Does Private School Choice Threaten Democracy? Evidence from Private and Public Schools in New York City and Dallas/Fort Worth

Abstract

A major concern in the ongoing debate over school choice is whether private schools help to increase their students’ levels of tolerance necessary for a functioning democracy in the United States. Over 40 years ago, scholars at the University of Minnesota created a survey which measured political knowledge, political tolerance, perceived threats from opposing groups, and support for democratic norms anchored in each respondent’s view of the political group they find most distasteful. In 1997, researchers at various universities used a similar survey instrument to derive responses from students in eighth-grade social studies classes who were enrolled in seven public and twenty-four private schools in New York City and Dallas/Fort Worth, Texas. Those original data remained archived and unexamined for decades. We analyze those data using Ordinary Least Squares (OLS) and rigorous Nearest Neighbor Matching (NNM) methods based on propensity scores. We find that students who attended private schools demonstrate higher levels of political knowledge and stronger support for democratic norms when compared to observationally similar students who attended public schools.

1. Introduction

The greatest insights of liberal theory are that the good is plural and that a civil society works best when all agree that government should be neutral among competing reasonable conceptions of the good. Whether we look to such past liberal thinkers as John Locke, John Stuart Mill, and Immanuel Kant or to more contemporary liberals such as Robert Nozick and John Rawls, we find claims that the willingness of citizens to tolerate conceptions of the good that differ from their own is critical to a liberal society. Liberal tolerance, the argument goes, not only helps keep society civil, but it also encourages cultural, religious, social, and political diversity. Although the normative goal of strict state neutrality regarding value-laden matters has been contested, tolerance remains central to most discussions of the foundations for free societies [1,2]. Liberal tolerance leads to respect for persons and groups whose conceptions of the good differ from our own [3].

In this study, we analyze original survey data that address the question of whether private schooling tends to produce higher, lower, or the same levels of student civic values including tolerance, political knowledge, and support for democratic norms. We first review the arguments concerning why public or private schools are more likely to develop tolerance and related democratic attitudes. Next, we describe past research on political tolerance, the data collection procedures, and the measures of the key variables used in this study. Then, we discuss the issue of selection bias, the problems it poses for studying the impacts of school types, and the statistical model used to correct those problems. In conclusion, we present the data analysis with a discussion of the results.

2. Arguments from the School Choice Debates

A major concern in the ongoing debate over school choice is whether private schools advance or undermine students’ levels of tolerance necessary for a functioning democracy in the United States. Neal [4] argues that private schools respond directly to parents’ and students’ preferences to increase students’ human capital and shape their moral values. Neither parents, nor government officials, would tolerate the continued operation of private schools that teach hate. As Glenn [5] (p. 343) states, “The idea that schools that undermine societal norms would flourish unchecked under a well-designed [private] school voucher program is a sort of ghost story to frighten the gullible”.

On the other hand, critics of private schooling claim that public schools are essential for democracy [6,7]. Likewise, supporters of public schooling believe that government-run public schools will emphasize the preparation of democratic citizens and are more diverse in terms of religion, culture, and ethnicity than private schools [8,9]. Gutmann [10] (p. 70) states, “…public, not private, schooling is … the primary means by which citizens can morally educate future citizens”.

American schools are places where students from diverse backgrounds can share their public values and learn political and religious tolerance [8,9,11,12,13]. The United States has relied on public schooling to establish a centralized set of societal traditions for its citizens [14,15]. Glenn [16] claims that the image of public schooling in America as an inclusive common school for students has been a “myth”. Public school boards often support the interests of the upper class, which directly neglects the interests of disadvantaged minority groups [17]. In addition, because American democracy is heterogeneous, it comprises diverse values and traditions. Berner [18] explains that public education should be as diverse as American democracy. She points out that the private sector is pluralistic and allows parents to choose schools that will reinforce parental values.

Most private schools have religious affiliations and sometimes require their students to learn about and participate in religious activities. For this reason, private school choice critics highlight schools run by fundamentalist Christians as instilling particular values that are inconsistent with shared social values [19]. Critics who perpetuate this stereotype of Christian fundamentalist schools claim such schools are less likely to encourage self-rule, tolerance, and respect for diversity among their students compared to traditional public schools [8,20,21,22,23]. In contrast, some scholars argue that, because Catholic schools stress the intrinsic value of each human being, in what Sikkink [24] calls “the hidden civic curriculum”, they are more likely than public schools to advance the development of community-oriented democratic citizens [25,26,27,28].

