Who Supports MO Private School Choice? Evidence from Likely Voters in Missouri

In the first half of 2021, 33 states considered new private school choice legislation. Despite this legislative interest, public opinion on private school choice remains varied. In this paper, we examine voter support for private school choice in Missouri, a politically conservative state that lacked a private school choice program until 2021. We conduct a survey experiment to gauge support for the expansion of private school choice to religious schools, assess preferences for universal or targeted student access, and explore opinions regarding the regulation of participating private schools. We find voters are significantly less likely to support voucher expansion when primed to consider that vouchers could be used at religiously affiliated schools. Political ideology, racial identity, college exposure, and family income are significant predictors of voucher support. Regulation support varies widely, with the strongest support for adherence to state testing requirements and the lowest support for waiving admissions requirements for voucher students. We conclude by discussing the implications of our findings for Missouri’s recently passed private school Education Policy Analysis Archives, Vol. 30 No. 81 2 choice program, particularly by considering ways in which political compromise may be reached to implement the program in a manner conscious of heterogeneous voter preferences.


Who Supports MO Private School Choice? Evidence from Likely Voters in Missouri
At the start of 2021, private school choice programs leveraging public funds to subsidize private school tuition were in operation or permitted by statute in 29 states, Washington, DC, and Puerto Rico (Ed Choice, 2021). Over the first half of the year, 33 states, including Missouri, considered new private school choice legislation, action precipitated in part by the COVID-19 pandemic and a heretofore unprecedented period of mass school closures (Educational Freedom Institute, 2021). Private school choice advocates gained additional leverage from a recent U.S. Supreme Court decision, Espinoza v. Montana Department of Revenue (2020), in which the Court decided the state of Montana could not prohibit private religious schools from receiving funds from a taxcredit scholarship program. Amid this multi-pronged contemporary push for the expansion of school choice options, however, public opinion on private school choice remains varied. Evidence unpacking this heterogeneous public opinion, therefore, may inform active state policy and legislative considerations, including both the establishment of private school choice programs and their statutory regulation.
Prior research has explored the relationship between support for private school choice and socio-demographic factors including political ideology, race and ethnicity, and religiosity, as well as the perceived quality of local public schools. In this paper, we explore the relationship between these demographic predictors and support for private school choice in Missouri, a politically conservative state that lacked a private school choice program until 2021. We also conduct a survey experiment to test voter support for the expansion of private school choice to religiously affiliated schools. Finally, we assess voter opinion on who should have access to private school choice and on the regulation of schools seeking to accept private school choice funding. Along each of these dimensions, Missouri presents an intriguing context to study the expansion of private school choice programs, particularly the heterogeneity of support among its electorate.
To do so, we leverage data from a representative survey of Missouri likely voters. We focus on the following research questions, each critical to understand support for private school choice in Missouri while also relevant to the dozens of additional states considering similar legislation: 1) Does support for private school vouchers vary by voter socio-demographic characteristics, residential location, and perceptions of public school quality? 2) Does the inclusion of schools with religious affiliation affect voters' support for private school vouchers, and do differences in support vary by voter characteristics? 3) Who do voters believe should have access to private school vouchers, and does opinion on voucher access vary by voter characteristics? 4) Do voters believe participating private schools should be regulated with respect to school admissions, tuition costs, religious activities, and state testing requirements?
To address our first two research questions, we estimate linear probability models predicting support for private school vouchers and then predicting support for vouchers when the question specifies vouchers can be used at religious schools. When survey question wording specifies the inclusion of religious private schools, support for vouchers declines sharply in nearly every demographic group, nine percentage points on average. We next estimate multinomial logit models predicting support for three types of vouchers: universal vouchers, vouchers for low-income students, and vouchers for students with disabilities. To address our fourth research question pertaining to private school choice program regulation, we estimate ordered logit models predicting the strength of support for each of four potential regulations. These hypothetical regulations would require participating private schools to 1) waive admissions requirements for voucher students; 2) accept a voucher to fully cover all costs of attendance; 3) follow all state testing and public reporting requirements required of public schools; and 4) allow voucher students to opt out of religious activities at religiously affiliated schools. We conclude by discussing the implications of our findings for Missouri's recently passed private school choice program, particularly by considering ways in which political compromise may be reached to implement the program in a manner conscious of heterogeneous voter preferences.

