-person instruction over the entire 2020–21 school year, we attempt to replicate and extend prior analyses. In many cases, our results reaffirm previous findings. For example, consistent with prior research (DeAngelis & Makridis, 2021; Grossmann et al., 2021; Harris & Oliver, 2021; Hartney & Finger, 2022), we find that local political partisan was a meaningful predictor of in-person instruction rates during AY2020-21. Specifically, communities with a larger proportion of Democratic voters were less likely to re-open. However, we add important nuance to this conclusion, demonstrating that the relationship between political partisanship and the rate of in-person instruction is reduced in magnitude when we consider AY2020-21 in its entirety (rather than focusing on the first few weeks of school in fall 2020). We also note that, under certain conditions, the relative threat of the pandemic—as measured by cumulative Covid case and death rates—is negatively associated with in-person instruction rates. Some observers have suggested that the availability of in-person instruction was simply a function of partisan politics (e.g., Bodenheimer, 2022; Chait, 2022; Murray, 2022; Nazaryan, 2020). We show that this was not the case.

We also contribute to this literature by embracing a broader definition of politics when considering the range of conflictual and coalitional elements that shaped educational leaders’ decisions about instructional modality. As part of this effort, we identify an as-yet-undiscussed factor that reliably predicts AY2020-21 in-person instruction rates: pre-pandemic local public support for increasing teachers’ salaries. The relationship between these two phenomena—public opinion on teachers’ salaries and in-person instruction rates—is modest in magnitude, and we do not identify a specific causal mechanism in this analysis. However, this relationship is robust to the inclusion of a large set of covariates as well as differences in question wording, date of survey administration, and sampling strategy across two nationally representative surveys. We speculate that this survey item captures some variation with respect to local support for educators that is not otherwise accounted for by a wide range of potentially confounding individual-level factors (such as political party identification, political ideology, and parental status) or county-level factors (such as local partisanship). It is possible that, compared to their counterparts in otherwise similar settings, educational leaders in communities with stronger pre-existing support for educators were better able to navigate the logistical and political challenges of re-opening schools for in-person learning. This interpretation would align with recent research revealing that pandemic-era test score declines were more modest in communities with higher institutional trust, as measured by voting rates and census response rates (Fahle et al., 2023). The inverse relationship would also be consistent with the results of our analysis: Leaders with greater administrative and political capabilities—which would prove useful when attempting to re-open schools—may have, prior to the pandemic, cultivated a deeper well of public support. Qualitative research on school re-opening decisions during AY2020-21 documents both the high degree of discretion afforded to local educational leaders as well as the importance of local political and cultural contexts, making it difficult to adjudicate between these potentially overlapping interpretations (Singer et al., 2022).

Evidence on School Closures and Re-Openings During the Pandemic

A nascent empirical literature has emerged describing the factors associated with districts’ decisions regarding instructional modality during AY2020-21. In spring 2020, at the onset of the pandemic, state-level officials made the initial decisions to close public elementary and secondary schools (Grossmann et al., 2021). Educators and school leaders scrambled to provide virtual instruction during these first chaotic months, with tremendous variation across communities in the quality and content of instruction (Henderson, Houston et al., 2020). The following summer, district-level officials strategized about whether to continue providing fully remote instruction, re-open schools for in-person instruction without restriction, or offer a hybrid option with a subset of students attending school in-person each day in order to facilitate adherence to public health guidance regarding the appropriate amount of physical distance between students.

Multiple research teams have demonstrated that school districts’ initial decisions in fall 2020 were either unrelated or only modestly related to local Covid case and death rates (DeAngelis & Makridis, 2021; Grossmann et al., 2021; Harris & Oliver, 2021; Hartney & Finger, 2022; Valant, 2020; but see Christian et al., 2022, for evidence from Ohio indicating that a marginal increase in new Covid cases reduced the probability that a district offered in-person instruction during the following week). Compared to the severity of the pandemic, political partisanship was a more reliable predictor of districts’ re-opening decisions (DeAngelis & Makridis, 2021; Grossmann et al., 2021; Haderlein et al., 2021; Harris & Oliver, 2021; Hartney & Finger, 2022; Valant, 2020). Independent of local Covid positivity rates, Republican-leaning districts were more likely to re-open for in-person instruction while Democratic-leaning districts were more likely to offer only remote learning options. In addition to partisanship, researchers have identified other factors associated with higher initial rates of in-person instruction: weaker local teachers’ unions (DeAngelis & Makridis, 2021; Grossmann et al., 2021; Harris & Oliver, 2021; Hartney & Finger, 2022), the racial/ethnic and economic composition of the community (Camp & Zamarro, 2022; Grossmann et al., 2021; Haderlein et al., 2021; Harris & Oliver, 2021; Hartney & Finger, 2022), and the availability of more alternatives to traditional district schools in the area (Hartney & Finger, 2022; but see Cohodes & Pitts, 2022, for evidence that charter schools were less likely to offer in-person instruction than their district counterparts). With a few exceptions (e.g., Christian et al., 2022; Harris & Oliver, 2021), the existing literature focuses primarily on the predictors of in-person instruction at the very beginning of AY2020-21.

Public opinion on the desirability of re-opening schools for in-person instruction was split along political and demographic lines, and there is evidence that decisions regarding school closures were at least somewhat aligned to local preferences. For instance, there was a stark partisan divide in support for re-opening schools during AY2020-21: Republicans expressed greater enthusiasm for resuming in-person instruction than Democrats (Collins, 2021; Haderlein et al., 2021). Survey researchers also observed differences in preferences about instructional modality by race/ethnicity (greater White support for in-person instruction), income (greater support for re-opening schools among more affluent individuals), and urbanicity (greater rural and suburban support for a return to the classroom) (Collins, 2021; Haderlein et al., 2021). When asked in January 2021 about the type of instruction their child was receiving versus the type of instruction that they preferred, fully three-quarters of a nationally representative sample of parents of school-age children indicated that their child was receiving the type of instruction that they wanted (Barnum, 2021; Haderlein et al., 2021). However, the causal direction—whether local leaders were responding to public opinion or whether public opinion was shaped by local decision-making—is unclear. When tracking parents’ attitudes toward instructional modality over time, preferences for remote-only instruction tended to decline after the child’s school re-opened, suggesting that public opinion was, in part, reacting to cues about the safety of in-person instruction implied by the decision to re-open or remain closed (Kogan, 2021).

Educational leadership at the local level has long been subject to political pressures (Hess, 1999; Howell, 2005)—a trend that only appears to be increasing in our polarized era (Henig et al., 2019; Houston, 2022). During the Covid-19 pandemic, these pressures intensified. Superintendents reported strong public reactions to operational decisions—whether to re-open for in-person instruction, require facemasks, mandate vaccines for teachers, etc.—that tracked closely with constituents’ partisan identities (Cash et al., 2022). School board meetings became battlegrounds for opposing political factions (Saul, 2021). In the midst of this turbulence, educational leaders faced a larger-than-usual list of logistical and administrative concerns, such as ensuring that students had access to adequate technology to participate in remote learning, tracking attendance across multiple modes of instruction, scheduling staggered in-person instruction for districts attempting a hybrid model, screening students for illness and enforcing quarantine policies, revamping sanitation procedures, renovating ventilation systems, maintaining physical distance between students in schools and on buses, administering assessments over the internet, and communicating with families about re-opening plans as well as any newly-developed resources available to help make up for lost learning time (Edgerton et al., 2021; Gill & Dusseault, 2021; Lake, 2021; Leung et al., 2021; McCann & Dusseault, 2021; Ondrasek et al., 2021). In short, an increase in the professional demands of district-level leadership coincided with an increase in the contentiousness of the political environment.

Our Contribution

Politics, at least in Lasswell’s (1936) classic definition, is about “who gets what, when, and how.” Laswell’s framing applies remarkably well to the issue of school re-opening decisions at the height of the Covid-19 pandemic: Which students, what kind of instruction, at what time, and through which medium? Moreover, the “who” in this formulation also ought to include the families of school-age children, the employees of the school system, and the other members of the communities in which the schools are located. Whether students should be learning from their homes or in their classrooms during the 2020–21 school year was an unavoidably political question.

The prior literature on this issue focuses on two elements of politics that are essential to any reasonably comprehensive account of 21st century education policy decision-making: the role of political partisanship and the influence of relevant special interest groups, namely teachers’ unions. We agree with our predecessors in this line of inquiry that both of these factors were important predictors of the availability of in-person instruction in fall 2020—often at the expense of more direct indicators of the relative threat of the Covid-19 virus. We build on this foundation in two ways. First, working in the same basic paradigm, we extend the timeline of the analysis to include the entirety of the 2020–21 school year. Second, we broaden the scope of our definition of politics. Recalling Stone’s (2001) work on local civic capacity with respect to school reform, we also think of politics as “the extent to which different sectors of the community—business, parents, educators, state and local officeholders, non-profits, and others—act in concert around a matter of community-wide import,” particularly in moments of crisis and distress (p. 596). With this expanded conceptualization of politics in mind, we seek to contribute to our collective understanding of the factors that made in-person instruction administratively manageable and politically feasible in the face of the significant, multiple, and overlapping logistical demands of pandemic-era schooling as well as the potentially treacherous cross-currents of partisan polarization and the reluctance of a core constituency, teachers, to return to the classroom in a moment of pervasive uncertainty regarding the threat of the pandemic.