The different ways public and private schools teach civics, including what can and cannot be discussed, are important factors for how schools shape students’ political and religious tolerance. Hirsch [29] theorizes that public schools could be more effective in teaching civics because they are more unified in selecting and teaching related subject matter to students, though they may not always avail themselves of that opportunity. In contrast, Berner [18] argues that private schools may be more willing to discuss controversial political topics than public schools are. Therefore, students in private schools who discuss controversial issues become more aware of and form their own opinions about American current events. Based on a theory of action grounded in the expectation that controversial moral questions will be more actively and inclusively discussed and debated in private schools than in public schools, supporters of school choice and private schooling assert that private schools can and often do outperform public schools in forming tolerant citizens [30,31,32].

3. Materials and Methods

The data for this study come from 2184 students in eighth-grade social studies classes in seven public and twenty-four private schools in New York City and Fort Worth, Texas, in 1997. The private schools include sixteen religious schools and eight secular schools. These schools were chosen using a stratified random procedure in both geographical locations that allowed for the inclusion of schools that were theoretically interesting and representative of the student population. We first assembled a list of all the public and private schools in both cities, then assigned each school to a strata based on whether it was public, private with a religious affiliation or identity, or private with a secular identity. We then randomly selected target schools from each strata to survey, oversampling both types of private schools because they had much smaller average enrollments than the public schools.

About 17 percent of the private school students surveyed attended evangelical schools, identified as such because they were Christian schools with evangelization-themed mission statements. The remaining 83 percent of the private school student sample was approximately balanced among students in Catholic schools, non-evangelical (i.e., mainline) Protestant schools, and secular private schools. These proportions represent a slight over-representation of evangelical, mainline Protestant, and secular private schools, given their shares of the private school sector in 1997, and a slight underrepresentation of Catholic schools [33,34].

All eighth-grade students in each selected school were targets to complete the survey. The schools sent informed consent forms home to parents, and 90% of the students returned signed forms to the school prior to survey day. On survey day, researchers visited the school and administered the paper-and-pencil survey to students with approved consent forms during their regularly scheduled social studies class. The surveys were anonymous but were coded by school identifier.

We surveyed eighth-grade students for several reasons. First, prior research suggests that as students enter their teenage years, their political knowledge, values, and attitudes begin to develop [35,36,37], aided by the fact that many public schools are required to teach American government or civics courses in the eighth grade. Second, a large percentage of private school students transfer into public schools for their high school years [25]. To capture the impact of private schools on most students who attended them before entering traditional public high schools, we must examine the students’ levels of political knowledge, perceived threats from opposing groups, support for democratic norms, and political tolerance prior to private school students entering public schools.

Evangelical Protestantism has a strong presence in the Fort Worth area. This fact enhances the internal validity of our tests of the relationship between evangelical schools and tolerance, since the evangelical schools in our sample are likely to be thoroughly and authentically evangelical in their beliefs and practices. The fact that our evangelical schools likely are strongly evangelical, however, probably limits the external validity of our findings, since evangelical Protestant schools in less strongly evangelical areas might not generate outcomes similar to those we observe here.

3.1. Measuring Political Tolerance

Sullivan et al. [38] defines political tolerance as an individual’s attitude or predisposition that mirrors the desire to support the granting of political rights to an objectionable group. You are not “tolerating” an extremist political group if you agree with their agenda. Extending political rights to a person’s least-liked political group flows from pre-existing beliefs about democratic norms and the perceived threat that the group introduces to individuals and society.

Sullivan et al. [38] and Marcus et al. [39] operationalize their conception of political tolerance using a survey that asked participants to choose the political group they liked the least. In this survey, political tolerance is the sum of an individual’s willingness to allow members of their least-liked group to make public speeches, to teach in public schools, and to hold public demonstrations, as well as their objection to the government outlawing or tapping the phone of the least-liked group. Students respond to these questions on either 5-item or 7-item Likert scales. Researchers then draw from those responses to construct summative indices to measure each respondent’s willingness to extend constitutionally guaranteed rights to their least-liked group [40]. The same survey also generates summative indices measuring political knowledge, general support for democratic norms, and each respondent’s level of perceived threat from their least-liked group [41]. (Appendix A 

Table A1. Survey questions that supported scale construction.