Background and Literature Review Private School Choice & Missouri Context
Private school choice can take several different forms, including tax-funded scholarships for private school tuition (i.e., school vouchers); tax credits for individuals and corporations donating to private school scholarship-granting organizations (i.e., tax-credit scholarships); and education savings accounts which provide funds to parents to pay for private school tuition, tutoring, or other education-related expenses. Many states operate more than one such program; voucher programs (29) and tax-credit scholarship programs (24) are the most common (Ed Choice, 2021).
Program eligibility may be "universal," open to all students, or limited to targeted groups, such as low-income students, students with disabilities, or students attending low-performing public schools. Additional regulations may be imposed on private schools participating in these programs, such as requiring schools to accept a voucher to fully cover all costs of attendance (National Conference of State Legislatures, 2014). In addition, so-called Blaine Amendments in state constitutions historically have prohibited school choice expansion to religiously affiliated schools (e.g., Heytens, 2000;Schwartz, 2008). Blaine Amendments derive their name from a federal constitutional amendment proposed in 1875 by Speaker of the House James G. Blaine-primarily motivated by anti-Catholic fervor-that would have prohibited the use of public funds at religiously affiliated schools. The amendment was defeated at the national level, but in the late 19 th and early 20 th centuries nearly 40 states, including Missouri, added Blaine amendments to their constitutions (Heytens, 2000;Schwartz, 2008). Historically, Blaine Amendments have stymied private religious schools from participating in publicly funded school choice programs. In Espinoza, however, the Court ruled parents participating in Montana's tax-credit scholarship program maintain the constitutional right to apply scholarships to tuition at private religious schools, notwithstanding the state's Blaine amendment.
Until 2021, Missouri's school choice landscape remained fairly limited. Charter schools were first permitted by law in 1998, but their operation is essentially limited to the boundaries of Kansas City Public Schools and St. Louis Public Schools, a restriction purported to support school desegregation (Mo. 89 th Gen. Assemb., 1998). Recent proposals to expand charter school operations statewide remain stalled. Alongside proposed charter school legislation, there have been nearly annual attempts to pass private school choice legislation. In 2013, the Governor signed Bryce's Law, creating a private school scholarship program available to students with disabilities (Mo. 97 th Gen. Assemb., 2012). Though originally conceived as a tax-credit scholarship program, its enacted version offered only a tax deduction for scholarship donations; no students participated in the program, however, and it sunsetted in 2019. While Bryce's Law never was fully implemented, its passage demonstrated initial political will in support of a private school choice program. The following year, the first "Empowerment Account" (education savings account) legislation was introduced, a proposal also restricted to students with disabilities (Mo. 97 th Gen. Assemb., 2013). Different iterations of Empowerment Account legislation have been introduced nearly every year since, often in the form of tax-credit scholarships for a targeted group of students, such as students with disabilities or foster care students.
More recently, proponents of private school choice in Missouri and elsewhere have leveraged pandemic-induced school shutdowns to argue for expanded school choice. In 2021, Missouri was one of 33 states, including neighboring states Arkansas and Kentucky, to consider new private school choice legislation (Educational Freedom Institute, 2021). Arkansas passed tax-credit scholarship legislation allowing low-income students to receive a scholarship to attend a private school (Ark. 93 rd Gen. Assemb., 2021). Kentucky's legislature overrode their Governor's veto of a tax-credit funded ESA program targeted to low-income students (Ky. Gen. Assemb., 2021). The number of ESA programs nationwide doubled to ten, including new programs in Indiana, Ohio, West Virginia, and New Hampshire. West Virginia's new program is one of the nation's most expansive, allowing most of its students to receive the average per-student state funding of $4,600 for private school tuition and other education expenses (Blad, 2021). On the other hand, private school choice legislation did not pass in Illinois, Iowa, Nebraska, or Oklahoma (Ill. 102 nd Gen. Assemb., 2021; Iowa 89 th Gen. Assemb., 2021; Neb. 107 th Leg., 2021; Okla. 57 th Leg., 2021).
In Missouri, Senate Bill 55 and House Bill 349 both sought to establish a tax-credit scholarship program called the "Missouri Empowerment Scholarship Accounts Program, MOScholars," available to students first starting school or those who attended a public school for at least one semester in the previous school year. Both bills would allow Missouri residents who donated toward scholarships to receive a tax-credit of up to 50% of their state tax liability. Students could use scholarships for tuition and fees at qualifying public and private schools; educational services like tutoring, curriculum, textbooks or other supplies; or educational therapies (Mo. 101 st Gen. Assemb., 2020a; Mo. 101 st Gen. Assemb., 2020b). House Bill 349 would grant students with disabilities or those who qualify for free or reduced-price lunch priority access to scholarships (Mo. 101 st Gen. Assemb., 2020a). House Bill 349 was signed into law on July 14, 2021 (Mo. 101 st Gen. Assemb., 2020a). Originally, the program was designed to accommodate up to $50 million in tax credits in its first year, subsequently increasing to $75 million annually. A last-minute amendment, however, reduced the tax credit limit to $25 million in the first year and $50 million in subsequent years (Mo. 101 st Gen. Assemb., 2020c). An additional legislative compromise limited access to the program to students living in one of the state's four charter counties or a city with a population of 30,000 or more. The program is set to take effect beginning in the 2022-23 school year.

