DCPS or Sidwell Friends? How Politician Schooling Choices Affect Voter Evaluations

Voters often rely on informational shortcuts, such as the background traits of politicians, to decide which candidates to support at the ballot box. One such background trait is family composition, particularly parental status. Research, however, has mostly overlooked whether the value-laden choices that politicians make regarding their families—like what neighborhoods they live in, where they worship, and what schools they send their children to—affect how constituents view them. We conduct a survey experiment in the U.S. that presents respondents with hypothetical biographies of politicians that randomly vary one of the most important decisions that politicians make regarding their families: whether to send them to public or private school. We find that: (1) voters are more inclined to vote for politicians with children in public school; and (2) this preference may be due to voters perceiving these politicians as both warmer and more committed to public services.

high-level contests. In Florida, for example, one Broward County School Board member's decision to enroll her son in prep school caused a former member to declare that " [n]o school board member should send her child to a private school for any reason" (Baron, 2013). In Wisconsin, a state school superintendent candidate was denounced for "mountain-sized hypocrisy" as a result of sending her own children to Catholic school while positioning herself as antischool choice (Bice, 2021). In Michigan, a trustee of a local board of education declared that a school board member who does not send his or her children to local public schools is akin "to the CEO of General Motors driving a Toyota or Detroit Tigers' Manager Jim Leyland rooting for the Chicago White Sox" (Hatter, 2006). Even outside the U.S., the issue has made headlines. In the U.K., for instance, a Labor Party cabinet minister's decision to enroll his child in a "private boarding school for pupils with special needs" led him to be lambasted by an MP within his own party for engaging in a "betrayal of the party's principles" (Hunt, 2007).
Are voters more likely to cast their ballot for politicians who send their children to public over private school? To answer this question, we fielded an original survey experiment conducted on a diverse national sample in the U.S. that presented respondents with biographies of hypothetical politicians running for office. A survey experiment has at least two advantages over observational studies. It isolates how the schooling choices of politicians affect voter evaluations apart from other traits that might both correlate with schooling choices and ballot box success. Additionally, it permits us to tease out the mechanisms that could explain why voters assess politicians differently based on where they send their children to school. We randomly varied whether two hypothetical candidates running for state office are parents-and, if so, whether they send their children to public or private school. We then asked about the likelihood of respondents supporting each profile of politician at the ballot box as well as respondent perceptions of the politicians. Results are consistent with expectations: Voters are significantly more inclined to vote for candidates who send their children to public instead of private school.
We explore two possible explanations for this finding. First, we theorize that voters may perceive politicians with public school children as "warmer." In the U.S., both public and private schools carry distinct connotations that speak to social class, mores, and other traits. Whereas private schools are typically the province of the elite, public schools serve the masses (Howard & Gaztambide-Fernandez, 2010;Kober, 2007). 3 As such, voters may perceive candidates with public school children as more likeable and friendly to the extent that the choice reflects being grounded and modest. Voters may also see such politicians as caring more about their community because they are invested in one of its main public institutions. They may even view such candidates as more conscientious, since they have aligned an important lifestyle choice with that of most constituents. We find evidence that respondents do view politicians with public school children as being more likeable, friendly, caring, and conscientioustraits that we use to create a measure of warmth. These judgments are most positive for respondents who presumably have the most in common with these politicians: public school parents themselves.
Second, we theorize that voters may also prefer politicians who send their children to public school because they view them as more dedicated to public services generally. Survey data reveal that overwhelming majorities of Americans favor increasing or preserving spending on public services. 4 This suggests that politicians who will follow through with maintaining or increasing government expenditures are likely to have electoral advantages. Politicians who send their children to public school may be more likely to signal a "credible commitment" (Elster, 1979;Holmes, 1988) to public spending because they have "skin in the game" insofar as they have tied the well-being of their family to a public institution. When candidates choose public goods themselves, voters may be more apt to believe that these politicians also value government funding because they have a personal incentive to maintain high-quality state provisions. We find evidence that respondents do view politicians with public school children as stronger advocates of public spending across a range of policy areas. This effect, however, is largest for the policy area with the most direct relevance to schooling choices: public education.
Our results add to a sizable literature on candidate experiments (Campbell & Cowley, 2014;Carnes & Sadin, 2015;Gift & Lastra-Anadón, 2018;Kahn, 1994;Sigelman et al., 1995) by examining how an important background traitspecifically, whether politicians send their children to public or private school-affects ballot box appeal. While prior research has analyzed the effect of family compositionincluding whether politicians have children-on electability, we find that decisions regarding families can condition how voters perceive politicians. We show that voters prefer politicians who send their children to public over private school, and that this preference may be attributable to both their perceived warmth and commitment to public services. Practically, our analysis sheds light on how politicians use their families in elections to shape their reputations (Lane, 2015;McElvoy, 2014). In the media and on the campaign trail, politicians are often seen with their families, which many seem to believe helps to cultivate their public images. We demonstrate that not only does having children matter to voters, but so do their parental choices.