We begin with a replication analysis of initial re-opening decisions in fall 2020 using an alternative measure of instructional modality: the estimated percentage of students learning in-person at the county level (much of the existing research on this issue focuses on dichotomous or trichotomous, rather than continuous, measures of instructional modality at the district level). We re-affirm many of the core findings of the prior literature. Specifically, we find that higher rates of in-person instruction occurred in communities that lean Republican and that have weaker local teachers’ unions. We note that our analysis, like the research we seek to emulate, relies on indirect and imperfect proxies of teachers’ union strength. Although there is little debate that teachers’ unions played an important role in the decisions regarding when and how to re-open schools for in-person instruction, their relative influence resists straightforward measurement. In contrast to some of the previous research, we also find evidence that, in certain circumstances, communities with higher Covid case and death rates had lower rates of in-person instruction. Although many of the individual studies we cite are candid about the difficulty of estimating these descriptive relationships accurately given the limited data available on instructional modality during the pandemic, the overarching inference that many observers have gleaned from this literature—that school closures and re-openings were exclusively about politics, not Covid—understates the sensitivity of these results to the myriad decisions that researchers must make when analyzing imperfect data.

We then extend this analysis beyond the first few weeks of AY2020-21 to identify the factors that predict the average rate of in-person instruction over the course of the entire school year. As we extend the time horizon, the magnitude of the relationship between partisan politics and in-person instruction is reduced in size. Conversely, the relationship between average student achievement and the rate of in-person instruction increases in magnitude and consistency across multiple model specifications. As the school year progressed, communities with a history of higher standardized test scores were more likely to re-open school buildings. Given the accumulating evidence on the detrimental effects of remote learning, this pattern has important and troubling implications for growing opportunity and achievement gaps between more and less privileged communities.

Finally, our analysis reveals an as-yet-undiscussed factor that reliably predicts in-person instruction rates during AY2020-21: local public support for teachers as measured by public opinion in favor of increasing teachers’ salaries. We argue that greater support for raising teachers’ salaries is not a consequence of re-opening local public schools. Rather, across multiple polls conducted at different times with different sampling strategies and with different question wordings, pre-pandemic public opinion regarding teachers’ salaries is consistently associated with higher pandemic-era in-person instruction rates. Moreover, insofar as public opinion on this issue serves as a proxy for local public support for teachers, it is a complicated one. We do not mean to imply that support for higher teachers’ salaries equates to complete deference toward teachers’ professional preferences. Indeed, as of August 2020, fully two-thirds of teachers indicated that they preferred online over in-person instruction (Kamenetz & Isensee, 2020). Rather, we argue that support for teachers, measured in this fashion, more closely approximates a willingness to provide educators with additional resources to accomplish their responsibilities.

We propose two potential interpretations of this relationship. First, in light of evidence describing the intense and often conflicting demands placed on teachers and school administrators during the pandemic, we speculate that educational leaders were better able to manage the logistical and political complexities of in-person instruction in communities with a greater support for educators—a specific form of local institutional trust, which Fahle et al. (2023) establish is associated with smaller post-pandemic test score declines. For example, scheduling staggered in-person instruction (if not every class and grade were to meet in school each day), maintaining physical distance between students in schools and busses, and enforcing face-mask mandates all required substantial community cooperation and coordination that may have been easier to marshal in a context with more congenial relationships between educators and families. Second, it could also be the case that more administratively and politically adept educational leaders had both stronger community support prior to the pandemic and more success at re-opening schools during AY2020-21. Leaders for whom the previous list of logistical challenges proved manageable may have cultivated strong local support for educators well before the pandemic began.

Data and Methods

Replication and Extension Analyses

Our replication and extension analyses are structured similarly to many of the studies we cite in the previous section. In brief, we regress a measure of school re-opening status on a large set of educational and demographic predictors with the objective of identifying factors that are associated with re-opening status, independent of all other included factors. While much of the earlier research adopts this same basic approach, each study employs slightly different variable choices and model specifications. As such, we use the terms “replication” and “extension” broadly. We do not seek to replicate or extend any one particular study; rather, we seek to re-examine the general conclusions of the prior literature using an analogous approach that incorporates an alternative measure of instructional modality collected over a longer period of time.

There is one way in which our analysis systematically differs from previously published studies. Most prior analyses were conducted at the district level, with the primary outcome of interest measured dichotomously (Was the school district open for in-person instruction or not?) or trichotomously (Was the school district offering in-person instruction, online instruction, or a hybrid option?). We focus instead on the county-level percentage of students receiving in-person instruction, measured as a continuous variable. This analytic choice has some strengths and weaknesses. By conducting the analysis at the county level, we are better able to align our empirical strategy with the most relevant unit of analysis for three theoretically important predictors of school re-opening decisions identified by earlier researchers: (1) Covid case and death rates, (2) political partisanship as measured by the local outcome of the most recent presidential election, and (3) teachers’ union strength. The first two factors, Covid case/death rates and partisan vote share, are measured and reported at the county level. The third factor, teachers’ union strength, is famously difficult to measure well (which we discuss at greater length in a later section). Our preferred approach—the percentage of labor union members among all employed adults—is also reported at the county level.

Our decision to focus on counties, however, is not without drawbacks. School boards and superintendents, not county leaders, were the officials vested with authority to make decisions regarding instructional modality. In most of the country, counties consist of multiple school districts, making it difficult to draw inferences about the specific decision-making process in a typical district. Therefore, the reader should view our replication and extension analyses as explorations into how larger communities—often composed of more than one school district—responded to the competing pressures of the pandemic when deciding whether to re-open schools for in-person instruction.

In-Person Instruction Rates in AY2020-21

Our data on in-person instruction rates among public school students during AY2020-21 come from the data aggregation firm Burbio (Burbio, 2022). During AY2020-21, Burbio audited over 1,200 public school districts representing 47% of the U.S. K-12 student enrollment in over 35,000 schools in 50 states. The Burbio data are more comprehensive with respect to school districts in more highly populated counties. For example, for the 232 most populated counties (representing 28% of the total student population), Burbio sampled school districts representing approximately 90% of students. For the next 228 most populated counties (representing 47% of the total student population), Burbio sampled 3-10 districts per county. For the remaining counties (representing 25% of the total student population), Burbio sampled a total of 130 districts. The districts least likely to be represented in the Burbio data are sparsely populated rural districts. Burbio’s audit team checked district websites, Facebook pages, local news stories, and other publicly available information every 72 hours for changes. School district learning modes were categorized as either traditional (in-person every day), hybrid (2–3 days in-person and 2–3 days virtual per week), or virtual. Burbio assigned a learning mode to a school district based on the most in-person option available to the general student population. Thus, if a district offered both traditional and virtual options, the district was categorized as traditional. If a district offered different learning modes by grade level, Burbio estimated the percent of students in each mode. These estimates were then used to generate county-level learning mode estimates, with each audited district in the county weighted by district enrollment. Burbio released these county-level estimates on a weekly basis from August 14, 2020 to June 25, 2021.

When conducting our replication analyses—re-examining the factors that predict school re-opening decisions in fall 2020—we rely on Burbio’s estimates of in-person instruction rates for the week of October 2, 2020 (i.e., at the beginning of AY2020-21 but after the initial chaos of the first days of school). When conducting our extension analyses—examining the factors that predict school re-opening decisions over the entirety of AY2020-21—we calculate the average weekly in-person instruction rate within each county over the full time period (August 14, 2020 to June 25, 2021). We also disaggregate the extension analyses by season: fall, winter, and spring (see Tables A1, A2, and A3 in the Appendix).

Teachers’ Union Strength

There is considerable debate in the prior literature over the best way to measure local teachers’ union strength (DeAngelis & Makridis, 2021; Hartney & Finger, 2022; Marianno et al., 2022; Strunk & Reardon, 2010). Many of the preferred metrics are only available at the state level or are only available for a handful of districts. For example, DeAngelis and Makridis (2021) employ three different state-level indicators: whether a school district is in a right-to-work state, a state-level ranking of teachers’ union strength generated by the Fordham Institute (Northern et al., 2012), and the percent of employed adults who are members of a labor union. They also use a county-level measure of the percent of employed adults who are members of a labor union based on proprietary data collected by the survey research firm Gallup. Hartney and Finger (2022) rely on student enrollment as a proxy, citing earlier research showing that teachers’ unions in larger districts exert more political influence and have more restrictive collective bargaining agreements (Moe, 2005, 2009; Rose & Sonstelie, 2010). They also consider whether or not a district has a collective bargaining agreement as well as an original measure of teachers’ union political activity—both of which are only available for a subset of districts. Marianno et al. (2022) employ a range of different measures: student enrollment, whether a district has a collective bargaining agreement, the length of a district’s collective bargaining agreement, union revenue, and the frequency with which the local teachers’ unions posted on Facebook.

Following Hartney and Finger (2022) and Marianno et al. (2022), we use the total number of students in the largest school district by enrollment in each county (collected from the National Center for Education Statistics’ Common Core of Data) as a measure of teachers’ union strength that is readily available for every county in the country. Multiple researchers argue that larger school districts tend to have stronger teachers’ unions (Hartney & Finger, 2022; Marianno et al., 2022; Moe, 2005, 2009; Rose & Sonstelie, 2010). This relationship, however, only holds within states. Across the country as a whole, many of the largest school systems are county-wide districts located in Southern states with regulatory environments that are less favorable to labor unions. Moreover, district size may have a separate and distinct relationship with instructional modality, independent of teachers’ union strength (if, e.g., the logistical complexity of in-person instruction was greater in districts with more students). In our interpretations of the relationships between district size and in-person instruction, it is not possible to adjudicate whether the relationship is driven primarily by teachers’ union strength or by student enrollment in its own right.

With these potential limitations in mind, we supplement our investigation with an additional measure: the U.S. Census Bureau Current Population Survey’s (CPS) estimate of the percentage of labor union members among all employed adults—similar to DeAngelis and Makridis (2021) supplemental measure—which is available for the 280 most populous counties (representing 64% of the total population nationwide). The CPS data on labor union membership offers a more direct gauge of union strength available for communities that represent approximately two-thirds of the U.S. population.