Who Supports Private School Choice?
Public support for private school choice varies along several dimensions, suggesting that legislation engineered to reflect prevailing statewide policy preferences must consider this heterogeneity. Previous studies have examined variation in support by socio-demographic characteristics such as political ideology, race and ethnicity, and religiosity, as well as the performance of local public schools (Bali, 2008;Brasington & Hite, 2014;Leal, 2004;Moe, 2001). To understand this heterogeneity, research typically leverages survey data or vote patterns on ballot initiatives. We review much of the relevant literature below.

Socio-Demographics
The annual Education Next survey of public opinion asks a nationally representative sample of adults about their support for school vouchers and tax-credit scholarships (Education Next, 2020b). In 2020, the same year we administered our survey, 51% of the general public indicated support for universal vouchers, up from 43% in 2013 (Education Next, 2020a). The 2020 poll found that Democrats were more likely to support low-income vouchers, while Republicans were more likely to support universal vouchers. Republican support for universal vouchers peaked at 64% in 2018, while support among Democrats has remained between 40% and 50% since 2013; conversely, 52% of Democrats and 45% of Republicans indicated support for vouchers targeted to low-income students. Forty percent of white respondents and 60% of Black respondents indicated support for vouchers. Leal (2004) found that Latinos and Blacks were more likely than whites to support vouchers, though there was some indication that Latino support could be attributed to Catholic religious affiliation. Leal (2004) found no differences by income or education level. Bali (2008) examined California exit poll data from 2000 pertaining to ballot initiatives that would eliminate bilingual education, create a statewide school voucher program, and reduce the majority needed for school bond passage. Bali found political ideology was the key predictor of support for each education initiative. Racially-based self-interest was also important, particularly when the initiative was racially significant. For example, Hispanic respondents were less likely to support eliminating bilingual education, and Black respondents were more likely to support the creation of a voucher program serving disadvantaged students. Respondents who were married or had children were more likely to support these initiatives. Local public school district conditions did not predict support for vouchers, but a religious background was related to support.  examined data from a survey of California potential voters and subsequent vote patterns on a universal voucher initiative, Proposition 38. Support among white households with children increased when the concentration of nonwhite students in their local public school increased. Conversely, support for vouchers among nonwhite families decreased as the nonwhite student share increased. Results were consistent when examining actual vote patterns; a higher school share of nonwhite students was associated with greater support for vouchers among white voters with children, with a much stronger effect for white households with children under 6 years old. The opposite was true for nonwhite households. Brasington and Hite (2014) examined support for a broad range of school choice options including district open enrollment, vouchers, tax-credit scholarships, and charter schools. To examine homeowner perspectives, they conducted a phone survey with heads of households of owner-occupied homes in Ohio. They found the strongest predictors of opposition to school choice were having a graduate degree and living in a high-performing public school district. Older residents, males, and very-liberal respondents all indicated strong opposition. Blue collar neighborhood residents and private school parents were the most likely to support school choice, followed by Black respondents and parents with young children. Individuals with some college exposure or a high school diploma were more likely to support school choice than those with college degrees. They did not find differences by income.

The Private School Choice Set
Little attention has been given to the relationship between the supply of potential choice options (the "private school choice set") and demand for private school choice policies, a possible additional source of heterogeneity in support for private school choice. In addition, the private school choice set may moderate the impact of other predictor covariates, potentially attenuating their effects. Studies that have considered this potential source of heterogeneity typically include basic controls for private school availability in their analyses. Brasington and Hite (2014) used distance to the closest private school and Bali (2008) included the percentage of private schools in a public school district boundary as control variables in their respective analyses. We employ a more robust method of measuring the private school choice set based on related literature on private school choice.