The Experiment
We conducted a survey experiment in the U.S. that randomly varied whether two hypothetical candidates running for governor had children and, if so, whether the children attended public or private school. By randomly varying details about children, while holding all else constant, we can reasonably attribute any differences in candidate assessments to that information.

Sample
We conducted the survey online through the Harvard Digital Lab for the Social Sciences (DLABSS) from April to August 2019. A well-established survey platform, the DLABSS panel is comprised of volunteers who sign up to complete questionnaires relevant to the social sciences (Strange et al., 2019). The sample is diverse with respect to standard demographic traits (age, race, gender, region, etc.) and is similar to samples fielded from other commonly-used online survey platforms like MTurk (see Appendix Table A1 for summary statistics). Considerable research has relied on DLABSS data (see http://dlabss.harvard.edu/results), and the panel has been externally validated on a number of well-known surveys, including ones using nationally-representative samples, such as the General Social Survey. 5 We analyze only respondents who self-report as U.S. residents. Our final sample size is 946 respondents.

Survey Instrument
Respondents each read vignettes of two hypothetical male candidates whose order of presentation was randomized (see Appendix). After reading details about a candidate, respondents answered questions about that candidate, then repeated this step for the next candidate. Respondents were randomly assigned to one of three groups: (1) a Control, where children were not mentioned; (2) a Public School treatment, where the politician had two children (a boy and a girl) who both attended public school; and (3) a Private School treatment, where the politician had two children (also a boy and a girl) who both attended private school. In addition, the vignettes provided other generic details about the candidates (age, work history, volunteerism/charitable work, government experience, and marital status). They also included a generic quote from the politicians explaining why voters should support them. To emphasize in the Control condition that the candidate did not have any children, we noted that the candidate and his wife lived with their dog, to imply that they did not live with anyone else. For consistency, in both the Public School and Private School conditions, we also said that the candidate and his wife lived with their dog, as well as their children. 6 We attempted to create a reasonable balance with respect to the content of the vignettes-one that provided sufficient detail such that the schooling treatment was not the obvious main focus, while also ensuring that the biographies were sufficiently parsimonious to reduce cognitive burdens and to make the politicians generic enough to ensure generalizability. The vignettes do lack extensive content-such as multi-faceted policy positions -that might be salient to voters in real campaigns. This could make the treatments stand out more and bias the results toward finding an effect of schooling choices. This challenge reflects ongoing debates in the literature about how detailed or (minimalist) politician depictions should be in candidate experiments. We aimed to make the level of detail in the biographies broadly consistent with related research (e.g., Campbell & Cowley, 2018;Gift & Lastra-Anadón, 2018, Iyengar & Westwood, 2015. In context of other information, it is not clear that schooling choices should be an overly blunt treatment. Nonetheless, by only manipulating parenthood and schooling choices, the vignettes do limit our ability to test, for example, whether our results are conditioned by various omitted variables (such as the candidate's partisanship, gender, etc.) that have been shown to matter in prior research, and to explicitly compare effect sizes to these variables.