Political Tolerance
Tap: The government should be able to tap the phones of members of your least-liked group.
Teach: Members of your least liked group should be allowed to teach in public schools.
Outlaw: The group you least like should be outlawed.
Speech: Members of your least liked group should be allowed to make a public speech.
Rally: Members of the group you like least should be allowed to hold public demonstrations or rallies.
Democratic Norms
Free Speech: I believe in free speech for everyone, no matter what their views might be.
Different View: Society shouldn’t have to put up with those who have political ideas that are extremely different than the majority.
Unpopular: It is refreshing to hear someone stand up for an unpopular view, even if most people find it offensive.
Extreme: Free speech is just not worth it if it means we have to put up with the danger of extremist political ideas.
Protect: No matter what a person’s political beliefs are, he or she is entitled to the same legal rights and protections as anyone else.
Perceived Threat
Show how you feel about {your least liked group} by marking one number for each pair of words.
Danger:→Safe→→→1→2→3→4→5→6→7→Dangerous
Bad:→→Good→→1→2→3→4→5→6→7→Bad
Threaten:→Non-Threatening 1→2→3→4→5→6→7→Threat
No trust:→Can be Trusted 1→2→3→4→5→6→7→Cannot be Trusted
Violence:→Non-Violent→1→2→3→4→5→6→7→Violent
Political Knowledge
What job or political office is now held by Al Gore?
Whose responsibility is it to determine if a law is constitutional or not? __President __Congress __Supreme Court?
How much of a majority is required for the U.S. Senate and House to override a presidential veto?
Do you happen to know which party has the most members in the House of Representatives?
Whose responsibility is it to nominate federal judges?
What are the first ten amendments to the constitution called?

provides the questions used in these measures).

Our survey also includes measures of socioeconomic variables, such as parents’ education, religion, and religiosity, as well as indicators of the student’s sex, current grades, and interest in politics (

Table 1. Summary statistics.

	(1)	(2)	(3)	(4)	(5)
Variables	N	Mean	Sd	Min	Max
Private	2313	0.316	0.465	0	1
Texas	2313	0.521	0.500	0	1
Interest in Politics	2294	1.774	0.574	1	3
Students’ Grades	2238	3.206	0.726	0	4
Black	2289	0.226	0.419	0	1
Hispanic	2289	0.288	0.453	0	1
Male	2276	0.499	0.500	0	1
Parents’ Education	1925	3.621	1.335	1	5
Willingness to make friends	2189	4.243	1.067	1	5
Fights between other groups	2193	3.451	1.382	1	5
Volunteer	2148	1.089	0.788	0	5
Political Knowledge	2252	1.888	1.640	0	6
Democratic Norms	2101	21.26	3.643	6	30
Perceived Threat	2188	32.77	6.825	6	42
Political Tolerance	2174	14.70	5.062	6	30
Evangelical	2313	0.057	0.231	0	1
Nonevangelical	2313	0.259	0.438	0	1

). To test the hypothesis that greater school diversity will lead to greater tolerance, we operationalize diversity by subtracting from 100 the percentage of the student population who are members of the largest ethnic group in the classroom. We then include a dummy variable to indicate whether a respondent is a member of the largest ethnic group in the school. To properly measure the interethnic climate in a school, we use Likert scale statements asking the level of agreement with, “Students make friends with students of other racial and ethnic groups at my current school” and “Fights often occur between different racial and ethnic groups at my current school”. In addition, because earlier research shows that volunteer work may help develop attitudes suitable for citizenship in a liberal democracy [26,42], we asked students whether they recently participated in voluntary community service work or charitable activities.

3.2. The Problem of Selection Bias

One challenge with surveying eighth graders is that young children may not have sufficiently structured political opinions to allow for valid and reliable measurements. Past research indicates that the reliability of the Sullivan et al. indices increases with the level of education of the respondents [36,39]. The Cronbach’s alpha coefficients measuring the internal consistency of our four scales are 0.67 for both political knowledge and political tolerance, 0.88 for threat score, and 0.56 for democratic norms. Because three of these coefficients are slightly lower than the 0.70 industry standard [43] (p. 85), we performed a confirmatory factor analysis (CFA) on the items in each scale and repeated all analyses using factor scores rather than summative indices as our measures of the dependent variables. The CFA (see Appendix B) shows that our measures are highly reliable; however, we note in the tables whenever the results of the two analyses differ.