Median Voter Theorem
We examine registered Missouri voters who self-reported they were likely to vote in the November 2020 election. This focus on likely voters attempts to capture the opinion of the state's median voter, as that voter expresses his or her political and policy stances through the ballot box in the electing of state legislators (i.e., Downs, 1957). In this manner, we are able to consider both the state's median voter, representing prevailing statewide average opinions, and to disentangle important heterogeneity in likely voter support. Examinations of median voters have a long history in education research across a range of different policy proposals. For example, demand for school spending and related tax increases often varies by voter income and the presence of school-aged children Corcoran & Evans, 2010;Fletcher & Kenny, 2008).  find that the interests of higher-income voters without school-age children and lowerincome voters converge in opposition to spending increases.
The median voter hypothesis likewise has been applied to voters' school choice preferences. For example, Chen and West (2000) theorize voter support for universal or targeted vouchers, concluding that support for vouchers likely is conditional on the median voter's assumed trajectory of educational costs (i.e., does the introduction of vouchers increase taxes?) and on arguments for improved educational efficiency (i.e., competition over school enrollments driving positive returns to school quality). Finally, legislators may be differentially beholden to median voter preferences, depending on their local contexts. Gerber and Lewis (2004) find that legislators more frequently adhere to median voter preferences in more homogeneous locales.

Regulations and Private School Participation
While proposals to enact private school choice programs often are intended to provide options for families, family demand for private school choice does not necessarily lead to private school supply. Proponents of private school choice programs have long argued that overregulation serves as a disincentive for private school participation (Stuit & Doan, 2013). Stuit and Doan's (2013) analysis of regulations and participation finds a modest, negative relationship between program rules and school participation. Other research uses surveys of private school leaders in choice settings to understand schools' decisions to enroll voucher students. One such study surveying leaders in Indiana, Florida, and Louisiana finds that concerns around program design and losing independence were the most common reason for schools to forgo participation (Kisida et al., 2015). Research in Indiana found that a desire to maintain independence is a common factor limiting participation (Austin, 2015). A study of other choice-rich environments finds that schools with more specialized educational settings (i.e., Montessori, special education, etc.) are less likely to participate in their state's programs (DeAngelis, 2019). Similarly, Sude et al. (2018) finds Washington, DC, Indiana, and Louisiana schools with high tuition rates and larger enrollment cohorts were less likely to participate in their respective local private school choice programs, while also finding differences in regulatory burdens across programs. Finally, Egalite et al. (2020) find North Carolina school leaders were most concerned about the introduction of additional regulations that would render participation in the state's voucher program less appealing.
For a private school choice program to have the intended impact, there would need to exist a sufficient supply of private school seats to meet the demand of allotted scholarship spaces. As shown in the existing literature on the supply side of school choice, programmatic regulations can hinder private schools' willingness to participate. With that in mind, our study examines what voters may desire in terms of regulation of a hypothetical private school choice program in Missouri. There exists a potential imbalance between what the general public desires from a regulation and accountability standpoint and what conditions most private schools would accept to participate.