Other County-Level Predictors

For our measures of county-level Covid cases and deaths, we use data collected by the New York Times. Specifically, we employ the cumulative number of cases and deaths in each county as of October 2, 2020 (for the replication analysis) or June 25, 2021 (for the extension analysis). We argue that cumulative cases and deaths—as opposed to the number of cases and deaths in the most recent week or month—more accurately capture communities’ experience with the virus and local perceptions of the virus’ threat to public health. For example, consider two hypothetical communities. Over the course of two months, there was a large spike in Covid cases and deaths in the first community, followed by a noteworthy decline. In the second community, cases and deaths remained at a lower level throughout the same time period. An analysis that relies on contemporaneous rather than cumulative values would be unable to distinguish between these two communities in the latter half of the analysis, despite the significant differences in threat from the virus—both actual and perceived.

For our measure of Trump 2020 presidential vote share by county, we rely on the MIT Election Lab. For our measures of average per-pupil spending, average teachers’ salaries, and the percentages of all K-12 students in each county that attend a private school or a charter school, we use the National Center for Education Statistics’ Common Core of Data (for the two fiscal variables, we focus on the values for the largest district by enrollment in each county). We use the Stanford Education Data Archive v4.1 for standardized county-level estimates of average test scores in reading and math in grades 3 to 8 from 2009 to 2019. For our measures of county demographics, we collect the following data from the U.S. Census Bureau’s American Community Survey 2015–2019 (5-year estimates): percentage of non-Hispanic White residents, median household income, population size, population density, percentage of residents under age 18, percentage of residents over age 25 with a BA degree or higher, percentage of residents over age 3 who are currently in school, and the Gini index (a common measure of income inequality). We opt for relatively simple measures of many of the relevant underlying constructs (e.g., using the percentage of non-Hispanic White residents to measure racial/ethnic demographic composition) in order to limit the number of independent variables in each analysis. To account for the fact that some counties contain multiple school districts while other counties’ borders are coterminous with a single district, each model that includes county-level covariates also controls for the number of school districts in the county as reported by the National Center for Education Statistics’ 2022 School District Geographic Relationship Files.

Empirical Strategy

To examine the predictors of AY2020-21 in-person instruction rates, we estimate eight variants of the following ordinary least squares (OLS) regression specification:

Y_{cs} = β X_{cs} + θ_{s} + ε_{cs},

(1)

where $Y$ is the rate of in-person instruction (either for the week of October 2, 2020 or the weekly average from August 14, 2020 to June 25, 2021) in county $c$ and state $s$ , $X$ is a vector of county-level predictors, $θ$ are state fixed effects, and $ε$ is the error term. To facilitate comparisons between the coefficients in Equation 1, we standardize each predictor in $X$ to a mean of zero and a standard deviation of one. In the Appendix, we disaggregate the extension analyses by season: fall, winter, and spring (see Tables A1, A2, and A3).

The eight variants of this specification feature each possible combination of the following: state fixed effects, county-level population weights, and our alternative measure of teachers’ union strength. The inclusion of state fixed effects and/or county population weights should not be conceived of as robustness checks on the same basic model. A crucial methodological concern for both our analyses and for all of the previous analyses that we seek to replicate is that the relevant independent variables—Covid case and death rates, partisan vote share, teachers’ union strength, etc.—could not plausibly be considered randomly distributed, even under strong assumptions regarding appropriate controls for observable differences. In the absence of exogenous variation, these permutations in model specification do not bring us closer to or further from an optimal specification that identifies unbiased estimates of the causal effects. Rather, each specification poses a slightly different observational research question. By incorporating state fixed effects into the model, we alter the analysis to capture the relationships between each of the various predictors and the rate of in-person instruction, on average, within each state. In the absence of state fixed effects, the analysis captures the same relationships, on average, across the country as a whole—a substantively different, but no less important, line of inquiry. When we incorporate county population weights, the results of the analysis better reflect the experience of the average individual living in a county with the characteristics represented in the model. In the absence of these weights, each county—representing a community of any size—is the fundamental unit of analysis. Thus, each of these approaches focuses on a slightly different set of comparisons. By presenting the results of multiple models, we demonstrate that the inferences one draws are dependent on these analytic decisions.

Public Opinion and In-Person Instruction Rates

We also advance this line of inquiry by expanding the range of potential predictors of in-person instruction. To do so, we merge the county-level dataset described above with two ongoing, nationally representative surveys of Americans’ attitudes toward education issues that were conducted before and during AY2020-21: the annual Education Next (EN) poll and the quarterly Murmuration poll. We consider two elements of public opinion that were consistently polled in both surveys: public support for increased education spending in general and public support for increased teachers’ salaries in particular.

Education Next Poll

The EN poll is designed by the Harvard Program on Education Policy and Governance and administered by Ipsos Public Affairs via its KnowledgePanel^®, a nationally representative panel of American adults who agree to participate in a limited number of online surveys. Ipsos provides internet access and/or an appropriate device to KnowledgePanel^® members who lack the necessary technology to participate. For individual surveys—like the EN poll—Ipsos samples respondents from the KnowledgePanel^®. Respondents can elect to complete the EN poll in English or Spanish. The polls from 2016 to 2021 collect precise location information for each participant, allowing us to link respondents to other data based on their geography. The EN poll also collects detailed demographic information for each participant (parental status, race/ethnicity, gender, political party identification, political ideology, age, family income, and educational attainment).

Each year, every EN poll participant is asked about education spending and teachers’ salaries. For each item, the participant is randomly assigned to receive one of two possible questions. Regarding education spending, the two options are as follows:

“According to the most recent available information, $[INSERT VALUE] is being spent each year per child attending public schools in your district. Do you think that government funding for public schools in your district should increase, decrease, or stay about the same?” (Response options: greatly increase, increase, stay about the same, decrease, greatly decrease)

“Do you think that government funding for public schools in your district should increase, decrease, or stay about the same?” (Response options: greatly increase, increase, stay about the same, decrease, greatly decrease)

Regarding teachers’ salaries, the two options are as follows:

“Public school teachers in your state are paid an average annual salary of $[INSERT VALUE]. Do you think that public school teacher salaries should increase, decrease, or stay about the same?” (Response options: greatly increase, increase, stay about the same, decrease, greatly decrease)

“Do you think that public school teacher salaries should increase, decrease, or stay about the same?” (Response options: greatly increase, increase, stay about the same, decrease, greatly decrease)

We code both survey items dichotomously (“increase/greatly increase” coded 1, otherwise 0). To maximize sample size for each item, we combine responses from both questions into a single variable and create a separate dummy variable indicating whether or not the participant received the first version of the question. We control for this dummy variable in every EN poll analysis.

Murmuration Poll

The Murmuration National Polling Project is a quarterly, nationally representative survey of America’s registered voters. Each iteration of the Murmuration poll includes about 1,500 live-caller surveys administered to roughly half landlines and half cell phones, with bilingual callers for Spanish-speaking respondents. In addition to using live telephone callers, Murmuration collects responses from approximately 1,500 voters through pre-registered panels of online participants. Similar to the EN poll, The Murmuration poll also collects precise location and demographic information for each participant (parental status, race/ethnicity, gender, and political party identification).

In the Murmuration poll, the question about education spending reads: “Would you be willing to pay more local taxes if the money went to increase funding to improve the public schools in your local area?” (Response options: yes, no, don’t know). The question about teachers’ salaries reads: “Do you support or oppose your state and local governments using more taxpayer money to increase pay for public school teachers?” (Response options: support, oppose, don’t know). We code both survey items dichotomously (spending: “yes” coded 1, otherwise 0; salaries: “support” coded 1, otherwise 0).

Empirical Strategy

To examine the relationship between public support for education/educators and county-level rates of in-person instruction during AY2020-21, we estimate variants of the following OLS regression specification:

Y_{c, t = AY 2020 - 21} = β_{1} Suppor t_{ict} + β_{2} X_{ict} + β_{3} Z_{c, t = AY 2020 - 21} + π_{t} + ε_{ict},

(2)

where $Y$ is the mean weekly rate of in-person instruction in county $c$ during AY2020-21, $Support$ equals 1 if survey respondent $i$ located in county $c$ supports increased education spending/teachers’ salaries (and 0 otherwise) in survey period $t$ , $X$ is a vector of individual-level characteristics for survey respondent $i$ in county $c$ and survey period $t$ ; $Z$ is a vector of county-level characteristics for county $c$ measured in AY2020-21 or the most recent available data collection period, $π$ are survey fixed effects, and $ε$ is the error term. We cluster standard errors at the county level.

Please note that Equation 2 does not include state fixed effects. Neither the EN poll nor the Murmuration poll is designed to be representative at the state level, so within-state analyses generate potentially inaccurate representations of public opinion dynamics. This has important implications for our primary measure of local teachers’ union strength (the number of students enrolled in the largest district in the county), which only captures this construct well when considered within a given state. We also consider our alternative measure of teachers’ union strength—which is not dependent on the inclusion of state fixed effects—in the analyses to follow.

Findings

Replication Analysis

When describing the results of our replication analysis, our goal is not to report and interpret each coefficient of every model. Rather, we seek to identify general patterns that emerge across multiple analytical approaches. Table 1 displays the results of eight variants of the same basic analysis: with and without state fixed effects, with and without county population weights, as well as with and without an alternative measure of local teachers’ union strength (which is available for the 280 most populous counties, representing about two-thirds of the U.S. population).

Table 1.

Fall 2020 Replication Analysis.