The self-selection of students into different types of schools typically presents the greatest difficulty in studying the causal impact of schools on educational outcomes. Suppose we find differences between students at public and private schools in their support for democratic norms or their level of tolerance. Do we understand the differences as a product of the schooling the students received or the product of their home environment? This self-selection problem might be especially acute in our sample, as neither New York nor Texas had policies to provide financial assistance to families to enroll their children in private schools, in 1997 or now. Any financial aid received by the private school students in our sample would have come exclusively from the schools themselves or private philanthropies. Thus, most of the private school students in our sample likely relied on their parents to pay some or all of the cost to attend such schools.

We hypothesize that many of the same factors that lead parents to choose private schools potentially affect their children’s political tolerance. In addition, there could be an important (but unmeasured) variable predicting whether parents choose a private school because of their intolerance of competing worldviews. For example, parents who choose a particular school might do so in the belief that only their worldview should be taught and tolerated. By contrast, parents who value tolerance might choose a public school in the belief that their child will interconnect with peers whose views differ from their own. Statistical analyses that do not correct for self-selection overestimate the impact of schools on tolerance [44,45].

We attempt to correct for likely selection bias by employing a nonexperimental approach known as matching. Specifically, we match observations using nearest-neighbor matching based on propensity scores while ensuring common support between the private and matched public school samples. The spirit of matching and other related methods is to emphasize direct comparisons free from functional form assumptions and extrapolation that can be introduced in standard ordinary least squares (OLS) regressions. Matching is also more flexible in its ability to provide different treatment effects, such as the average treatment effect (ATE), average treated on treated (ATT), and average treated on control (ATC).

Matching, like including control variables in a regression equation, can decrease or even eliminate estimation bias due to observable factors, so long as those observable characteristics are measured and included in the matching protocol or regression controls. Matching, like control variables, cannot control for unobservable factors that might bias effect estimates so it does not guarantee that identified relationships are causal. Matching students on key observable characteristics has an advantage over simply including control variables for those same factors because the basic concept behind matching is intuitive: make the two comparison groups look as similar as possible to each other. Matching with common support provides the additional benefit of excluding from the analysis comparison group observations that hardly look at all like the “treatment” students, in this case students attending private schools. Thus, at least in expectation, matching with a common support requirement provides a slightly stronger bulwark against significant selection bias problems than do simple multivariate regressions with control variables.

To estimate the average treatment effect of attending a private school on students’ political knowledge and tolerance, support for democratic norms, and perceived threat from their least-liked group compared to students in a public school, we first run a logit model to predict the probability that students will select private schooling, conditional on a set of variables available. The logit model (Equation (1)) allows us to calculate propensity scores for each student, as:

$$P\left(X\right)=G(Private=1|X_{ij})$$

(1)

where $X_{ij}$ is a vector of students’ individual characteristics used to predict the probability they will select into private schooling. The vector of characteristics includes the current state of the student (New York or Texas), the student’s race and sex, whether the student is interested in politics, the highest level of education their parents have completed, and their current grades.

We then use OLS regression and the calculated propensity scores to match private and public school students one-to-one based on their profiles as captured by the propensity score. A private school student with a specific “likelihood of selecting into private school” based on key characteristics is matched with a public school student with a similar “likelihood of selecting into private school” based on the same set of characteristics. The ATE on the dependent variable (Equation (2)) is then estimated:

$$ToleranceScor{e}_i=B_0+B_{1}Privat{e}_i+B_2{X}_i+u_i$$

(2)

where $B_0$ is the constant, $B_1$ is the ATE of attending a private school compared to a public school, $Privat{e}_i$ is a private school indicator variable, $X_i$ is the same vector of individual characteristics used in Equation (1), and $B_2$ is their estimated effect on the dependent variable.

3.3. Common Support

The goal of matching is to close the doors to confounds that might otherwise generate biased effect estimates, using the matching variables indexed in Equation (1). To have confidence in our findings, we must have common support or overlap between the characteristics of the treated (private school) and comparison (public school) groups. There are multiple ways to study the degree of overlap between groups. We inspect the distribution of propensity scores and combine both groups to determine where most of the overlap occurs.

Figure 1 shows that most of the common support between treated and comparison groups lies between the 0.2 and 0.6 propensity score range where the two distributions substantially overlap. This is not surprising, as students with high values of the factors that predict selection into private schooling are more likely to actually be in private school and students with low values of the factors that predict selection into private schooling are more likely to actually be in public schools. We would question the usefulness of our matching variables if that was not the case. To ensure unbiased estimates, we restrict our sample to only observations within the propensity score range of 0.2 to 0.6. Private and public school students with propensity scores outside that range lack “common support” and therefore are excluded from our study.