SLU/YouGov Poll
Saint Louis University (SLU) partnered with a professional polling firm YouGov to survey 931 Missouri likely voters in September and October of 2020. YouGov uses its own survey panel, to which respondents must opt in. Using self-reported demographic characteristics such as age, race, gender, and education level, the sample is weighted to reflect the characteristics of the state's registered voters as reported in the 2018 Current Population Survey.
The SLU/YouGov Poll (Poll) included a variety of private school choice questions, including items pertaining to general support for a tax-funded private school choice program, as well as support for a tax-funded private school choice program that would permit enrollment in private religious schools. 1 Respondents were asked whether they support a universal voucher program, a voucher for low-income students only, a voucher available only to students with an Individualized Education Plan (IEP), or oppose vouchers. For the hypothetical low-income voucher, our poll presented an income threshold of 250% of the federally determined poverty level, an amount between that of the Milwaukee (WI) Parental Choice Program; the Opportunity Scholarship Program in Washington, DC; and the Indiana School Choice Program.
Along with questions exploring general support for private school choice programs, the Poll presented a hypothetical scenario in which Missouri had a private school choice program and voters were asked to express their preferences regarding the regulations to which participating private schools must adhere. This included schools waiving admissions requirements for voucher students, accepting the voucher amount to fully cover attendance costs, allowing voucher students to opt out of religious activities, and requiring private schools to participate in state testing and accountability programs required of public schools.
The Poll also collected data on a rich set of covariates that have been found to be related to support for private school choice, such as political ideology, race and ethnicity, and religiosity. Table  1 compares the characteristics of our sample of Missouri likely voters with the characteristics of the Missouri adult population.
Poll respondents provided their home zip code, which allowed us to calculate each respondent's Zip Code Tabulation Area (ZCTA) centroid to approximate their residence and identify a local private school choice set. We accessed Missouri ZCTA geometric data from the U.S. Census Bureau, along with 2019 American Community Survey 5-year estimates for school-aged population by ZCTA. We used geoprocessing to draw 5-and 10-mile radii around each respondent's ZCTA centroid and then assigned every private school in Missouri to each 5-or 10-mile radius in which it was located.
We drew data on Missouri private schools from the National Center for Education Statistics Private School Universe Survey (PSS; U.S. Department of Education, 2021). There were 508 private schools in Missouri in the 2017-18 school year, the most recent data year available. We aggregated the count of private schools within each respondent's 5-and 10-mile radius and the count of types of private schools within each radius. The 5-mile crows-flight measure is used by Hart (2014) when estimating the private school choice set of private school scholarship users in Florida. Egalite and Mills (2021) and Figlio and Hart (2014) used 5-and 10-mile radii from public schools when identifying the private schools that may have a competitive effect on each school. This is consistent with earlier competitive effects studies that also used 5-and 10-mile radii (Carnoy et al., 2007;Greene & Marsh, 2009;Greene & Winters, 2007;Lubienski et al., 2009;Winters & Greene, 2011). These earlier studies identified the density (count) of schools within each radius, as well as the distance to the closest private school. Egalite and Mills (2021) and Figlio and Hart (2014) added two additional measures, diversity (count of types of private schools) and concentration using a modified Herfindahl Index. 2 Religious school diversity can be indexed in broad categories such as Catholic, other religious, and nonsectarian. Alternatively, Figlio and Hart (2014) used 10 categories based on selfreported orientation. The PSS includes 30 religious orientation categories. We explored the Missouri PSS data and identified the most frequently reported religious orientations: Catholic (38%), Nonsectarian (17%), Christian (no denomination; 12%), Lutheran (11%), Amish (7%), and Baptist (5%). We grouped all remaining schools into a category labeled Other religious (10%). Each respondent was assigned a diversity measure of 0 to 7 for their respective 5-and 10-mile radii based on the count of types of schools within each radius. Access to private schools varied greatly across the state among Poll respondents, ranging from zero to 59 private schools in a 5-mile radius and zero to 119 private schools in a 10-mile radius. We recognize a few limitations of our Poll dataset. First, we lack information on respondents' home ownership status and, though we know whether they have school-aged children, we lack data on whether these children attend or have attended private school. Next, respondents' ZCTA centroids do not perfectly approximate their residences, which may impact estimates of the private school choice set. In addition, this study does not account for the accessibility (seat availability or admissions practices) or desirability of private schools within each respondent's private school choice set, which may be particularly relevant for parents. Finally, we elect to focus on the perspectives of likely voters. While voter opinions certainly are relevant for policymakers who wish to understand support for and opposition to private school choice legislation, the sample of likely voters is not representative of all adults in the state, nor is it representative of adults with school-aged children, the group most immediately affected by private school choice legislation. If any groups are underrepresented in voter registration rolls those groups will be similarly underrepresented in our data.

Empirical Approach
To address our first two research questions, we estimate linear probability models predicting support for private school vouchers and then predicting support for vouchers when the question specifies that vouchers can be used at religious schools. Our base model predicts support for private school choice based on an individual's political ideology: (1) = 0 + 1 + + , where Yi denotes the likelihood that individual i supports vouchers. is assumed to be a linear function of political ideology, . is the idiosyncratic error term. represents a vector of respondent characteristics; we first exclude these traits, then layer them on to the model in iterative specifications: (2) Ni, a binary indicator that takes a value of one if individual i is nonwhite; (3) Ci, a binary indicator that takes a value of one if individual i has children currently in school; and (4) Ri, a binary indicator that takes a value of one if individual i responds that religion is "Very Important" or "Somewhat Important." The fully specified and preferred model (5) also includes Si, a vector of other socio-demographic covariates for individual i, including age, gender, college exposure, and income.
(5) = 0 + 1 + 2 + 3 + 4 + 5 + 6 + 7 + 8 + 9 5 + + We include squared and cubic terms for age because of the observed nonlinear relationship between age and support for private school choice. We also control for the perceived quality Q of local l and Missouri m public schools. Uc is county urbanicity, and Pi5 represents a choice vector including distance to the closest private school and the density and diversity of private schools in a 5-mile radius. Fi represents a regional fixed effect that accounts for the fact that respondents in St. Louis City and Kansas City have access to public school choice in the form of charter schools. 3 To address the third research question pertaining to voucher access, we estimate multinomial logit models predicting support for each voucher type. Finally, to address our fourth research question pertaining to private school choice program regulation, we estimate ordered logit models predicting the strength of support for each of four potential regulations. In each of these models we include all covariates included in equation (5), with the exception of the higher order age terms.