	Outcome: % in–person (October 2, 2020)
	1	2	3	4	5	6	7	8
Covid cases	8.34* (1.02)	9.61* (1.20)	13.78* (4.51)	9.09* (4.13)	2.36* (0.88)	−3.76* (1.18)	−5.89 (5.12)	−14.33* (5.32)
Covid deaths	0.47 (1.01)	−2.81* (0.97)	−8.46* (3.42)	−6.58* (2.85)	−2.13* (0.83)	1.15 (1.04)	−2.53 (4.01)	1.21 (3.77)
Trump 2020	6.28* (0.88)	10.21* (0.99)	19.29* (3.55)	18.08* (4.34)	13.00* (2.10)	13.38* (1.84)	22.44* (7.85)	25.07* (7.78)
Student enrollment	−0.87 (1.43)	−1.46* (0.32)	−0.02 (1.61)	−1.26 (0.90)	−2.76* (1.14)	−3.19* (0.33)	−1.86 (1.54)	−3.27* (1.04)
Labor union membership			−3.72 (2.17)	−6.76* (2.64)			−0.37 (1.76)	0.21 (2.16)
Per–pupil spending	−1.30 (0.85)	−3.66* (0.77)	0.41 (2.04)	0.87 (1.93)	−0.69 (0.68)	4.05* (0.72)	4.17* (1.83)	4.83* (1.82)
Avg teacher salary	−1.80* (0.88)	−3.65* (0.60)	−1.47 (2.21)	−3.43 (2.12)	0.25 (1.53)	−1.55 (1.11)	0.78 (4.20)	−1.12 (3.63)
Avg test scores	15.82* (5.39)	38.55* (4.97)	47.26* (19.29)	30.11 (17.48)	−5.42 (4.86)	−9.59 (4.92)	−17.90 (19.78)	−44.40* (18.00)
Private market share	−2.62* (0.83)	−2.99* (0.82)	−6.92* (3.06)	−8.80* (2.97)	0.24 (0.66)	−0.54 (0.70)	−3.33 (2.72)	−8.40* (2.82)
Charter market share	−0.85 (0.85)	2.15* (0.65)	0.51 (2.07)	0.15 (1.94)	−0.38 (0.68)	−3.35* (0.64)	−3.51 (2.02)	−5.64* (2.03)
White, Non–Hispanic	−0.91 (1.33)	2.02 (1.18)	2.36 (4.59)	2.78 (3.97)	2.25 (1.45)	8.27* (1.34)	5.54 (5.01)	4.07 (4.68)
Median HH income	−8.01* (1.50)	−5.40* (1.21)	−3.93 (4.09)	−4.43 (3.77)	−2.46 (1.30)	−0.36 (1.14)	0.51 (3.87)	2.03 (3.53)
Other covariates	X	X	X	X	X	X	X	X
State fixed effects					X	X	X	X
Population weights		X		X		X		X
$R^{2}$	.10	.23	.35	.36	.52	.59	.71	.73
Adjusted $R^{2}$	.09	.23	.30	.31	.50	.58	.62	.65
$N$ (Counties)	3,080	3,080	261	261	3,080	3,080	261	261

Note. Each cell reports OLS coefficients with robust standard errors in parentheses (clustered at the county level); other covariates include county population, population density, percentage under age 18, percentage of adults with college degrees, percentage of school–age children in school, Gini index, and the number of school districts in the county; all predictors are standardized ( $μ = 0, σ = 1$ ).

p < .05.

The first and most important finding is just how sensitive the results are to model specification. Depending on the model, factors that other scholars have identified as important predictors of initial in-person instruction rates fluctuate from positive to negative, significant to non-significant. This is not to suggest that there are no meaningful predictors of in-person instruction in fall 2020. Rather, we seek to impress upon the reader the extent to which minor analytic decisions—which may have nontrivial implications for the interpretation of the results—are important in the pursuit of this line of inquiry. We describe the general patterns that we observe below and conclude with our synthesis and interpretation of those patterns.

Some of the prior research suggests that Covid cases and deaths were essentially unrelated to in-person instruction in fall 2020. This is not, in fact, what we observe. In a county-level analysis (but omitting county population weights, thereby making “county” the simple unit of analysis—i.e., Model 1 in Table 1), a one standard deviation increase in the Covid case rate is associated with, on average, an 8.34 percentage-point increase in the in-person instruction rate, after adjusting for a range of other potentially relevant factors. When we include county population weights (which places greater emphasis on more populous counties, making the results more representative of the average student’s experience) and/or restrict our analysis to the most populous counties, this relationship increases in magnitude (mean $β$ in Models 2–4: 10.83). However, when we introduce state fixed effects (thereby altering our analysis to consider the average relationship between Covid case rates and in-person instruction rates within states), this relationship is generally negative (mean $β$ in Models 5–8: −5.41). This pattern suggests that, overall, counties in states with higher Covid case rates also had higher in-person instruction rates. However, within a given state, counties with higher Covid case rates had lower in-person instruction rates.

The relationship between Covid deaths and in-person instruction follows a different pattern. In the absence of state fixed effects, there is generally a negative relationship between Covid death rate and in-person instruction (mean $β$ in Models 1–4: −4.35). However, this finding only holds when thinking about the country as a whole. Our within-state analyses generate mostly non-significant results.

Our replication analysis resoundingly affirms the prior literature’s conclusion that political partisanship was a reliable predictor of in-person instruction in fall 2020. Regardless of model specification, county-level Trump 2020 presidential election vote share is positively and consistently related to in-person instruction rates (mean $β$ in Models 1–8: 15.97). More Republican counties were more likely to re-open for in-person instruction in fall 2020—both within a given state and across the country, both when placing equal weight on all counties and when placing greater weight on more populous counties.

We also observe a relatively consistent relationship between local teachers’ union strength and in-person instruction rates. Our first approach employs the number of students in the largest district by enrollment within the county. As previously discussed, this measure is most effective at capturing differences in the relative influence of teachers’ unions between communities in the same state. Accordingly—and in line with the prior literature—this measure reveals a consistently negative relationship between teachers’ union strength and in-person instruction rates in the models that employ state fixed effects (mean $β$ in Models 5–8: −2.77). In the absence of state fixed effects, this relationship is modest in size and typically non-significant.

It is important to note that district size in itself may also have its own relationship with the rate of in-person instruction, distinct from the predictive power of teachers’ union strength, for which it serves as a proxy. With this in mind, we employ a second measure of teachers’ union strength: the percentage of employees who belong to any labor union. This measure offers a more precise indicator of local labor union power, albeit not limited to teachers. While it is only available for the 280 most populous counties in the country, these counties constitute 64% of the U.S. population. Most states include at least one of these populous counties, but the geographic dispersion makes within-state analyses less relevant. Models 3–4 indicate that higher union membership rates are also negatively related to in-person instruction rates, although the magnitude of the coefficient exceeds the conventional threshold for statistical significance in Model 4 only (mean $β$ in Models 3–4: −5.24).

There is less prior research on the relationship between educational inputs/outputs and initial in-person instruction rates in fall 2020 (although such factors are often included as control variables in other researchers’ analyses). Within a given a state, we generally observe a positive relationship between per-pupil spending and in-person instruction rates (mean $β$ in Models 5–8: 3.09). However, this pattern does not hold across the country as a whole. We do not observe a consistent relationship between average teachers’ salaries and in-person instruction rates. With respect to student achievement, counties with higher average test scores had higher in-person instruction rates (mean $β$ in Models 1–4: 32.94). Yet, once we look within states, this relationship reverses and is not always statistically significant (mean $β$ in Models 5–8: −19.33).

In contrast to prior findings, we note that counties with larger private school market shares generally had lower rates of in-person instruction—although this finding, like many of the others, is dependent on the model specification (mean $β$ in Models 1–8: −4.17). Similarly, the direction and statistical significance of the relationship between charter school market share and in-person instruction rates both vary from model to model (mean $β$ in Models 1–8: −1.37).

Lastly, we also consider how county demographics—specifically race and income—are related to fall 2020 rates of in-person instruction. After adjusting for all of the aforementioned factors, neither county-level percentage of non-Hispanic White residents nor median household income are consistently related to school re-opening status.

In short, the results of our replication analysis are largely aligned with two of the major findings of the prior literature: both political partisanship and local teachers’ union strength are meaningful predictors of fall 2020 in-person instruction rates. In the case of political partisanship, the magnitude of the coefficient is reliably larger than the coefficients on other independent variables. However, we also add valuable nuance to one of the broad conclusions of the existing research: that initial re-opening decisions were primarily about politics and not the threat of the pandemic. When comparing counties in the same state—in other words, thinking about the decisions of educational leaders all subject to the same statewide policies and procedures—counties with higher Covid case rates had lower in-person instruction rates after adjusting for many other potentially relevant factors. This suggests that, within the constraints set at the state level, local educational leaders appear to have incorporated both political and public health concerns into their decision-making processes. Moreover, when comparing counties across the whole country—an empirical strategy that may also reveal state or regional differences unaccounted for by county-level covariates—counties with higher Covid death rates also had lower in-person instruction rates. Since we generally do not observe the same pattern within states, this suggests that large swaths of the country that disproportionately suffered high death tolls during the first wave of the pandemic were less likely to re-open schools for in-person instruction in fall 2020. This dynamic also suggests that local educational leaders were attending to both political and public health concerns.

Extension Analysis

One of the limitations of the existing research on school re-opening decisions is that most published studies focus on educational leaders’ initial decisions at the beginning of AY2020-21. However, the decision to re-open schools for in-person instruction was not a discrete choice at a single point in time; it was updated over and over again as the school year progressed. In this section, we report the results of our analysis of the predictors of county-level mean weekly in-person instruction rates over the course of the entire academic year. Table 2 displays these estimates. When describing the results of our extension analysis, we focus primarily on instances where the general patterns clearly diverge from those observed in the replication analysis.

Table 2.

AY2020-21 Extension Analysis.