4. Results

We first examine the relationship between school type and the choice of least-liked group. We next examine the bivariate relationships between school types and the students’ scores on the political knowledge, perceived threat, democratic norms, and political tolerance scales. We then estimate our multivariate models to isolate the effects of the various school types. The analysis includes four comparisons: all private school students with public school students, evangelical school students with public school students, evangelical school students with students in non-evangelical private schools (secular private schools), and secular private school students with public school students. Because the New York sample does not include evangelical schools, we use only the Texas sample for the second and third comparisons. Only the New York sample includes students in Catholic schools. Our main results all use the matched samples with common support. See Appendix B 

Table A3. OLS regression estimates comparing the main private vs. public results with the results estimated on the full private vs. public sample.

	(1)	(2)
Dependent Variables	Private vs. Public (Main Results)	Private vs. Public (Full Sample)
Political Knowledge	0.525 ***	0.585 ***
	(0.131)	(0.125)
Perceived Threat	−0.377	−0.858 *
	(0.570)	(0.492)
Democratic Norms	1.338 ***	1.669 ***
	(0.348)	(0.395)
Political Tolerance	0.399	0.162
	(0.502)	(0.505)
Observations	748–801	1668–1774

Standard errors in parentheses, *** p < 0.01, * p < 0.1.

for a comparison of the main private vs. public results with the results estimated on the full private vs. public sample.

4.1. Bivariate Relationships

We first categorize our students into three school types (public, non-evangelical private, and evangelical private) and then examine which groups the students are likely to pick as their least-liked group. We listed eight groups as possible targets of dislike in 1997: American Nazis, the Ku Klux Klan (KKK), advocates for the rights of women, advocates for the rights of racial and ethnic minorities, Christian fundamentalists, atheists, groups that are “pro-choice” on abortion, and groups that are “pro-life” on abortion. In addition, we create a dummy variable that takes the value of 1 if the least-liked group is a right-wing group or takes the value of 0 if the least-liked group is not clearly a right-wing group. American Nazis, the KKK, and Christian fundamentalists take the value of 1, while the rest of the groups take the value of 0, since the abortion question did not align closely with one’s political ideology in the 1990s. We include this dummy variable as the dependent variable in a logit model while controlling for student characteristics. We find that Black students are more likely to select right-wing groups as their least-liked group when compared to White students. This estimate is statistically significant at the 99% confidence level.

Table 2. Selection of least-liked group by type of school.

Variables	All Public	Evangelical Private	Nonevangelical Private
Ku Klux Klan or Nazis	81%	50%	83%
Feminists or Pro-Choice	2%	10%	2%
Religious fundamentalists Atheists Rights for minorities	1% 8% 8%	6% 5% 29%	1% 6% 8%
Observations	1444	124	567

shows that more than 80 percent of students in public and non-evangelical private schools chose either the American Nazis or the KKK as their least-liked group while only 50 percent of evangelical students selected these right-wing extremist groups. In sharp contrast, about 29 percent of students in evangelical private schools chose advocates of rights for ethnic minorities as their least-liked group, while 8 percent of public and non-evangelical private students chose this left-wing group. Limiting the analysis to White students does not change the pattern. The only difference is that the selection of the American Nazis or the KKK increases among evangelical students from 37 percent to over 40 percent.

The willingness of evangelical students to identify advocates for greater political equality as their least-liked group rather than groups known for their willingness to repress minorities is consistent with the claims of some observers that evangelical schools teach intolerance, or at least did in the 1990s. However, many bivariate comparisons between school types and the four dependent variables in

Table 3. Mean scores on the dependent variables by type of school.

Variables	All Public	Evangelical Private	Nonevangelical Private
Political Knowledge	1.58	3.41 *	2.33 *
Perceived Threat	33.05	26.38	33.46
Democratic Norms	20.71	21.58 *	22.59 *
Political Tolerance	14.58	15.36	14.89
Observations	1418–1526	121–130	556–596

* Indicates that the difference between the private school and public school was significant at p < 0.05.

do not support that expectation. Private school students score significantly higher than public school students on political knowledge and support for democratic norms, and students in evangelical schools score significantly lower than public school students in the threat they perceive from their least-liked group. Private school students also have higher average scores on the political tolerance scale, but the difference is not statistically significant at conventional confidence levels. Difference of means tests between evangelical school students and those enrolled in secular private schools show that evangelicals have significantly higher scores on political knowledge and significantly lower scores on perceived threat.