Voucher Support
In Figure 1 we explore support for voucher use at private schools and the difference in support when we include the phrase "including private religious schools." We then supplement these unconditional cross-tabulations with the series of linear probability model specifications described above.
There was a statistically significant reduction in support for every subgroup when the question mentioned the inclusion of religious schools, though the change in support was only marginally significant for conservatives. Respondents who identified as liberal and those who were not religious were approximately 20 percentage points less likely to indicate support for vouchers if they could be used at religious schools. The decline in support was significantly different between subgroups in four categories: political ideology, religiosity, college exposure, and perception of Missouri public schools.

Support for Private School Vouchers, by Inclusion of Private Religious Schools
Note: Significance tests indicate whether the change in support was statistically significant within in each subgroup and whether the change was significantly different between subgroups within each category. * p<0.1. ** p<0.05. *** p<0.01.   In our fully specified model (column 5) in Table 2, we consider the effects of key covariates, combining the indicated preferences for voucher support for all students, low-income students, and students with disabilities into a variable indicating support for some type of voucher. Relative to conservatives, the modal Missouri voter, liberals were 16 percentage points less likely to support vouchers. Nonwhite respondents were 22 percentage points more likely than white respondents to indicate support. Parents of school-aged children were eight percentage points more likely to support vouchers, though having children in school was not a statistically significant predictor of voucher support after including other covariates in the model. Those with at least some college exposure were 13 percentage points more likely to support vouchers than those who had no college experience. Finally, those whose family income was less than $60,000, approximately Missouri's median income, were 12 percentage points more likely than those with higher incomes to support vouchers.
Next, we estimated the same models examining support for private school vouchers with the stated inclusion of religious schools. Consistent with the unconditional cross-tabulations above, support for vouchers is lower under this condition, with notable differences in support between subgroups. Relative to conservatives, moderates were nine percentage points less likely and liberals 28 percentage points less likely to support vouchers if vouchers could be used at religious schools. Nonwhite respondents were 19 percentage points more likely than white respondents to indicate support for vouchers under this condition. Parents were nine percentage points more likely to support vouchers, though this finding was marginally significant. Support for vouchers among religious respondents increased from a seven percentage point premium in Table 2 (though not statistically significant) to 18 percentage points when the question indicated that vouchers could be used at religious schools. Those with at least some college exposure were 11 percentage points more likely to support vouchers than those who had no college experience, and those with below-median incomes were 11 percentage points more likely than those with higher incomes to support voucher use at religious schools.

Robustness Checks
We estimated the fully specified models for the choice questions excluding and including the religious component with an additional covariate controlling for child population density per ZCTA square mile. The results were essentially identical, and the addition of this covariate did not increase either model's adjusted R-squared. Thus, we did not include this covariate in our preferred models. We also estimated logit models for both versions of the question, again finding the results essentially unchanged.
Eighty-eight Poll respondents answered "Prefer not to say" when asked about their family income level. Six of these were panel respondents who had answered the income question in a previous Poll iteration; we used their previous responses to fill in the missing values. As a robustness check, we used a Markov Chain Monte Carlo simulation to impute missing income data for the remaining 82 Poll respondents and reran the fully specified models with this data. The findings were robust to the inclusion of the imputed values (full results are available upon request). Conservative political ideology, nonwhite racial identity, college exposure, and below-median income all remained significant predictors of support for private school vouchers. The coefficients for religiosity and perception of local public schools both gained marginal significance. The gap in support for vouchers between those with incomes below and above $60,000 decreased from 12 percentage points to 10 with the inclusion of the imputed values for income. On the question including private religious schools, having children in school and perceiving Missouri public schools to be excellent or good were no longer marginally significant predictors. The gap in support for vouchers between those with incomes below and above $60,000 also decreased from 11 percentage points to nine with the inclusion of the imputed values for income.