	Outcome: Weekly average % in–person (AY2020–21)
	1	2	3	4	5	6	7	8
Covid cases	4.31* (0.62)	7.92* (0.69)	7.20* (2.73)	9.62* (2.22)	2.26* (0.50)	3.27* (0.59)	1.33 (2.62)	0.14 (2.29)
Covid deaths	5.26* (0.64)	−0.18 (0.92)	−4.08 (3.57)	−11.98* (3.09)	1.25* (0.50)	0.92 (0.74)	−4.82 (3.34)	−7.80* (3.08)
Trump 2020	4.24* (0.58)	7.33* (0.70)	11.46* (2.63)	13.80* (2.97)	8.74* (1.33)	9.41* (1.16)	12.31* (5.26)	18.01* (4.75)
Student enrollment	1.04 (0.95)	−0.81* (0.23)	−0.22 (1.21)	−1.81* (0.63)	−0.98 (0.72)	−1.60* (0.21)	−1.11 (1.03)	−2.68* (0.62)
Labor union membership			−4.05* (1.60)	−6.32* (1.82)			−0.57 (1.16)	0.41 (1.29)
Per–pupil spending	−1.32* (0.56)	−6.48* (0.54)	−3.39* (1.50)	−3.58* (1.32)	−0.41 (0.43)	−0.97* (0.45)	−0.43 (1.20)	−0.79 (1.08)
Avg teacher salary	−0.73 (0.59)	−4.81* (0.43)	−2.00 (1.64)	−4.61* (1.47)	1.32 (0.97)	−0.98 (0.71)	2.83 (2.75)	−0.25 (2.20)
Avg test scores	13.53* (3.61)	48.69* (3.47)	62.50* (13.83)	78.04* (11.76)	−1.32 (3.08)	16.79* (3.10)	30.80* (12.89)	34.04* (10.69)
Private market share	−2.08* (0.55)	−1.21* (0.58)	−2.35 (2.25)	−0.72 (2.03)	−0.37 (0.42)	0.60 (0.44)	1.37 (1.79)	−0.17 (1.67)
Charter market share	0.13 (0.54)	3.45* (0.46)	3.57* (1.53)	3.88* (1.34)	0.36 (0.41)	0.75 (0.40)	1.05 (1.33)	0.55 (1.21)
White, Non–Hispanic	−0.34 (0.85)	−0.93 (0.81)	0.47 (3.34)	−2.89 (2.56)	3.58* (0.88)	5.20* (0.81)	4.34 (3.15)	−1.25 (2.58)
Median HH income	−6.97* (1.00)	−7.02* (0.85)	−5.06 (3.03)	−5.10 (2.59)	−3.01* (0.82)	−1.32 (0.70)	−2.38 (2.56)	−0.57 (2.08)
Other covariates	X	X	X	X	X	X	X	X
State fixed effects					X	X	X	X
Population weights		X		X		X		X
$R^{2}$	.19	.47	.48	.65	.61	.78	.82	.89
Adjusted $R^{2}$	.19	.47	.44	.63	.60	.77	.76	.86
$N$ (Counties)	3,085	3,085	261	261	3,085	3,085	261	261

p < .05.

Notably, the relationship between political partisanship and the in-person instruction rate is reduced in size across nearly every model specification (mean $β$ in Models 1–8: 10.66). While the local partisan landscape may have been an influential guidepost for educational leaders’ initial decisions in fall 2020, its salience appears to have waned as the school year wore on. By contrast, the relationship between teachers’ union strength and in-person instruction remains relatively constant. When measured by student enrollment, the average coefficient in the models with state fixed effects declines modestly from −2.77 to −1.59. When measured by the labor union membership rate, the average coefficient in the relevant models is basically unchanged: from −5.24 to −5.19.

A different dynamic emerges with respect to average student achievement. In the replication analysis, average test scores are positively associated with in-person instruction rates across the country as a whole, but not in the within-state analyses. In the extension analysis, we observe this relationship in nearly every model. Moreover, the magnitude of the relationship is substantively large. On average across all eight models, a one standard deviation increase in average test scores is associated with about a 35 percentage-point increase in the in-person instruction rate during the school year. In fall 2020, higher-scoring states were more likely to re-open for in-person instruction after adjusting for other factors, but this pattern did not translate to higher-scoring counties within a given state tending toward more in-person instruction. As the school year continued, however, counties with a history of higher test scores were more likely to have students return to their classrooms—no matter the geographic lens through which one analyzes the data.

In our replication analysis, the relationship between per-pupil spending and in-person instruction rates is modest in magnitude and inconsistent. When considering AY2020-21 in its entirety, we note that, after adjusting for other potentially relevant factors, higher spending counties typically had lower in-person instruction rates. Yet this pattern only holds across the country as a whole (mean $β$ in Models 1–4: −3.69), not within states (mean $β$ in Models 5–8: −0.65). This suggests that relatively high-spending states were less likely to re-open for in-person instruction during AY2020-21.

Tables A1, A2, and A3 in the Appendix break down the extension analysis seasonally for fall (August 2020 to November 2020), winter (December 2020 to February 2021), and spring (March 2020 to June 2020). This allows us to observe the trajectory of these changes with greater clarity. The reduction in the magnitude of the relationship between political partisanship and in-person instruction is approximately linear (mean $β$ in Models 1–8 in fall: 15.57; winter: 11.92; spring: 4.81). Meanwhile, most of the growth in the relationship between student achievement and in-person instruction had already occurred by winter (mean $β$ in Models 1–8 in fall: 18.57; winter: 40.75; spring: 48.18), indicating that the advantages of in-person instruction began accruing disproportionately to students in already high-performing districts relatively early in AY2020-21.

Public Opinion and In-Person Instruction Rates

By extending our analysis of AY2020-21 school re-opening decisions through the entirety of the school year, we provide a more comprehensive picture of the educational, social, economic, political, and public health factors that were associated with in-person instruction rates during this pivotal time. But there are more general limitations to this type of analysis that cannot be remedied merely by expanding the timeline. The approach that appears in much of the prior literature and that we have also employed thus far tells us relatively little about the factors that made in-person instruction administratively manageable and politically feasible in the face of the significant, multiple, and overlapping demands of pandemic-era schooling. In this section, we report the results of a novel empirical strategy that merges the county-level data described above with geo-coded public opinion data collected before and during AY2020-21.

We begin by briefly describing the unadjusted individual-level and county-level predictors of support for increased teachers’ salaries (see Tables A4 and A5 in the Appendix). Individuals who support raising teachers’ salaries are more likely be parents/caretakers of young children, members of a racial/ethnic group other than white, females, Democrats, liberals, younger, more affluent (peaking in the $50,000–$100,000 range), and to have completed more years of formal schooling. Such individuals are also more likely to live in counties with a lower Covid death rate, a lower Trump vote share in 2020, a lower labor union membership rate, a lower average teachers’ salary, a higher charter school market share, and a lower proportion of white residents. In all subsequent analyses, we include controls for these factors as well as other variables described in our Data and Methods section.

The results displayed in Table 3 document the relationship between individual-level support for increased teachers’ salaries as measured in the annual EN poll and local in-person instruction rates. Models 1–6 display the results of a series of analyses that merge a single year of EN poll data with the pandemic-era county characteristics described above. For example, in Model 1 we observe that, on average, an individual who supported increased teachers’ salaries in 2016 lived in a county that went on to have about a three percentage-point higher AY2020-21 in-person instruction rate than their contemporaries who did not support increased teachers’ salaries—even after adjusting for a variety of individual-level and county-level characteristics. This relationship is remarkably consistent regardless of whether one considers pre-pandemic polling data (2016–19) or pandemic-era polling data (2020–21). Given the relatively time-invariant nature of this relationship, we pool these six survey waves together to generate a more precise estimate of this relationship (see Model 7, in which the relevant coefficient is modestly reduced to 2.56). Due to the increased sample size of this pooled analysis, we are also able to incorporate our alternative measure of local teachers’ union strength (see Model 8, in which the relevant coefficient is again modestly reduced to 2.20).

Table 3.

Support for Increased Teacher Salaries and Local AY2020-21 In-Person Rates—EN Poll.

	Outcome: Weekly average % in-person (AY2020-21)
	1	2	3	4	5	6	7	8
Support increased teacher salaries	3.38* (1.32)	1.95* (0.89)	3.38* (0.97)	2.20* (1.06)	2.17* (0.93)	3.58* (1.62)	2.56* (0.68)	2.20* (0.83)
Covariates	X	X	X	X	X	X	X	X
Survey FE							X	X
Union M’ship								X
Time period(s)	2016	2017	2018	2019	2020	2021	All years	All years
$R^{2}$	.49	.48	.48	.49	.49	.48	.48	.66
Adjusted $R^{2}$	.48	.48	.47	.48	.48	.46	.48	.66
$N$ (Respondents)	1,982	3,851	4,253	2,875	4,079	1,346	18,386	8,018
$N$ (Counties)	779	1,122	1,150	905	1,074	621	1,790	261

Note. Each cell reports OLS coefficients with robust standard errors in parentheses (clustered at the county level); county-level covariates include Covid cases, Covid deaths, Trump vote share, student enrollment, per-pupil spending, average teacher salary, average test scores, private market share, charter market share, percent non-Hispanic White, median household income, population, population density, percentage under age 18, percentage of adults with college degrees, percentage of school-age children in school, Gini index, and the number of school districts in the county; individual-level covariates include parent status, race/ethnicity, gender, political party, political ideology, age, family income, educational attainment, and EN poll question type.

p < .05.

One might be concerned that this relationship is an artifact of the question wording or sampling strategy of the EN poll. We replicate our analyses using data from the quarterly Murmuration poll, which features a nationally representative sample of registered voters rather than all adults and which asks about attitudes toward teachers’ salaries with a distinctly different question. These results can be found in Table A6 in the Appendix. While we observe more variation in the magnitude of the relationship between support for increased teachers’ salaries and local in-person instruction rates when focusing on different quarterly Murmuration poll data, the results of the two pooled analyses (Models 7–8) are consistent with their EN poll counterparts (with coefficients of 2.63 and 2.56, respectively).