4.2. Political Knowledge

Table 4. Effects of school type on political knowledge.

	(1)	(2)	(3)	(4)
Variables	Private vs. Public	Evangelical Students vs. Secular Private (Texas)	Evangelical Students vs. Public (Texas)	Secular Private Students vs. Public (Texas)
Private	0.525 ***
	(0.131)
Evangelical		1.249 ***	2.093 ***
		(0.196)	(0.397)
Nonevangelical/Secular				0.263 ***
				(0.310)
Observations	801	341	690	803

Standard errors in parentheses, *** p < 0.01.

presents the average effect of different types of schools on the political knowledge outcome variable. The comparison of all private school students with matched public school students (column 1) shows that private school students have higher political knowledge than their public school counterparts. This estimate is statistically significant at the 99% confidence level. The effect size (ES) is 0.32 standard deviations. When we compare students in evangelical schools with students in secular private schools (column 2), we see that evangelicals have significantly higher political knowledge (ES = 0.76). Evangelical students, when compared to all public school students (column 3), have substantially higher political knowledge (ES = 1.28). We also find that secular private school students, when compared to public school students, have slightly higher political knowledge, but this estimate is not statistically significant at conventional confidence levels.

4.3. Perceived Threat

Sullivan et al. showed that the lower the threat that an individual perceives from their least-liked group, the more likely the individual is to grant civil liberties to members of the least-liked group [39]. Column 1 in

Table 5. Effects of school type on perceived threat.

	(1)	(2)	(3)	(4)
Variables	Private vs. Public	Evangelical Students vs. Secular Private (Texas)	Evangelical Students vs. Public (Texas)	Secular Private Students vs. Public (Texas)
Private	−0.377
	(0.570)
Evangelical		−6.952 ***	−9.529 ***
		(1.288)	(1.577)
Nonevangelical/Secular				2.503 *
				(1.533)
Observations	786	338	670	780

Standard errors in parentheses, *** p < 0.01, * p < 0.1.

shows that when comparing all public and private schools in our sample, students in private schools perceive fewer threats from their least-liked group than public school students, but this estimate is not statistically significant at conventional confidence levels. When we compare evangelical private school students to other private and public school students (columns 2 and 3), we find that evangelicals perceive significantly fewer threats than do other private and public school students (ES = −1.02 and −1.39). Public school students, when compared to secular private school students, have higher levels of perceived threats (ES = 0.37).

4.4. Support for Democratic Norms

Do public and private schools differ in their abilities to impact the democratic norms of their students? Column 1 in

Table 6. Effects of school type on democratic norms.

	(1)	(2)	(3)	(4)
Variables	Private vs. Public	Evangelical Students vs. Secular Private (Texas)	Evangelical Students vs. Public (Texas)	Secular Private Students vs. Public (Texas)
Private	1.338 ***
	(0.348)
Evangelical		0.305	1.262
		(0.599)	(0.861)
Nonevangelical/Secular				1.735 ***
				(0.545)
Observations	748	334	666	778

Standard errors in parentheses, *** p < 0.01.

shows that private school students have significantly higher support for democratic norms than public school students (ES = 0.37). This estimate was statistically significant at the 99% confidence level. When we compare evangelical and other private school students (column 2), evangelical students have higher support for democratic norms, but this estimate is not statistically significant at conventional confidence levels. Furthermore, evangelical students, when compared to public school students, have higher levels of support for democratic norms; however, this estimate is not statistically significant at conventional confidence levels. Also, secular private school students have higher support for democratic norms when compared to public school students (ES = 0.48).

4.5. Political Tolerance

Column 1 in

Table 7. Effects of school type on political tolerance.

	(1)	(2)	(3)	(4)
Variables	Private vs. Public	Evangelical Students vs. Secular Private (Texas)	Evangelical Students vs. Public (Texas)	Secular Private Students vs. Public (Texas)
Private	0.399
	(0.502)
Evangelical		1.547 ***	0.628
		(0.561)	(0.905)
Nonevangelical/Secular				−2.274 ***
				(0.674)
Observations	782	332	666	786

Standard errors in parentheses, *** p < 0.01.

shows the effect of school type on political tolerance based on all private school students compared to matched public school students. Private school students have slightly higher levels of political tolerance, but this estimate is not statistically significant at conventional confidence levels. Surprisingly, we find that evangelical private school students (column 2) have significantly higher levels of political tolerance than other private school students (ES = 0.31). Column 3 presents the regression estimate between evangelical and public school students. Evangelical students have a slightly higher tolerance for their least-liked group, but this estimate is not statistically significant at conventional confidence levels. Secular private school students have lower political tolerance levels when compared to public school students (ES = 0.45).