Voucher Access
Next, we examined which students respondents believed should have access to school vouchers. Among respondents who supported vouchers of some type, liberals were 30 percentage points less likely than conservatives to prefer universal vouchers and 28 percentage points more likely to prefer vouchers for low-income students (see Table A1 in the appendix). Liberals who supported voucher use at religious schools were also significantly less likely than conservatives to prefer universal rather than targeted vouchers and significantly more likely to prefer vouchers for students with disabilities under this condition. Nonwhite respondents were marginally more likely than white respondents to prefer universal vouchers, a finding which did not hold when examining voucher application at religious schools. Religious respondents were five percentage points less likely than those who were not religious to prefer vouchers for students with disabilities, but there were no differences between religious and non-religious respondents who supported vouchers being used at religious schools.
When considering predicted marginal probabilities by political ideology, we observed approximately two-thirds of conservatives but only one-third of liberals supported universal vouchers, while one-third of conservatives and two-thirds of liberals supported vouchers for lowincome students (Table 3). Support for voucher programs limited only to students with disabilities ranged from three to six percent and did not differ significantly by political ideology. The distribution of support for each type of voucher program was the same for conservatives when considering the inclusion of religious schools. However, liberals were nine percentage points more likely to support universal programs if they supported voucher use at religious schools and 21 percentage points less likely to support vouchers only for low-income students. Support for voucher programs targeted only to students with disabilities was highest among liberals who supported voucher use at religious schools. Approximately two-thirds of nonwhite respondents and only 53% of white respondents were predicted to support universal vouchers. Support for universal vouchers was slightly less than 50% for respondents who perceived their local public schools to be excellent or good, while 61% of respondents who lacked a positive perception of their local public schools were likely to support universal voucher programs.

Voucher Regulations
The next series of analyses examined the regulation of participating private schools. Figure 2 shows unconditional cross-tabulations of support for each of four regulations. We supplement these cross-tabulations with regression estimates (available in the appendix).

Admissions
First, we examined support for a regulation requiring private schools to waive admissions requirements for voucher students. Among those regulations we examined, overall support was predicted to be lowest for this regulation, at 36% (see Table A2 in the appendix). Nonwhite respondents and respondents with school-aged children were significantly more likely than white respondents or respondents without children, respectively, to indicate support. Respondents with some college experience were significantly less likely than those with no college experience to support a requirement that would waive admissions criteria. Both moderate and liberal respondents were significantly more likely than conservatives to support this regulation, with liberal support predicted to be twice as high as conservative support (54% versus 27%).

Full Cost
The second proposed regulation would require participating private schools to accept vouchers to fully cover all costs of attendance for voucher students. Overall support for this regulation was predicted to be 60% (see Table A3 in the appendix). Predicted support was 15 percentage points higher for nonwhite than white respondents (74% versus 59%) and 12 percentage points higher for respondents with children than those without (69% versus 57%). Respondents with some college experience were 12 percentage points more likely than those without college experience to express support for the "full cost" requirement. In addition, respondents with incomes less than $60,000 were also significantly more likely than higher-income respondents to support this regulation (by eight percentage points). There were no significant differences by political ideology in support for the full cost regulation, with 60% or more of all three groups in support.

Opt Out
A third proposed regulation would require private schools participating in a voucher program to allow voucher students to opt out of religious activities. Our model predicted only 41% of likely voters would support such a regulation, though both moderate and liberal respondents were significantly more likely than conservatives to express support for this regulation (see Table A4 in the appendix). Predicted support was 51 percentage points higher for liberals than for conservatives. Religious respondents were 23 percentage points less likely than those who were not religious to support this regulation.

Testing
A final proposed regulation would require participating private schools to follow all Missouri state testing and public reporting requirements. Support was highest for this regulation across the sample, with a predicted likelihood of 78% of likely voters supporting such a requirement (see Table  A5 in the appendix). Liberal respondents were significantly more likely than conservative respondents to support this regulation (84% versus 73%).