We also present these results visually in Figure 1, which displays the unadjusted differences over time in support for increased teachers’ salaries between individuals who live in counties where schools were mostly closed during AY2020-21 (<50% in-person on average) and individuals who live in counties where schools were mostly open during AY2020-21 (≥50% in-person on average). Even without statistical adjustments, this relationship is vivid using the EN poll data as well as visually detectable using the Murmuration poll data.

In short, there appears to be something noteworthy about communities with greater support for increased teachers’ salaries that may have facilitated higher rates of AY2020-21 in-person instruction, but which is conceptually distinct from the other factors known to be associated with pandemic-era school re-opening decisions. We speculate that communities with a deeper well of public support for educators—even if those communities faced an equivalent threat from Covid, share a common political persuasion, feature a comparable demographic composition, or support a local public school system that is similar across multiple dimensions—were better able to navigate the many inter-connected and overlapping challenges inherent in the task of bringing students back into school buildings during the first full school year of the pandemic. The causal direction here, however, is not entirely clear. It could also be the case that a history of strong educational leadership that predated the pandemic both cultivated robust community support and facilitated the difficult task of re-opening schools.

The relationship that we document above is specific to attitudes about teachers’ salaries. We also conduct a parallel set of analyses that explore the relationship between support for increased education spending in general and local AY2020-21 in-person instruction rates (see Tables A7 and A8 in the Appendix). Across a variety of model specifications, this analogous relationship is reduced in size and, with few exceptions, non-significant. It is not, therefore, the case that communities with greater generalized public support for public education that were more likely to return to in-person instruction. Rather, communities with greater public support for public educators were more likely to bring students back into the classroom.

Lastly, we consider the extent to which the relationship between support for increased teachers’ salaries and in-person instruction rates varies along three theoretically important dimensions: Covid case rates, political partisanship, and average teachers’ salaries. The first two heterogeneity analyses are motivated by the core findings from our replication and extension analyses that both politics and the threat of the virus to public health factored into school re-opening decisions. It is therefore important to investigate whether public support for educators potentially played a mediating role in either of these two dynamics. The third heterogeneity analysis is motivated by the nature of our measure of public support for educators. We seek to understand whether the relationship between support for increased teachers’ salaries and in-person instruction rates is particularly pronounced in communities with relatively high or low teacher compensation. Table 4 displays the results of a series of analyses in which the dichotomous indicator for support for increased teachers’ salaries is sequentially interacted with the measures representing these three other factors.

Table 4.

Heterogeneity Analyses.

	Outcome: Weekly average % in-person (AY2020-21)
	1	2	3	4	5	6
Support increased salaries (0/1)	2.55* (0.68)	3.57* (0.75)	3.43* (0.79)	2.65* (0.59)	2.83* (0.66)	2.83* (0.69)
Covid cases	5.84* (1.52)	6.33* (1.47)	6.29* (1.46)	7.92* (1.46)	8.35* (1.42)	8.34* (1.41)
Trump 2020	7.97* (1.40)	6.46* (1.40)	7.95* (1.40)	7.58* (1.22)	7.18* (1.24)	7.57* (1.22)
Average teachers’ salary	−4.89* (1.38)	−4.88* (1.38)	−4.20* (1.39)	−3.90* (1.32)	−3.89* (1.32)	−3.66* (1.38)
Support × Covid cases	0.74 (0.82)			0.56 (0.64)
Support × Trump 2020		2.31* (0.82)			0.54 (0.75)
Support × average salary			−1.25* (0.51)			−0.34 (0.46)
Other covariates	X	X	X	X	X	X
Survey FE	X	X	X	X	X	X
Poll	EN	EN	EN	Murm.	Murm.	Murm.
Time periods	2016–2021	2016–2021	2016–2021	Sp19-Wi21	Sp19-Wi21	Sp19-Wi21
$R^{2}$	.48	.48	.48	.46	.46	.46
Adjusted $R^{2}$	.48	.48	.48	.46	.46	.46
$N$ (Respondents)	18,386	18,386	18,386	13,267	13,267	13,267
$N$ (Counties)	1,790	1,790	1,790	2,008	2,008	2,008

Note. Each cell reports OLS coefficients with robust standard errors in parentheses (clustered at the county level); county-level covariates include Covid deaths, student enrollment, per-pupil spending, average test scores, private market share, charter market share, percent non-Hispanic White, median household income, population, population density, percentage under age 18, percentage of adults with college degrees, percentage of school-age children in school, Gini index, and the number of school districts in the county; individual-level covariates include parent status, race/ethnicity, gender, political party, political ideology, age, family income, educational attainment, and EN poll question type (for EN poll analyses); all county-level predictors are standardized ( $μ = 0, σ = 1$ ).

p < .05.

We observe two instances of heterogeneity across these three dimensions. When analyzing data from the EN poll, it appears that the magnitude of the relationship between support for increased teachers’ salaries and in-person instruction rates is greater for individuals in counties with a higher Trump vote share in the 2020 presidential election, after adjusting for other county-level characteristics. In other words, in Republican-leaning communities, greater public support for educators may have played a more prominent role in facilitating the return to in-person instruction. Conversely, it may also be the case that, in Democratic-leaning communities, the relative absence of public support for educators further reduced the likelihood of bringing students back into their classrooms. We adjudicate between these two potential interpretations empirically in Table A9, where we interact support for higher teachers’ salaries with quartiles of county-level Trump vote share. We find that the positive relationship is driven entirely by individuals living in Republican-leaning communities, especially among those living in counties in the highest quartile of Trump vote share (in which the magnitude of the combined coefficient is 5.46).

The second instance of heterogeneity is rooted in local average teachers’ salaries. In the EN data, the magnitude of the relationship between support for increased teachers’ salaries and in-person instruction rates is greater for individuals in communities with below-average salaries, controlling for other county-level characteristics. We also break down this analysis using quartiles of local average teachers’ salaries (Table A9). The positive relationship is driven by individuals living in communities in the bottom three quartiles of average teachers’ salaries. The relationship is essentially reduced to zero among individuals living in communities at the upper range of teacher compensation.

Despite the potential theoretical importance of these patterns of heterogeneity, we are hesitant to draw strong inferences from these analyses. The patterns noted above only appear when analyzing the EN poll data. They do not replicate in the analogous analyses with Murmuration poll data.

Conclusion

Why did some communities’ public schools re-open for in-person instruction during AY2020-21 and others did not? By re-analyzing initial decisions in fall 2020 using a continuous measure of county-level in-person instruction rates and by extending this analysis to incorporate the entire 2020–21 school year, we add valuable nuance to the prevailing narrative that school re-opening decisions primarily reflected political partisanship rather than the local risks of Covid. We reaffirm the previous finding that local political partisanship was an important predictor of fall 2020 school re-opening decisions (DeAngelis & Makridis, 2021; Grossmann et al., 2021; Haderlein et al., 2021; Harris & Oliver, 2021; Hartney & Finger, 2022; Valant, 2020). We also demonstrate that Covid case and death rates played an important role in school re-opening decisions, consistent with recent evidence from Ohio (Christian et al., 2022). When controlling for state-of-residence and when placing greater weight on more populous communities, counties with higher Covid case rates tended to have lower in-person instruction rates. In other words, within a given state, the average student living in a county with greater exposure to the virus was less likely to attend school in-person. Moreover, across the country as a whole—but not within states—counties with higher Covid death rates were also less likely to return to in-person instruction. This suggests that entire states and regions that experienced the most severe effects of the first wave of the pandemic subsequently had lower in-person instruction rates. These findings attest to the fact that educational leaders responded to both political and public health pressures when deciding to keep schools closed or re-open them for in-person instruction. Lastly, in contrast to some previous research, we do not observe a consistent relationship between local racial/ethnic and economic compositions and in-person instruction rates after adjusting for other county-level characteristics (Camp & Zamarro, 2022; Grossmann et al., 2021; Haderlein et al., 2021; Harris & Oliver, 2021; Hartney & Finger, 2022).

As AY2020-21 proceeded, the decision-making context repeatedly shifted. Covid positivity rates during the first wave of the pandemic peaked in January 2021 (Moser, 2021). Only a few months later, vaccines became widely available to the public (Weintraub & Weise, 2021). The risks posed by the pandemic were in constant flux. Much of the prior literature is focused on school districts’ initial re-opening decisions in fall 2020, but there is less evidence available regarding how the predictors of in-person instruction changed over time. We build on this body of research by conducting a similar set of analyses that alters the dependent variable to capture the average weekly in-person instruction rate during AY2020-21. We find that the magnitude of the relationship between political partisanship and in-person instruction is reduced when considering the entire academic year. Instead, we begin to see a pattern in which communities with a history of higher standardized test scores were notably more likely to re-open school buildings than their lower-achieving counterparts. This pattern may help us understand the widening test score gaps that have emerged in the wake of the pandemic (Kuhfeld et al., 2022; NAEP, 2022a, 2022b, 2022c).

One of our central arguments rests on the acknowledgment that many minor technical decisions that researchers make when analyzing their data—often in good faith and with reasonable justifications—can have nontrivial consequences for the interpretation of the results. Our analysis is not immune from the same general critique. Our choices to focus on county as the primary unit of analysis (rather than school, district, or state) and to rely on Burbio’s measures of in-person instruction (as opposed to some of the alternative data sources available) undoubtedly have a non-zero bearing on our estimates of the relationships that we present here. For example, our use of the Burbio data makes it difficult to examine re-opening dynamics in the country’s most sparsely populated rural communities. Our objective is not to offer the sole, definitive account of the predictors of in-person instruction during the 2020–21 academic year. That goal exceeds the upper limits of possibility for a single research project. Rather, our contribution adds valuable complexity and nuance to the existing literature.