4.6. Considering Possible Mechanisms

What specific differences in the experiences of students in public, evangelical private, and secular private schools might explain these differences in civic outcomes? School climate, experience volunteering, and the level of perceived threat from one’s least-liked group are three possible mechanisms to examine [26,46]. The first and second interethnic climate variables are a measure of the willingness of students to make friends with other groups and how often students see fights among other groups. Volunteering is a measure of how often students do volunteer work outside of school. Perceived threat was described previously as the extent to which a student feels threatened by members of their least-liked political group.

To test for specific mechanisms, we include the school climate and volunteering variables in the regression estimates of the effect of school type on political knowledge and support for democratic norms. Then we include the perceived threat variable in the regression estimates of the effect of school type on political tolerance.

Table 8. Test of mechanisms for effects of school type on political knowledge.

	(1)	(2)	(3)	(4)
Variables	Private vs. Public	Evangelical Students vs. Secular Private (Texas)	Evangelical Students vs. Public (Texas)	Secular Private Students vs. Public (Texas)
Private	0.231
	(0.159)
Evangelical		1.735 ***	2.329 ***
		(0.260)	(0.309)
Nonevangelical/Secular				0.335
				(0.284)
Observations	746	333	643	756

Standard errors in parentheses, *** p < 0.01.

shows the ATE of including both climate and volunteer variables. When we include the school climate and volunteering variables in the analysis, the positive private schooling effect on political knowledge shrinks to insignificance. The same thing happens when we add the climate and volunteering variables when estimating the effects of secular private schooling on political knowledge, compared to public schooling. These results suggest that a positive school climate and volunteering by students might be the mechanisms by which private schools improve political knowledge. However, columns 2 and 3 in Table 8 show evidence that evangelical students, when compared to other public and private school students, still have significantly higher levels of political knowledge (effect sizes = 1.06 and 1.42) that are not explained away or diminished at all by including school climate and volunteering variables in the regressions. Thus, school climate and volunteering do not appear to be consistent mechanisms behind the political knowledge effects of private, evangelical, and secular schools.

We then turn our attention to the inclusion of the school climate and volunteering variables in the estimation of the effects of school type on support for democratic norms. Column 1 in

Table 9. Test of mechanisms for effects of school type on democratic norms.

	(1)	(2)	(3)	(4)
Variables	Private vs. Public	Evangelical Students vs. Secular Private (Texas)	Evangelical Students vs. Public (Texas)	Secular Private Students vs. Public (Texas)
Private	1.134 **
	(0.499)
Evangelical		0.696	0.846
		(0.884)	(1.665)
Nonevangelical/Secular				1.516 ***
				(0.695)
Observations	722	326	633	745

Standard errors in parentheses, *** p < 0.01, ** p < 0.05.

shows that private school students have significantly higher levels of support for democratic norms when compared to public schools, controlling for school conditions and volunteering (ES = 0.31). When we compare evangelical and private school students, we find that evangelicals have higher support for democratic norms, but this estimate is not statistically significant at conventional confidence levels. Further evidence shows that evangelicals have slightly higher support for democratic norms when compared to public school students, but this estimate also is not statistically significant at conventional confidence levels. Secular private school students have higher support for democratic norms when compared to public school students, controlling for schooling conditions and volunteering, and this estimate is statistically significant at the 95% confidence level (ES = 0.42). These school type effects are almost identical to the effects when these potential mechanism variables are not included in the regressions, indicating that school climate and volunteering are likely not mechanisms driving the effects of different types of private schools on support for democratic norms.

Table 10. Test of mechanisms for effects of school type on political tolerance.