Discussion and Conclusions
In this study, we leverage polling data from a politically conservative state to describe the expressed opinions of likely voters regarding private school choice programs and their regulation, particularly relevant during a period when a majority of states are considering such legislation. Consistent with previous research, we find that conservative political ideology, nonwhite racial identity, college exposure, and below-median family income predict support for private school vouchers. As seen in national polls and other survey studies, liberal respondents are significantly less likely than conservatives to prefer universal vouchers and more likely to prefer vouchers for lowincome students. Notably, we find that support for school vouchers declines sharply in nearly every demographic group when respondents are primed to consider that vouchers could be used at private religious schools. It is important for policymakers to consider who supports private school choice under this condition, in light of the Supreme Court ruling in Espinoza allowing religious private school participation in a tax-funded private school choice program.
We acknowledge the unique historical moment in which our poll was administered. The survey was conducted in the early months of the 2020-21 school year, the first full school year to take place during the COVID-19 pandemic. We lack pre-pandemic Missouri polling data on private school choice and thus cannot identify a possible pandemic effect on support for vouchers. However, historical data from Education Next suggests that any impact from the pandemic may be heterogeneous. The national Education Next poll found that while support for universal vouchers decreased among both Republicans and Democrats from 2019 to 2020, support for vouchers for low-income students stayed the same for Democrats and increased for Republicans during the early part of the pandemic, when compared with the previous year (Education Next, 2020a). Notwithstanding potential pandemic effects, we contend that our poll results maintain significant policy relevance, as public opinion at this moment will likely impact policy decisions pertaining to legislation with lasting impact.
We add to the literature on support for private school choice by examining support for four potential regulations. These findings are especially relevant as Missouri considers the design and implementation of its new private school choice program, which will operate in the state's most populous and relatively less conservative counties and cities. We find strong support among all key subgroups considered for requiring participating private schools to follow state testing and public reporting requirements. Support is lowest for requiring private schools to waive admissions requirements for scholarship students. We also note important heterogeneity in voter opinion. Moderates and liberals are more supportive than conservatives of waiving admissions requirements for voucher students and allowing voucher students to opt out of religious activities. Nonwhite respondents and respondents with school-aged children are significantly more likely than white respondents or respondents without children, respectively, to support regulations that would increase private school accessibility by requiring schools to waive admissions requirements and to accept a voucher to fully cover all costs of attendance. Respondents with some college exposure are less likely to support a regulation requiring schools to waive admission requirements but more likely to support requiring schools to accept a voucher to fully cover all costs. Religious respondents are significantly less likely to indicate support for a proposed regulation allowing voucher students to opt out of religious activities.
These findings suggest potential avenues for political coalition-building around private school choice. HB 349 was passed by a Republican-majority legislature, with no Democratic legislators supporting the bill in its final votes in the House and Senate (Mo. 101st Gen. Assemb., 2021a, 2021b. However, opposition from Democratic legislators may be out of step with the desires of a key portion of their constituency. In particular, while nonwhite voters are primarily located in less conservative portions of the state, these voters are even more likely than conservatives to express support for vouchers. Previous literature also suggests that eligible Black students are more likely than eligible white students to participate in private school choice (see for example Campbell et al., 2005;Figlio et al., 2010;Fleming et al, 2015;Goldhaber, et al., 1999;Howell, 2004;Witte & Thorn, 1996).
Much remains to be determined about the rules of the state's new tax-credit scholarship program, and it may be in the best interest of both major political parties to work toward a bipartisan agreement on these regulations. In late 2021, a group linked to the Missouri National Education Association (MNEA) was circulating a ballot initiative that would prohibit the implementation of the program (Weinberg, 2021), and it is likely the program will continue to face opposition unless a bipartisan compromise is reached regarding its design. There is broad support among both conservatives and liberals for requiring participating private schools to take part in the state standardized testing and reporting programs used for public schools, and this could be an area of agreement for legislators as they debate the design of the program. Moderates and liberals are more likely than conservatives to support requiring participating private schools to waive admissions requirements for voucher students, a possible area in which Republican legislators may compromise with their colleagues across the aisle.
Our findings also suggest fruitful avenues of additional inquiry. First, we plan to survey a representative sample of Missouri adults with school-aged children to explore support for choice among those most immediately affected. In addition, we plan to survey the universe of private school leaders in Missouri, which will allow us to better control for the unique private school context of each poll respondent. Data from private school leaders will also provide additional insight into a potential mismatch between voter support for regulations and the regulatory burden private schools are willing to accept to participate in a private school choice program. Finally, we intend to give further consideration to the potential impact of the private school choice set on support for private school choice. Many supporters of choice have little access to private schools, and we hope to further explore the relationship between private school access, political ideology, urbanicity, and support for choice.
Evan Rhinesmith, PhD, is the executive director of the Policy Research in Missouri Education (PRiME) Center at the Saint Louis University School of Education. His research focuses on postsecondary access and outcomes, college remediation, and teacher quality and evaluation.