We also seek to broaden the conceptualization of politics that we apply to this question. We merged our data on in-person instruction rates with two ongoing, nationally representative public opinion polls that were conducted before and during the time period in question. We find that pre-pandemic public attitudes about teachers’ salaries have a modest positive relationship with AY2020-21 in-person instruction rates. Specifically, the average U.S. adult who thinks that teachers ought to be paid more is also more likely to live in a county where kids were learning in-person for a larger proportion of the school year. This relationship is remarkably consistent across different sampling strategies, date (and even year) of survey administration, and question wording. There appears to be something noteworthy about places with greater support for local educators that may have contributed to higher rates of in-person instruction. Given the enormous challenges of in-person schooling during a pandemic, we speculate that such environments lowered the social, economic, and political transaction costs of bringing students and teachers back to their classrooms. Another potential explanation—which is not necessarily mutually exclusive with the first—is that local educational leaders with exceptional administrative and political skills had both A) a well-earned history of community support and B) greater success at re-opening schools in a challenging context. These proposed mechanisms are speculative. Our data do not include the information on educational leaders’ perceptions of community support necessary to test these hypotheses empirically. Additional research—both quantitative and qualitative—is necessary to provide greater clarity regarding the complex interplay between educational leaders and local communities during times of crisis.

Footnotes

Appendix

Table A9.

Heterogeneity Analysis Extension.

	Outcome: Weekly average % in-person (AY2020-21)
	1	2
Support increased salaries (0/1)	−0.55 (1.18)	2.65* (1.17)
Covid cases	6.24*(1.45)	5.27* (1.24)
Trump 2020 (continuous)		7.98* (1.37)
Trump 2020 (second quartile)	−5.21 (3.60)
Trump 2020 (third quartile)	1.53 (3.32)
Trump 2020 (fourth Quartile)	12.76* (2.89)
Average teachers’ salary (continuous)	−4.55* (1.16)
Average teachers’ salary (second quartile)		−0.92 (2.82)
Average teachers’ salary (third quartile)		−21.29* (3.38)
Average teachers’ salary (fourth quartile)		−25.86* (3.88)
Support × Trump 2020 (second quartile)	0.68 (1.62)
Support × Trump 2020 (third quartile)	3.08 (1.76)
Support × Trump 2020 (fourth quartile)	6.01* (1.61)
Support × average teachers’ salary (second quartile)		−1.21 (1.66)
Support × average teachers’ salary (third quartile)		0.29 (1.66)
Support × average teachers’ salary (fourth quartile)		−4.01* (1.42)
Other covariates	X	X
Survey FE	X	X
Poll	EN	EN
Time periods	2016–2021	2016–2021
$R^{2}$	.49
Adjusted $R^{2}$	.49
$N$ (Respondents)	18,386	18,386
$N$ (Counties)	1,790	1,790

Note. Each cell reports OLS coefficients with robust standard errors in parentheses (clustered at the county level); county-level covariates include Covid deaths, student enrollment, per-pupil spending, average test scores, private market share, charter market share, percent non-Hispanic White, median household income, population, population density, percentage under age 18, percentage of adults with college degrees, percentage of school-age children in school, Gini index, and the number of school districts in the county; individual-level covariates include parent status, race/ethnicity, gender, political party, political ideology, age, family income, educational attainment, and EN poll question type; all county-level predictors are standardized ( $μ = 0, σ = 1$ ); second, third, and fourth quartiles compared to the first quartile.

p < .05.

Acknowledgements

The authors thank the Harvard Program on Education Policy and Governance and Murmuration for providing access to the Education Next Poll and the Murmuration National Polling Project data. We would also like to thank the participants of the “Exploring Drivers of Education Policy Decisions” panel at the Association for Education Finance and Policy conference as well as the participants of the George Mason University EdPolicyForward Research Workshop for their thoughtful and constructive feedback.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

David M. Houston

Author Biographies

David M. Houston is an Assistant Professor of Education at George Mason University. He studies education politics, governance, and public opinion.

Matthew P. Steinberg, PhD, is the Managing Director of Research and Evaluation at Accelerate, a recently launched national non-profit focused on building the nation's knowledge of the educational strategies and practices that significantly improve student learning. His research focuses on educator evaluation and human capital, school safety and student discipline, accountability and urban school reform, and school finance.

References

American Academy of Pediatrics, American Academic of Child and Adolescent Psychiatry, & Children’s Hospital Association. (2021). AAP-AACAP-CHA declaration of a national emergency in child and adolescent mental health. American Academy of Pediatrics. https://www.aap.org/en/advocacy/child-and-adolescent-healthy-mental-development/aap-aacap-cha-declaration-of-a-national-emergency-in-child-and-adolescent-mental-health/

Bacher-Hicks

Goodman

Green

J. G.

Holt

M. K.

(2022). The COVID-19 pandemic disrupted both school bullying and cyberbullying. AER: Insights, 4(3), 353–370.

Barnum

(2021). Polls show most—though not all—parents are getting the type of instruction they want for their kids. Chalkbeat. https://www.chalkbeat.org/2021/2/18/22289735/parents-polls-schools-opening-remote.

Bodenheimer

(2022). Hoe school closures made me question my progressive politics. Politico. https://www.politico.com/news/magazine/2022/01/11/oakland-lefty-my-whole-life-school-closures-triggered-an-identity-crisis-526860.

Burbio. (2022). Burbio’s school opening tracker methodology. Author. https://about.burbio.com/school-opening-tracker-methodology.

Cain Miller

Sanger-Katz

. (2022). School is back in person, but the five-day school week often isn’t. New York Times. https://www.nytimes.com/2022/02/28/upshot/schools-covid-closings.html.

Camp

A. M.

Zamarro

(2022). Determinants of ethnic differences in school modality choices during the COVID-19 crisis. Educational Researcher, 51(1), 6–16.

Cash

Brinkmann

Price

(2022). Superintendents’ leadership during the pandemic. In Marshall

D. T.

(Ed.), COVID-19 and the classroom: How schools navigated the great disruption. Rowman & Littlefield, pp. 123–144.

Centers for Disease Control and Prevention. (2021). Science brief: Transmission of SARS-CoV-2 in K-12 schools and early care and education programs—updated. Centers for Disease Control and Prevention. https://stacks.cdc.gov/view/cdc/104303.

10.

Chait

(2022). School closures were a catastrophic error. Progressives still haven’t reckoned with it. New York Magazine. https://nymag.com/intelligencer/article/progressives-must-reckon-with-the-school-closing-catastrophe.html.

11.

Christian

Jacob

Singleton

J. D.

(2022). Assessing school district decision-making: Evidence from the COVID-19 pandemic (EdWorking Paper 22-660). Annenberg Institute at Brown University. https://doi.org/10.26300/w9hv-n258.

12.

Cohodes

Pitts

(2022). A new COVID-19 data source for answering emerging pandemic questions. Center for Reinventing Public Education, https://crpe.org/wp-content/uploads/final-COVID-school-data-hub-blog-final.pdf.

13.

Collins

J. E.

(2021). Who wants to reopen schools in a pandemic? Explaining public preferences for reopening schools and public compliance with reopening orders during the COVID-19 pandemic (EdWorking Paper No. 21-448). https://www.edworkingpapers.com/sites/default/files/ai21448.pdf.

14.

DeAngelis

C. A.

Makridis

(2021). Are school reopening decisions related to union influence? Social Science Quarterly, 102(5), 2266–2284.

15.

Edgerton

A. K.

Ondrasek

Truong

O’Neal

(2021). New York City public schools: Supporting school reopening with a focus on testing and tracing. Learning Policy Institute. https://learningpolicyinstitute.org/product/safe-school-reopening-nyc-brief.

16.

Fahle

E. M.

Kane

T. J.

Patterson

Reardon

S. F.

Staiger

D. O.

Stuart

E. A.

(2023). School district and community factors associated with learning loss during the COVID-19 pandemic. The Center for Education Policy Research at Harvard University. https://cepr.harvard.edu/sites/hwpi.harvard.edu/files/cepr/files/explaining_covid_losses_5.23.pdf

17.

Gill

Dusseault

(2021). As urban districts prepare to reopen, most are not doing enough to communicate how they will keep students and teachers safe. Center for Reinventing Public Education. https://crpe.org/as-urban-districts-prepare-to-reopen-most-are-not-doing-enough-to-communicate-how-they-will-keep-students-and-teachers-safe/.

18.

Goldhaber

Kane

T. J.

McEachin

Morton

Patterson

Staiger

D. O.

(2022). The consequences of remote and hybrid instruction during the pandemic. Harvard University Center for Education Policy Research. https://cepr.harvard.edu/files/cepr/files/5-4.pdf?m=1651690491

19.

Gross

Opalka

Gundapaneni

(2021). School reopening trends offer districts the opportunity to start planning beyond the pandemic. Center for Reinventing Public Education. https://crpe.org/wp-content/uploads/final_rep_sample_brief_march_2021.pdf

20.

Grossmann

Reckhow

Strunk

K. O.

Turner

(2021). All states close but red districts reopen: The politics of in-person schooling during the COVID-19 pandemic. Educational Researcher, 50(9), 637–648.

21.

Haderlein

S. K.

Saavedra

A. R.

Polikoff

M. S.

Silver

Rapaport

Garland

(2021). Disparities in educational access in the time of COVID: Evidence from a nationally representative panel of American families. AERA Open, 7(1), 1–21.

22.

Halloran

Jack

Okun

J. C.

Oster

(2021). Pandemic schooling mode and student test scores: Evidence from US states (NBER Working Paper 29497). https://www.nber.org/papers/w29497

23.