	(1)	(2)	(3)	(4)
Variables	Private vs. Public	Evangelical Students vs. Secular Private (Texas)	Evangelical Students vs. Public (Texas)	Secular Private Students vs. Public (Texas)
Private	0.244
	(0.494)
Evangelical		0.189	−2.825 ***
		(0.678)	(0.978)
Nonevangelical/Secular				−2.281 ***
				(0.756)
Observations	753	328	637	755

Standard errors in parentheses, *** p < 0.01.

shows the effect of including students’ perceived threat in the estimation of the effects of school type on political tolerance. Column 1 shows that private school students have higher levels of political tolerance, but this estimate is not statistically significant at conventional confidence levels. Evangelical private school students have higher tolerance levels when compared to secular private school students. Surprisingly, in column 3, we see that evangelical students compared to public school students have lower levels of tolerance (ES = −0.56). This estimate indicates that lower levels of perceived threat is a possible mechanism driving the effect of evangelical private schools. Lastly, secular private school students have lower levels of political tolerance when compared to public school students (ES = 0.45).

Finally, as a robustness check, we replicate the entire analysis with a more expansive statistical model that includes school fixed effects (results available by request). None of our substantive findings change with school fixed effects included, though sample power decreases slightly.

5. Discussion

Our results indicate that private schools impart significantly more knowledge about the American government and support for democratic norms to their students than public schools do. Furthermore, we find some evidence suggesting that school climate and volunteer work could be mechanisms for advancing support for democratic norms in private schooling, but those findings are not consistent across school types or outcomes. Additionally, we find slight evidence that perceived threat is a possible mechanism for driving the effect of students’ political tolerance levels. Greater ethnic diversity in schools is not associated with an increase in either political tolerance (see Appendix B 

Table A4. OLS regression estimates for democratic norms and political tolerance.

	(1)	(2)	(3)
Variables	Democratic Norms	Political Tolerance	Democratic Norms
Private	1.404 ***	0.446	1.262 ***
	(0.216)	(0.299)	(0.224)
Texas	−0.838 ***	0.038	−0.691 ***
	(0.175)	(0.255)	(0.176)
Black	−0.318	−0.688 **	−0.327
	(0.215)	(0.303)	(0.215)
Hispanic	−0.385 *	−1.011 ***	−0.512 **
	(0.228)	(0.313)	(0.227)
Male	−0.898 ***	0.228	−0.745 ***
	(0.169)	(0.246)	(0.171)
Parents’ Education	0.0001	−0.337 ***	−0.014
	(0.073)	(0.106)	(0.073)
Interest in Politics	0.722 ***	0.334	0.636 ***
	(0.155)	(0.229)	(0.155)
Students’ Grades	0.538 ***	−0.459 ***	0.471 ***
	(0.138)	(0.176)	(0.136)
Willing to make friends			0.513 ***
			(0.094)
Fights between other groups			0.096
			(0.067)
Constant	18.91 ***	16.84 ***	16.73 ***
	(0.602)	(0.816)	(0.725)
Observations	1667	1719	1643
R-squared	0.114	0.017	0.137

Robust standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1.

) or support for democratic norms (see Appendix B Table A4). Interethnic friendships increase support for democratic norms (see Appendix B Table A4), but the reported incidence of interethnic friendships is actually higher in private schools despite their lower ethnic diversity.

In short, many of the differences found in this study favor private schooling even though we restrict our sample to public school students with propensity scores indicating that they are observationally similar to private school students. Opponents of school choice continue to claim that private schooling will undermine American democracy [6], but our analysis here does not support that claim. Nevertheless, our results are not necessarily causal, as we cannot directly control for unmeasured factors, not captured by our propensity scores, which might affect both private school enrollment and civic outcomes. Our evidence also is limited to New York City and the Dallas/Fort Worth area and is “vintage”, since it was collected in 1997. The late 1990s were a period of continued political mobilization of evangelical Christians under the banner of the “Moral Majority”, a historical and social phenomenon that might explain why evangelicals evidenced the highest level of political knowledge among our comparison groups. The political commitments and values of evangelical Christians in the U.S. likely have changed since 2016, when Donald Trump disrupted traditional partisan alignments. The racially charged murder of George Floyd in 2020 and the seminal 2022 Supreme Court ruling in Dobbs v. Jackson that returned abortion policy decisions to the individual states likely have modified the views of Americans regarding their least-liked political groups. Fresh research on private schooling and civic values clearly is desirable.

Acknowledging those study limitations, our evidence suggests that restricting private school choice initiatives might have a negative impact on students’ civic outcomes. Our new results from old data add to the evidentiary record suggesting that private schools tend to do as well or even better at preparing young people for democratic citizenship.