Dependent Variable
Our dependent variables were the candidate's ballot box appeal, warmth, and commitment to public services. We measured ballot box appeal by asking how likely the respondent would be to vote for the candidate. Answer choices were coded on a 1 to 5 scale, where 1 denoted that respondents would be "Extremely unlikely" to vote for the candidate and 5 "Extremely likely." We measured perceived warmth on a 1 to 5 scale by creating an average index incorporating likeability, conscientiousness, friendliness, and caringness. 7 Respondents answered how well or poorly each of the words "likeable," "conscientious," "friendly," and "caring" describe the candidate. 8 We measured perceived commitment to public spending on a 1 to 5 scale averaged across multiple policy areas. For each policy area, we asked, "If this candidate were elected governor, how likely do you think he would be to advocate for more funding for public [schools/anti-poverty programs/transportation/health initiatives]?" 9

Empirics
We estimate the effect of where politicians send their children to school using the following model: Yi where Public refers to the experimental condition where the candidate has children enrolled in public school, and NonParent refers to the no-children condition. The omitted term is the private school condition. For all the models, we report simulated predicted values carried out using Zelig (Choirat et al., 2020;Imai et al., 2008). 10 As robustness checks, we also reestimate the linear models with standard individual-level covariates (Appendix Table A4) 11 ; using only the first encountered vignette for each respondent (Appendix Table A5) 12 ; and using ordered probit models without covariates to adjust for underlying nonlinearities in the dependent variables (Appendix Table A6). 13  Appendix Table A3 for the results as a table). For each dependent variable-ballot box appeal, perceived warmth, and perceived commitment to public spending-we find that respondents give public school candidates the highest ratings. The greatest differences in evaluations are between the public and private school candidates, and in all cases, these differences are statistically significant. For ballot box appeal, respondents rate candidates who send their children to public school 3.26, meaning that they are slightly more than "neither likely nor unlikely" to vote for them. That rating drops to 2.94 when politicians send their children to private school (slightly less than "neither likely nor unlikely"). 14 The rating of non-parent candidates lies in the middle, at 3.1. We theorize that politicians who send their children to public schools have greater ballot box appeal because they are viewed as warmer and more committed to public services. For perceived warmth, the rating for public school candidates is 3.55, compared to 3.42 for private school candidates. 15 For perceived support for public spending, those ratings are 3.52 and 3.09, respectively. 16

Conditional Effects
We also probe whether our mechanisms have a stronger effect under certain conditions. In Figure 2, we analyze whether respondents who send their children to public schools view politicians with public school children as particularly warm. Restricting the data to only respondents with children, we find that public school respondents rate public school candidates as 3.58 on the warmth scale, compared to 3.4 for private school candidates. Appendix Table A9 confirms that the interaction between the public school treatment and the respondent being a public school parent is positive and statistically significant. In Figure 3, we analyze whether voters perceive politicians with public school children as more likely to invest in public schools than any other policy area. We find that, for public school politicians, perceived support for public schools is 3.87 on the spending scale, compared to Predicted values range from 1 to 5, corresponding to the Likert scales described above. Dots represent the median predicted value. The displayed results are calculated by simulating predicted values using models that regress each dependent variable on the public school and non-parent conditions, with no controls. Models are displayed in Appendix Table A3. Predicted values range from 1 to 5, corresponding to the Likert scales described above. Dots represent the median predicted value. The displayed results are calculated by simulating predicted values using a model that regresses the warmth index on the non-parent and public school conditions, and interacting the public school condition with whether the respondent sends his or her children to public school, with no controls. Models are displayed in Appendix Table A9. Predicted values range from 1 to 5, corresponding to the Likert scales described above. Dots represent the median predicted value. The displayed results are calculated by simulating predicted values using models that regress perceived support for each policy area on the public school and non-parent conditions, with no controls. Models are displayed in Appendix Table A8. 3.49 for public health, 3.3 for public transportation, and 3.43 for anti-poverty programs. Tests for the equality of coefficients across the underlying models for each spending area confirm that the effect of the public school treatment is largest when the outcome is perceived support for public school spending. 17