Harris

D. N.

Oliver

D. M.

(2021). Why did so many public schools stay remote during the COVID crisis?REACH National Center for Research on Education Access and Choice. https://www.reachcentered.org/publications/why-did-so-many-public-schools-stay-remote-during-the-covid-crisis

24.

Hartney

M. T.

Finger

L. K.

(2022). Politics, markets, and pandemics: Public education’s response to COVID-19. Perspectives on Politics, 20(2), 457–473.

25.

Henderson

M. B.

Houston

D. M.

Peterson

P. E.

Shakeel

M. D.

West

M. R.

(2020). Amid pandemic, support soars for online learning, parent poll shows. Education Next. https://www.educationnext.org/amid-pandemic-support-soars-online-learning-parent-poll-shows-2020-education-next-survey-public-opinion/

26.

Henderson

M. B.

Houston

D. M.

Peterson

P. E.

West

M. R.

(2021). Parent poll reveals support for school covid-safety measures despite vaccine hesitancy, partisan polarization. Education Next. https://www.educationnext.org/parent-poll-reveals-support-school-covid-safety-measures-despite-vaccine-hesitancy-partisan-polarization/

27.

Henderson

M. B.

Peterson

P. E.

West

M. R.

(2020). Pandemic parent survey finds perverse pattern: Students are more likely to be attending school in person where covid is spreading more rapidly. Education Next. https://www.educationnext.org/pandemic-parent-survey-finds-perverse-pattern-students-more-likely-to-be-attending-school-in-person-where-covid-is-spreading-more-rapidly/

28.

Henig

J. R.

Jacobsen

Reckhow

(2019). Outside money in school board elections: The nationalization of education politics. Harvard Education Press.

29.

Hess

F. M.

(1999). Spinning wheels: The politics of urban school reform. Brookings Institution Press.

30.

Houston

D. M.

(2022). Polarization, partisan sorting, and the politics of education (EdWorking Paper No. 22-690). https://www.edworkingpapers.com/ai22-690

31.

Howell

W. G.

(2005). Besieged: School boards and the future of education politics. Brookings Institution Press.

32.

Kamenetz

Isensee

(2020). Most teachers concerned about in-person school; 2 in 3 want to start the year online. NPR. https://www.npr.org/2020/08/06/898584176/most-teachers-concerned-about-in-person-school-2-in-3-want-to-start-the-year-onl

33.

Kogan

(2021). What’s behind racial differences in attitudes toward school reopening (and what to do about them). American Enterprise Institute. https://www.aei.org/wp-content/uploads/2021/04/What%E2%80%99s-behind-racial-differences-in-attitudes-toward-school-reopening.pdf?x91208

34.

Kuhfeld

Soland

Lewis

Morton

(2022). The pandemic has had devastating impacts on learning. What will it take to help students catch up?Brookings Institution. https://www.brookings.edu/blog/brown-center-chalkboard/2022/03/03/the-pandemic-has-had-devastating-impacts-on-learning-what-will-it-take-to-help-students-catch-up/

35.

Lake

(2021). What a diverse group of 20 school districts are doing right in their COVID-19 reopening plans. Center for Reinventing Public Education. https://crpe.org/what-a-diverse-group-of-20-school-districts-are-doing-right-in-their-covid-19-reopening-plans/

36.

Lake

Dusseault

(2020). We reviewed the school reopening plans for 106 districts around the country. Here’s how they square with reality. Center for Reinventing Public Education. https://crpe.org/we-reviewed-the-school-reopening-plans-for-106-districts-around-the-country-heres-how-they-square-with-reality/

37.

Lasswell

H. D.

(1936). Politics: Who gets what, when, how. Whittlesey House.

38.

Leung

O’Neal

Ondrasek

Melnick

(2021). San Diego County: Mobilizing technical assistance, partnerships, and data to support school reopening. Learning Policy Institute. https://learningpolicyinstitute.org/product/safe-school-reopening-san-diego-brief

39.

Lewis

Kuhfeld

Ruzek

McEachin

(2021). Learning during COVID-19: Reading and math achievement in the 2020-21 school year. NWEA. https://www.nwea.org/uploads/2021/07/Learning-during-COVID-19-Reading-and-math-achievement-in-the-2020-2021-school-year.research-brief-1.pdf

40.

Mansoor

(2020). Dismissing ‘flexible’ CDC guidelines, education secretary Betsy DeVos doubles down on pushing schools to reopen. Time. https://time.com/5865987/betsy-devos-school-reopening-coronavirus/

41.

Marianno

B. D.

Hemphill

A. A.

Loures-Elias

A. P. S.

Garcia

Cooper

Coombes

(2022). Power in a pandemic: Teachers’ unions and their responses to school reopening. AERA Open, 8(1), 1–16.

42.

McCann

Dusseault

(2021). Districts quiet on how they plan to help students recover from a year of continued disruption. Center for Reinventing Public Education. https://crpe.org/districts-quiet-on-how-they-plan-to-help-students-recover-from-a-year-of-continued-disruption/

43.

Menasce Horowitz

. (2020). Republicans, Democrats differ over factors K-12 schools should consider in deciding whether to reopen. Pew Research Center. https://www.pewresearch.org/fact-tank/2020/08/05/republicans-democrats-differ-over-factors-k-12-schools-should-consider-in-deciding-whether-to-reopen/

44.

Milosh

Painter

Sonin

Van Dijcke

Wright

A. L.

(2021). Unmasking partisanship: Polarization undermines public response to collective risk. Journal of Public Economics, 204(204), 1–8.

45.

Moe

T. M.

(2005). Teachers unions and school board elections. In Howell

W. G.

(Ed.), Beseiged: School boards and the future of education politics. Brookings Institution Press, pp. 254–287.

46.

Moe

T. M.

(2009). Collective bargaining and the performance of the public schools. American Journal of Political Science, 53(1), 156–174.

47.

Moser

(2021). The deadliest month yet: January Saw one-fifth of all American COVID-19 deaths to date. The Atlantic. https://www.theatlantic.com/health/archive/2021/02/january-pandemic-deadliest-month/617898/.

48.

Murray

S. H.

(2022). Politics closed American schools last year. Now politics are keeping them open. The Week. https://theweek.com/coronavirus/1008970/politics-closed-american-schools-last-year-now-politics-are-keeping-them-open.

49.

National Assessment of Educational Progress. (2022a). Largest score declines in NAEP mathematics in grades 4 and 8 since initial assessments in 1990. U.S. Department of Education. https://www.nationsreportcard.gov/highlights/mathematics/2022/

50.

National Assessment of Educational Progress. (2022b). Scores decline in NAEP reading at grades 4 and 8 compared to 2019. U.S. Department of Education. https://www.nationsreportcard.gov/highlights/reading/2022/

51.

National Assessment of Educational Progress. (2022c). U.S. History score continues decline begun in 2014. U.S. Department of Education. https://www.nationsreportcard.gov/highlights/ushistory/2022/

52.

Nazaryan

(2020). The coronavirus closed schools. Our diseased politics is keeping them closed. Yahoo! News. https://www.yahoo.com/video/the-coronavirus-closed-schools-our-diseased-politics-has-kept-them-closed-170004563.html.

53.

Northern

A. M.

Scull

Zeehandelaar

(2012). How strong are US teacher unions? A state-by-state comparison. Thomas B. Fordham Institute. https://fordhaminstitute.org/national/research/how-strong-are-us-teacher-unions-state-state-comparison

54.

Ondrasek

Truong

Edgerton

A. K.

(2021). Marin County: Leveraging education and public health partnerships to support school reopening. Learning Policy Institute. https://learningpolicyinstitute.org/product/safe-school-reopening-marin-county-brief

55.

Rose

Sonstelie

(2010). School board politics, school district size, and the bargaining power of teachers’ unions. Journal of Urban Economics, 67(3), 438–450.

56.

Saul

(2021). How a school district got caught in Virginia’s political maelstrom. New York Times. https://www.nytimes.com/2021/11/14/us/loudoun-county-school-board-va.html.

57.

Singer

Marsh

J. A.

Menefee-Libey

Alonso

Bradley

Tracy

(2022). The politics of school reopening during COVID-19: A multiple case study of five urban districts in the 2020-21 school year (EdWorkingPaper No. 22-605). Annenberg Institute at Brown University. https://www.edworkingpapers.com/ai22-605.

58.

Stone

C. N.

(2001). Civic capacity and urban education. Urban Affairs Review, 36(5), 595–619.

59.

Strunk

K. O.

Reardon

S. F.

(2010). Measuring the strength of teachers’ unions: An empirical application of the partial independence item response approach. Journal of Educational and Behavioral Statistics, 35(6), 629–670.

60.

Valant

(2020). School reopening plans linked to politics rather than public health. Brookings Institution. https://www.brookings.edu/blog/brown-center-chalkboard/2020/07/29/school-reopening-plans-linked-to-politics-rather-than-public-health/

61.

Weintraub

Weise

(2021). America reaches milestone with COVID-19 vaccine widely to those who want it, but hesitancy still casts a shadow. USA Today. https://www.usatoday.com/in-depth/news/health/2021/04/19/covid-vaccine-tracker-available-widely-hesitancy/7187382002/

62.

Yard

Radhakrishnan

Ballesteros

M. F.

Sheppard

Gates

Stein

Hartnett

Kite-Powell

Rodgers

Adjemian

Ehlman

D. C.

Holland

Idaikkadar

Ivey-Stephenson

Martinez

Law

Stone

D. M.

(2021). Emergency department visits for suspect suicide attempts among persons aged 12-25 years before and during the COVID-19 pandemic—United States, January 2019–May 2021. Centers for Disease Control and Prevention Morbidity and Mortality Weekly Report. https://www.cdc.gov/mmwr/volumes/70/wr/mm7024e1.htm