Separating Partisanship and Schooling Choices
One potential confounder is that voters use politician schooling decisions as a heuristic for other background traits that are excluded from our biographies-mostly notably, partisanship, since this is often linked to educational policy agendas. The Democratic Party, for example, is aligned more with teachers unions and against school choice, so voters may assume that politicians who send their children to public school are more likely to be Democrats and cast their ballot accordingly. Although we do not manipulate candidate partisanship due to concerns that it would overwhelm the evaluations, 18 we reestimate the main ballot box appeal model but include controls for whether the respondent perceives the candidate to be a Democrat, the respondent's own partisanship, and an interaction of the two. As anticipated, Appendix Table A10 shows that, although the magnitude of the public school parent coefficient decreases, it remains positive and statistically significant. This improves confidence in our results. Even if some respondents infer the partisanship of candidates based on their schooling choices and then evaluate them according to whether this partisanship matches their own, this does not fully explain why politicians with public school children have more ballot box appeal.

Conclusion
It is widely believed that the decisions politicians make regarding their families can influence how voters perceive them. We analyzed how one of the most visible and important such decisions-whether politicians send their children to public or private school-can shape their ballot box appeal. To do so, we fielded an original survey experiment among a nationwide sample in the U.S. that presented biographies of hypothetical politicians, and then randomly varied whether they had children and enrolled them in public or private school. We found that voters were more likely to support politicians who send their children to public over private school. Additionally, respondents perceived politicians with public school children as warmer and more committed to public services, which may explain their greater ballot box appeal. Results also point to notable conditional effects. Respondents with public school children find politicians who also send their children to public school especially warm. Moreover, respondents perceive politicians who send their children to public school as particularly committed to spending on public education relative to other policy areas.
These findings may have implications for understanding how politicians use family members strategically in campaigns. They also contribute to a growing literature on candidate experiments. We are, to our knowledge, the first to analyze how the actual decisions that politicians make regarding their families shape voter evaluations. This is a promising avenue for inquiry, as there is much we do not know about how other family choices-such as whether politicians live in wealthy suburban enclaves or working-class neighborhoods, whether families are active in local place of worship, and so on-shape voter assessments. Future research could test our main predictions using observational data. Scholars could also analyze whether other background traits of politicians-such as their partisanship, gender, or race-interact with parental status and schooling decisions to affect whether voters view them favorably. Another extension could be to see whether results our replicate in nationallevel campaigns (e.g., for president) and district-or local-level campaigns (e.g., for U.S. or state representatives or school boards). Our study suggests that the decisions politicians make regarding their families can affect how voters assess them.

Survey Vignettes
The text of the survey vignettes was as follows: George Jones Is Running for Governor George Jones, a 42-year-old business analyst, has announced his candidacy for governor. Jones has an extensive history of working for his community. Prior to declaring his candidacy, Jones held roles in state and local government. Additionally, he has devoted his time to numerous charitable causes. "I want a government that improves the lives of real people," Jones said at a political rally. "I know how to solve tough problems." Jones married his wife, Suzy, 13 years ago. They reside with their dog, Fido [Prime]. Primes   Note. Standard errors in parentheses. Models do not include controls. † p < .1 , * p < .05 , ** p < .01, *** p < .001 .
12. Each respondent received two randomized vignettes, but there was some attrition on the second, perhaps due to fatigue. If we look only at the first vignette respondents read, missingness on the voting question is similar across the three treatments (8.3% of the control, 6.7% of the private school treatment, and 8.7% of the public school treatment). However, nonresponse increases for answers in response to the second vignette, particularly for those receiving the control. This may explain discrepancies in missingness across the treatment groups in balance Appendix Table A2. To ensure this is not biasing our results, we reestimate our main specifications with only the first treatment vignette for each respondent. The results are substantively similar. 13. Appendix Table A2 shows that our treatment groups are balanced across all relevant covariates. In our specifications without controls, we drop between 5% and 11% of our observations due to nonresponse on our DVs. 14. All numbers reported in the text are derived from the coefficients in the underlying models contained in the relevant Appendix Tables. 15. Results are consistent for disaggregated indicators of perceived warmth (Appendix Table A7). 16. Results are consistent for disaggregated indicators of perceived support for public spending (Appendix Table A8). 17. This relies on estimations from Appendix Table A8. 18. See McGraw (2011) for the pros and cons of excluding partisanship in candidate experiments.